diff --git a/.github/workflows/integration-test.yml b/.github/workflows/integration-test.yml
index b52f72d6..fa4666cc 100644
--- a/.github/workflows/integration-test.yml
+++ b/.github/workflows/integration-test.yml
@@ -53,7 +53,8 @@ jobs:
             libcurl4-openssl-dev \
             pkg-config \
             libsasl2-dev \
-            protobuf-compiler
+            protobuf-compiler \
+            libprotobuf-dev
 
       - name: Cache Cargo
         uses: Swatinem/rust-cache@v2
diff --git a/.github/workflows/verify-package.yml b/.github/workflows/verify-package.yml
index 085ac61d..bd7dfd51 100644
--- a/.github/workflows/verify-package.yml
+++ b/.github/workflows/verify-package.yml
@@ -52,7 +52,8 @@ jobs:
             libcurl4-openssl-dev \
             pkg-config \
             libsasl2-dev \
-            protobuf-compiler
+            protobuf-compiler \
+            libprotobuf-dev
 
       - name: Cache Cargo
         uses: Swatinem/rust-cache@v2
diff --git a/Cargo.lock b/Cargo.lock
index b8edca1b..18b01c0d 100644
--- a/Cargo.lock
+++ b/Cargo.lock
@@ -132,6 +132,30 @@ version = "1.0.102"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "7f202df86484c868dbad7eaa557ef785d5c66295e41b460ef922eca0723b842c"
 
+[[package]]
+name = "apache-avro"
+version = "0.21.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "36fa98bc79671c7981272d91a8753a928ff6a1cd8e4f20a44c45bd5d313840bf"
+dependencies = [
+ "bigdecimal",
+ "bon",
+ "digest",
+ "log",
+ "miniz_oxide",
+ "num-bigint",
+ "quad-rand",
+ "rand 0.9.2",
+ "regex-lite",
+ "serde",
+ "serde_bytes",
+ "serde_json",
+ "strum 0.27.2",
+ "strum_macros 0.27.2",
+ "thiserror 2.0.18",
+ "uuid",
+]
+
 [[package]]
 name = "ar_archive_writer"
 version = "0.5.1"
@@ -147,6 +171,15 @@ version = "1.4.2"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "c3d036a3c4ab069c7b410a2ce876bd74808d2d0888a82667669f8e783a898bf1"
 
+[[package]]
+name = "arc-swap"
+version = "1.9.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "6a3a1fd6f75306b68087b831f025c712524bcb19aad54e557b1129cfa0a2b207"
+dependencies = [
+ "rustversion",
+]
+
 [[package]]
 name = "arrayref"
 version = "0.3.9"
@@ -165,19 +198,40 @@ version = "55.2.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "f3f15b4c6b148206ff3a2b35002e08929c2462467b62b9c02036d9c34f9ef994"
 dependencies = [
- "arrow-arith",
+ "arrow-arith 55.2.0",
  "arrow-array 55.2.0",
  "arrow-buffer 55.2.0",
  "arrow-cast 55.2.0",
- "arrow-csv",
+ "arrow-csv 55.2.0",
  "arrow-data 55.2.0",
  "arrow-ipc 55.2.0",
- "arrow-json",
- "arrow-ord",
- "arrow-row",
+ "arrow-json 55.2.0",
+ "arrow-ord 55.2.0",
+ "arrow-row 55.2.0",
  "arrow-schema 55.2.0",
  "arrow-select 55.2.0",
- "arrow-string",
+ "arrow-string 55.2.0",
+]
+
+[[package]]
+name = "arrow"
+version = "57.3.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "e4754a624e5ae42081f464514be454b39711daae0458906dacde5f4c632f33a8"
+dependencies = [
+ "arrow-arith 57.3.0",
+ "arrow-array 57.3.0",
+ "arrow-buffer 57.3.0",
+ "arrow-cast 57.3.0",
+ "arrow-csv 57.3.0",
+ "arrow-data 57.3.0",
+ "arrow-ipc 57.3.0",
+ "arrow-json 57.3.0",
+ "arrow-ord 57.3.0",
+ "arrow-row 57.3.0",
+ "arrow-schema 57.3.0",
+ "arrow-select 57.3.0",
+ "arrow-string 57.3.0",
 ]
 
 [[package]]
@@ -194,6 +248,20 @@ dependencies = [
  "num",
 ]
 
+[[package]]
+name = "arrow-arith"
+version = "57.3.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "f7b3141e0ec5145a22d8694ea8b6d6f69305971c4fa1c1a13ef0195aef2d678b"
+dependencies = [
+ "arrow-array 57.3.0",
+ "arrow-buffer 57.3.0",
+ "arrow-data 57.3.0",
+ "arrow-schema 57.3.0",
+ "chrono",
+ "num-traits",
+]
+
 [[package]]
 name = "arrow-array"
 version = "52.2.0"
@@ -227,6 +295,25 @@ dependencies = [
  "num",
 ]
 
+[[package]]
+name = "arrow-array"
+version = "57.3.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "4c8955af33b25f3b175ee10af580577280b4bd01f7e823d94c7cdef7cf8c9aef"
+dependencies = [
+ "ahash",
+ "arrow-buffer 57.3.0",
+ "arrow-data 57.3.0",
+ "arrow-schema 57.3.0",
+ "chrono",
+ "chrono-tz",
+ "half",
+ "hashbrown 0.16.1",
+ "num-complex",
+ "num-integer",
+ "num-traits",
+]
+
 [[package]]
 name = "arrow-buffer"
 version = "52.2.0"
@@ -249,6 +336,18 @@ dependencies = [
  "num",
 ]
 
+[[package]]
+name = "arrow-buffer"
+version = "57.3.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "c697ddca96183182f35b3a18e50b9110b11e916d7b7799cbfd4d34662f2c56c2"
+dependencies = [
+ "bytes",
+ "half",
+ "num-bigint",
+ "num-traits",
+]
+
 [[package]]
 name = "arrow-cast"
 version = "52.2.0"
@@ -290,6 +389,28 @@ dependencies = [
  "ryu",
 ]
 
+[[package]]
+name = "arrow-cast"
+version = "57.3.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "646bbb821e86fd57189c10b4fcdaa941deaf4181924917b0daa92735baa6ada5"
+dependencies = [
+ "arrow-array 57.3.0",
+ "arrow-buffer 57.3.0",
+ "arrow-data 57.3.0",
+ "arrow-ord 57.3.0",
+ "arrow-schema 57.3.0",
+ "arrow-select 57.3.0",
+ "atoi",
+ "base64",
+ "chrono",
+ "comfy-table",
+ "half",
+ "lexical-core 1.0.6",
+ "num-traits",
+ "ryu",
+]
+
 [[package]]
 name = "arrow-csv"
 version = "55.2.0"
@@ -305,6 +426,21 @@ dependencies = [
  "regex",
 ]
 
+[[package]]
+name = "arrow-csv"
+version = "57.3.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "8da746f4180004e3ce7b83c977daf6394d768332349d3d913998b10a120b790a"
+dependencies = [
+ "arrow-array 57.3.0",
+ "arrow-cast 57.3.0",
+ "arrow-schema 57.3.0",
+ "chrono",
+ "csv",
+ "csv-core",
+ "regex",
+]
+
 [[package]]
 name = "arrow-data"
 version = "52.2.0"
@@ -329,6 +465,19 @@ dependencies = [
  "num",
 ]
 
+[[package]]
+name = "arrow-data"
+version = "57.3.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "1fdd994a9d28e6365aa78e15da3f3950c0fdcea6b963a12fa1c391afb637b304"
+dependencies = [
+ "arrow-buffer 57.3.0",
+ "arrow-schema 57.3.0",
+ "half",
+ "num-integer",
+ "num-traits",
+]
+
 [[package]]
 name = "arrow-ipc"
 version = "52.2.0"
@@ -354,7 +503,23 @@ dependencies = [
  "arrow-data 55.2.0",
  "arrow-schema 55.2.0",
  "flatbuffers 25.12.19",
- "lz4_flex",
+ "lz4_flex 0.11.6",
+]
+
+[[package]]
+name = "arrow-ipc"
+version = "57.3.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "abf7df950701ab528bf7c0cf7eeadc0445d03ef5d6ffc151eaae6b38a58feff1"
+dependencies = [
+ "arrow-array 57.3.0",
+ "arrow-buffer 57.3.0",
+ "arrow-data 57.3.0",
+ "arrow-schema 57.3.0",
+ "arrow-select 57.3.0",
+ "flatbuffers 25.12.19",
+ "lz4_flex 0.12.1",
+ "zstd",
 ]
 
 [[package]]
@@ -379,6 +544,30 @@ dependencies = [
  "simdutf8",
 ]
 
+[[package]]
+name = "arrow-json"
+version = "57.3.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "0ff8357658bedc49792b13e2e862b80df908171275f8e6e075c460da5ee4bf86"
+dependencies = [
+ "arrow-array 57.3.0",
+ "arrow-buffer 57.3.0",
+ "arrow-cast 57.3.0",
+ "arrow-data 57.3.0",
+ "arrow-schema 57.3.0",
+ "chrono",
+ "half",
+ "indexmap 2.13.1",
+ "itoa",
+ "lexical-core 1.0.6",
+ "memchr",
+ "num-traits",
+ "ryu",
+ "serde_core",
+ "serde_json",
+ "simdutf8",
+]
+
 [[package]]
 name = "arrow-ord"
 version = "55.2.0"
@@ -392,6 +581,19 @@ dependencies = [
  "arrow-select 55.2.0",
 ]
 
+[[package]]
+name = "arrow-ord"
+version = "57.3.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "f7d8f1870e03d4cbed632959498bcc84083b5a24bded52905ae1695bd29da45b"
+dependencies = [
+ "arrow-array 57.3.0",
+ "arrow-buffer 57.3.0",
+ "arrow-data 57.3.0",
+ "arrow-schema 57.3.0",
+ "arrow-select 57.3.0",
+]
+
 [[package]]
 name = "arrow-row"
 version = "55.2.0"
@@ -405,6 +607,19 @@ dependencies = [
  "half",
 ]
 
+[[package]]
+name = "arrow-row"
+version = "57.3.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "18228633bad92bff92a95746bbeb16e5fc318e8382b75619dec26db79e4de4c0"
+dependencies = [
+ "arrow-array 57.3.0",
+ "arrow-buffer 57.3.0",
+ "arrow-data 57.3.0",
+ "arrow-schema 57.3.0",
+ "half",
+]
+
 [[package]]
 name = "arrow-schema"
 version = "52.2.0"
@@ -421,6 +636,17 @@ dependencies = [
  "serde_json",
 ]
 
+[[package]]
+name = "arrow-schema"
+version = "57.3.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "8c872d36b7bf2a6a6a2b40de9156265f0242910791db366a2c17476ba8330d68"
+dependencies = [
+ "bitflags 2.11.0",
+ "serde_core",
+ "serde_json",
+]
+
 [[package]]
 name = "arrow-select"
 version = "52.2.0"
@@ -449,6 +675,20 @@ dependencies = [
  "num",
 ]
 
+[[package]]
+name = "arrow-select"
+version = "57.3.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "68bf3e3efbd1278f770d67e5dc410257300b161b93baedb3aae836144edcaf4b"
+dependencies = [
+ "ahash",
+ "arrow-array 57.3.0",
+ "arrow-buffer 57.3.0",
+ "arrow-data 57.3.0",
+ "arrow-schema 57.3.0",
+ "num-traits",
+]
+
 [[package]]
 name = "arrow-string"
 version = "55.2.0"
@@ -466,6 +706,35 @@ dependencies = [
  "regex-syntax",
 ]
 
+[[package]]
+name = "arrow-string"
+version = "57.3.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "85e968097061b3c0e9fe3079cf2e703e487890700546b5b0647f60fca1b5a8d8"
+dependencies = [
+ "arrow-array 57.3.0",
+ "arrow-buffer 57.3.0",
+ "arrow-data 57.3.0",
+ "arrow-schema 57.3.0",
+ "arrow-select 57.3.0",
+ "memchr",
+ "num-traits",
+ "regex",
+ "regex-syntax",
+]
+
+[[package]]
+name = "async-channel"
+version = "2.5.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "924ed96dd52d1b75e9c1a3e6275715fd320f5f9439fb5a4a11fa51f4221158d2"
+dependencies = [
+ "concurrent-queue",
+ "event-listener-strategy",
+ "futures-core",
+ "pin-project-lite",
+]
+
 [[package]]
 name = "async-compression"
 version = "0.4.19"
@@ -483,6 +752,28 @@ dependencies = [
  "zstd-safe",
 ]
 
+[[package]]
+name = "async-lock"
+version = "3.4.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "290f7f2596bd5b78a9fec8088ccd89180d7f9f55b94b0576823bbbdc72ee8311"
+dependencies = [
+ "event-listener",
+ "event-listener-strategy",
+ "pin-project-lite",
+]
+
+[[package]]
+name = "async-recursion"
+version = "1.1.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "3b43422f69d8ff38f95f1b2bb76517c91589a924d1559a0e935d7c8ce0274c11"
+dependencies = [
+ "proc-macro2",
+ "quote",
+ "syn 2.0.117",
+]
+
 [[package]]
 name = "async-stream"
 version = "0.3.6"
@@ -502,7 +793,7 @@ checksum = "c7c24de15d275a1ecfd47a380fb4d5ec9bfe0933f309ed5e705b775596a3574d"
 dependencies = [
  "proc-macro2",
  "quote",
- "syn",
+ "syn 2.0.117",
 ]
 
 [[package]]
@@ -513,7 +804,16 @@ checksum = "9035ad2d096bed7955a320ee7e2230574d28fd3c3a0f186cbea1ff3c7eed5dbb"
 dependencies = [
  "proc-macro2",
  "quote",
- "syn",
+ "syn 2.0.117",
+]
+
+[[package]]
+name = "async_cell"
+version = "0.2.3"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "447ab28afbb345f5408b120702a44e5529ebf90b1796ec76e9528df8e288e6c2"
+dependencies = [
+ "loom",
 ]
 
 [[package]]
@@ -538,93 +838,471 @@ source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "c08606f8c3cbf4ce6ec8e28fb0014a2c086708fe954eaa885384a6165172e7e8"
 
 [[package]]
-name = "axum"
-version = "0.7.9"
+name = "aws-config"
+version = "1.8.14"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "edca88bc138befd0323b20752846e6587272d3b03b0343c8ea28a6f819e6e71f"
+checksum = "8a8fc176d53d6fe85017f230405e3255cedb4a02221cb55ed6d76dccbbb099b2"
 dependencies = [
- "async-trait",
- "axum-core",
+ "aws-credential-types",
+ "aws-runtime",
+ "aws-sdk-sso",
+ "aws-sdk-ssooidc",
+ "aws-sdk-sts",
+ "aws-smithy-async",
+ "aws-smithy-http",
+ "aws-smithy-json",
+ "aws-smithy-runtime",
+ "aws-smithy-runtime-api",
+ "aws-smithy-types",
+ "aws-types",
  "bytes",
- "futures-util",
- "http",
- "http-body",
- "http-body-util",
- "itoa",
- "matchit",
- "memchr",
- "mime",
- "percent-encoding",
- "pin-project-lite",
- "rustversion",
- "serde",
- "sync_wrapper",
- "tower 0.5.3",
- "tower-layer",
- "tower-service",
+ "fastrand",
+ "hex",
+ "http 1.4.0",
+ "ring",
+ "time",
+ "tokio",
+ "tracing",
+ "url",
+ "zeroize",
 ]
 
 [[package]]
-name = "axum-core"
-version = "0.4.5"
+name = "aws-credential-types"
+version = "1.2.13"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "09f2bd6146b97ae3359fa0cc6d6b376d9539582c7b4220f041a33ec24c226199"
+checksum = "6d203b0bf2626dcba8665f5cd0871d7c2c0930223d6b6be9097592fea21242d0"
 dependencies = [
- "async-trait",
- "bytes",
- "futures-util",
- "http",
- "http-body",
- "http-body-util",
- "mime",
- "pin-project-lite",
- "rustversion",
- "sync_wrapper",
- "tower-layer",
- "tower-service",
+ "aws-smithy-async",
+ "aws-smithy-runtime-api",
+ "aws-smithy-types",
+ "zeroize",
 ]
 
 [[package]]
-name = "base64"
-version = "0.22.1"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "72b3254f16251a8381aa12e40e3c4d2f0199f8c6508fbecb9d91f575e0fbb8c6"
-
-[[package]]
-name = "bigdecimal"
-version = "0.4.10"
+name = "aws-lc-rs"
+version = "1.16.3"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "4d6867f1565b3aad85681f1015055b087fcfd840d6aeee6eee7f2da317603695"
+checksum = "0ec6fb3fe69024a75fa7e1bfb48aa6cf59706a101658ea01bfd33b2b248a038f"
 dependencies = [
- "autocfg",
- "libm",
- "num-bigint",
- "num-integer",
- "num-traits",
+ "aws-lc-sys",
+ "zeroize",
 ]
 
 [[package]]
-name = "bincode"
-version = "2.0.1"
+name = "aws-lc-sys"
+version = "0.40.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "36eaf5d7b090263e8150820482d5d93cd964a81e4019913c972f4edcc6edb740"
+checksum = "f50037ee5e1e41e7b8f9d161680a725bd1626cb6f8c7e901f91f942850852fe7"
 dependencies = [
- "bincode_derive",
- "serde",
- "unty",
+ "cc",
+ "cmake",
+ "dunce",
+ "fs_extra",
 ]
 
 [[package]]
-name = "bincode_derive"
-version = "2.0.1"
+name = "aws-runtime"
+version = "1.7.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "bf95709a440f45e986983918d0e8a1f30a9b1df04918fc828670606804ac3c09"
+checksum = "ede2ddc593e6c8acc6ce3358c28d6677a6dc49b65ba4b37a2befe14a11297e75"
 dependencies = [
- "virtue",
+ "aws-credential-types",
+ "aws-sigv4",
+ "aws-smithy-async",
+ "aws-smithy-http",
+ "aws-smithy-runtime",
+ "aws-smithy-runtime-api",
+ "aws-smithy-types",
+ "aws-types",
+ "bytes",
+ "bytes-utils",
+ "fastrand",
+ "http 1.4.0",
+ "http-body 1.0.1",
+ "percent-encoding",
+ "pin-project-lite",
+ "tracing",
+ "uuid",
 ]
 
 [[package]]
-name = "bindgen"
+name = "aws-sdk-sso"
+version = "1.95.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "00c5ff27c6ba2cbd95e6e26e2e736676fdf6bcf96495b187733f521cfe4ce448"
+dependencies = [
+ "aws-credential-types",
+ "aws-runtime",
+ "aws-smithy-async",
+ "aws-smithy-http",
+ "aws-smithy-json",
+ "aws-smithy-observability",
+ "aws-smithy-runtime",
+ "aws-smithy-runtime-api",
+ "aws-smithy-types",
+ "aws-types",
+ "bytes",
+ "fastrand",
+ "http 0.2.12",
+ "http 1.4.0",
+ "regex-lite",
+ "tracing",
+]
+
+[[package]]
+name = "aws-sdk-ssooidc"
+version = "1.97.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "4d186f1e5a3694a188e5a0640b3115ccc6e084d104e16fd6ba968dca072ffef8"
+dependencies = [
+ "aws-credential-types",
+ "aws-runtime",
+ "aws-smithy-async",
+ "aws-smithy-http",
+ "aws-smithy-json",
+ "aws-smithy-observability",
+ "aws-smithy-runtime",
+ "aws-smithy-runtime-api",
+ "aws-smithy-types",
+ "aws-types",
+ "bytes",
+ "fastrand",
+ "http 0.2.12",
+ "http 1.4.0",
+ "regex-lite",
+ "tracing",
+]
+
+[[package]]
+name = "aws-sdk-sts"
+version = "1.99.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "9acba7c62f3d4e2408fa998a3a8caacd8b9a5b5549cf36e2372fbdae329d5449"
+dependencies = [
+ "aws-credential-types",
+ "aws-runtime",
+ "aws-smithy-async",
+ "aws-smithy-http",
+ "aws-smithy-json",
+ "aws-smithy-observability",
+ "aws-smithy-query",
+ "aws-smithy-runtime",
+ "aws-smithy-runtime-api",
+ "aws-smithy-types",
+ "aws-smithy-xml",
+ "aws-types",
+ "fastrand",
+ "http 0.2.12",
+ "http 1.4.0",
+ "regex-lite",
+ "tracing",
+]
+
+[[package]]
+name = "aws-sigv4"
+version = "1.4.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "37411f8e0f4bea0c3ca0958ce7f18f6439db24d555dbd809787262cd00926aa9"
+dependencies = [
+ "aws-credential-types",
+ "aws-smithy-http",
+ "aws-smithy-runtime-api",
+ "aws-smithy-types",
+ "bytes",
+ "form_urlencoded",
+ "hex",
+ "hmac",
+ "http 0.2.12",
+ "http 1.4.0",
+ "percent-encoding",
+ "sha2",
+ "time",
+ "tracing",
+]
+
+[[package]]
+name = "aws-smithy-async"
+version = "1.2.13"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "5cc50d0f63e714784b84223abd7abbc8577de8c35d699e0edd19f0a88a08ae13"
+dependencies = [
+ "futures-util",
+ "pin-project-lite",
+ "tokio",
+]
+
+[[package]]
+name = "aws-smithy-http"
+version = "0.63.5"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "d619373d490ad70966994801bc126846afaa0d1ee920697a031f0cf63f2568e7"
+dependencies = [
+ "aws-smithy-runtime-api",
+ "aws-smithy-types",
+ "bytes",
+ "bytes-utils",
+ "futures-core",
+ "futures-util",
+ "http 1.4.0",
+ "http-body 1.0.1",
+ "http-body-util",
+ "percent-encoding",
+ "pin-project-lite",
+ "pin-utils",
+ "tracing",
+]
+
+[[package]]
+name = "aws-smithy-http-client"
+version = "1.1.11"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "00ccbb08c10f6bcf912f398188e42ee2eab5f1767ce215a02a73bc5df1bbdd95"
+dependencies = [
+ "aws-smithy-async",
+ "aws-smithy-runtime-api",
+ "aws-smithy-types",
+ "h2",
+ "http 1.4.0",
+ "hyper",
+ "hyper-rustls",
+ "hyper-util",
+ "pin-project-lite",
+ "rustls",
+ "rustls-native-certs",
+ "rustls-pki-types",
+ "tokio",
+ "tokio-rustls",
+ "tower 0.5.3",
+ "tracing",
+]
+
+[[package]]
+name = "aws-smithy-json"
+version = "0.62.4"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "27b3a779093e18cad88bbae08dc4261e1d95018c4c5b9356a52bcae7c0b6e9bb"
+dependencies = [
+ "aws-smithy-types",
+]
+
+[[package]]
+name = "aws-smithy-observability"
+version = "0.2.5"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "4d3f39d5bb871aaf461d59144557f16d5927a5248a983a40654d9cf3b9ba183b"
+dependencies = [
+ "aws-smithy-runtime-api",
+]
+
+[[package]]
+name = "aws-smithy-query"
+version = "0.60.14"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "05f76a580e3d8f8961e5d48763214025a2af65c2fa4cd1fb7f270a0e107a71b0"
+dependencies = [
+ "aws-smithy-types",
+ "urlencoding",
+]
+
+[[package]]
+name = "aws-smithy-runtime"
+version = "1.10.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "22ccf7f6eba8b2dcf8ce9b74806c6c185659c311665c4bf8d6e71ebd454db6bf"
+dependencies = [
+ "aws-smithy-async",
+ "aws-smithy-http",
+ "aws-smithy-http-client",
+ "aws-smithy-observability",
+ "aws-smithy-runtime-api",
+ "aws-smithy-types",
+ "bytes",
+ "fastrand",
+ "http 0.2.12",
+ "http 1.4.0",
+ "http-body 0.4.6",
+ "http-body 1.0.1",
+ "http-body-util",
+ "pin-project-lite",
+ "pin-utils",
+ "tokio",
+ "tracing",
+]
+
+[[package]]
+name = "aws-smithy-runtime-api"
+version = "1.11.5"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "b4af6e5def28be846479bbeac55aa4603d6f7986fc5da4601ba324dd5d377516"
+dependencies = [
+ "aws-smithy-async",
+ "aws-smithy-types",
+ "bytes",
+ "http 0.2.12",
+ "http 1.4.0",
+ "pin-project-lite",
+ "tokio",
+ "tracing",
+ "zeroize",
+]
+
+[[package]]
+name = "aws-smithy-types"
+version = "1.4.5"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "8ca2734c16913a45343b37313605d84e7d8b34a4611598ce1d25b35860a2bed3"
+dependencies = [
+ "base64-simd",
+ "bytes",
+ "bytes-utils",
+ "http 0.2.12",
+ "http 1.4.0",
+ "http-body 0.4.6",
+ "http-body 1.0.1",
+ "http-body-util",
+ "itoa",
+ "num-integer",
+ "pin-project-lite",
+ "pin-utils",
+ "ryu",
+ "serde",
+ "time",
+]
+
+[[package]]
+name = "aws-smithy-xml"
+version = "0.60.14"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "b53543b4b86ed43f051644f704a98c7291b3618b67adf057ee77a366fa52fcaa"
+dependencies = [
+ "xmlparser",
+]
+
+[[package]]
+name = "aws-types"
+version = "1.3.13"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "0470cc047657c6e286346bdf10a8719d26efd6a91626992e0e64481e44323e96"
+dependencies = [
+ "aws-credential-types",
+ "aws-smithy-async",
+ "aws-smithy-runtime-api",
+ "aws-smithy-types",
+ "rustc_version",
+ "tracing",
+]
+
+[[package]]
+name = "axum"
+version = "0.7.9"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "edca88bc138befd0323b20752846e6587272d3b03b0343c8ea28a6f819e6e71f"
+dependencies = [
+ "async-trait",
+ "axum-core",
+ "bytes",
+ "futures-util",
+ "http 1.4.0",
+ "http-body 1.0.1",
+ "http-body-util",
+ "itoa",
+ "matchit",
+ "memchr",
+ "mime",
+ "percent-encoding",
+ "pin-project-lite",
+ "rustversion",
+ "serde",
+ "sync_wrapper",
+ "tower 0.5.3",
+ "tower-layer",
+ "tower-service",
+]
+
+[[package]]
+name = "axum-core"
+version = "0.4.5"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "09f2bd6146b97ae3359fa0cc6d6b376d9539582c7b4220f041a33ec24c226199"
+dependencies = [
+ "async-trait",
+ "bytes",
+ "futures-util",
+ "http 1.4.0",
+ "http-body 1.0.1",
+ "http-body-util",
+ "mime",
+ "pin-project-lite",
+ "rustversion",
+ "sync_wrapper",
+ "tower-layer",
+ "tower-service",
+]
+
+[[package]]
+name = "backon"
+version = "1.6.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "cffb0e931875b666fc4fcb20fee52e9bbd1ef836fd9e9e04ec21555f9f85f7ef"
+dependencies = [
+ "fastrand",
+ "gloo-timers",
+ "tokio",
+]
+
+[[package]]
+name = "base64"
+version = "0.22.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "72b3254f16251a8381aa12e40e3c4d2f0199f8c6508fbecb9d91f575e0fbb8c6"
+
+[[package]]
+name = "base64-simd"
+version = "0.8.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "339abbe78e73178762e23bea9dfd08e697eb3f3301cd4be981c0f78ba5859195"
+dependencies = [
+ "outref",
+ "vsimd",
+]
+
+[[package]]
+name = "bigdecimal"
+version = "0.4.10"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "4d6867f1565b3aad85681f1015055b087fcfd840d6aeee6eee7f2da317603695"
+dependencies = [
+ "autocfg",
+ "libm",
+ "num-bigint",
+ "num-integer",
+ "num-traits",
+ "serde",
+]
+
+[[package]]
+name = "bincode"
+version = "2.0.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "36eaf5d7b090263e8150820482d5d93cd964a81e4019913c972f4edcc6edb740"
+dependencies = [
+ "bincode_derive",
+ "serde",
+ "unty",
+]
+
+[[package]]
+name = "bincode_derive"
+version = "2.0.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "bf95709a440f45e986983918d0e8a1f30a9b1df04918fc828670606804ac3c09"
+dependencies = [
+ "virtue",
+]
+
+[[package]]
+name = "bindgen"
 version = "0.65.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "cfdf7b466f9a4903edc73f95d6d2bcd5baf8ae620638762244d3f60143643cc5"
@@ -641,7 +1319,7 @@ dependencies = [
  "regex",
  "rustc-hash 1.1.0",
  "shlex",
- "syn",
+ "syn 2.0.117",
 ]
 
 [[package]]
@@ -659,7 +1337,7 @@ dependencies = [
  "regex",
  "rustc-hash 2.1.2",
  "shlex",
- "syn",
+ "syn 2.0.117",
 ]
 
 [[package]]
@@ -683,6 +1361,27 @@ dependencies = [
  "typenum",
 ]
 
+[[package]]
+name = "bitpacking"
+version = "0.9.3"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "96a7139abd3d9cebf8cd6f920a389cf3dc9576172e32f4563f188cae3c3eb019"
+dependencies = [
+ "crunchy",
+]
+
+[[package]]
+name = "bitvec"
+version = "1.0.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "1bc2832c24239b0141d5674bb9174f9d68a8b5b3f2753311927c172ca46f7e9c"
+dependencies = [
+ "funty",
+ "radium",
+ "tap",
+ "wyz",
+]
+
 [[package]]
 name = "blake2"
 version = "0.10.6"
@@ -715,6 +1414,31 @@ dependencies = [
  "generic-array",
 ]
 
+[[package]]
+name = "bon"
+version = "3.9.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "f47dbe92550676ee653353c310dfb9cf6ba17ee70396e1f7cf0a2020ad49b2fe"
+dependencies = [
+ "bon-macros",
+ "rustversion",
+]
+
+[[package]]
+name = "bon-macros"
+version = "3.9.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "519bd3116aeeb42d5372c29d982d16d0170d3d4a5ed85fc7dd91642ffff3c67c"
+dependencies = [
+ "darling",
+ "ident_case",
+ "prettyplease",
+ "proc-macro2",
+ "quote",
+ "rustversion",
+ "syn 2.0.117",
+]
+
 [[package]]
 name = "brotli"
 version = "8.0.2"
@@ -745,6 +1469,12 @@ dependencies = [
  "allocator-api2",
 ]
 
+[[package]]
+name = "bytemuck"
+version = "1.25.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "c8efb64bd706a16a1bdde310ae86b351e4d21550d98d056f22f8a7f7a2183fec"
+
 [[package]]
 name = "byteorder"
 version = "1.5.0"
@@ -757,6 +1487,16 @@ version = "1.11.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "1e748733b7cbc798e1434b6ac524f0c1ff2ab456fe201501e6497c8417a4fc33"
 
+[[package]]
+name = "bytes-utils"
+version = "0.1.4"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "7dafe3a8757b027e2be6e4e5601ed563c55989fcf1546e933c66c8eb3a058d35"
+dependencies = [
+ "bytes",
+ "either",
+]
+
 [[package]]
 name = "bzip2"
 version = "0.5.2"
@@ -866,13 +1606,19 @@ dependencies = [
  "shlex",
 ]
 
+[[package]]
+name = "census"
+version = "0.4.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "4f4c707c6a209cbe82d10abd08e1ea8995e9ea937d2550646e02798948992be0"
+
 [[package]]
 name = "cexpr"
 version = "0.6.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "6fac387a98bb7c37292057cffc56d62ecb629900026402633ae9160df93a8766"
 dependencies = [
- "nom",
+ "nom 7.1.3",
 ]
 
 [[package]]
@@ -887,6 +1633,12 @@ version = "0.1.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "fd16c4719339c4530435d38e511904438d07cce7950afa3718a84ac36c10e89e"
 
+[[package]]
+name = "cfg_aliases"
+version = "0.2.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "613afe47fcd5fac7ccf1db93babcb082c5994d996f20b8b159f2ad1658eb5724"
+
 [[package]]
 name = "chrono"
 version = "0.4.44"
@@ -896,6 +1648,7 @@ dependencies = [
  "iana-time-zone",
  "js-sys",
  "num-traits",
+ "serde",
  "wasm-bindgen",
  "windows-link",
 ]
@@ -952,7 +1705,7 @@ dependencies = [
  "heck",
  "proc-macro2",
  "quote",
- "syn",
+ "syn 2.0.117",
 ]
 
 [[package]]
@@ -1005,6 +1758,21 @@ dependencies = [
  "unicode-width 0.2.2",
 ]
 
+[[package]]
+name = "concurrent-queue"
+version = "2.5.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "4ca0197aee26d1ae37445ee532fefce43251d24cc7c166799f4d46817f1d3973"
+dependencies = [
+ "crossbeam-utils",
+]
+
+[[package]]
+name = "const-oid"
+version = "0.9.6"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "c2459377285ad874054d797f3ccebf984978aa39129f6eafde5cdc8315b612f8"
+
 [[package]]
 name = "const-random"
 version = "0.1.18"
@@ -1031,6 +1799,16 @@ version = "0.4.2"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "3d52eff69cd5e647efe296129160853a42795992097e8af39800e1060caeea9b"
 
+[[package]]
+name = "core-foundation"
+version = "0.10.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "b2a6cd9ae233e7f62ba4e9353e81a88df7fc8a5987b8d445b4d90c879bd156f6"
+dependencies = [
+ "core-foundation-sys",
+ "libc",
+]
+
 [[package]]
 name = "core-foundation-sys"
 version = "0.8.7"
@@ -1205,6 +1983,15 @@ version = "0.128.4"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "903adeaf4938e60209a97b53a2e4326cd2d356aab9764a1934630204bae381c9"
 
+[[package]]
+name = "crc32c"
+version = "0.6.8"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "3a47af21622d091a8f0fb295b88bc886ac74efcc613efc19f5d0b21de5c89e47"
+dependencies = [
+ "rustc_version",
+]
+
 [[package]]
 name = "crc32fast"
 version = "1.5.0"
@@ -1237,8 +2024,27 @@ dependencies = [
 name = "crossbeam-epoch"
 version = "0.9.18"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "5b82ac4a3c2ca9c3460964f020e1402edd5753411d7737aa39c3714ad1b5420e"
+checksum = "5b82ac4a3c2ca9c3460964f020e1402edd5753411d7737aa39c3714ad1b5420e"
+dependencies = [
+ "crossbeam-utils",
+]
+
+[[package]]
+name = "crossbeam-queue"
+version = "0.3.12"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "0f58bbc28f91df819d0aa2a2c00cd19754769c2fad90579b3592b1c9ba7a3115"
+dependencies = [
+ "crossbeam-utils",
+]
+
+[[package]]
+name = "crossbeam-skiplist"
+version = "0.1.3"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "df29de440c58ca2cc6e587ec3d22347551a32435fbde9d2bff64e78a9ffa151b"
 dependencies = [
+ "crossbeam-epoch",
  "crossbeam-utils",
 ]
 
@@ -1323,6 +2129,40 @@ dependencies = [
  "windows-sys 0.59.0",
 ]
 
+[[package]]
+name = "darling"
+version = "0.23.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "25ae13da2f202d56bd7f91c25fba009e7717a1e4a1cc98a76d844b65ae912e9d"
+dependencies = [
+ "darling_core",
+ "darling_macro",
+]
+
+[[package]]
+name = "darling_core"
+version = "0.23.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "9865a50f7c335f53564bb694ef660825eb8610e0a53d3e11bf1b0d3df31e03b0"
+dependencies = [
+ "ident_case",
+ "proc-macro2",
+ "quote",
+ "strsim",
+ "syn 2.0.117",
+]
+
+[[package]]
+name = "darling_macro"
+version = "0.23.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "ac3984ec7bd6cfa798e62b4a642426a5be0e68f9401cfc2a01e3fa9ea2fcdb8d"
+dependencies = [
+ "darling_core",
+ "quote",
+ "syn 2.0.117",
+]
+
 [[package]]
 name = "dashmap"
 version = "6.1.0"
@@ -1342,36 +2182,36 @@ name = "datafusion"
 version = "48.0.1"
 source = "git+https://github.com/FunctionStream/datafusion?branch=48.0.1%2Ffs#c3b2ac13dabd4145f21045e315529b917079b7d1"
 dependencies = [
- "arrow",
+ "arrow 55.2.0",
  "arrow-ipc 55.2.0",
  "arrow-schema 55.2.0",
  "async-trait",
  "bytes",
  "bzip2",
  "chrono",
- "datafusion-catalog",
- "datafusion-catalog-listing",
- "datafusion-common",
- "datafusion-common-runtime",
- "datafusion-datasource",
- "datafusion-datasource-csv",
- "datafusion-datasource-json",
+ "datafusion-catalog 48.0.1",
+ "datafusion-catalog-listing 48.0.1",
+ "datafusion-common 48.0.1",
+ "datafusion-common-runtime 48.0.1",
+ "datafusion-datasource 48.0.1",
+ "datafusion-datasource-csv 48.0.1",
+ "datafusion-datasource-json 48.0.1",
  "datafusion-datasource-parquet",
- "datafusion-execution",
- "datafusion-expr",
- "datafusion-expr-common",
- "datafusion-functions",
- "datafusion-functions-aggregate",
- "datafusion-functions-nested",
- "datafusion-functions-table",
- "datafusion-functions-window",
- "datafusion-optimizer",
- "datafusion-physical-expr",
- "datafusion-physical-expr-common",
- "datafusion-physical-optimizer",
- "datafusion-physical-plan",
- "datafusion-session",
- "datafusion-sql",
+ "datafusion-execution 48.0.1",
+ "datafusion-expr 48.0.1",
+ "datafusion-expr-common 48.0.1",
+ "datafusion-functions 48.0.1",
+ "datafusion-functions-aggregate 48.0.1",
+ "datafusion-functions-nested 48.0.1",
+ "datafusion-functions-table 48.0.1",
+ "datafusion-functions-window 48.0.1",
+ "datafusion-optimizer 48.0.1",
+ "datafusion-physical-expr 48.0.1",
+ "datafusion-physical-expr-common 48.0.1",
+ "datafusion-physical-optimizer 48.0.1",
+ "datafusion-physical-plan 48.0.1",
+ "datafusion-session 48.0.1",
+ "datafusion-sql 48.0.1",
  "flate2",
  "futures",
  "itertools 0.14.0",
@@ -1381,7 +2221,7 @@ dependencies = [
  "parquet",
  "rand 0.9.2",
  "regex",
- "sqlparser",
+ "sqlparser 0.55.0",
  "tempfile",
  "tokio",
  "url",
@@ -1390,23 +2230,97 @@ dependencies = [
  "zstd",
 ]
 
+[[package]]
+name = "datafusion"
+version = "52.5.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "7541353e77dc7262b71ca27be07d8393661737e3a73b5d1b1c6f7d814c64fa2a"
+dependencies = [
+ "arrow 57.3.0",
+ "arrow-schema 57.3.0",
+ "async-trait",
+ "bytes",
+ "chrono",
+ "datafusion-catalog 52.5.0",
+ "datafusion-catalog-listing 52.5.0",
+ "datafusion-common 52.5.0",
+ "datafusion-common-runtime 52.5.0",
+ "datafusion-datasource 52.5.0",
+ "datafusion-datasource-arrow",
+ "datafusion-datasource-csv 52.5.0",
+ "datafusion-datasource-json 52.5.0",
+ "datafusion-execution 52.5.0",
+ "datafusion-expr 52.5.0",
+ "datafusion-expr-common 52.5.0",
+ "datafusion-functions 52.5.0",
+ "datafusion-functions-aggregate 52.5.0",
+ "datafusion-functions-nested 52.5.0",
+ "datafusion-functions-table 52.5.0",
+ "datafusion-functions-window 52.5.0",
+ "datafusion-optimizer 52.5.0",
+ "datafusion-physical-expr 52.5.0",
+ "datafusion-physical-expr-adapter",
+ "datafusion-physical-expr-common 52.5.0",
+ "datafusion-physical-optimizer 52.5.0",
+ "datafusion-physical-plan 52.5.0",
+ "datafusion-session 52.5.0",
+ "datafusion-sql 52.5.0",
+ "futures",
+ "itertools 0.14.0",
+ "log",
+ "object_store",
+ "parking_lot",
+ "rand 0.9.2",
+ "regex",
+ "sqlparser 0.59.0",
+ "tempfile",
+ "tokio",
+ "url",
+ "uuid",
+]
+
 [[package]]
 name = "datafusion-catalog"
 version = "48.0.1"
 source = "git+https://github.com/FunctionStream/datafusion?branch=48.0.1%2Ffs#c3b2ac13dabd4145f21045e315529b917079b7d1"
 dependencies = [
- "arrow",
+ "arrow 55.2.0",
+ "async-trait",
+ "dashmap",
+ "datafusion-common 48.0.1",
+ "datafusion-common-runtime 48.0.1",
+ "datafusion-datasource 48.0.1",
+ "datafusion-execution 48.0.1",
+ "datafusion-expr 48.0.1",
+ "datafusion-physical-expr 48.0.1",
+ "datafusion-physical-plan 48.0.1",
+ "datafusion-session 48.0.1",
+ "datafusion-sql 48.0.1",
+ "futures",
+ "itertools 0.14.0",
+ "log",
+ "object_store",
+ "parking_lot",
+ "tokio",
+]
+
+[[package]]
+name = "datafusion-catalog"
+version = "52.5.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "9997731f90fa5398ef831ad0e69600f92c861b79c0d38bd1a29b6f0e3a0ce4c8"
+dependencies = [
+ "arrow 57.3.0",
  "async-trait",
  "dashmap",
- "datafusion-common",
- "datafusion-common-runtime",
- "datafusion-datasource",
- "datafusion-execution",
- "datafusion-expr",
- "datafusion-physical-expr",
- "datafusion-physical-plan",
- "datafusion-session",
- "datafusion-sql",
+ "datafusion-common 52.5.0",
+ "datafusion-common-runtime 52.5.0",
+ "datafusion-datasource 52.5.0",
+ "datafusion-execution 52.5.0",
+ "datafusion-expr 52.5.0",
+ "datafusion-physical-expr 52.5.0",
+ "datafusion-physical-plan 52.5.0",
+ "datafusion-session 52.5.0",
  "futures",
  "itertools 0.14.0",
  "log",
@@ -1420,30 +2334,53 @@ name = "datafusion-catalog-listing"
 version = "48.0.1"
 source = "git+https://github.com/FunctionStream/datafusion?branch=48.0.1%2Ffs#c3b2ac13dabd4145f21045e315529b917079b7d1"
 dependencies = [
- "arrow",
+ "arrow 55.2.0",
  "async-trait",
- "datafusion-catalog",
- "datafusion-common",
- "datafusion-datasource",
- "datafusion-execution",
- "datafusion-expr",
- "datafusion-physical-expr",
- "datafusion-physical-expr-common",
- "datafusion-physical-plan",
- "datafusion-session",
+ "datafusion-catalog 48.0.1",
+ "datafusion-common 48.0.1",
+ "datafusion-datasource 48.0.1",
+ "datafusion-execution 48.0.1",
+ "datafusion-expr 48.0.1",
+ "datafusion-physical-expr 48.0.1",
+ "datafusion-physical-expr-common 48.0.1",
+ "datafusion-physical-plan 48.0.1",
+ "datafusion-session 48.0.1",
  "futures",
  "log",
  "object_store",
  "tokio",
 ]
 
+[[package]]
+name = "datafusion-catalog-listing"
+version = "52.5.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "2b30a3dd50dec860c9559275c8d97d9de602e611237a6ecfbda0b3b63b872352"
+dependencies = [
+ "arrow 57.3.0",
+ "async-trait",
+ "datafusion-catalog 52.5.0",
+ "datafusion-common 52.5.0",
+ "datafusion-datasource 52.5.0",
+ "datafusion-execution 52.5.0",
+ "datafusion-expr 52.5.0",
+ "datafusion-physical-expr 52.5.0",
+ "datafusion-physical-expr-adapter",
+ "datafusion-physical-expr-common 52.5.0",
+ "datafusion-physical-plan 52.5.0",
+ "futures",
+ "itertools 0.14.0",
+ "log",
+ "object_store",
+]
+
 [[package]]
 name = "datafusion-common"
 version = "48.0.1"
 source = "git+https://github.com/FunctionStream/datafusion?branch=48.0.1%2Ffs#c3b2ac13dabd4145f21045e315529b917079b7d1"
 dependencies = [
  "ahash",
- "arrow",
+ "arrow 55.2.0",
  "arrow-ipc 55.2.0",
  "base64",
  "half",
@@ -1455,7 +2392,29 @@ dependencies = [
  "parquet",
  "paste",
  "recursive",
- "sqlparser",
+ "sqlparser 0.55.0",
+ "tokio",
+ "web-time",
+]
+
+[[package]]
+name = "datafusion-common"
+version = "52.5.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "d551054acec0398ca604512310b77ce05c46f66e54b54d48200a686e385cca4e"
+dependencies = [
+ "ahash",
+ "arrow 57.3.0",
+ "arrow-ipc 57.3.0",
+ "chrono",
+ "half",
+ "hashbrown 0.16.1",
+ "indexmap 2.13.1",
+ "libc",
+ "log",
+ "object_store",
+ "paste",
+ "sqlparser 0.59.0",
  "tokio",
  "web-time",
 ]
@@ -1470,25 +2429,36 @@ dependencies = [
  "tokio",
 ]
 
+[[package]]
+name = "datafusion-common-runtime"
+version = "52.5.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "567d40e285f5b79f8737b576605721cd6c1133b5d2b00bdbd5d9838d90d0812f"
+dependencies = [
+ "futures",
+ "log",
+ "tokio",
+]
+
 [[package]]
 name = "datafusion-datasource"
 version = "48.0.1"
 source = "git+https://github.com/FunctionStream/datafusion?branch=48.0.1%2Ffs#c3b2ac13dabd4145f21045e315529b917079b7d1"
 dependencies = [
- "arrow",
+ "arrow 55.2.0",
  "async-compression",
  "async-trait",
  "bytes",
  "bzip2",
  "chrono",
- "datafusion-common",
- "datafusion-common-runtime",
- "datafusion-execution",
- "datafusion-expr",
- "datafusion-physical-expr",
- "datafusion-physical-expr-common",
- "datafusion-physical-plan",
- "datafusion-session",
+ "datafusion-common 48.0.1",
+ "datafusion-common-runtime 48.0.1",
+ "datafusion-execution 48.0.1",
+ "datafusion-expr 48.0.1",
+ "datafusion-physical-expr 48.0.1",
+ "datafusion-physical-expr-common 48.0.1",
+ "datafusion-physical-plan 48.0.1",
+ "datafusion-session 48.0.1",
  "flate2",
  "futures",
  "glob",
@@ -1505,24 +2475,100 @@ dependencies = [
  "zstd",
 ]
 
+[[package]]
+name = "datafusion-datasource"
+version = "52.5.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "27d2668f51b3b30befae2207472569e37807fdedd1d14da58acc6f8ca6257eae"
+dependencies = [
+ "arrow 57.3.0",
+ "async-trait",
+ "bytes",
+ "chrono",
+ "datafusion-common 52.5.0",
+ "datafusion-common-runtime 52.5.0",
+ "datafusion-execution 52.5.0",
+ "datafusion-expr 52.5.0",
+ "datafusion-physical-expr 52.5.0",
+ "datafusion-physical-expr-adapter",
+ "datafusion-physical-expr-common 52.5.0",
+ "datafusion-physical-plan 52.5.0",
+ "datafusion-session 52.5.0",
+ "futures",
+ "glob",
+ "itertools 0.14.0",
+ "log",
+ "object_store",
+ "rand 0.9.2",
+ "tokio",
+ "url",
+]
+
+[[package]]
+name = "datafusion-datasource-arrow"
+version = "52.5.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "e02e1b3e3a8ec55f1f62de4252b0407c8567363d056078769a197e24fc834a0f"
+dependencies = [
+ "arrow 57.3.0",
+ "arrow-ipc 57.3.0",
+ "async-trait",
+ "bytes",
+ "datafusion-common 52.5.0",
+ "datafusion-common-runtime 52.5.0",
+ "datafusion-datasource 52.5.0",
+ "datafusion-execution 52.5.0",
+ "datafusion-expr 52.5.0",
+ "datafusion-physical-expr-common 52.5.0",
+ "datafusion-physical-plan 52.5.0",
+ "datafusion-session 52.5.0",
+ "futures",
+ "itertools 0.14.0",
+ "object_store",
+ "tokio",
+]
+
 [[package]]
 name = "datafusion-datasource-csv"
 version = "48.0.1"
 source = "git+https://github.com/FunctionStream/datafusion?branch=48.0.1%2Ffs#c3b2ac13dabd4145f21045e315529b917079b7d1"
 dependencies = [
- "arrow",
+ "arrow 55.2.0",
+ "async-trait",
+ "bytes",
+ "datafusion-catalog 48.0.1",
+ "datafusion-common 48.0.1",
+ "datafusion-common-runtime 48.0.1",
+ "datafusion-datasource 48.0.1",
+ "datafusion-execution 48.0.1",
+ "datafusion-expr 48.0.1",
+ "datafusion-physical-expr 48.0.1",
+ "datafusion-physical-expr-common 48.0.1",
+ "datafusion-physical-plan 48.0.1",
+ "datafusion-session 48.0.1",
+ "futures",
+ "object_store",
+ "regex",
+ "tokio",
+]
+
+[[package]]
+name = "datafusion-datasource-csv"
+version = "52.5.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "b559d7bf87d4f900f847baba8509634f838d9718695389e903604cdcccdb01f3"
+dependencies = [
+ "arrow 57.3.0",
  "async-trait",
  "bytes",
- "datafusion-catalog",
- "datafusion-common",
- "datafusion-common-runtime",
- "datafusion-datasource",
- "datafusion-execution",
- "datafusion-expr",
- "datafusion-physical-expr",
- "datafusion-physical-expr-common",
- "datafusion-physical-plan",
- "datafusion-session",
+ "datafusion-common 52.5.0",
+ "datafusion-common-runtime 52.5.0",
+ "datafusion-datasource 52.5.0",
+ "datafusion-execution 52.5.0",
+ "datafusion-expr 52.5.0",
+ "datafusion-physical-expr-common 52.5.0",
+ "datafusion-physical-plan 52.5.0",
+ "datafusion-session 52.5.0",
  "futures",
  "object_store",
  "regex",
@@ -1534,45 +2580,67 @@ name = "datafusion-datasource-json"
 version = "48.0.1"
 source = "git+https://github.com/FunctionStream/datafusion?branch=48.0.1%2Ffs#c3b2ac13dabd4145f21045e315529b917079b7d1"
 dependencies = [
- "arrow",
+ "arrow 55.2.0",
  "async-trait",
  "bytes",
- "datafusion-catalog",
- "datafusion-common",
- "datafusion-common-runtime",
- "datafusion-datasource",
- "datafusion-execution",
- "datafusion-expr",
- "datafusion-physical-expr",
- "datafusion-physical-expr-common",
- "datafusion-physical-plan",
- "datafusion-session",
+ "datafusion-catalog 48.0.1",
+ "datafusion-common 48.0.1",
+ "datafusion-common-runtime 48.0.1",
+ "datafusion-datasource 48.0.1",
+ "datafusion-execution 48.0.1",
+ "datafusion-expr 48.0.1",
+ "datafusion-physical-expr 48.0.1",
+ "datafusion-physical-expr-common 48.0.1",
+ "datafusion-physical-plan 48.0.1",
+ "datafusion-session 48.0.1",
  "futures",
  "object_store",
  "serde_json",
  "tokio",
 ]
 
+[[package]]
+name = "datafusion-datasource-json"
+version = "52.5.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "250e2d7591ba8b638f063854650faa40bca4e8bd4059b2ece8836f6388d02db4"
+dependencies = [
+ "arrow 57.3.0",
+ "async-trait",
+ "bytes",
+ "datafusion-common 52.5.0",
+ "datafusion-common-runtime 52.5.0",
+ "datafusion-datasource 52.5.0",
+ "datafusion-execution 52.5.0",
+ "datafusion-expr 52.5.0",
+ "datafusion-physical-expr-common 52.5.0",
+ "datafusion-physical-plan 52.5.0",
+ "datafusion-session 52.5.0",
+ "futures",
+ "object_store",
+ "tokio",
+]
+
 [[package]]
 name = "datafusion-datasource-parquet"
 version = "48.0.1"
 source = "git+https://github.com/FunctionStream/datafusion?branch=48.0.1%2Ffs#c3b2ac13dabd4145f21045e315529b917079b7d1"
 dependencies = [
- "arrow",
+ "arrow 55.2.0",
  "async-trait",
  "bytes",
- "datafusion-catalog",
- "datafusion-common",
- "datafusion-common-runtime",
- "datafusion-datasource",
- "datafusion-execution",
- "datafusion-expr",
- "datafusion-functions-aggregate",
- "datafusion-physical-expr",
- "datafusion-physical-expr-common",
- "datafusion-physical-optimizer",
- "datafusion-physical-plan",
- "datafusion-session",
+ "datafusion-catalog 48.0.1",
+ "datafusion-common 48.0.1",
+ "datafusion-common-runtime 48.0.1",
+ "datafusion-datasource 48.0.1",
+ "datafusion-execution 48.0.1",
+ "datafusion-expr 48.0.1",
+ "datafusion-functions-aggregate 48.0.1",
+ "datafusion-physical-expr 48.0.1",
+ "datafusion-physical-expr-common 48.0.1",
+ "datafusion-physical-optimizer 48.0.1",
+ "datafusion-physical-plan 48.0.1",
+ "datafusion-session 48.0.1",
  "futures",
  "itertools 0.14.0",
  "log",
@@ -1588,15 +2656,42 @@ name = "datafusion-doc"
 version = "48.0.1"
 source = "git+https://github.com/FunctionStream/datafusion?branch=48.0.1%2Ffs#c3b2ac13dabd4145f21045e315529b917079b7d1"
 
+[[package]]
+name = "datafusion-doc"
+version = "52.5.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "b9496cb0db222dbb9a3735760ceca7fc56f35e1d5502c38d0caa77a81e9c1f6a"
+
 [[package]]
 name = "datafusion-execution"
 version = "48.0.1"
 source = "git+https://github.com/FunctionStream/datafusion?branch=48.0.1%2Ffs#c3b2ac13dabd4145f21045e315529b917079b7d1"
 dependencies = [
- "arrow",
+ "arrow 55.2.0",
+ "dashmap",
+ "datafusion-common 48.0.1",
+ "datafusion-expr 48.0.1",
+ "futures",
+ "log",
+ "object_store",
+ "parking_lot",
+ "rand 0.9.2",
+ "tempfile",
+ "url",
+]
+
+[[package]]
+name = "datafusion-execution"
+version = "52.5.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "dc45d23c516ed8d3637751e44e09e21b45b3f58b473c802dddd1f1ad4fe435ff"
+dependencies = [
+ "arrow 57.3.0",
+ "async-trait",
+ "chrono",
  "dashmap",
- "datafusion-common",
- "datafusion-expr",
+ "datafusion-common 52.5.0",
+ "datafusion-expr 52.5.0",
  "futures",
  "log",
  "object_store",
@@ -1611,19 +2706,41 @@ name = "datafusion-expr"
 version = "48.0.1"
 source = "git+https://github.com/FunctionStream/datafusion?branch=48.0.1%2Ffs#c3b2ac13dabd4145f21045e315529b917079b7d1"
 dependencies = [
- "arrow",
+ "arrow 55.2.0",
  "chrono",
- "datafusion-common",
- "datafusion-doc",
- "datafusion-expr-common",
- "datafusion-functions-aggregate-common",
- "datafusion-functions-window-common",
- "datafusion-physical-expr-common",
+ "datafusion-common 48.0.1",
+ "datafusion-doc 48.0.1",
+ "datafusion-expr-common 48.0.1",
+ "datafusion-functions-aggregate-common 48.0.1",
+ "datafusion-functions-window-common 48.0.1",
+ "datafusion-physical-expr-common 48.0.1",
  "indexmap 2.13.1",
  "paste",
  "recursive",
  "serde_json",
- "sqlparser",
+ "sqlparser 0.55.0",
+]
+
+[[package]]
+name = "datafusion-expr"
+version = "52.5.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "63dd30526d2db4fda6440806a41e4676334a94bc0596cc9cc2a0efed20ef2c44"
+dependencies = [
+ "arrow 57.3.0",
+ "async-trait",
+ "chrono",
+ "datafusion-common 52.5.0",
+ "datafusion-doc 52.5.0",
+ "datafusion-expr-common 52.5.0",
+ "datafusion-functions-aggregate-common 52.5.0",
+ "datafusion-functions-window-common 52.5.0",
+ "datafusion-physical-expr-common 52.5.0",
+ "indexmap 2.13.1",
+ "itertools 0.14.0",
+ "paste",
+ "serde_json",
+ "sqlparser 0.59.0",
 ]
 
 [[package]]
@@ -1631,8 +2748,21 @@ name = "datafusion-expr-common"
 version = "48.0.1"
 source = "git+https://github.com/FunctionStream/datafusion?branch=48.0.1%2Ffs#c3b2ac13dabd4145f21045e315529b917079b7d1"
 dependencies = [
- "arrow",
- "datafusion-common",
+ "arrow 55.2.0",
+ "datafusion-common 48.0.1",
+ "indexmap 2.13.1",
+ "itertools 0.14.0",
+ "paste",
+]
+
+[[package]]
+name = "datafusion-expr-common"
+version = "52.5.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "1b486b5f6255d40976b88bb83813b0d035a8333e0ec39864824e78068cf42fa6"
+dependencies = [
+ "arrow 57.3.0",
+ "datafusion-common 52.5.0",
  "indexmap 2.13.1",
  "itertools 0.14.0",
  "paste",
@@ -1643,22 +2773,53 @@ name = "datafusion-functions"
 version = "48.0.1"
 source = "git+https://github.com/FunctionStream/datafusion?branch=48.0.1%2Ffs#c3b2ac13dabd4145f21045e315529b917079b7d1"
 dependencies = [
- "arrow",
+ "arrow 55.2.0",
  "arrow-buffer 55.2.0",
  "base64",
  "blake2",
  "blake3",
  "chrono",
- "datafusion-common",
- "datafusion-doc",
- "datafusion-execution",
- "datafusion-expr",
- "datafusion-expr-common",
- "datafusion-macros",
+ "datafusion-common 48.0.1",
+ "datafusion-doc 48.0.1",
+ "datafusion-execution 48.0.1",
+ "datafusion-expr 48.0.1",
+ "datafusion-expr-common 48.0.1",
+ "datafusion-macros 48.0.1",
+ "hex",
+ "itertools 0.14.0",
+ "log",
+ "md-5",
+ "rand 0.9.2",
+ "regex",
+ "sha2",
+ "unicode-segmentation",
+ "uuid",
+]
+
+[[package]]
+name = "datafusion-functions"
+version = "52.5.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "07356c94118d881130dd0ffbff127540407d969c8978736e324edcd6c41cd48f"
+dependencies = [
+ "arrow 57.3.0",
+ "arrow-buffer 57.3.0",
+ "base64",
+ "blake2",
+ "blake3",
+ "chrono",
+ "chrono-tz",
+ "datafusion-common 52.5.0",
+ "datafusion-doc 52.5.0",
+ "datafusion-execution 52.5.0",
+ "datafusion-expr 52.5.0",
+ "datafusion-expr-common 52.5.0",
+ "datafusion-macros 52.5.0",
  "hex",
  "itertools 0.14.0",
  "log",
  "md-5",
+ "num-traits",
  "rand 0.9.2",
  "regex",
  "sha2",
@@ -1672,15 +2833,36 @@ version = "48.0.1"
 source = "git+https://github.com/FunctionStream/datafusion?branch=48.0.1%2Ffs#c3b2ac13dabd4145f21045e315529b917079b7d1"
 dependencies = [
  "ahash",
- "arrow",
- "datafusion-common",
- "datafusion-doc",
- "datafusion-execution",
- "datafusion-expr",
- "datafusion-functions-aggregate-common",
- "datafusion-macros",
- "datafusion-physical-expr",
- "datafusion-physical-expr-common",
+ "arrow 55.2.0",
+ "datafusion-common 48.0.1",
+ "datafusion-doc 48.0.1",
+ "datafusion-execution 48.0.1",
+ "datafusion-expr 48.0.1",
+ "datafusion-functions-aggregate-common 48.0.1",
+ "datafusion-macros 48.0.1",
+ "datafusion-physical-expr 48.0.1",
+ "datafusion-physical-expr-common 48.0.1",
+ "half",
+ "log",
+ "paste",
+]
+
+[[package]]
+name = "datafusion-functions-aggregate"
+version = "52.5.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "b644f9cf696df9233ce6958b9807666d78563b56f923267474dd6c07795f1f8f"
+dependencies = [
+ "ahash",
+ "arrow 57.3.0",
+ "datafusion-common 52.5.0",
+ "datafusion-doc 52.5.0",
+ "datafusion-execution 52.5.0",
+ "datafusion-expr 52.5.0",
+ "datafusion-functions-aggregate-common 52.5.0",
+ "datafusion-macros 52.5.0",
+ "datafusion-physical-expr 52.5.0",
+ "datafusion-physical-expr-common 52.5.0",
  "half",
  "log",
  "paste",
@@ -1692,10 +2874,23 @@ version = "48.0.1"
 source = "git+https://github.com/FunctionStream/datafusion?branch=48.0.1%2Ffs#c3b2ac13dabd4145f21045e315529b917079b7d1"
 dependencies = [
  "ahash",
- "arrow",
- "datafusion-common",
- "datafusion-expr-common",
- "datafusion-physical-expr-common",
+ "arrow 55.2.0",
+ "datafusion-common 48.0.1",
+ "datafusion-expr-common 48.0.1",
+ "datafusion-physical-expr-common 48.0.1",
+]
+
+[[package]]
+name = "datafusion-functions-aggregate-common"
+version = "52.5.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "c1de2deaaabe8923ce9ea9f29c47bbb4ee14f67ea2fe1ab5398d9bbebcf86e56"
+dependencies = [
+ "ahash",
+ "arrow 57.3.0",
+ "datafusion-common 52.5.0",
+ "datafusion-expr-common 52.5.0",
+ "datafusion-physical-expr-common 52.5.0",
 ]
 
 [[package]]
@@ -1703,16 +2898,39 @@ name = "datafusion-functions-nested"
 version = "48.0.1"
 source = "git+https://github.com/FunctionStream/datafusion?branch=48.0.1%2Ffs#c3b2ac13dabd4145f21045e315529b917079b7d1"
 dependencies = [
- "arrow",
- "arrow-ord",
- "datafusion-common",
- "datafusion-doc",
- "datafusion-execution",
- "datafusion-expr",
- "datafusion-functions",
- "datafusion-functions-aggregate",
- "datafusion-macros",
- "datafusion-physical-expr-common",
+ "arrow 55.2.0",
+ "arrow-ord 55.2.0",
+ "datafusion-common 48.0.1",
+ "datafusion-doc 48.0.1",
+ "datafusion-execution 48.0.1",
+ "datafusion-expr 48.0.1",
+ "datafusion-functions 48.0.1",
+ "datafusion-functions-aggregate 48.0.1",
+ "datafusion-macros 48.0.1",
+ "datafusion-physical-expr-common 48.0.1",
+ "itertools 0.14.0",
+ "log",
+ "paste",
+]
+
+[[package]]
+name = "datafusion-functions-nested"
+version = "52.5.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "552f8d92e4331ee91d23c02d12bb6acf32cbfd5215117e01c0fb63cd4b15af1a"
+dependencies = [
+ "arrow 57.3.0",
+ "arrow-ord 57.3.0",
+ "datafusion-common 52.5.0",
+ "datafusion-doc 52.5.0",
+ "datafusion-execution 52.5.0",
+ "datafusion-expr 52.5.0",
+ "datafusion-expr-common 52.5.0",
+ "datafusion-functions 52.5.0",
+ "datafusion-functions-aggregate 52.5.0",
+ "datafusion-functions-aggregate-common 52.5.0",
+ "datafusion-macros 52.5.0",
+ "datafusion-physical-expr-common 52.5.0",
  "itertools 0.14.0",
  "log",
  "paste",
@@ -1723,12 +2941,28 @@ name = "datafusion-functions-table"
 version = "48.0.1"
 source = "git+https://github.com/FunctionStream/datafusion?branch=48.0.1%2Ffs#c3b2ac13dabd4145f21045e315529b917079b7d1"
 dependencies = [
- "arrow",
+ "arrow 55.2.0",
+ "async-trait",
+ "datafusion-catalog 48.0.1",
+ "datafusion-common 48.0.1",
+ "datafusion-expr 48.0.1",
+ "datafusion-physical-plan 48.0.1",
+ "parking_lot",
+ "paste",
+]
+
+[[package]]
+name = "datafusion-functions-table"
+version = "52.5.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "970fd0cdd3df8802b9a9975ff600998289ba9d46682a4f7285cba4820c9ada78"
+dependencies = [
+ "arrow 57.3.0",
  "async-trait",
- "datafusion-catalog",
- "datafusion-common",
- "datafusion-expr",
- "datafusion-physical-plan",
+ "datafusion-catalog 52.5.0",
+ "datafusion-common 52.5.0",
+ "datafusion-expr 52.5.0",
+ "datafusion-physical-plan 52.5.0",
  "parking_lot",
  "paste",
 ]
@@ -1738,14 +2972,32 @@ name = "datafusion-functions-window"
 version = "48.0.1"
 source = "git+https://github.com/FunctionStream/datafusion?branch=48.0.1%2Ffs#c3b2ac13dabd4145f21045e315529b917079b7d1"
 dependencies = [
- "arrow",
- "datafusion-common",
- "datafusion-doc",
- "datafusion-expr",
- "datafusion-functions-window-common",
- "datafusion-macros",
- "datafusion-physical-expr",
- "datafusion-physical-expr-common",
+ "arrow 55.2.0",
+ "datafusion-common 48.0.1",
+ "datafusion-doc 48.0.1",
+ "datafusion-expr 48.0.1",
+ "datafusion-functions-window-common 48.0.1",
+ "datafusion-macros 48.0.1",
+ "datafusion-physical-expr 48.0.1",
+ "datafusion-physical-expr-common 48.0.1",
+ "log",
+ "paste",
+]
+
+[[package]]
+name = "datafusion-functions-window"
+version = "52.5.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "40b4c21a7c8a986a1866c0a87ab756d0bbf7b5f41f306009fa2d9af79c52ed31"
+dependencies = [
+ "arrow 57.3.0",
+ "datafusion-common 52.5.0",
+ "datafusion-doc 52.5.0",
+ "datafusion-expr 52.5.0",
+ "datafusion-functions-window-common 52.5.0",
+ "datafusion-macros 52.5.0",
+ "datafusion-physical-expr 52.5.0",
+ "datafusion-physical-expr-common 52.5.0",
  "log",
  "paste",
 ]
@@ -1755,8 +3007,18 @@ name = "datafusion-functions-window-common"
 version = "48.0.1"
 source = "git+https://github.com/FunctionStream/datafusion?branch=48.0.1%2Ffs#c3b2ac13dabd4145f21045e315529b917079b7d1"
 dependencies = [
- "datafusion-common",
- "datafusion-physical-expr-common",
+ "datafusion-common 48.0.1",
+ "datafusion-physical-expr-common 48.0.1",
+]
+
+[[package]]
+name = "datafusion-functions-window-common"
+version = "52.5.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "b1210ad73b8b3211aeaf4a42bef9bd7a2b7fce3ec119a478831f18c6ff7f7b93"
+dependencies = [
+ "datafusion-common 52.5.0",
+ "datafusion-physical-expr-common 52.5.0",
 ]
 
 [[package]]
@@ -1764,9 +3026,20 @@ name = "datafusion-macros"
 version = "48.0.1"
 source = "git+https://github.com/FunctionStream/datafusion?branch=48.0.1%2Ffs#c3b2ac13dabd4145f21045e315529b917079b7d1"
 dependencies = [
- "datafusion-expr",
+ "datafusion-expr 48.0.1",
+ "quote",
+ "syn 2.0.117",
+]
+
+[[package]]
+name = "datafusion-macros"
+version = "52.5.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "aaa566a963013a38681ad82a727a654bc7feb19632426aea8c3412d415d200c5"
+dependencies = [
+ "datafusion-doc 52.5.0",
  "quote",
- "syn",
+ "syn 2.0.117",
 ]
 
 [[package]]
@@ -1774,11 +3047,11 @@ name = "datafusion-optimizer"
 version = "48.0.1"
 source = "git+https://github.com/FunctionStream/datafusion?branch=48.0.1%2Ffs#c3b2ac13dabd4145f21045e315529b917079b7d1"
 dependencies = [
- "arrow",
+ "arrow 55.2.0",
  "chrono",
- "datafusion-common",
- "datafusion-expr",
- "datafusion-physical-expr",
+ "datafusion-common 48.0.1",
+ "datafusion-expr 48.0.1",
+ "datafusion-physical-expr 48.0.1",
  "indexmap 2.13.1",
  "itertools 0.14.0",
  "log",
@@ -1787,18 +3060,37 @@ dependencies = [
  "regex-syntax",
 ]
 
+[[package]]
+name = "datafusion-optimizer"
+version = "52.5.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "ff9aa82b240252a88dee118372f9b9757c545ab9e53c0736bebab2e7da0ef1f2"
+dependencies = [
+ "arrow 57.3.0",
+ "chrono",
+ "datafusion-common 52.5.0",
+ "datafusion-expr 52.5.0",
+ "datafusion-expr-common 52.5.0",
+ "datafusion-physical-expr 52.5.0",
+ "indexmap 2.13.1",
+ "itertools 0.14.0",
+ "log",
+ "regex",
+ "regex-syntax",
+]
+
 [[package]]
 name = "datafusion-physical-expr"
 version = "48.0.1"
 source = "git+https://github.com/FunctionStream/datafusion?branch=48.0.1%2Ffs#c3b2ac13dabd4145f21045e315529b917079b7d1"
 dependencies = [
  "ahash",
- "arrow",
- "datafusion-common",
- "datafusion-expr",
- "datafusion-expr-common",
- "datafusion-functions-aggregate-common",
- "datafusion-physical-expr-common",
+ "arrow 55.2.0",
+ "datafusion-common 48.0.1",
+ "datafusion-expr 48.0.1",
+ "datafusion-expr-common 48.0.1",
+ "datafusion-functions-aggregate-common 48.0.1",
+ "datafusion-physical-expr-common 48.0.1",
  "half",
  "hashbrown 0.14.5",
  "indexmap 2.13.1",
@@ -1808,55 +3100,128 @@ dependencies = [
  "petgraph 0.8.3",
 ]
 
+[[package]]
+name = "datafusion-physical-expr"
+version = "52.5.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "7d48022b8af9988c1d852644f9e8b5584c490659769a550c5e8d39457a1da0a5"
+dependencies = [
+ "ahash",
+ "arrow 57.3.0",
+ "datafusion-common 52.5.0",
+ "datafusion-expr 52.5.0",
+ "datafusion-expr-common 52.5.0",
+ "datafusion-functions-aggregate-common 52.5.0",
+ "datafusion-physical-expr-common 52.5.0",
+ "half",
+ "hashbrown 0.16.1",
+ "indexmap 2.13.1",
+ "itertools 0.14.0",
+ "parking_lot",
+ "paste",
+ "petgraph 0.8.3",
+ "tokio",
+]
+
+[[package]]
+name = "datafusion-physical-expr-adapter"
+version = "52.5.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "ae7a8abc0b4fe624000972a9b145b30b7f1b680bffaa950ea53f78d9b21c27c3"
+dependencies = [
+ "arrow 57.3.0",
+ "datafusion-common 52.5.0",
+ "datafusion-expr 52.5.0",
+ "datafusion-functions 52.5.0",
+ "datafusion-physical-expr 52.5.0",
+ "datafusion-physical-expr-common 52.5.0",
+ "itertools 0.14.0",
+]
+
 [[package]]
 name = "datafusion-physical-expr-common"
 version = "48.0.1"
 source = "git+https://github.com/FunctionStream/datafusion?branch=48.0.1%2Ffs#c3b2ac13dabd4145f21045e315529b917079b7d1"
 dependencies = [
  "ahash",
- "arrow",
- "datafusion-common",
- "datafusion-expr-common",
+ "arrow 55.2.0",
+ "datafusion-common 48.0.1",
+ "datafusion-expr-common 48.0.1",
  "hashbrown 0.14.5",
  "itertools 0.14.0",
 ]
 
+[[package]]
+name = "datafusion-physical-expr-common"
+version = "52.5.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "147253ca3e6b9d59c162de64c02800973018660e13340dd1886dd038d17ac429"
+dependencies = [
+ "ahash",
+ "arrow 57.3.0",
+ "chrono",
+ "datafusion-common 52.5.0",
+ "datafusion-expr-common 52.5.0",
+ "hashbrown 0.16.1",
+ "indexmap 2.13.1",
+ "itertools 0.14.0",
+ "parking_lot",
+]
+
 [[package]]
 name = "datafusion-physical-optimizer"
 version = "48.0.1"
 source = "git+https://github.com/FunctionStream/datafusion?branch=48.0.1%2Ffs#c3b2ac13dabd4145f21045e315529b917079b7d1"
 dependencies = [
- "arrow",
- "datafusion-common",
- "datafusion-execution",
- "datafusion-expr",
- "datafusion-expr-common",
- "datafusion-physical-expr",
- "datafusion-physical-expr-common",
- "datafusion-physical-plan",
+ "arrow 55.2.0",
+ "datafusion-common 48.0.1",
+ "datafusion-execution 48.0.1",
+ "datafusion-expr 48.0.1",
+ "datafusion-expr-common 48.0.1",
+ "datafusion-physical-expr 48.0.1",
+ "datafusion-physical-expr-common 48.0.1",
+ "datafusion-physical-plan 48.0.1",
  "itertools 0.14.0",
  "log",
  "recursive",
 ]
 
+[[package]]
+name = "datafusion-physical-optimizer"
+version = "52.5.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "689156bb2282107b6239db8d7ef44b4dab10a9b33d3491a0c74acac5e4fedd72"
+dependencies = [
+ "arrow 57.3.0",
+ "datafusion-common 52.5.0",
+ "datafusion-execution 52.5.0",
+ "datafusion-expr 52.5.0",
+ "datafusion-expr-common 52.5.0",
+ "datafusion-physical-expr 52.5.0",
+ "datafusion-physical-expr-common 52.5.0",
+ "datafusion-physical-plan 52.5.0",
+ "datafusion-pruning",
+ "itertools 0.14.0",
+]
+
 [[package]]
 name = "datafusion-physical-plan"
 version = "48.0.1"
 source = "git+https://github.com/FunctionStream/datafusion?branch=48.0.1%2Ffs#c3b2ac13dabd4145f21045e315529b917079b7d1"
 dependencies = [
  "ahash",
- "arrow",
- "arrow-ord",
+ "arrow 55.2.0",
+ "arrow-ord 55.2.0",
  "arrow-schema 55.2.0",
  "async-trait",
  "chrono",
- "datafusion-common",
- "datafusion-common-runtime",
- "datafusion-execution",
- "datafusion-expr",
- "datafusion-functions-window-common",
- "datafusion-physical-expr",
- "datafusion-physical-expr-common",
+ "datafusion-common 48.0.1",
+ "datafusion-common-runtime 48.0.1",
+ "datafusion-execution 48.0.1",
+ "datafusion-expr 48.0.1",
+ "datafusion-functions-window-common 48.0.1",
+ "datafusion-physical-expr 48.0.1",
+ "datafusion-physical-expr-common 48.0.1",
  "futures",
  "half",
  "hashbrown 0.14.5",
@@ -1868,19 +3233,50 @@ dependencies = [
  "tokio",
 ]
 
+[[package]]
+name = "datafusion-physical-plan"
+version = "52.5.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "68253dc0ee5330aa558b2549c9b0da5af9fc17d753ae73022939014ad616fc28"
+dependencies = [
+ "ahash",
+ "arrow 57.3.0",
+ "arrow-ord 57.3.0",
+ "arrow-schema 57.3.0",
+ "async-trait",
+ "datafusion-common 52.5.0",
+ "datafusion-common-runtime 52.5.0",
+ "datafusion-execution 52.5.0",
+ "datafusion-expr 52.5.0",
+ "datafusion-functions 52.5.0",
+ "datafusion-functions-aggregate-common 52.5.0",
+ "datafusion-functions-window-common 52.5.0",
+ "datafusion-physical-expr 52.5.0",
+ "datafusion-physical-expr-common 52.5.0",
+ "futures",
+ "half",
+ "hashbrown 0.16.1",
+ "indexmap 2.13.1",
+ "itertools 0.14.0",
+ "log",
+ "parking_lot",
+ "pin-project-lite",
+ "tokio",
+]
+
 [[package]]
 name = "datafusion-proto"
 version = "48.0.1"
 source = "git+https://github.com/FunctionStream/datafusion?branch=48.0.1%2Ffs#c3b2ac13dabd4145f21045e315529b917079b7d1"
 dependencies = [
- "arrow",
+ "arrow 55.2.0",
  "chrono",
- "datafusion",
- "datafusion-common",
- "datafusion-expr",
+ "datafusion 48.0.1",
+ "datafusion-common 48.0.1",
+ "datafusion-expr 48.0.1",
  "datafusion-proto-common",
  "object_store",
- "prost",
+ "prost 0.13.5",
 ]
 
 [[package]]
@@ -1888,9 +3284,26 @@ name = "datafusion-proto-common"
 version = "48.0.1"
 source = "git+https://github.com/FunctionStream/datafusion?branch=48.0.1%2Ffs#c3b2ac13dabd4145f21045e315529b917079b7d1"
 dependencies = [
- "arrow",
- "datafusion-common",
- "prost",
+ "arrow 55.2.0",
+ "datafusion-common 48.0.1",
+ "prost 0.13.5",
+]
+
+[[package]]
+name = "datafusion-pruning"
+version = "52.5.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "0fcad240a54d0b1d3e8f668398900260a53122d522b2102ab57218590decacd6"
+dependencies = [
+ "arrow 57.3.0",
+ "datafusion-common 52.5.0",
+ "datafusion-datasource 52.5.0",
+ "datafusion-expr-common 52.5.0",
+ "datafusion-physical-expr 52.5.0",
+ "datafusion-physical-expr-common 52.5.0",
+ "datafusion-physical-plan 52.5.0",
+ "itertools 0.14.0",
+ "log",
 ]
 
 [[package]]
@@ -1898,16 +3311,16 @@ name = "datafusion-session"
 version = "48.0.1"
 source = "git+https://github.com/FunctionStream/datafusion?branch=48.0.1%2Ffs#c3b2ac13dabd4145f21045e315529b917079b7d1"
 dependencies = [
- "arrow",
+ "arrow 55.2.0",
  "async-trait",
  "dashmap",
- "datafusion-common",
- "datafusion-common-runtime",
- "datafusion-execution",
- "datafusion-expr",
- "datafusion-physical-expr",
- "datafusion-physical-plan",
- "datafusion-sql",
+ "datafusion-common 48.0.1",
+ "datafusion-common-runtime 48.0.1",
+ "datafusion-execution 48.0.1",
+ "datafusion-expr 48.0.1",
+ "datafusion-physical-expr 48.0.1",
+ "datafusion-physical-plan 48.0.1",
+ "datafusion-sql 48.0.1",
  "futures",
  "itertools 0.14.0",
  "log",
@@ -1916,20 +3329,51 @@ dependencies = [
  "tokio",
 ]
 
+[[package]]
+name = "datafusion-session"
+version = "52.5.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "f58e83a68bb67007a8fcbf005c44cefe441270c7ee7f6dee10c0e0109b556f6d"
+dependencies = [
+ "async-trait",
+ "datafusion-common 52.5.0",
+ "datafusion-execution 52.5.0",
+ "datafusion-expr 52.5.0",
+ "datafusion-physical-plan 52.5.0",
+ "parking_lot",
+]
+
 [[package]]
 name = "datafusion-sql"
 version = "48.0.1"
 source = "git+https://github.com/FunctionStream/datafusion?branch=48.0.1%2Ffs#c3b2ac13dabd4145f21045e315529b917079b7d1"
 dependencies = [
- "arrow",
+ "arrow 55.2.0",
  "bigdecimal",
- "datafusion-common",
- "datafusion-expr",
+ "datafusion-common 48.0.1",
+ "datafusion-expr 48.0.1",
  "indexmap 2.13.1",
  "log",
  "recursive",
  "regex",
- "sqlparser",
+ "sqlparser 0.55.0",
+]
+
+[[package]]
+name = "datafusion-sql"
+version = "52.5.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "be53e9eb55db0fbb8980bb6d87f2435b0524acf4c718ed54a57cabbb299b2ab3"
+dependencies = [
+ "arrow 57.3.0",
+ "bigdecimal",
+ "chrono",
+ "datafusion-common 52.5.0",
+ "datafusion-expr 52.5.0",
+ "indexmap 2.13.1",
+ "log",
+ "regex",
+ "sqlparser 0.59.0",
 ]
 
 [[package]]
@@ -1941,6 +3385,26 @@ dependencies = [
  "uuid",
 ]
 
+[[package]]
+name = "deepsize"
+version = "0.2.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "1cdb987ec36f6bf7bfbea3f928b75590b736fc42af8e54d97592481351b2b96c"
+dependencies = [
+ "deepsize_derive",
+]
+
+[[package]]
+name = "deepsize_derive"
+version = "0.1.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "990101d41f3bc8c1a45641024377ee284ecc338e5ecf3ea0f0e236d897c72796"
+dependencies = [
+ "proc-macro2",
+ "quote",
+ "syn 1.0.109",
+]
+
 [[package]]
 name = "deranged"
 version = "0.5.8"
@@ -1948,6 +3412,7 @@ source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "7cd812cc2bc1d69d4764bd80df88b4317eaef9e773c75226407d9bc0876b211c"
 dependencies = [
  "powerfmt",
+ "serde_core",
 ]
 
 [[package]]
@@ -1957,6 +3422,7 @@ source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "9ed9a281f7bc9b7576e61468ba615a66a5c8cfdff42420a70aa82701a3b1e292"
 dependencies = [
  "block-buffer",
+ "const-oid",
  "crypto-common",
  "subtle",
 ]
@@ -1971,6 +3437,27 @@ dependencies = [
  "dirs-sys-next",
 ]
 
+[[package]]
+name = "dirs"
+version = "6.0.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "c3e8aa94d75141228480295a7d0e7feb620b1a5ad9f12bc40be62411e38cce4e"
+dependencies = [
+ "dirs-sys",
+]
+
+[[package]]
+name = "dirs-sys"
+version = "0.5.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "e01a3366d27ee9890022452ee61b2b63a67e6f13f58900b651ff5665f0bb1fab"
+dependencies = [
+ "libc",
+ "option-ext",
+ "redox_users 0.5.2",
+ "windows-sys 0.61.2",
+]
+
 [[package]]
 name = "dirs-sys-next"
 version = "0.1.2"
@@ -1978,7 +3465,7 @@ source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "4ebda144c4fe02d1f7ea1a7d9641b6fc6b580adcfa024ae48797ecdeb6825b4d"
 dependencies = [
  "libc",
- "redox_users",
+ "redox_users 0.4.6",
  "winapi",
 ]
 
@@ -1990,7 +3477,16 @@ checksum = "97369cbbc041bc366949bc74d34658d6cda5621039731c6310521892a3a20ae0"
 dependencies = [
  "proc-macro2",
  "quote",
- "syn",
+ "syn 2.0.117",
+]
+
+[[package]]
+name = "dlv-list"
+version = "0.5.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "442039f5147480ba31067cb00ada1adae6892028e40e45fc5de7b7df6dcc1b5f"
+dependencies = [
+ "const-random",
 ]
 
 [[package]]
@@ -2002,6 +3498,12 @@ dependencies = [
  "litrs",
 ]
 
+[[package]]
+name = "downcast-rs"
+version = "2.0.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "117240f60069e65410b3ae1bb213295bd828f707b5bec6596a1afc8793ce0cbc"
+
 [[package]]
 name = "duct"
 version = "0.13.7"
@@ -2014,6 +3516,12 @@ dependencies = [
  "shared_child",
 ]
 
+[[package]]
+name = "dunce"
+version = "1.0.5"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "92773504d58c093f6de2459af4af33faa518c13451eb8f2b5698ed3d36e7c813"
+
 [[package]]
 name = "either"
 version = "1.15.0"
@@ -2082,12 +3590,51 @@ version = "3.3.2"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "dea2df4cf52843e0452895c455a1a2cfbb842a1e7329671acf418fdc53ed4c59"
 
+[[package]]
+name = "ethnum"
+version = "1.5.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "ca81e6b4777c89fd810c25a4be2b1bd93ea034fbe58e6a75216a34c6b82c539b"
+
+[[package]]
+name = "event-listener"
+version = "5.4.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "e13b66accf52311f30a0db42147dadea9850cb48cd070028831ae5f5d4b856ab"
+dependencies = [
+ "concurrent-queue",
+ "parking",
+ "pin-project-lite",
+]
+
+[[package]]
+name = "event-listener-strategy"
+version = "0.5.4"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "8be9f3dfaaffdae2972880079a491a1a8bb7cbed0b8dd7a347f668b4150a3b93"
+dependencies = [
+ "event-listener",
+ "pin-project-lite",
+]
+
 [[package]]
 name = "fallible-iterator"
 version = "0.3.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "2acce4a10f12dc2fb14a218589d4f1f62ef011b2d0cc4b3cb1bba8e94da14649"
 
+[[package]]
+name = "fast-float2"
+version = "0.2.3"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "f8eb564c5c7423d25c886fb561d1e4ee69f72354d16918afa32c08811f6b6a55"
+
+[[package]]
+name = "fastdivide"
+version = "0.4.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "9afc2bd4d5a73106dd53d10d73d3401c2f32730ba2c0b93ddb888a8983680471"
+
 [[package]]
 name = "fastrand"
 version = "2.4.0"
@@ -2166,6 +3713,12 @@ version = "0.1.5"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "d9c4f5dac5e15c24eb999c26181a6ca40b39fe946cbe4c263c7209467bc83af2"
 
+[[package]]
+name = "foldhash"
+version = "0.2.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "77ce24cb58228fbb8aa041425bb1050850ac19177686ea6e0f41a70416f56fdb"
+
 [[package]]
 name = "form_urlencoded"
 version = "1.2.2"
@@ -2186,39 +3739,80 @@ dependencies = [
  "windows-sys 0.59.0",
 ]
 
+[[package]]
+name = "fs4"
+version = "0.8.4"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "f7e180ac76c23b45e767bd7ae9579bc0bb458618c4bc71835926e098e61d15f8"
+dependencies = [
+ "rustix 0.38.44",
+ "windows-sys 0.52.0",
+]
+
+[[package]]
+name = "fs_extra"
+version = "1.3.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "42703706b716c37f96a77aea830392ad231f44c9e9a67872fa5548707e11b11c"
+
+[[package]]
+name = "fsst"
+version = "4.0.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "2195cc7f87e84bd695586137de99605e7e9579b26ec5e01b82960ddb4d0922f2"
+dependencies = [
+ "arrow-array 57.3.0",
+ "rand 0.9.2",
+]
+
+[[package]]
+name = "fst"
+version = "0.4.7"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "7ab85b9b05e3978cc9a9cf8fea7f01b494e1a09ed3037e16ba39edc7a29eb61a"
+dependencies = [
+ "utf8-ranges",
+]
+
 [[package]]
 name = "function-stream"
 version = "0.6.0"
 dependencies = [
  "ahash",
  "anyhow",
- "arrow",
+ "apache-avro",
+ "arrow 55.2.0",
  "arrow-array 55.2.0",
+ "arrow-array 57.3.0",
  "arrow-ipc 55.2.0",
- "arrow-json",
+ "arrow-ipc 57.3.0",
+ "arrow-json 55.2.0",
  "arrow-schema 55.2.0",
  "async-trait",
  "base64",
  "bincode",
+ "bytes",
  "chrono",
  "crossbeam-channel",
- "datafusion",
- "datafusion-common",
- "datafusion-execution",
- "datafusion-expr",
- "datafusion-physical-expr",
+ "datafusion 48.0.1",
+ "datafusion-common 48.0.1",
+ "datafusion-execution 48.0.1",
+ "datafusion-expr 48.0.1",
+ "datafusion-physical-expr 48.0.1",
  "datafusion-proto",
  "futures",
  "governor",
  "itertools 0.14.0",
+ "lance",
  "log",
  "lru",
  "num_cpus",
+ "object_store",
  "parking_lot",
  "parquet",
  "petgraph 0.7.1",
  "proctitle",
- "prost",
+ "prost 0.13.5",
  "protocol",
  "rand 0.8.5",
  "rdkafka",
@@ -2227,8 +3821,8 @@ dependencies = [
  "serde_json",
  "serde_json_path",
  "serde_yaml",
- "sqlparser",
- "strum",
+ "sqlparser 0.55.0",
+ "strum 0.26.3",
  "tempfile",
  "thiserror 2.0.18",
  "tokio",
@@ -2259,6 +3853,12 @@ dependencies = [
  "tonic",
 ]
 
+[[package]]
+name = "funty"
+version = "2.0.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "e6d5a32815ae3f33302d95fdcb2ce17862f8c65363dcfd29360480ba1001fc9c"
+
 [[package]]
 name = "futures"
 version = "0.3.32"
@@ -2315,7 +3915,7 @@ checksum = "e835b70203e41293343137df5c0664546da5745f82ec9b84d40be8336958447b"
 dependencies = [
  "proc-macro2",
  "quote",
- "syn",
+ "syn 2.0.117",
 ]
 
 [[package]]
@@ -2367,6 +3967,21 @@ dependencies = [
  "serde_json",
 ]
 
+[[package]]
+name = "generator"
+version = "0.8.8"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "52f04ae4152da20c76fe800fa48659201d5cf627c5149ca0b707b69d7eef6cf9"
+dependencies = [
+ "cc",
+ "cfg-if",
+ "libc",
+ "log",
+ "rustversion",
+ "windows-link",
+ "windows-result",
+]
+
 [[package]]
 name = "generic-array"
 version = "0.14.7"
@@ -2384,8 +3999,10 @@ source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "ff2abc00be7fca6ebc474524697ae276ad847ad0a6b3faa4bcb027e9a4614ad0"
 dependencies = [
  "cfg-if",
+ "js-sys",
  "libc",
  "wasi",
+ "wasm-bindgen",
 ]
 
 [[package]]
@@ -2432,6 +4049,18 @@ version = "0.3.3"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "0cc23270f6e1808e30a928bdc84dea0b9b4136a8bc82338574f23baf47bbd280"
 
+[[package]]
+name = "gloo-timers"
+version = "0.3.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "bbb143cf96099802033e0d4f4963b19fd2e0b728bcf076cd9cf7f6634f092994"
+dependencies = [
+ "futures-channel",
+ "futures-core",
+ "js-sys",
+ "wasm-bindgen",
+]
+
 [[package]]
 name = "governor"
 version = "0.8.1"
@@ -2466,7 +4095,7 @@ dependencies = [
  "fnv",
  "futures-core",
  "futures-sink",
- "http",
+ "http 1.4.0",
  "indexmap 2.13.1",
  "slab",
  "tokio",
@@ -2510,7 +4139,7 @@ checksum = "9229cfe53dfd69f0609a49f65461bd93001ea1ef889cd5529dd176593f5338a1"
 dependencies = [
  "allocator-api2",
  "equivalent",
- "foldhash",
+ "foldhash 0.1.5",
  "serde",
 ]
 
@@ -2519,6 +4148,11 @@ name = "hashbrown"
 version = "0.16.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "841d1cc9bed7f9236f321df977030373f4a4163ae1a7dbfe1a51a2c1a51d9100"
+dependencies = [
+ "allocator-api2",
+ "equivalent",
+ "foldhash 0.2.0",
+]
 
 [[package]]
 name = "heck"
@@ -2538,6 +4172,15 @@ version = "0.4.3"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "7f24254aa9a54b5c858eaee2f5bccdb46aaf0e486a595ed5fd8f86ba55232a70"
 
+[[package]]
+name = "hmac"
+version = "0.12.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "6c49c37c09c17a53d937dfbb742eb3a961d65a994e6bcdcf37e7399d0cc8ab5e"
+dependencies = [
+ "digest",
+]
+
 [[package]]
 name = "home"
 version = "0.5.12"
@@ -2547,6 +4190,23 @@ dependencies = [
  "windows-sys 0.61.2",
 ]
 
+[[package]]
+name = "htmlescape"
+version = "0.3.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "e9025058dae765dee5070ec375f591e2ba14638c63feff74f13805a72e523163"
+
+[[package]]
+name = "http"
+version = "0.2.12"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "601cbb57e577e2f5ef5be8e7b83f0f63994f25aa94d673e54a92d5c516d101f1"
+dependencies = [
+ "bytes",
+ "fnv",
+ "itoa",
+]
+
 [[package]]
 name = "http"
 version = "1.4.0"
@@ -2557,6 +4217,17 @@ dependencies = [
  "itoa",
 ]
 
+[[package]]
+name = "http-body"
+version = "0.4.6"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "7ceab25649e9960c0311ea418d17bee82c0dcec1bd053b5f9a66e265a693bed2"
+dependencies = [
+ "bytes",
+ "http 0.2.12",
+ "pin-project-lite",
+]
+
 [[package]]
 name = "http-body"
 version = "1.0.1"
@@ -2564,7 +4235,7 @@ source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "1efedce1fb8e6913f23e0c92de8e62cd5b772a67e7b3946df930a62566c93184"
 dependencies = [
  "bytes",
- "http",
+ "http 1.4.0",
 ]
 
 [[package]]
@@ -2575,8 +4246,8 @@ checksum = "b021d93e26becf5dc7e1b75b1bed1fd93124b374ceb73f43d4d4eafec896a64a"
 dependencies = [
  "bytes",
  "futures-core",
- "http",
- "http-body",
+ "http 1.4.0",
+ "http-body 1.0.1",
  "pin-project-lite",
 ]
 
@@ -2609,15 +4280,33 @@ dependencies = [
  "futures-channel",
  "futures-core",
  "h2",
- "http",
- "http-body",
+ "http 1.4.0",
+ "http-body 1.0.1",
  "httparse",
  "httpdate",
  "itoa",
  "pin-project-lite",
  "smallvec",
  "tokio",
- "want",
+ "want",
+]
+
+[[package]]
+name = "hyper-rustls"
+version = "0.27.7"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "e3c93eb611681b207e1fe55d5a71ecf91572ec8a6705cdb6857f7d8d5242cf58"
+dependencies = [
+ "http 1.4.0",
+ "hyper",
+ "hyper-util",
+ "rustls",
+ "rustls-native-certs",
+ "rustls-pki-types",
+ "tokio",
+ "tokio-rustls",
+ "tower-service",
+ "webpki-roots",
 ]
 
 [[package]]
@@ -2639,13 +4328,16 @@ version = "0.1.20"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "96547c2556ec9d12fb1578c4eaf448b04993e7fb79cbaad930a656880a6bdfa0"
 dependencies = [
+ "base64",
  "bytes",
  "futures-channel",
  "futures-util",
- "http",
- "http-body",
+ "http 1.4.0",
+ "http-body 1.0.1",
  "hyper",
+ "ipnet",
  "libc",
+ "percent-encoding",
  "pin-project-lite",
  "socket2 0.6.3",
  "tokio",
@@ -2653,6 +4345,15 @@ dependencies = [
  "tracing",
 ]
 
+[[package]]
+name = "hyperloglogplus"
+version = "0.4.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "621debdf94dcac33e50475fdd76d34d5ea9c0362a834b9db08c3024696c1fbe3"
+dependencies = [
+ "serde",
+]
+
 [[package]]
 name = "iana-time-zone"
 version = "0.1.65"
@@ -2765,6 +4466,12 @@ version = "2.3.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "3d3067d79b975e8844ca9eb072e16b31c3c1c36928edf9c6789548c524d0d954"
 
+[[package]]
+name = "ident_case"
+version = "1.0.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "b9e0384b61958566e926dc50660321d12159025e767c18e043daf26b70104c39"
+
 [[package]]
 name = "idna"
 version = "1.1.0"
@@ -2777,169 +4484,698 @@ dependencies = [
 ]
 
 [[package]]
-name = "idna_adapter"
-version = "1.2.1"
+name = "idna_adapter"
+version = "1.2.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "3acae9609540aa318d1bc588455225fb2085b9ed0c4f6bd0d9d5bcd86f1a0344"
+dependencies = [
+ "icu_normalizer",
+ "icu_properties",
+]
+
+[[package]]
+name = "im-rc"
+version = "15.1.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "af1955a75fa080c677d3972822ec4bad316169ab1cfc6c257a942c2265dbe5fe"
+dependencies = [
+ "bitmaps",
+ "rand_core 0.6.4",
+ "rand_xoshiro 0.6.0",
+ "sized-chunks",
+ "typenum",
+ "version_check",
+]
+
+[[package]]
+name = "indexmap"
+version = "1.9.3"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "bd070e393353796e801d209ad339e89596eb4c8d430d18ede6a1cced8fafbd99"
+dependencies = [
+ "autocfg",
+ "hashbrown 0.12.3",
+]
+
+[[package]]
+name = "indexmap"
+version = "2.13.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "45a8a2b9cb3e0b0c1803dbb0758ffac5de2f425b23c28f518faabd9d805342ff"
+dependencies = [
+ "equivalent",
+ "hashbrown 0.16.1",
+ "serde",
+ "serde_core",
+]
+
+[[package]]
+name = "integer-encoding"
+version = "3.0.4"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "8bb03732005da905c88227371639bf1ad885cc712789c011c31c5fb3ab3ccf02"
+
+[[package]]
+name = "inventory"
+version = "0.3.24"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "a4f0c30c76f2f4ccee3fe55a2435f691ca00c0e4bd87abe4f4a851b1d4dac39b"
+dependencies = [
+ "rustversion",
+]
+
+[[package]]
+name = "io-extras"
+version = "0.18.4"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "2285ddfe3054097ef4b2fe909ef8c3bcd1ea52a8f0d274416caebeef39f04a65"
+dependencies = [
+ "io-lifetimes",
+ "windows-sys 0.59.0",
+]
+
+[[package]]
+name = "io-lifetimes"
+version = "2.0.4"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "06432fb54d3be7964ecd3649233cddf80db2832f47fec34c01f65b3d9d774983"
+
+[[package]]
+name = "ipnet"
+version = "2.12.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "d98f6fed1fde3f8c21bc40a1abb88dd75e67924f9cffc3ef95607bad8017f8e2"
+
+[[package]]
+name = "iri-string"
+version = "0.7.12"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "25e659a4bb38e810ebc252e53b5814ff908a8c58c2a9ce2fae1bbec24cbf4e20"
+dependencies = [
+ "memchr",
+ "serde",
+]
+
+[[package]]
+name = "is-terminal"
+version = "0.4.17"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "3640c1c38b8e4e43584d8df18be5fc6b0aa314ce6ebf51b53313d4306cca8e46"
+dependencies = [
+ "hermit-abi",
+ "libc",
+ "windows-sys 0.61.2",
+]
+
+[[package]]
+name = "is_terminal_polyfill"
+version = "1.70.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "a6cb138bb79a146c1bd460005623e142ef0181e3d0219cb493e02f7d08a35695"
+
+[[package]]
+name = "itertools"
+version = "0.13.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "413ee7dfc52ee1a4949ceeb7dbc8a33f2d6c088194d9f922fb8318faf1f01186"
+dependencies = [
+ "either",
+]
+
+[[package]]
+name = "itertools"
+version = "0.14.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "2b192c782037fadd9cfa75548310488aabdbf3d2da73885b31bd0abd03351285"
+dependencies = [
+ "either",
+]
+
+[[package]]
+name = "itoa"
+version = "1.0.18"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "8f42a60cbdf9a97f5d2305f08a87dc4e09308d1276d28c869c684d7777685682"
+
+[[package]]
+name = "ittapi"
+version = "0.4.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "6b996fe614c41395cdaedf3cf408a9534851090959d90d54a535f675550b64b1"
+dependencies = [
+ "anyhow",
+ "ittapi-sys",
+ "log",
+]
+
+[[package]]
+name = "ittapi-sys"
+version = "0.4.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "52f5385394064fa2c886205dba02598013ce83d3e92d33dbdc0c52fe0e7bf4fc"
+dependencies = [
+ "cc",
+]
+
+[[package]]
+name = "jiff"
+version = "0.2.23"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "1a3546dc96b6d42c5f24902af9e2538e82e39ad350b0c766eb3fbf2d8f3d8359"
+dependencies = [
+ "jiff-static",
+ "jiff-tzdb-platform",
+ "log",
+ "portable-atomic",
+ "portable-atomic-util",
+ "serde_core",
+ "windows-sys 0.61.2",
+]
+
+[[package]]
+name = "jiff-static"
+version = "0.2.23"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "2a8c8b344124222efd714b73bb41f8b5120b27a7cc1c75593a6ff768d9d05aa4"
+dependencies = [
+ "proc-macro2",
+ "quote",
+ "syn 2.0.117",
+]
+
+[[package]]
+name = "jiff-tzdb"
+version = "0.1.6"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "c900ef84826f1338a557697dc8fc601df9ca9af4ac137c7fb61d4c6f2dfd3076"
+
+[[package]]
+name = "jiff-tzdb-platform"
+version = "0.1.3"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "875a5a69ac2bab1a891711cf5eccbec1ce0341ea805560dcd90b7a2e925132e8"
+dependencies = [
+ "jiff-tzdb",
+]
+
+[[package]]
+name = "jobserver"
+version = "0.1.34"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "9afb3de4395d6b3e67a780b6de64b51c978ecf11cb9a462c66be7d4ca9039d33"
+dependencies = [
+ "getrandom 0.3.4",
+ "libc",
+]
+
+[[package]]
+name = "js-sys"
+version = "0.3.94"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "2e04e2ef80ce82e13552136fabeef8a5ed1f985a96805761cbb9a2c34e7664d9"
+dependencies = [
+ "cfg-if",
+ "futures-util",
+ "once_cell",
+ "wasm-bindgen",
+]
+
+[[package]]
+name = "jsonb"
+version = "0.5.6"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "eb98fb29636087c40ad0d1274d9a30c0c1e83e03ae93f6e7e89247b37fcc6953"
+dependencies = [
+ "byteorder",
+ "ethnum",
+ "fast-float2",
+ "itoa",
+ "jiff",
+ "nom 8.0.0",
+ "num-traits",
+ "ordered-float 5.3.0",
+ "rand 0.9.2",
+ "serde",
+ "serde_json",
+ "zmij",
+]
+
+[[package]]
+name = "lance"
+version = "4.0.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "efe6c3ddd79cdfd2b7e1c23cafae52806906bc40fbd97de9e8cf2f8c7a75fc04"
+dependencies = [
+ "arrow 57.3.0",
+ "arrow-arith 57.3.0",
+ "arrow-array 57.3.0",
+ "arrow-buffer 57.3.0",
+ "arrow-ipc 57.3.0",
+ "arrow-ord 57.3.0",
+ "arrow-row 57.3.0",
+ "arrow-schema 57.3.0",
+ "arrow-select 57.3.0",
+ "async-recursion",
+ "async-trait",
+ "async_cell",
+ "aws-credential-types",
+ "byteorder",
+ "bytes",
+ "chrono",
+ "crossbeam-skiplist",
+ "dashmap",
+ "datafusion 52.5.0",
+ "datafusion-expr 52.5.0",
+ "datafusion-functions 52.5.0",
+ "datafusion-physical-expr 52.5.0",
+ "datafusion-physical-plan 52.5.0",
+ "deepsize",
+ "either",
+ "futures",
+ "half",
+ "humantime",
+ "itertools 0.13.0",
+ "lance-arrow",
+ "lance-core",
+ "lance-datafusion",
+ "lance-encoding",
+ "lance-file",
+ "lance-index",
+ "lance-io",
+ "lance-linalg",
+ "lance-namespace",
+ "lance-table",
+ "log",
+ "moka",
+ "object_store",
+ "permutation",
+ "pin-project",
+ "prost 0.14.3",
+ "prost-types 0.14.3",
+ "rand 0.9.2",
+ "roaring",
+ "semver",
+ "serde",
+ "serde_json",
+ "snafu",
+ "tantivy",
+ "tokio",
+ "tokio-stream",
+ "tokio-util",
+ "tracing",
+ "url",
+ "uuid",
+]
+
+[[package]]
+name = "lance-arrow"
+version = "4.0.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "3acae9609540aa318d1bc588455225fb2085b9ed0c4f6bd0d9d5bcd86f1a0344"
+checksum = "5d9f5d95bdda2a2b790f1fb8028b5b6dcf661abeb3133a8bca0f3d24b054af87"
 dependencies = [
- "icu_normalizer",
- "icu_properties",
+ "arrow-array 57.3.0",
+ "arrow-buffer 57.3.0",
+ "arrow-cast 57.3.0",
+ "arrow-data 57.3.0",
+ "arrow-ord 57.3.0",
+ "arrow-schema 57.3.0",
+ "arrow-select 57.3.0",
+ "bytes",
+ "futures",
+ "getrandom 0.2.17",
+ "half",
+ "jsonb",
+ "num-traits",
+ "rand 0.9.2",
 ]
 
 [[package]]
-name = "im-rc"
-version = "15.1.0"
+name = "lance-bitpacking"
+version = "4.0.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "af1955a75fa080c677d3972822ec4bad316169ab1cfc6c257a942c2265dbe5fe"
+checksum = "f827d6ab9f8f337a9509d5ad66a12f3314db8713868260521c344ef6135eb4e4"
 dependencies = [
- "bitmaps",
- "rand_core 0.6.4",
- "rand_xoshiro",
- "sized-chunks",
- "typenum",
- "version_check",
+ "arrayref",
+ "paste",
+ "seq-macro",
 ]
 
 [[package]]
-name = "indexmap"
-version = "1.9.3"
+name = "lance-core"
+version = "4.0.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "bd070e393353796e801d209ad339e89596eb4c8d430d18ede6a1cced8fafbd99"
+checksum = "0f1e25df6a79bf72ee6bcde0851f19b1cd36c5848c1b7db83340882d3c9fdecb"
 dependencies = [
- "autocfg",
- "hashbrown 0.12.3",
+ "arrow-array 57.3.0",
+ "arrow-buffer 57.3.0",
+ "arrow-schema 57.3.0",
+ "async-trait",
+ "byteorder",
+ "bytes",
+ "chrono",
+ "datafusion-common 52.5.0",
+ "datafusion-sql 52.5.0",
+ "deepsize",
+ "futures",
+ "itertools 0.13.0",
+ "lance-arrow",
+ "libc",
+ "log",
+ "mock_instant",
+ "moka",
+ "num_cpus",
+ "object_store",
+ "pin-project",
+ "prost 0.14.3",
+ "rand 0.9.2",
+ "roaring",
+ "serde_json",
+ "snafu",
+ "tempfile",
+ "tokio",
+ "tokio-stream",
+ "tokio-util",
+ "tracing",
+ "url",
 ]
 
 [[package]]
-name = "indexmap"
-version = "2.13.1"
+name = "lance-datafusion"
+version = "4.0.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "45a8a2b9cb3e0b0c1803dbb0758ffac5de2f425b23c28f518faabd9d805342ff"
+checksum = "93146de8ae720cb90edef81c2f2d0a1b065fc2f23ecff2419546f389b0fa70a4"
 dependencies = [
- "equivalent",
- "hashbrown 0.16.1",
- "serde",
- "serde_core",
+ "arrow 57.3.0",
+ "arrow-array 57.3.0",
+ "arrow-buffer 57.3.0",
+ "arrow-ord 57.3.0",
+ "arrow-schema 57.3.0",
+ "arrow-select 57.3.0",
+ "async-trait",
+ "chrono",
+ "datafusion 52.5.0",
+ "datafusion-common 52.5.0",
+ "datafusion-functions 52.5.0",
+ "datafusion-physical-expr 52.5.0",
+ "futures",
+ "jsonb",
+ "lance-arrow",
+ "lance-core",
+ "lance-datagen",
+ "log",
+ "pin-project",
+ "prost 0.14.3",
+ "prost-build 0.14.3",
+ "snafu",
+ "tokio",
+ "tracing",
 ]
 
 [[package]]
-name = "integer-encoding"
-version = "3.0.4"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "8bb03732005da905c88227371639bf1ad885cc712789c011c31c5fb3ab3ccf02"
-
-[[package]]
-name = "inventory"
-version = "0.3.24"
+name = "lance-datagen"
+version = "4.0.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "a4f0c30c76f2f4ccee3fe55a2435f691ca00c0e4bd87abe4f4a851b1d4dac39b"
+checksum = "ccec8ce4d8e0a87a99c431dab2364398029f2ffb649c1a693c60c79e05ed30dd"
 dependencies = [
- "rustversion",
+ "arrow 57.3.0",
+ "arrow-array 57.3.0",
+ "arrow-cast 57.3.0",
+ "arrow-schema 57.3.0",
+ "chrono",
+ "futures",
+ "half",
+ "hex",
+ "rand 0.9.2",
+ "rand_distr 0.5.1",
+ "rand_xoshiro 0.7.0",
+ "random_word",
 ]
 
 [[package]]
-name = "io-extras"
-version = "0.18.4"
+name = "lance-encoding"
+version = "4.0.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "2285ddfe3054097ef4b2fe909ef8c3bcd1ea52a8f0d274416caebeef39f04a65"
+checksum = "5c1aec0bbbac6bce829bc10f1ba066258126100596c375fb71908ecf11c2c2a5"
 dependencies = [
- "io-lifetimes",
- "windows-sys 0.59.0",
+ "arrow-arith 57.3.0",
+ "arrow-array 57.3.0",
+ "arrow-buffer 57.3.0",
+ "arrow-cast 57.3.0",
+ "arrow-data 57.3.0",
+ "arrow-schema 57.3.0",
+ "arrow-select 57.3.0",
+ "bytemuck",
+ "byteorder",
+ "bytes",
+ "fsst",
+ "futures",
+ "hex",
+ "hyperloglogplus",
+ "itertools 0.13.0",
+ "lance-arrow",
+ "lance-bitpacking",
+ "lance-core",
+ "log",
+ "lz4",
+ "num-traits",
+ "prost 0.14.3",
+ "prost-build 0.14.3",
+ "prost-types 0.14.3",
+ "rand 0.9.2",
+ "snafu",
+ "strum 0.26.3",
+ "tokio",
+ "tracing",
+ "xxhash-rust",
+ "zstd",
 ]
 
 [[package]]
-name = "io-lifetimes"
-version = "2.0.4"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "06432fb54d3be7964ecd3649233cddf80db2832f47fec34c01f65b3d9d774983"
-
-[[package]]
-name = "ipnet"
-version = "2.12.0"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "d98f6fed1fde3f8c21bc40a1abb88dd75e67924f9cffc3ef95607bad8017f8e2"
-
-[[package]]
-name = "is-terminal"
-version = "0.4.17"
+name = "lance-file"
+version = "4.0.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "3640c1c38b8e4e43584d8df18be5fc6b0aa314ce6ebf51b53313d4306cca8e46"
+checksum = "14a8c548804f5b17486dc2d3282356ed1957095a852780283bc401fdd69e9075"
 dependencies = [
- "hermit-abi",
- "libc",
- "windows-sys 0.61.2",
+ "arrow-arith 57.3.0",
+ "arrow-array 57.3.0",
+ "arrow-buffer 57.3.0",
+ "arrow-data 57.3.0",
+ "arrow-schema 57.3.0",
+ "arrow-select 57.3.0",
+ "async-recursion",
+ "async-trait",
+ "byteorder",
+ "bytes",
+ "datafusion-common 52.5.0",
+ "deepsize",
+ "futures",
+ "lance-arrow",
+ "lance-core",
+ "lance-encoding",
+ "lance-io",
+ "log",
+ "num-traits",
+ "object_store",
+ "prost 0.14.3",
+ "prost-build 0.14.3",
+ "prost-types 0.14.3",
+ "snafu",
+ "tokio",
+ "tracing",
 ]
 
 [[package]]
-name = "is_terminal_polyfill"
-version = "1.70.2"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "a6cb138bb79a146c1bd460005623e142ef0181e3d0219cb493e02f7d08a35695"
-
-[[package]]
-name = "itertools"
-version = "0.13.0"
+name = "lance-index"
+version = "4.0.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "413ee7dfc52ee1a4949ceeb7dbc8a33f2d6c088194d9f922fb8318faf1f01186"
+checksum = "2da212f0090ea59f79ac3686660f596520c167fe1cb5f408900cf71d215f0e03"
 dependencies = [
- "either",
+ "arrow 57.3.0",
+ "arrow-arith 57.3.0",
+ "arrow-array 57.3.0",
+ "arrow-ord 57.3.0",
+ "arrow-schema 57.3.0",
+ "arrow-select 57.3.0",
+ "async-channel",
+ "async-recursion",
+ "async-trait",
+ "bitpacking",
+ "bitvec",
+ "bytes",
+ "chrono",
+ "crossbeam-queue",
+ "datafusion 52.5.0",
+ "datafusion-common 52.5.0",
+ "datafusion-expr 52.5.0",
+ "datafusion-physical-expr 52.5.0",
+ "datafusion-sql 52.5.0",
+ "deepsize",
+ "dirs",
+ "fst",
+ "futures",
+ "half",
+ "itertools 0.13.0",
+ "jsonb",
+ "lance-arrow",
+ "lance-core",
+ "lance-datafusion",
+ "lance-datagen",
+ "lance-encoding",
+ "lance-file",
+ "lance-io",
+ "lance-linalg",
+ "lance-table",
+ "libm",
+ "log",
+ "ndarray",
+ "num-traits",
+ "object_store",
+ "prost 0.14.3",
+ "prost-build 0.14.3",
+ "prost-types 0.14.3",
+ "rand 0.9.2",
+ "rand_distr 0.5.1",
+ "rangemap",
+ "rayon",
+ "roaring",
+ "serde",
+ "serde_json",
+ "smallvec",
+ "snafu",
+ "tantivy",
+ "tempfile",
+ "tokio",
+ "tracing",
+ "twox-hash",
+ "uuid",
 ]
 
 [[package]]
-name = "itertools"
-version = "0.14.0"
+name = "lance-io"
+version = "4.0.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "2b192c782037fadd9cfa75548310488aabdbf3d2da73885b31bd0abd03351285"
+checksum = "41d958eb4b56f03bbe0f5f85eb2b4e9657882812297b6f711f201ffc995f259f"
 dependencies = [
- "either",
+ "arrow 57.3.0",
+ "arrow-arith 57.3.0",
+ "arrow-array 57.3.0",
+ "arrow-buffer 57.3.0",
+ "arrow-cast 57.3.0",
+ "arrow-data 57.3.0",
+ "arrow-schema 57.3.0",
+ "arrow-select 57.3.0",
+ "async-recursion",
+ "async-trait",
+ "aws-config",
+ "aws-credential-types",
+ "byteorder",
+ "bytes",
+ "chrono",
+ "deepsize",
+ "futures",
+ "http 1.4.0",
+ "lance-arrow",
+ "lance-core",
+ "lance-namespace",
+ "log",
+ "object_store",
+ "object_store_opendal",
+ "opendal",
+ "path_abs",
+ "pin-project",
+ "prost 0.14.3",
+ "rand 0.9.2",
+ "serde",
+ "snafu",
+ "tempfile",
+ "tokio",
+ "tracing",
+ "url",
 ]
 
 [[package]]
-name = "itoa"
-version = "1.0.18"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "8f42a60cbdf9a97f5d2305f08a87dc4e09308d1276d28c869c684d7777685682"
-
-[[package]]
-name = "ittapi"
-version = "0.4.0"
+name = "lance-linalg"
+version = "4.0.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "6b996fe614c41395cdaedf3cf408a9534851090959d90d54a535f675550b64b1"
+checksum = "0285b70da35def7ed95e150fae1d5308089554e1290470403ed3c50cb235bc5e"
 dependencies = [
- "anyhow",
- "ittapi-sys",
- "log",
+ "arrow-array 57.3.0",
+ "arrow-buffer 57.3.0",
+ "arrow-schema 57.3.0",
+ "cc",
+ "deepsize",
+ "half",
+ "lance-arrow",
+ "lance-core",
+ "num-traits",
+ "rand 0.9.2",
 ]
 
 [[package]]
-name = "ittapi-sys"
-version = "0.4.0"
+name = "lance-namespace"
+version = "4.0.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "52f5385394064fa2c886205dba02598013ce83d3e92d33dbdc0c52fe0e7bf4fc"
+checksum = "5f78e2a828b654e062a495462c6e3eb4fcf0e7e907d761b8f217fc09ccd3ceac"
 dependencies = [
- "cc",
+ "arrow 57.3.0",
+ "async-trait",
+ "bytes",
+ "lance-core",
+ "lance-namespace-reqwest-client",
+ "serde",
+ "snafu",
 ]
 
 [[package]]
-name = "jobserver"
-version = "0.1.34"
+name = "lance-namespace-reqwest-client"
+version = "0.6.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "9afb3de4395d6b3e67a780b6de64b51c978ecf11cb9a462c66be7d4ca9039d33"
+checksum = "ee2e48de899e2931afb67fcddd0a08e439bf5d8b6ea2a2ed9cb8f4df669bd5cc"
 dependencies = [
- "getrandom 0.3.4",
- "libc",
+ "reqwest",
+ "serde",
+ "serde_json",
+ "serde_repr",
+ "url",
 ]
 
 [[package]]
-name = "js-sys"
-version = "0.3.94"
+name = "lance-table"
+version = "4.0.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "2e04e2ef80ce82e13552136fabeef8a5ed1f985a96805761cbb9a2c34e7664d9"
+checksum = "3df9c4adca3eb2074b3850432a9fb34248a3d90c3d6427d158b13ff9355664ee"
 dependencies = [
- "cfg-if",
- "futures-util",
- "once_cell",
- "wasm-bindgen",
+ "arrow 57.3.0",
+ "arrow-array 57.3.0",
+ "arrow-buffer 57.3.0",
+ "arrow-ipc 57.3.0",
+ "arrow-schema 57.3.0",
+ "async-trait",
+ "byteorder",
+ "bytes",
+ "chrono",
+ "deepsize",
+ "futures",
+ "lance-arrow",
+ "lance-core",
+ "lance-file",
+ "lance-io",
+ "log",
+ "object_store",
+ "prost 0.14.3",
+ "prost-build 0.14.3",
+ "prost-types 0.14.3",
+ "rand 0.9.2",
+ "rangemap",
+ "roaring",
+ "semver",
+ "serde",
+ "serde_json",
+ "snafu",
+ "tokio",
+ "tracing",
+ "url",
+ "uuid",
 ]
 
 [[package]]
@@ -2966,6 +5202,12 @@ version = "0.1.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "09edd9e8b54e49e587e4f6295a7d29c3ea94d469cb40ab8ca70b288248a81db2"
 
+[[package]]
+name = "levenshtein_automata"
+version = "0.2.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "0c2cdeb66e45e9f36bfad5bbdb4d2384e70936afbee843c6f6543f0c551ebb25"
+
 [[package]]
 name = "lexical-core"
 version = "0.8.5"
@@ -3185,6 +5427,19 @@ version = "0.4.29"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "5e5032e24019045c762d3c0f28f5b6b8bbf38563a65908389bf7978758920897"
 
+[[package]]
+name = "loom"
+version = "0.7.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "419e0dc8046cb947daa77eb95ae174acfbddb7673b4151f56d1eed8e93fbfaca"
+dependencies = [
+ "cfg-if",
+ "generator",
+ "scoped-tls",
+ "tracing",
+ "tracing-subscriber",
+]
+
 [[package]]
 name = "lru"
 version = "0.12.5"
@@ -3194,6 +5449,21 @@ dependencies = [
  "hashbrown 0.15.5",
 ]
 
+[[package]]
+name = "lru-slab"
+version = "0.1.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "112b39cec0b298b6c1999fee3e31427f74f676e4cb9879ed1a121b43661a4154"
+
+[[package]]
+name = "lz4"
+version = "1.28.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "a20b523e860d03443e98350ceaac5e71c6ba89aea7d960769ec3ce37f4de5af4"
+dependencies = [
+ "lz4-sys",
+]
+
 [[package]]
 name = "lz4-sys"
 version = "1.11.1+lz4-1.10.0"
@@ -3206,9 +5476,18 @@ dependencies = [
 
 [[package]]
 name = "lz4_flex"
-version = "0.11.6"
+version = "0.11.6"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "373f5eceeeab7925e0c1098212f2fbc4d416adec9d35051a6ab251e824c1854a"
+dependencies = [
+ "twox-hash",
+]
+
+[[package]]
+name = "lz4_flex"
+version = "0.12.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "373f5eceeeab7925e0c1098212f2fbc4d416adec9d35051a6ab251e824c1854a"
+checksum = "98c23545df7ecf1b16c303910a69b079e8e251d60f7dd2cc9b4177f2afaf1746"
 dependencies = [
  "twox-hash",
 ]
@@ -3248,6 +5527,19 @@ version = "0.7.3"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "0e7465ac9959cc2b1404e8e2367b43684a6d13790fe23056cc8c6c5a6b7bcb94"
 
+[[package]]
+name = "matrixmultiply"
+version = "0.3.10"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "a06de3016e9fae57a36fd14dba131fccf49f74b40b7fbdb472f96e361ec71a08"
+dependencies = [
+ "autocfg",
+ "num_cpus",
+ "once_cell",
+ "rawpointer",
+ "thread-tree",
+]
+
 [[package]]
 name = "maybe-owned"
 version = "0.3.4"
@@ -3264,6 +5556,15 @@ dependencies = [
  "digest",
 ]
 
+[[package]]
+name = "measure_time"
+version = "0.9.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "51c55d61e72fc3ab704396c5fa16f4c184db37978ae4e94ca8959693a235fc0e"
+dependencies = [
+ "log",
+]
+
 [[package]]
 name = "memchr"
 version = "2.8.0"
@@ -3279,12 +5580,31 @@ dependencies = [
  "rustix 1.1.4",
 ]
 
+[[package]]
+name = "memmap2"
+version = "0.9.10"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "714098028fe011992e1c3962653c96b2d578c4b4bce9036e15ff220319b1e0e3"
+dependencies = [
+ "libc",
+]
+
 [[package]]
 name = "mime"
 version = "0.3.17"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "6877bb514081ee2a7ff5ef9de3281f14a4dd4bceac4c09388074a6b5df8a139a"
 
+[[package]]
+name = "mime_guess"
+version = "2.0.5"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "f7c44f8e672c00fe5308fa235f821cb4198414e1c77935c1ab6948d3fd78550e"
+dependencies = [
+ "mime",
+ "unicase",
+]
+
 [[package]]
 name = "minimal-lexical"
 version = "0.2.1"
@@ -3312,12 +5632,59 @@ dependencies = [
  "windows-sys 0.61.2",
 ]
 
+[[package]]
+name = "mock_instant"
+version = "0.6.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "dce6dd36094cac388f119d2e9dc82dc730ef91c32a6222170d630e5414b956e6"
+
+[[package]]
+name = "moka"
+version = "0.12.15"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "957228ad12042ee839f93c8f257b62b4c0ab5eaae1d4fa60de53b27c9d7c5046"
+dependencies = [
+ "async-lock",
+ "crossbeam-channel",
+ "crossbeam-epoch",
+ "crossbeam-utils",
+ "equivalent",
+ "event-listener",
+ "futures-util",
+ "parking_lot",
+ "portable-atomic",
+ "smallvec",
+ "tagptr",
+ "uuid",
+]
+
 [[package]]
 name = "multimap"
 version = "0.10.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "1d87ecb2933e8aeadb3e3a02b828fed80a7528047e68b4f424523a0981a3a084"
 
+[[package]]
+name = "murmurhash32"
+version = "0.3.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "2195bf6aa996a481483b29d62a7663eed3fe39600c460e323f8ff41e90bdd89b"
+
+[[package]]
+name = "ndarray"
+version = "0.16.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "882ed72dce9365842bf196bdeedf5055305f11fc8c03dee7bb0194a6cad34841"
+dependencies = [
+ "matrixmultiply",
+ "num-complex",
+ "num-integer",
+ "num-traits",
+ "portable-atomic",
+ "portable-atomic-util",
+ "rawpointer",
+]
+
 [[package]]
 name = "nibble_vec"
 version = "0.1.0"
@@ -3335,7 +5702,7 @@ checksum = "ab2156c4fce2f8df6c499cc1c763e4394b7482525bf2a9701c9d79d215f519e4"
 dependencies = [
  "bitflags 2.11.0",
  "cfg-if",
- "cfg_aliases",
+ "cfg_aliases 0.1.1",
  "libc",
 ]
 
@@ -3355,6 +5722,15 @@ dependencies = [
  "minimal-lexical",
 ]
 
+[[package]]
+name = "nom"
+version = "8.0.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "df9761775871bdef83bee530e60050f7e54b1105350d6884eb0fb4f46c2f9405"
+dependencies = [
+ "memchr",
+]
+
 [[package]]
 name = "nonzero_ext"
 version = "0.3.0"
@@ -3392,6 +5768,7 @@ checksum = "a5e44f723f1133c9deac646763579fdb3ac745e418f2a7af9cd0c431da1f20b9"
 dependencies = [
  "num-integer",
  "num-traits",
+ "serde",
 ]
 
 [[package]]
@@ -3479,7 +5856,7 @@ dependencies = [
  "proc-macro-crate",
  "proc-macro2",
  "quote",
- "syn",
+ "syn 2.0.117",
 ]
 
 [[package]]
@@ -3501,14 +5878,26 @@ source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "fbfbfff40aeccab00ec8a910b57ca8ecf4319b335c542f2edcd19dd25a1e2a00"
 dependencies = [
  "async-trait",
+ "base64",
  "bytes",
  "chrono",
+ "form_urlencoded",
  "futures",
- "http",
+ "http 1.4.0",
+ "http-body-util",
  "humantime",
+ "hyper",
  "itertools 0.14.0",
+ "md-5",
  "parking_lot",
  "percent-encoding",
+ "quick-xml 0.38.4",
+ "rand 0.9.2",
+ "reqwest",
+ "ring",
+ "serde",
+ "serde_json",
+ "serde_urlencoded",
  "thiserror 2.0.18",
  "tokio",
  "tracing",
@@ -3518,6 +5907,22 @@ dependencies = [
  "web-time",
 ]
 
+[[package]]
+name = "object_store_opendal"
+version = "0.55.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "113ab0769e972eee585e57407b98de08bda5354fa28e8ba4d89038d6cb6a8991"
+dependencies = [
+ "async-trait",
+ "bytes",
+ "chrono",
+ "futures",
+ "object_store",
+ "opendal",
+ "pin-project",
+ "tokio",
+]
+
 [[package]]
 name = "once_cell"
 version = "1.21.4"
@@ -3530,6 +5935,47 @@ version = "1.70.2"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "384b8ab6d37215f3c5301a95a4accb5d64aa607f1fcb26a11b5303878451b4fe"
 
+[[package]]
+name = "oneshot"
+version = "0.1.13"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "269bca4c2591a28585d6bf10d9ed0332b7d76900a1b02bec41bdc3a2cdcda107"
+
+[[package]]
+name = "opendal"
+version = "0.55.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "d075ab8a203a6ab4bc1bce0a4b9fe486a72bf8b939037f4b78d95386384bc80a"
+dependencies = [
+ "anyhow",
+ "backon",
+ "base64",
+ "bytes",
+ "crc32c",
+ "futures",
+ "getrandom 0.2.17",
+ "http 1.4.0",
+ "http-body 1.0.1",
+ "jiff",
+ "log",
+ "md-5",
+ "percent-encoding",
+ "quick-xml 0.38.4",
+ "reqsign",
+ "reqwest",
+ "serde",
+ "serde_json",
+ "tokio",
+ "url",
+ "uuid",
+]
+
+[[package]]
+name = "openssl-probe"
+version = "0.2.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "7c87def4c32ab89d880effc9e097653c8da5d6ef28e6b539d313baaacfbafcbe"
+
 [[package]]
 name = "openssl-sys"
 version = "0.9.112"
@@ -3542,6 +5988,12 @@ dependencies = [
  "vcpkg",
 ]
 
+[[package]]
+name = "option-ext"
+version = "0.2.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "04744f49eae99ab78e0d5c0b603ab218f515ea8cfe5a456d7629ad883a3b6e7d"
+
 [[package]]
 name = "ordered-float"
 version = "2.10.1"
@@ -3551,6 +6003,25 @@ dependencies = [
  "num-traits",
 ]
 
+[[package]]
+name = "ordered-float"
+version = "5.3.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "b7d950ca161dc355eaf28f82b11345ed76c6e1f6eb1f4f4479e0323b9e2fbd0e"
+dependencies = [
+ "num-traits",
+]
+
+[[package]]
+name = "ordered-multimap"
+version = "0.7.3"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "49203cdcae0030493bad186b28da2fa25645fa276a51b6fec8010d281e02ef79"
+dependencies = [
+ "dlv-list",
+ "hashbrown 0.14.5",
+]
+
 [[package]]
 name = "os_pipe"
 version = "1.2.3"
@@ -3561,6 +6032,27 @@ dependencies = [
  "windows-sys 0.61.2",
 ]
 
+[[package]]
+name = "outref"
+version = "0.5.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "1a80800c0488c3a21695ea981a54918fbb37abf04f4d0720c453632255e2ff0e"
+
+[[package]]
+name = "ownedbytes"
+version = "0.9.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "2fbd56f7631767e61784dc43f8580f403f4475bd4aaa4da003e6295e1bab4a7e"
+dependencies = [
+ "stable_deref_trait",
+]
+
+[[package]]
+name = "parking"
+version = "2.2.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "f38d5652c16fde515bb1ecef450ab0f6a219d619a7274976324d5e377f7dceba"
+
 [[package]]
 name = "parking_lot"
 version = "0.12.5"
@@ -3606,7 +6098,7 @@ dependencies = [
  "futures",
  "half",
  "hashbrown 0.15.5",
- "lz4_flex",
+ "lz4_flex 0.11.6",
  "num",
  "num-bigint",
  "object_store",
@@ -3626,6 +6118,18 @@ version = "1.0.15"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "57c0d7b74b563b49d38dae00a0c37d4d6de9b432382b2892f0574ddcae73fd0a"
 
+[[package]]
+name = "path_abs"
+version = "0.5.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "05ef02f6342ac01d8a93b65f96db53fe68a92a15f41144f97fb00a9e669633c3"
+dependencies = [
+ "serde",
+ "serde_derive",
+ "std_prelude",
+ "stfu8",
+]
+
 [[package]]
 name = "peeking_take_while"
 version = "0.1.2"
@@ -3638,6 +6142,12 @@ version = "2.3.2"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "9b4f627cb1b25917193a259e49bdad08f671f8d9708acfd5fe0a8c1455d87220"
 
+[[package]]
+name = "permutation"
+version = "0.4.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "df202b0b0f5b8e389955afd5f27b007b00fb948162953f1db9c70d2c7e3157d7"
+
 [[package]]
 name = "petgraph"
 version = "0.6.5"
@@ -3705,7 +6215,7 @@ checksum = "d9b20ed30f105399776b9c883e68e536ef602a16ae6f596d2c473591d6ad64c6"
 dependencies = [
  "proc-macro2",
  "quote",
- "syn",
+ "syn 2.0.117",
 ]
 
 [[package]]
@@ -3714,6 +6224,12 @@ version = "0.2.17"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "a89322df9ebe1c1578d689c92318e070967d1042b512afbe49518723f4e6d5cd"
 
+[[package]]
+name = "pin-utils"
+version = "0.1.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "8b870d8c151b6f2fb93e84a13146138f05d02ed11c7e7c54f8826aaaf7c9f184"
+
 [[package]]
 name = "pkg-config"
 version = "0.3.32"
@@ -3726,6 +6242,15 @@ version = "1.13.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "c33a9471896f1c69cecef8d20cbe2f7accd12527ce60845ff44c153bb2a21b49"
 
+[[package]]
+name = "portable-atomic-util"
+version = "0.2.7"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "c2a106d1259c23fac8e543272398ae0e3c0b8d33c88ed73d0cc71b0f1d902618"
+dependencies = [
+ "portable-atomic",
+]
+
 [[package]]
 name = "postcard"
 version = "1.1.3"
@@ -3769,7 +6294,7 @@ source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "479ca8adacdd7ce8f1fb39ce9ecccbfe93a3f1344b3d0d97f20bc0196208f62b"
 dependencies = [
  "proc-macro2",
- "syn",
+ "syn 2.0.117",
 ]
 
 [[package]]
@@ -3808,7 +6333,17 @@ source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "2796faa41db3ec313a31f7624d9286acf277b52de526150b7e69f3debf891ee5"
 dependencies = [
  "bytes",
- "prost-derive",
+ "prost-derive 0.13.5",
+]
+
+[[package]]
+name = "prost"
+version = "0.14.3"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "d2ea70524a2f82d518bce41317d0fae74151505651af45faf1ffbd6fd33f0568"
+dependencies = [
+ "bytes",
+ "prost-derive 0.14.3",
 ]
 
 [[package]]
@@ -3824,10 +6359,29 @@ dependencies = [
  "once_cell",
  "petgraph 0.7.1",
  "prettyplease",
- "prost",
- "prost-types",
+ "prost 0.13.5",
+ "prost-types 0.13.5",
+ "regex",
+ "syn 2.0.117",
+ "tempfile",
+]
+
+[[package]]
+name = "prost-build"
+version = "0.14.3"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "343d3bd7056eda839b03204e68deff7d1b13aba7af2b2fd16890697274262ee7"
+dependencies = [
+ "heck",
+ "itertools 0.14.0",
+ "log",
+ "multimap",
+ "petgraph 0.8.3",
+ "prettyplease",
+ "prost 0.14.3",
+ "prost-types 0.14.3",
  "regex",
- "syn",
+ "syn 2.0.117",
  "tempfile",
 ]
 
@@ -3841,7 +6395,20 @@ dependencies = [
  "itertools 0.14.0",
  "proc-macro2",
  "quote",
- "syn",
+ "syn 2.0.117",
+]
+
+[[package]]
+name = "prost-derive"
+version = "0.14.3"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "27c6023962132f4b30eb4c172c91ce92d933da334c59c23cddee82358ddafb0b"
+dependencies = [
+ "anyhow",
+ "itertools 0.14.0",
+ "proc-macro2",
+ "quote",
+ "syn 2.0.117",
 ]
 
 [[package]]
@@ -3850,7 +6417,16 @@ version = "0.13.5"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "52c2c1bf36ddb1a1c396b3601a3cec27c2462e45f07c386894ec3ccf5332bd16"
 dependencies = [
- "prost",
+ "prost 0.13.5",
+]
+
+[[package]]
+name = "prost-types"
+version = "0.14.3"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "8991c4cbdb8bc5b11f0b074ffe286c30e523de90fee5ba8132f1399f23cb3dd7"
+dependencies = [
+ "prost 0.14.3",
 ]
 
 [[package]]
@@ -3859,7 +6435,7 @@ version = "0.1.0"
 dependencies = [
  "env_logger",
  "log",
- "prost",
+ "prost 0.13.5",
  "serde",
  "tonic",
  "tonic-build",
@@ -3888,29 +6464,110 @@ dependencies = [
 ]
 
 [[package]]
-name = "pulley-macros"
-version = "41.0.4"
+name = "pulley-macros"
+version = "41.0.4"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "56000349b6896e3d44286eb9c330891237f40b27fd43c1ccc84547d0b463cb40"
+dependencies = [
+ "proc-macro2",
+ "quote",
+ "syn 2.0.117",
+]
+
+[[package]]
+name = "quad-rand"
+version = "0.2.3"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "5a651516ddc9168ebd67b24afd085a718be02f8858fe406591b013d101ce2f40"
+
+[[package]]
+name = "quanta"
+version = "0.12.6"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "f3ab5a9d756f0d97bdc89019bd2e4ea098cf9cde50ee7564dde6b81ccc8f06c7"
+dependencies = [
+ "crossbeam-utils",
+ "libc",
+ "once_cell",
+ "raw-cpuid",
+ "wasi",
+ "web-sys",
+ "winapi",
+]
+
+[[package]]
+name = "quick-xml"
+version = "0.37.5"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "331e97a1af0bf59823e6eadffe373d7b27f485be8748f71471c662c1f269b7fb"
+dependencies = [
+ "memchr",
+ "serde",
+]
+
+[[package]]
+name = "quick-xml"
+version = "0.38.4"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "b66c2058c55a409d601666cffe35f04333cf1013010882cec174a7467cd4e21c"
+dependencies = [
+ "memchr",
+ "serde",
+]
+
+[[package]]
+name = "quinn"
+version = "0.11.9"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "b9e20a958963c291dc322d98411f541009df2ced7b5a4f2bd52337638cfccf20"
+dependencies = [
+ "bytes",
+ "cfg_aliases 0.2.1",
+ "pin-project-lite",
+ "quinn-proto",
+ "quinn-udp",
+ "rustc-hash 2.1.2",
+ "rustls",
+ "socket2 0.6.3",
+ "thiserror 2.0.18",
+ "tokio",
+ "tracing",
+ "web-time",
+]
+
+[[package]]
+name = "quinn-proto"
+version = "0.11.14"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "56000349b6896e3d44286eb9c330891237f40b27fd43c1ccc84547d0b463cb40"
+checksum = "434b42fec591c96ef50e21e886936e66d3cc3f737104fdb9b737c40ffb94c098"
 dependencies = [
- "proc-macro2",
- "quote",
- "syn",
+ "bytes",
+ "getrandom 0.3.4",
+ "lru-slab",
+ "rand 0.9.2",
+ "ring",
+ "rustc-hash 2.1.2",
+ "rustls",
+ "rustls-pki-types",
+ "slab",
+ "thiserror 2.0.18",
+ "tinyvec",
+ "tracing",
+ "web-time",
 ]
 
 [[package]]
-name = "quanta"
-version = "0.12.6"
+name = "quinn-udp"
+version = "0.5.14"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "f3ab5a9d756f0d97bdc89019bd2e4ea098cf9cde50ee7564dde6b81ccc8f06c7"
+checksum = "addec6a0dcad8a8d96a771f815f0eaf55f9d1805756410b39f5fa81332574cbd"
 dependencies = [
- "crossbeam-utils",
+ "cfg_aliases 0.2.1",
  "libc",
  "once_cell",
- "raw-cpuid",
- "wasi",
- "web-sys",
- "winapi",
+ "socket2 0.6.3",
+ "tracing",
+ "windows-sys 0.60.2",
 ]
 
 [[package]]
@@ -3934,6 +6591,12 @@ version = "6.0.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "f8dcc9c7d52a811697d2151c701e0d08956f92b0e24136cf4cf27b57a6a0d9bf"
 
+[[package]]
+name = "radium"
+version = "0.7.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "dc33ff2d4973d518d823d61aa239014831e521c75da58e3df4840d3f47749d09"
+
 [[package]]
 name = "radix_trie"
 version = "0.2.1"
@@ -4003,6 +6666,26 @@ dependencies = [
  "getrandom 0.3.4",
 ]
 
+[[package]]
+name = "rand_distr"
+version = "0.4.3"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "32cb0b9bc82b0a0876c2dd994a7e7a2683d3e7390ca40e6886785ef0c7e3ee31"
+dependencies = [
+ "num-traits",
+ "rand 0.8.5",
+]
+
+[[package]]
+name = "rand_distr"
+version = "0.5.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "6a8615d50dcf34fa31f7ab52692afec947c4dd0ab803cc87cb3b0b4570ff7463"
+dependencies = [
+ "num-traits",
+ "rand 0.9.2",
+]
+
 [[package]]
 name = "rand_xoshiro"
 version = "0.6.0"
@@ -4012,6 +6695,34 @@ dependencies = [
  "rand_core 0.6.4",
 ]
 
+[[package]]
+name = "rand_xoshiro"
+version = "0.7.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "f703f4665700daf5512dcca5f43afa6af89f09db47fb56be587f80636bda2d41"
+dependencies = [
+ "rand_core 0.9.5",
+]
+
+[[package]]
+name = "random_word"
+version = "0.5.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "e47a395bdb55442b883c89062d6bcff25dc90fa5f8369af81e0ac6d49d78cf81"
+dependencies = [
+ "ahash",
+ "brotli",
+ "paste",
+ "rand 0.9.2",
+ "unicase",
+]
+
+[[package]]
+name = "rangemap"
+version = "1.7.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "973443cf09a9c8656b574a866ab68dfa19f0867d0340648c7d2f6a71b8a8ea68"
+
 [[package]]
 name = "raw-cpuid"
 version = "11.6.0"
@@ -4021,6 +6732,12 @@ dependencies = [
  "bitflags 2.11.0",
 ]
 
+[[package]]
+name = "rawpointer"
+version = "0.2.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "60a357793950651c4ed0f3f52338f53b2f809f32d83a07f72909fa13e4c6c1e3"
+
 [[package]]
 name = "rayon"
 version = "1.11.0"
@@ -4092,7 +6809,7 @@ source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "76009fbe0614077fc1a2ce255e3a1881a2e3a3527097d5dc6d8212c585e7e38b"
 dependencies = [
  "quote",
- "syn",
+ "syn 2.0.117",
 ]
 
 [[package]]
@@ -4115,6 +6832,17 @@ dependencies = [
  "thiserror 1.0.69",
 ]
 
+[[package]]
+name = "redox_users"
+version = "0.5.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "a4e608c6638b9c18977b00b475ac1f28d14e84b27d8d42f70e0bf1e3dec127ac"
+dependencies = [
+ "getrandom 0.2.17",
+ "libredox",
+ "thiserror 2.0.18",
+]
+
 [[package]]
 name = "regalloc2"
 version = "0.13.5"
@@ -4153,12 +6881,117 @@ dependencies = [
  "regex-syntax",
 ]
 
+[[package]]
+name = "regex-lite"
+version = "0.1.9"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "cab834c73d247e67f4fae452806d17d3c7501756d98c8808d7c9c7aa7d18f973"
+
 [[package]]
 name = "regex-syntax"
 version = "0.8.10"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "dc897dd8d9e8bd1ed8cdad82b5966c3e0ecae09fb1907d58efaa013543185d0a"
 
+[[package]]
+name = "reqsign"
+version = "0.16.5"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "43451dbf3590a7590684c25fb8d12ecdcc90ed3ac123433e500447c7d77ed701"
+dependencies = [
+ "anyhow",
+ "async-trait",
+ "base64",
+ "chrono",
+ "form_urlencoded",
+ "getrandom 0.2.17",
+ "hex",
+ "hmac",
+ "home",
+ "http 1.4.0",
+ "log",
+ "percent-encoding",
+ "quick-xml 0.37.5",
+ "rand 0.8.5",
+ "reqwest",
+ "rust-ini",
+ "serde",
+ "serde_json",
+ "sha1",
+ "sha2",
+ "tokio",
+]
+
+[[package]]
+name = "reqwest"
+version = "0.12.28"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "eddd3ca559203180a307f12d114c268abf583f59b03cb906fd0b3ff8646c1147"
+dependencies = [
+ "base64",
+ "bytes",
+ "encoding_rs",
+ "futures-core",
+ "futures-util",
+ "h2",
+ "http 1.4.0",
+ "http-body 1.0.1",
+ "http-body-util",
+ "hyper",
+ "hyper-rustls",
+ "hyper-util",
+ "js-sys",
+ "log",
+ "mime",
+ "mime_guess",
+ "percent-encoding",
+ "pin-project-lite",
+ "quinn",
+ "rustls",
+ "rustls-native-certs",
+ "rustls-pki-types",
+ "serde",
+ "serde_json",
+ "serde_urlencoded",
+ "sync_wrapper",
+ "tokio",
+ "tokio-rustls",
+ "tokio-util",
+ "tower 0.5.3",
+ "tower-http",
+ "tower-service",
+ "url",
+ "wasm-bindgen",
+ "wasm-bindgen-futures",
+ "wasm-streams",
+ "web-sys",
+ "webpki-roots",
+]
+
+[[package]]
+name = "ring"
+version = "0.17.14"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "a4689e6c2294d81e88dc6261c768b63bc4fcdb852be6d1352498b114f61383b7"
+dependencies = [
+ "cc",
+ "cfg-if",
+ "getrandom 0.2.17",
+ "libc",
+ "untrusted",
+ "windows-sys 0.52.0",
+]
+
+[[package]]
+name = "roaring"
+version = "0.11.3"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "8ba9ce64a8f45d7fc86358410bb1a82e8c987504c0d4900e9141d69a9f26c885"
+dependencies = [
+ "bytemuck",
+ "byteorder",
+]
+
 [[package]]
 name = "rocksdb"
 version = "0.21.0"
@@ -4169,6 +7002,26 @@ dependencies = [
  "librocksdb-sys",
 ]
 
+[[package]]
+name = "rust-ini"
+version = "0.21.3"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "796e8d2b6696392a43bea58116b667fb4c29727dc5abd27d6acf338bb4f688c7"
+dependencies = [
+ "cfg-if",
+ "ordered-multimap",
+]
+
+[[package]]
+name = "rust-stemmers"
+version = "1.2.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "e46a2036019fdb888131db7a4c847a1063a7493f971ed94ea82c67eada63ca54"
+dependencies = [
+ "serde",
+ "serde_derive",
+]
+
 [[package]]
 name = "rustc-demangle"
 version = "0.1.27"
@@ -4232,6 +7085,55 @@ dependencies = [
  "rustix 1.1.4",
 ]
 
+[[package]]
+name = "rustls"
+version = "0.23.37"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "758025cb5fccfd3bc2fd74708fd4682be41d99e5dff73c377c0646c6012c73a4"
+dependencies = [
+ "aws-lc-rs",
+ "once_cell",
+ "ring",
+ "rustls-pki-types",
+ "rustls-webpki",
+ "subtle",
+ "zeroize",
+]
+
+[[package]]
+name = "rustls-native-certs"
+version = "0.8.3"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "612460d5f7bea540c490b2b6395d8e34a953e52b491accd6c86c8164c5932a63"
+dependencies = [
+ "openssl-probe",
+ "rustls-pki-types",
+ "schannel",
+ "security-framework",
+]
+
+[[package]]
+name = "rustls-pki-types"
+version = "1.14.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "be040f8b0a225e40375822a563fa9524378b9d63112f53e19ffff34df5d33fdd"
+dependencies = [
+ "web-time",
+ "zeroize",
+]
+
+[[package]]
+name = "rustls-webpki"
+version = "0.103.11"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "20a6af516fea4b20eccceaf166e8aa666ac996208e8a644ce3ef5aa783bc7cd4"
+dependencies = [
+ "aws-lc-rs",
+ "ring",
+ "rustls-pki-types",
+ "untrusted",
+]
+
 [[package]]
 name = "rustversion"
 version = "1.0.22"
@@ -4287,12 +7189,50 @@ dependencies = [
  "pkg-config",
 ]
 
+[[package]]
+name = "schannel"
+version = "0.1.29"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "91c1b7e4904c873ef0710c1f407dde2e6287de2bebc1bbbf7d430bb7cbffd939"
+dependencies = [
+ "windows-sys 0.61.2",
+]
+
+[[package]]
+name = "scoped-tls"
+version = "1.0.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "e1cf6437eb19a8f4a6cc0f7dca544973b0b78843adbfeb3683d1a94a0024a294"
+
 [[package]]
 name = "scopeguard"
 version = "1.2.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "94143f37725109f92c262ed2cf5e59bce7498c01bcc1502d7b9afe439a4e9f49"
 
+[[package]]
+name = "security-framework"
+version = "3.7.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "b7f4bc775c73d9a02cde8bf7b2ec4c9d12743edf609006c7facc23998404cd1d"
+dependencies = [
+ "bitflags 2.11.0",
+ "core-foundation",
+ "core-foundation-sys",
+ "libc",
+ "security-framework-sys",
+]
+
+[[package]]
+name = "security-framework-sys"
+version = "2.17.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "6ce2691df843ecc5d231c0b14ece2acc3efb62c0a398c7e1d875f3983ce020e3"
+dependencies = [
+ "core-foundation-sys",
+ "libc",
+]
+
 [[package]]
 name = "semver"
 version = "1.0.28"
@@ -4319,6 +7259,16 @@ dependencies = [
  "serde_derive",
 ]
 
+[[package]]
+name = "serde_bytes"
+version = "0.11.19"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "a5d440709e79d88e51ac01c4b72fc6cb7314017bb7da9eeff678aa94c10e3ea8"
+dependencies = [
+ "serde",
+ "serde_core",
+]
+
 [[package]]
 name = "serde_core"
 version = "1.0.228"
@@ -4336,7 +7286,7 @@ checksum = "d540f220d3187173da220f885ab66608367b6574e925011a9353e4badda91d79"
 dependencies = [
  "proc-macro2",
  "quote",
- "syn",
+ "syn 2.0.117",
 ]
 
 [[package]]
@@ -4359,7 +7309,7 @@ source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "b992cea3194eea663ba99a042d61cea4bd1872da37021af56f6a37e0359b9d33"
 dependencies = [
  "inventory",
- "nom",
+ "nom 7.1.3",
  "regex",
  "serde",
  "serde_json",
@@ -4399,7 +7349,18 @@ checksum = "aafbefbe175fa9bf03ca83ef89beecff7d2a95aaacd5732325b90ac8c3bd7b90"
 dependencies = [
  "proc-macro2",
  "quote",
- "syn",
+ "syn 2.0.117",
+]
+
+[[package]]
+name = "serde_repr"
+version = "0.1.20"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "175ee3e80ae9982737ca543e96133087cbd9a485eecc3bc4de9c1a37b47ea59c"
+dependencies = [
+ "proc-macro2",
+ "quote",
+ "syn 2.0.117",
 ]
 
 [[package]]
@@ -4411,6 +7372,18 @@ dependencies = [
  "serde_core",
 ]
 
+[[package]]
+name = "serde_urlencoded"
+version = "0.7.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "d3491c14715ca2294c4d6a88f15e84739788c1d030eed8c110436aafdaa2f3fd"
+dependencies = [
+ "form_urlencoded",
+ "itoa",
+ "ryu",
+ "serde",
+]
+
 [[package]]
 name = "serde_yaml"
 version = "0.9.34+deprecated"
@@ -4424,6 +7397,17 @@ dependencies = [
  "unsafe-libyaml",
 ]
 
+[[package]]
+name = "sha1"
+version = "0.10.6"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "e3bf829a2d51ab4a5ddf1352d8470c140cadc8301b2ae1789db023f01cedd6ba"
+dependencies = [
+ "cfg-if",
+ "cpufeatures 0.2.17",
+ "digest",
+]
+
 [[package]]
 name = "sha2"
 version = "0.10.9"
@@ -4520,6 +7504,15 @@ dependencies = [
  "typenum",
 ]
 
+[[package]]
+name = "sketches-ddsketch"
+version = "0.3.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "0c6f73aeb92d671e0cc4dca167e59b2deb6387c375391bc99ee743f326994a2b"
+dependencies = [
+ "serde",
+]
+
 [[package]]
 name = "slab"
 version = "0.4.12"
@@ -4535,6 +7528,27 @@ dependencies = [
  "serde",
 ]
 
+[[package]]
+name = "snafu"
+version = "0.9.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "d1d4bced6a69f90b2056c03dcff2c4737f98d6fb9e0853493996e1d253ca29c6"
+dependencies = [
+ "snafu-derive",
+]
+
+[[package]]
+name = "snafu-derive"
+version = "0.9.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "54254b8531cafa275c5e096f62d48c81435d1015405a91198ddb11e967301d40"
+dependencies = [
+ "heck",
+ "proc-macro2",
+ "quote",
+ "syn 2.0.117",
+]
+
 [[package]]
 name = "snap"
 version = "1.1.1"
@@ -4577,7 +7591,28 @@ source = "git+https://github.com/FunctionStream/sqlparser-rs?branch=0.58.0%2Ffs#
 dependencies = [
  "log",
  "recursive",
- "sqlparser_derive",
+ "sqlparser_derive 0.3.0 (git+https://github.com/FunctionStream/sqlparser-rs?branch=0.58.0%2Ffs)",
+]
+
+[[package]]
+name = "sqlparser"
+version = "0.59.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "4591acadbcf52f0af60eafbb2c003232b2b4cd8de5f0e9437cb8b1b59046cc0f"
+dependencies = [
+ "log",
+ "sqlparser_derive 0.3.0 (registry+https://github.com/rust-lang/crates.io-index)",
+]
+
+[[package]]
+name = "sqlparser_derive"
+version = "0.3.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "da5fc6819faabb412da764b99d3b713bb55083c11e7e0c00144d386cd6a1939c"
+dependencies = [
+ "proc-macro2",
+ "quote",
+ "syn 2.0.117",
 ]
 
 [[package]]
@@ -4587,7 +7622,7 @@ source = "git+https://github.com/FunctionStream/sqlparser-rs?branch=0.58.0%2Ffs#
 dependencies = [
  "proc-macro2",
  "quote",
- "syn",
+ "syn 2.0.117",
 ]
 
 [[package]]
@@ -4615,6 +7650,18 @@ version = "1.1.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "a2eb9349b6444b326872e140eb1cf5e7c522154d69e7a0ffb0fb81c06b37543f"
 
+[[package]]
+name = "std_prelude"
+version = "0.2.12"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "8207e78455ffdf55661170876f88daf85356e4edd54e0a3dbc79586ca1e50cbe"
+
+[[package]]
+name = "stfu8"
+version = "0.2.7"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "e51f1e89f093f99e7432c491c382b88a6860a5adbe6bf02574bf0a08efff1978"
+
 [[package]]
 name = "strsim"
 version = "0.11.1"
@@ -4627,9 +7674,15 @@ version = "0.26.3"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "8fec0f0aef304996cf250b31b5a10dee7980c85da9d759361292b8bca5a18f06"
 dependencies = [
- "strum_macros",
+ "strum_macros 0.26.4",
 ]
 
+[[package]]
+name = "strum"
+version = "0.27.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "af23d6f6c1a224baef9d3f61e287d2761385a5b88fdab4eb4c6f11aeb54c4bcf"
+
 [[package]]
 name = "strum_macros"
 version = "0.26.4"
@@ -4640,7 +7693,19 @@ dependencies = [
  "proc-macro2",
  "quote",
  "rustversion",
- "syn",
+ "syn 2.0.117",
+]
+
+[[package]]
+name = "strum_macros"
+version = "0.27.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "7695ce3845ea4b33927c055a39dc438a45b059f7c1b3d91d38d10355fb8cbca7"
+dependencies = [
+ "heck",
+ "proc-macro2",
+ "quote",
+ "syn 2.0.117",
 ]
 
 [[package]]
@@ -4649,6 +7714,17 @@ version = "2.6.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "13c2bddecc57b384dee18652358fb23172facb8a2c51ccc10d74c157bdea3292"
 
+[[package]]
+name = "syn"
+version = "1.0.109"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "72b64191b275b66ffe2469e8af2c1cfe3bafa67b529ead792a6d0160888b4237"
+dependencies = [
+ "proc-macro2",
+ "quote",
+ "unicode-ident",
+]
+
 [[package]]
 name = "syn"
 version = "2.0.117"
@@ -4665,6 +7741,9 @@ name = "sync_wrapper"
 version = "1.0.2"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "0bf256ce5efdfa370213c1dabab5935a12e49f2c58d15e9eac2870d3b4f27263"
+dependencies = [
+ "futures-core",
+]
 
 [[package]]
 name = "synstructure"
@@ -4674,7 +7753,7 @@ checksum = "728a70f3dbaf5bab7f0c4b1ac8d7ae5ea60a4b5549c8a5914361c99147a709d2"
 dependencies = [
  "proc-macro2",
  "quote",
- "syn",
+ "syn 2.0.117",
 ]
 
 [[package]]
@@ -4693,6 +7772,164 @@ dependencies = [
  "winx",
 ]
 
+[[package]]
+name = "tagptr"
+version = "0.2.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "7b2093cf4c8eb1e67749a6762251bc9cd836b6fc171623bd0a9d324d37af2417"
+
+[[package]]
+name = "tantivy"
+version = "0.24.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "64a966cb0e76e311f09cf18507c9af192f15d34886ee43d7ba7c7e3803660c43"
+dependencies = [
+ "aho-corasick",
+ "arc-swap",
+ "base64",
+ "bitpacking",
+ "bon",
+ "byteorder",
+ "census",
+ "crc32fast",
+ "crossbeam-channel",
+ "downcast-rs",
+ "fastdivide",
+ "fnv",
+ "fs4",
+ "htmlescape",
+ "hyperloglogplus",
+ "itertools 0.14.0",
+ "levenshtein_automata",
+ "log",
+ "lru",
+ "lz4_flex 0.11.6",
+ "measure_time",
+ "memmap2",
+ "once_cell",
+ "oneshot",
+ "rayon",
+ "regex",
+ "rust-stemmers",
+ "rustc-hash 2.1.2",
+ "serde",
+ "serde_json",
+ "sketches-ddsketch",
+ "smallvec",
+ "tantivy-bitpacker",
+ "tantivy-columnar",
+ "tantivy-common",
+ "tantivy-fst",
+ "tantivy-query-grammar",
+ "tantivy-stacker",
+ "tantivy-tokenizer-api",
+ "tempfile",
+ "thiserror 2.0.18",
+ "time",
+ "uuid",
+ "winapi",
+]
+
+[[package]]
+name = "tantivy-bitpacker"
+version = "0.8.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "1adc286a39e089ae9938935cd488d7d34f14502544a36607effd2239ff0e2494"
+dependencies = [
+ "bitpacking",
+]
+
+[[package]]
+name = "tantivy-columnar"
+version = "0.5.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "6300428e0c104c4f7db6f95b466a6f5c1b9aece094ec57cdd365337908dc7344"
+dependencies = [
+ "downcast-rs",
+ "fastdivide",
+ "itertools 0.14.0",
+ "serde",
+ "tantivy-bitpacker",
+ "tantivy-common",
+ "tantivy-sstable",
+ "tantivy-stacker",
+]
+
+[[package]]
+name = "tantivy-common"
+version = "0.9.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "e91b6ea6090ce03dc72c27d0619e77185d26cc3b20775966c346c6d4f7e99d7f"
+dependencies = [
+ "async-trait",
+ "byteorder",
+ "ownedbytes",
+ "serde",
+ "time",
+]
+
+[[package]]
+name = "tantivy-fst"
+version = "0.5.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "d60769b80ad7953d8a7b2c70cdfe722bbcdcac6bccc8ac934c40c034d866fc18"
+dependencies = [
+ "byteorder",
+ "regex-syntax",
+ "utf8-ranges",
+]
+
+[[package]]
+name = "tantivy-query-grammar"
+version = "0.24.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "e810cdeeebca57fc3f7bfec5f85fdbea9031b2ac9b990eb5ff49b371d52bbe6a"
+dependencies = [
+ "nom 7.1.3",
+ "serde",
+ "serde_json",
+]
+
+[[package]]
+name = "tantivy-sstable"
+version = "0.5.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "709f22c08a4c90e1b36711c1c6cad5ae21b20b093e535b69b18783dd2cb99416"
+dependencies = [
+ "futures-util",
+ "itertools 0.14.0",
+ "tantivy-bitpacker",
+ "tantivy-common",
+ "tantivy-fst",
+ "zstd",
+]
+
+[[package]]
+name = "tantivy-stacker"
+version = "0.5.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "2bcdebb267671311d1e8891fd9d1301803fdb8ad21ba22e0a30d0cab49ba59c1"
+dependencies = [
+ "murmurhash32",
+ "rand_distr 0.4.3",
+ "tantivy-common",
+]
+
+[[package]]
+name = "tantivy-tokenizer-api"
+version = "0.5.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "dfa942fcee81e213e09715bbce8734ae2180070b97b33839a795ba1de201547d"
+dependencies = [
+ "serde",
+]
+
+[[package]]
+name = "tap"
+version = "1.0.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "55937e1799185b12863d447f42597ed69d9928686b8d88a1df17376a097d8369"
+
 [[package]]
 name = "target-lexicon"
 version = "0.13.5"
@@ -4747,7 +7984,7 @@ checksum = "4fee6c4efc90059e10f81e6d42c60a18f76588c3d74cb83a0b242a2b6c7504c1"
 dependencies = [
  "proc-macro2",
  "quote",
- "syn",
+ "syn 2.0.117",
 ]
 
 [[package]]
@@ -4758,7 +7995,16 @@ checksum = "ebc4ee7f67670e9b64d05fa4253e753e016c6c95ff35b89b7941d6b856dec1d5"
 dependencies = [
  "proc-macro2",
  "quote",
- "syn",
+ "syn 2.0.117",
+]
+
+[[package]]
+name = "thread-tree"
+version = "0.3.3"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "ffbd370cb847953a25954d9f63e14824a36113f8c72eecf6eccef5dc4b45d630"
+dependencies = [
+ "crossbeam-channel",
 ]
 
 [[package]]
@@ -4778,7 +8024,7 @@ checksum = "7e54bc85fc7faa8bc175c4bab5b92ba8d9a3ce893d0e9f42cc455c8ab16a9e09"
 dependencies = [
  "byteorder",
  "integer-encoding",
- "ordered-float",
+ "ordered-float 2.10.1",
 ]
 
 [[package]]
@@ -4831,6 +8077,21 @@ dependencies = [
  "zerovec",
 ]
 
+[[package]]
+name = "tinyvec"
+version = "1.11.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "3e61e67053d25a4e82c844e8424039d9745781b3fc4f32b8d55ed50f5f667ef3"
+dependencies = [
+ "tinyvec_macros",
+]
+
+[[package]]
+name = "tinyvec_macros"
+version = "0.1.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "1f3ccbac311fea05f86f61904b462b55fb3df8837a366dfc601a0161d0532f20"
+
 [[package]]
 name = "tokio"
 version = "1.51.0"
@@ -4855,7 +8116,17 @@ checksum = "385a6cb71ab9ab790c5fe8d67f1645e6c450a7ce006a33de03daa956cf70a496"
 dependencies = [
  "proc-macro2",
  "quote",
- "syn",
+ "syn 2.0.117",
+]
+
+[[package]]
+name = "tokio-rustls"
+version = "0.26.4"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "1729aa945f29d91ba541258c8df89027d5792d85a8841fb65e8bf0f4ede4ef61"
+dependencies = [
+ "rustls",
+ "tokio",
 ]
 
 [[package]]
@@ -4954,15 +8225,15 @@ dependencies = [
  "base64",
  "bytes",
  "h2",
- "http",
- "http-body",
+ "http 1.4.0",
+ "http-body 1.0.1",
  "http-body-util",
  "hyper",
  "hyper-timeout",
  "hyper-util",
  "percent-encoding",
  "pin-project",
- "prost",
+ "prost 0.13.5",
  "socket2 0.5.10",
  "tokio",
  "tokio-stream",
@@ -4980,10 +8251,10 @@ checksum = "9557ce109ea773b399c9b9e5dca39294110b74f1f342cb347a80d1fce8c26a11"
 dependencies = [
  "prettyplease",
  "proc-macro2",
- "prost-build",
- "prost-types",
+ "prost-build 0.13.5",
+ "prost-types 0.13.5",
  "quote",
- "syn",
+ "syn 2.0.117",
 ]
 
 [[package]]
@@ -5016,6 +8287,30 @@ dependencies = [
  "futures-util",
  "pin-project-lite",
  "sync_wrapper",
+ "tokio",
+ "tower-layer",
+ "tower-service",
+]
+
+[[package]]
+name = "tower-http"
+version = "0.6.8"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "d4e6559d53cc268e5031cd8429d05415bc4cb4aefc4aa5d6cc35fbf5b924a1f8"
+dependencies = [
+ "async-compression",
+ "bitflags 2.11.0",
+ "bytes",
+ "futures-core",
+ "futures-util",
+ "http 1.4.0",
+ "http-body 1.0.1",
+ "http-body-util",
+ "iri-string",
+ "pin-project-lite",
+ "tokio",
+ "tokio-util",
+ "tower 0.5.3",
  "tower-layer",
  "tower-service",
 ]
@@ -5063,7 +8358,7 @@ checksum = "7490cfa5ec963746568740651ac6781f701c9c5ea257c58e057f3ba8cf69e8da"
 dependencies = [
  "proc-macro2",
  "quote",
- "syn",
+ "syn 2.0.117",
 ]
 
 [[package]]
@@ -5116,6 +8411,9 @@ name = "twox-hash"
 version = "2.1.2"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "9ea3136b675547379c4bd395ca6b938e5ad3c3d20fad76e7fe85f9e0d011419c"
+dependencies = [
+ "rand 0.9.2",
+]
 
 [[package]]
 name = "typenum"
@@ -5165,6 +8463,12 @@ version = "0.2.11"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "673aac59facbab8a9007c7f6108d11f63b603f7cabff99fabf650fea5c32b861"
 
+[[package]]
+name = "untrusted"
+version = "0.9.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "8ecb6da28b8a351d773b68d5825ac39017e680750f980f3a1a85cd8dd28a47c1"
+
 [[package]]
 name = "unty"
 version = "0.0.4"
@@ -5183,6 +8487,18 @@ dependencies = [
  "serde",
 ]
 
+[[package]]
+name = "urlencoding"
+version = "2.1.3"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "daf8dba3b7eb870caf1ddeed7bc9d2a049f3cfdfae7cb521b087cc33ae4c49da"
+
+[[package]]
+name = "utf8-ranges"
+version = "1.0.5"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "7fcfc827f90e53a02eaef5e535ee14266c1d569214c6aa70133a624d8a3164ba"
+
 [[package]]
 name = "utf8_iter"
 version = "1.0.4"
@@ -5203,6 +8519,7 @@ checksum = "5ac8b6f42ead25368cf5b098aeb3dc8a1a2c05a3eee8a9a1a68c640edbfc79d9"
 dependencies = [
  "getrandom 0.4.2",
  "js-sys",
+ "serde_core",
  "wasm-bindgen",
 ]
 
@@ -5230,6 +8547,12 @@ version = "0.0.18"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "051eb1abcf10076295e815102942cc58f9d5e3b4560e46e53c21e8ff6f3af7b1"
 
+[[package]]
+name = "vsimd"
+version = "0.8.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "5c3082ca00d5a5ef149bb8b555a72ae84c9c59f7250f013ac822ac2e49b19c64"
+
 [[package]]
 name = "walkdir"
 version = "2.5.0"
@@ -5315,7 +8638,7 @@ dependencies = [
  "bumpalo",
  "proc-macro2",
  "quote",
- "syn",
+ "syn 2.0.117",
  "wasm-bindgen-shared",
 ]
 
@@ -5391,6 +8714,19 @@ dependencies = [
  "wasmparser 0.244.0",
 ]
 
+[[package]]
+name = "wasm-streams"
+version = "0.4.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "15053d8d85c7eccdbefef60f06769760a563c7f0a9d6902a13d35c7800b0ad65"
+dependencies = [
+ "futures-util",
+ "js-sys",
+ "wasm-bindgen",
+ "wasm-bindgen-futures",
+ "web-sys",
+]
+
 [[package]]
 name = "wasmparser"
 version = "0.243.0"
@@ -5551,7 +8887,7 @@ dependencies = [
  "anyhow",
  "proc-macro2",
  "quote",
- "syn",
+ "syn 2.0.117",
  "wasmtime-internal-component-util",
  "wasmtime-internal-wit-bindgen",
  "wit-parser 0.243.0",
@@ -5665,7 +9001,7 @@ checksum = "70f8b9796a3f0451a7b702508b303d654de640271ac80287176de222f187a237"
 dependencies = [
  "proc-macro2",
  "quote",
- "syn",
+ "syn 2.0.117",
 ]
 
 [[package]]
@@ -5793,6 +9129,15 @@ dependencies = [
  "wasm-bindgen",
 ]
 
+[[package]]
+name = "webpki-roots"
+version = "1.0.7"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "52f5ee44c96cf55f1b349600768e3ece3a8f26010c05265ab73f945bb1a2eb9d"
+dependencies = [
+ "rustls-pki-types",
+]
+
 [[package]]
 name = "wiggle"
 version = "41.0.4"
@@ -5817,7 +9162,7 @@ dependencies = [
  "heck",
  "proc-macro2",
  "quote",
- "syn",
+ "syn 2.0.117",
  "witx",
 ]
 
@@ -5829,7 +9174,7 @@ checksum = "fea2aea744eded58ae092bf57110c27517dab7d5a300513ff13897325c5c5021"
 dependencies = [
  "proc-macro2",
  "quote",
- "syn",
+ "syn 2.0.117",
  "wiggle-generate",
 ]
 
@@ -5905,7 +9250,7 @@ checksum = "053e2e040ab57b9dc951b72c264860db7eb3b0200ba345b4e4c3b14f67855ddf"
 dependencies = [
  "proc-macro2",
  "quote",
- "syn",
+ "syn 2.0.117",
 ]
 
 [[package]]
@@ -5916,7 +9261,7 @@ checksum = "3f316c4a2570ba26bbec722032c4099d8c8bc095efccdc15688708623367e358"
 dependencies = [
  "proc-macro2",
  "quote",
- "syn",
+ "syn 2.0.117",
 ]
 
 [[package]]
@@ -6163,7 +9508,7 @@ dependencies = [
  "heck",
  "indexmap 2.13.1",
  "prettyplease",
- "syn",
+ "syn 2.0.117",
  "wasm-metadata",
  "wit-bindgen-core",
  "wit-component",
@@ -6179,7 +9524,7 @@ dependencies = [
  "prettyplease",
  "proc-macro2",
  "quote",
- "syn",
+ "syn 2.0.117",
  "wit-bindgen-core",
  "wit-bindgen-rust",
 ]
@@ -6257,6 +9602,21 @@ version = "0.6.3"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "1ffae5123b2d3fc086436f8834ae3ab053a283cfac8fe0a0b8eaae044768a4c4"
 
+[[package]]
+name = "wyz"
+version = "0.5.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "05f360fc0b24296329c78fda852a1e9ae82de9cf7b27dae4b7f62f118f77b9ed"
+dependencies = [
+ "tap",
+]
+
+[[package]]
+name = "xmlparser"
+version = "0.13.6"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "66fee0b777b0f5ac1c69bb06d361268faafa61cd4682ae064a171c16c433e9e4"
+
 [[package]]
 name = "xxhash-rust"
 version = "0.8.15"
@@ -6291,7 +9651,7 @@ checksum = "de844c262c8848816172cef550288e7dc6c7b7814b4ee56b3e1553f275f1858e"
 dependencies = [
  "proc-macro2",
  "quote",
- "syn",
+ "syn 2.0.117",
  "synstructure",
 ]
 
@@ -6312,7 +9672,7 @@ checksum = "70e3cd084b1788766f53af483dd21f93881ff30d7320490ec3ef7526d203bad4"
 dependencies = [
  "proc-macro2",
  "quote",
- "syn",
+ "syn 2.0.117",
 ]
 
 [[package]]
@@ -6332,10 +9692,16 @@ checksum = "11532158c46691caf0f2593ea8358fed6bbf68a0315e80aae9bd41fbade684a1"
 dependencies = [
  "proc-macro2",
  "quote",
- "syn",
+ "syn 2.0.117",
  "synstructure",
 ]
 
+[[package]]
+name = "zeroize"
+version = "1.8.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "b97154e67e32c85465826e8bcc1c59429aaaf107c1e4a9e53c8d8ccd5eff88d0"
+
 [[package]]
 name = "zerotrie"
 version = "0.2.4"
@@ -6366,7 +9732,7 @@ checksum = "625dc425cab0dca6dc3c3319506e6593dcb08a9f387ea3b284dbd52a92c40555"
 dependencies = [
  "proc-macro2",
  "quote",
- "syn",
+ "syn 2.0.117",
 ]
 
 [[package]]
diff --git a/Cargo.toml b/Cargo.toml
index 531601d0..eebf2a6c 100644
--- a/Cargo.toml
+++ b/Cargo.toml
@@ -52,6 +52,8 @@ arrow-array = "55"
 arrow-ipc = "55"
 arrow-schema = { version = "55", features = ["serde"] }
 parquet = "55"
+object_store = { version = "0.12.5", features = ["aws"] }
+bytes = "1"
 futures = "0.3"
 serde_json_path = "0.7"
 xxhash-rust = { version = "0.8", features = ["xxh3"] }
@@ -63,6 +65,7 @@ itertools = "0.14"
 strum = { version = "0.26", features = ["derive"] }
 
 arrow-json = {version = '55.2.0'}
+apache-avro = "0.21"
 datafusion = {git = 'https://github.com/FunctionStream/datafusion', branch = '48.0.1/fs'}
 datafusion-common = {git = 'https://github.com/FunctionStream/datafusion', branch = '48.0.1/fs'}
 datafusion-execution = {git = 'https://github.com/FunctionStream/datafusion', branch = '48.0.1/fs'}
@@ -74,6 +77,9 @@ sqlparser = { git = "https://github.com/FunctionStream/sqlparser-rs", branch = "
 
 ahash = "0.8"
 governor = "0.8.0"
+lance = { version = "4.0.0", default-features = false, features = ["aws"] }
+arrow-array-lance = { package = "arrow-array", version = "57.3.0" }
+arrow-ipc-lance = { package = "arrow-ipc", version = "57.3.0" }
 
 [features]
 default = ["incremental-cache", "python"]
diff --git a/Makefile b/Makefile
index 87a2339a..78138dae 100644
--- a/Makefile
+++ b/Makefile
@@ -9,8 +9,6 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
-
-
 APP_NAME    := function-stream
 VERSION     := $(shell grep '^version' Cargo.toml | head -1 | awk -F '"' '{print $$2}')
 DATE        := $(shell date -u +"%Y-%m-%dT%H:%M:%SZ")
@@ -29,30 +27,22 @@ endif
 OS          := $(shell uname -s | tr '[:upper:]' '[:lower:]')
 OS_NAME     := $(shell uname -s)
 
-# 2. Configure RUSTFLAGS and target triple per platform
+# 2. Configure target triple per platform
 DIST_ROOT   := dist
 ifeq ($(OS_NAME), Linux)
     TRIPLE := $(ARCH)-unknown-linux-gnu
     STATIC_FLAGS :=
 else ifeq ($(OS_NAME), Darwin)
-    # macOS: strip symbols but keep dynamic linking (Apple system restriction)
     TRIPLE := $(ARCH)-apple-darwin
     STATIC_FLAGS :=
 else ifneq (,$(findstring MINGW,$(OS_NAME))$(findstring MSYS,$(OS_NAME)))
-    # Windows (Git Bash / MSYS2): static-link MSVC runtime
     TRIPLE := $(ARCH)-pc-windows-msvc
     STATIC_FLAGS := -C target-feature=+crt-static
 else
-    # Fallback
     TRIPLE := $(ARCH)-unknown-linux-gnu
     STATIC_FLAGS :=
 endif
 
-# 3. Aggressive optimization flags
-# opt-level=z  : size-oriented, minimize binary footprint
-# strip=symbols: remove debug symbol table at link time
-# Note: panic=abort is intentionally omitted to preserve stack unwinding
-#       for better fault tolerance in the streaming runtime
 OPTIMIZE_FLAGS := -C opt-level=z -C strip=symbols $(STATIC_FLAGS)
 
 TARGET_DIR  := target/$(TRIPLE)/release
@@ -110,7 +100,7 @@ help:
 	(printf "$(C_Y)[!] Auto-installing target toolchain for $(OS_NAME): $(TRIPLE)$(C_0)\n" && \
 	 rustup target add $(TRIPLE))
 
-# 5. Build targets (depend on .ensure-target for automatic toolchain setup)
+# 5. Build targets
 build: .check-env .ensure-target .build-wasm
 	$(call log,BUILD,Rust Full [$(OS_NAME) / $(TRIPLE)])
 	@RUSTFLAGS="$(OPTIMIZE_FLAGS)" \
@@ -128,24 +118,24 @@ build: .check-env .ensure-target .build-wasm
 
 build-lite: .check-env .ensure-target
 	$(call log,BUILD,Rust Lite [$(OS_NAME) / $(TRIPLE)])
-	@RUSTFLAGS="$(OPTIMIZE_FLAGS)" \
+	@RUSTFLAGS="$(INDUSTRIAL_RUSTFLAGS)" \
 	cargo build --release \
-		--target $(TRIPLE) \
-		--no-default-features \
-		--features incremental-cache \
-		--quiet
+	   --target $(TRIPLE) \
+	   --no-default-features \
+	   --features incremental-cache \
+	   --quiet
 	$(call log,BUILD,CLI for dist)
-	@RUSTFLAGS="$(OPTIMIZE_FLAGS)" \
+	@RUSTFLAGS="$(INDUSTRIAL_RUSTFLAGS)" \
 	cargo build --release \
-		--target $(TRIPLE) \
-		-p function-stream-cli \
-		--quiet
+	   --target $(TRIPLE) \
+	   -p function-stream-cli \
+	   --quiet
 	$(call success,Target: $(TARGET_DIR)/$(APP_NAME) $(TARGET_DIR)/cli)
 
 .build-wasm:
 	$(call log,WASM,Building Python Runtime using $(PYTHON_EXEC))
 	@cd $(PYTHON_ROOT)/functionstream-runtime && \
-		PYTHONPATH=../functionstream-api:../functionstream-api-advanced ../../$(PYTHON_EXEC) build.py > /dev/null
+	   PYTHONPATH=../functionstream-api:../functionstream-api-advanced ../../$(PYTHON_EXEC) build.py > /dev/null
 	@[ -f "$(WASM_SOURCE)" ] || (printf "$(C_R)[X] WASM Build Failed$(C_0)\n" && exit 1)
 
 dist: build
@@ -223,9 +213,9 @@ docker:
 docker-run:
 	$(call log,DOCKER,Starting Container)
 	@docker run --rm -it \
-		-p 8080:8080 \
-		-v $(CURDIR)/logs:/app/logs \
-		$(IMAGE_NAME)
+	   -p 8080:8080 \
+	   -v $(CURDIR)/logs:/app/logs \
+	   $(IMAGE_NAME)
 
 docker-push:
 	$(call log,DOCKER,Pushing $(IMAGE_NAME))
diff --git a/README-zh.md b/README-zh.md
index 05fd5fc2..5d38c020 100644
--- a/README-zh.md
+++ b/README-zh.md
@@ -204,6 +204,7 @@ function-stream-<version>/
 | 文档                                                                     | 描述                       |
 |------------------------------------------------------------------------|--------------------------|
 | [Streaming SQL 使用指南](docs/streaming-sql-guide-zh.md)                   | 声明式 SQL 实时流处理指南         |
+| [Streaming SQL Source/Sink 文档](docs/streaming-sql/README-zh.md)          | Source/Sink 能力与参数速查         |
 | [连接器、格式与类型参考](docs/connectors-and-formats-zh.md)                       | 支持的 Source/Sink、格式与数据类型  |
 | [服务端配置与运维指南](docs/server-configuration-zh.md)                          | 服务端配置与运维操作              |
 | [Function 任务配置规范](docs/function-configuration-zh.md)        | 任务定义规范           |
diff --git a/docs/streaming-sql-guide-zh.md b/docs/streaming-sql-guide-zh.md
index 5721971c..216bf557 100644
--- a/docs/streaming-sql-guide-zh.md
+++ b/docs/streaming-sql-guide-zh.md
@@ -90,6 +90,8 @@ flowchart LR
 | **水位线 (Watermark)** | `AS <column> - INTERVAL ...` | 对**迟到、乱序**数据的容忍度；时间推进由水位线驱动，**过度迟到**的事件会被安全丢弃。 |
 
 > **完整参考**：支持的连接器、数据格式和 SQL 数据类型，请参阅 [连接器、格式与类型参考](connectors-and-formats-zh.md)。
+>
+> **Source / Sink 专项说明**：请参阅 [Streaming SQL Connector 文档](streaming-sql/README-zh.md)（含 Kafka Source 与 filesystem/s3/delta/iceberg/lanceDB Sink）。
 
 ---
 
diff --git a/docs/streaming-sql/README-zh.md b/docs/streaming-sql/README-zh.md
new file mode 100644
index 00000000..563588f1
--- /dev/null
+++ b/docs/streaming-sql/README-zh.md
@@ -0,0 +1,13 @@
+# Streaming SQL Connector 文档
+
+本目录提供 Streaming SQL 的 Source / Sink 专项文档，推荐配合 `CREATE STREAMING TABLE ... AS SELECT ...` 使用。
+
+## 目录
+
+- [Source 文档](Source/README-zh.md)
+- [Sink 文档](Sink/README-zh.md)
+
+## 使用建议
+
+1. 先用 `CREATE TABLE ... WITH (...)` 注册 Source（当前仅 Kafka）。
+2. 再用 `CREATE STREAMING TABLE ... WITH (...) AS SELECT ...` 创建持续运行的 Pipeline 并写入 Sink。
diff --git a/docs/streaming-sql/README.md b/docs/streaming-sql/README.md
new file mode 100644
index 00000000..8c3777cf
--- /dev/null
+++ b/docs/streaming-sql/README.md
@@ -0,0 +1,14 @@
+# Streaming SQL Connector Docs
+
+This directory contains Source/Sink focused docs for Streaming SQL, intended to be used with `CREATE STREAMING TABLE ... AS SELECT ...`.
+
+## Index
+
+- [Source Docs](Source/README.md)
+- [Sink Docs](Sink/README.md)
+
+## Recommended workflow
+
+1. Register sources using `CREATE TABLE ... WITH (...)` (currently Kafka source).
+2. Build a continuous pipeline using `CREATE STREAMING TABLE ... WITH (...) AS SELECT ...` and write to sinks.
+
diff --git a/docs/streaming-sql/Sink/README-zh.md b/docs/streaming-sql/Sink/README-zh.md
new file mode 100644
index 00000000..56e9af97
--- /dev/null
+++ b/docs/streaming-sql/Sink/README-zh.md
@@ -0,0 +1,31 @@
+# Streaming SQL Sink 文档
+
+本目录聚焦 Streaming SQL 的下游写出能力（Sink）。
+
+## 支持矩阵
+
+| Connector | 支持格式 |
+
+|---|---|
+| `kafka` | `json` / `raw_string` / `raw_bytes`（沿用 Kafka Sink 编码能力） |
+| `filesystem` | `csv` / `parquet` / `json`(JSONL) / `avro` / `orc` |
+| `s3` | `csv` / `parquet` |
+| `delta` | `csv` / `parquet` / `json`(JSONL) / `avro` / `orc` |
+| `iceberg` | `csv` / `parquet` |
+| `lanceDB` | `lance` |
+
+## 文档列表
+
+- [Kafka Sink](kafka-sink-zh.md)
+- [Filesystem Sink](filesystem-sink-zh.md)
+- [S3 Sink](s3-sink-zh.md)
+- [Delta Sink](delta-sink-zh.md)
+- [Iceberg Sink](iceberg-sink-zh.md)
+- [LanceDB Sink](lancedb-sink-zh.md)
+
+## 通用约定
+
+- 在 `CREATE STREAMING TABLE ... WITH (...) AS SELECT ...` 中通过 `WITH` 指定 `connector` 与 `format`。
+- Sink 场景建议显式指定 `type='sink'`。
+- 仅 `lanceDB` connector 允许 `format='lance'`；其余 Sink connector 不支持 `lance`。
+- `format='json'` 的文件类 Sink 输出为 JSON Lines（NDJSON，`.jsonl`）。
diff --git a/docs/streaming-sql/Sink/README.md b/docs/streaming-sql/Sink/README.md
new file mode 100644
index 00000000..63d73458
--- /dev/null
+++ b/docs/streaming-sql/Sink/README.md
@@ -0,0 +1,31 @@
+# Streaming SQL Sink Docs
+
+This directory documents sink connectors for Streaming SQL.
+
+## Support matrix
+
+| Connector | Supported formats |
+|---|---|
+| `kafka` | `json` / `raw_string` / `raw_bytes` |
+| `filesystem` | `csv` / `parquet` / `json`(JSONL) / `avro` / `orc` |
+| `s3` | `csv` / `parquet` |
+| `delta` | `csv` / `parquet` / `json`(JSONL) / `avro` / `orc` |
+| `iceberg` | `csv` / `parquet` |
+| `lanceDB` | `lance` |
+
+## Documents
+
+- [Kafka Sink](kafka-sink.md)
+- [Filesystem Sink](filesystem-sink.md)
+- [S3 Sink](s3-sink.md)
+- [Delta Sink](delta-sink.md)
+- [Iceberg Sink](iceberg-sink.md)
+- [LanceDB Sink](lancedb-sink.md)
+
+## Notes
+
+- Configure sink connectors via `WITH (...)` in `CREATE STREAMING TABLE ... AS SELECT ...`.
+- Use `type='sink'` explicitly for sink tables.
+- Only `lanceDB` accepts `format='lance'`.
+- For file-like sinks, `format='json'` is written as JSON Lines (NDJSON, `.jsonl`).
+
diff --git a/docs/streaming-sql/Sink/delta-sink-zh.md b/docs/streaming-sql/Sink/delta-sink-zh.md
new file mode 100644
index 00000000..4e973b2c
--- /dev/null
+++ b/docs/streaming-sql/Sink/delta-sink-zh.md
@@ -0,0 +1,33 @@
+# Delta Sink
+
+`delta` connector 对应 Delta 数据湖写出通道。
+
+## 支持格式
+
+- `csv`
+- `parquet`
+- `json`（写出为 JSON Lines / NDJSON，文件后缀 `.jsonl`）
+- `avro`
+- `orc`
+
+## 常用 WITH 参数
+
+- `connector='delta'`
+- `type='sink'`
+- `format='csv'|'parquet'|'json'|'avro'|'orc'`
+- `path='/data/delta/orders'`（本地）或对象存储前缀
+- 可选 S3 参数：`s3.bucket` / `s3.region` / `s3.endpoint` / AKSK
+- `parquet.compression`（仅 `parquet`）
+
+## 示例（CREATE STREAMING TABLE）
+
+```sql
+CREATE STREAMING TABLE st_delta_parquet
+WITH (
+  connector='delta',
+  type='sink',
+  format='parquet',
+  path='/tmp/delta_orders'
+) AS
+SELECT * FROM src_kafka_orders;
+```
diff --git a/docs/streaming-sql/Sink/delta-sink.md b/docs/streaming-sql/Sink/delta-sink.md
new file mode 100644
index 00000000..3d3bbd6c
--- /dev/null
+++ b/docs/streaming-sql/Sink/delta-sink.md
@@ -0,0 +1,21 @@
+# Delta Sink
+
+`delta` sink is the Delta Lake write path.
+
+## Supported formats
+
+- `csv`
+- `parquet`
+- `json` (written as JSON Lines / NDJSON, `.jsonl`)
+- `avro`
+- `orc`
+
+## Common `WITH` options
+
+- `connector='delta'`
+- `type='sink'`
+- `format='csv'|'parquet'|'json'|'avro'|'orc'`
+- `path='/data/delta/orders'` (local) or object-store prefix
+- Optional S3 options: `s3.bucket` / `s3.region` / `s3.endpoint` / access keys
+- `parquet.compression` (only for `parquet`)
+
diff --git a/docs/streaming-sql/Sink/filesystem-sink-zh.md b/docs/streaming-sql/Sink/filesystem-sink-zh.md
new file mode 100644
index 00000000..e60f4bf2
--- /dev/null
+++ b/docs/streaming-sql/Sink/filesystem-sink-zh.md
@@ -0,0 +1,33 @@
+# Filesystem Sink
+
+`filesystem` 用于将流数据落到本地文件系统目录。
+
+## 支持格式
+
+- `csv`
+- `parquet`
+- `json`（写出为 JSON Lines / NDJSON，文件后缀 `.jsonl`）
+- `avro`
+- `orc`
+
+## 常用 WITH 参数
+
+- `connector='filesystem'`
+- `type='sink'`
+- `format='csv'|'parquet'|'json'|'avro'|'orc'`
+- `path='/path/to/output'`（或 `sink.path`）
+- `parquet.compression`（仅 `parquet` 生效）
+
+## 示例（CREATE STREAMING TABLE）
+
+```sql
+CREATE STREAMING TABLE st_fs_parquet
+WITH (
+  connector='filesystem',
+  type='sink',
+  format='parquet',
+  path='/tmp/fs_orders',
+  'parquet.compression'='zstd'
+) AS
+SELECT * FROM src_kafka_orders;
+```
diff --git a/docs/streaming-sql/Sink/filesystem-sink.md b/docs/streaming-sql/Sink/filesystem-sink.md
new file mode 100644
index 00000000..3eb825c0
--- /dev/null
+++ b/docs/streaming-sql/Sink/filesystem-sink.md
@@ -0,0 +1,20 @@
+# Filesystem Sink
+
+`filesystem` sink writes streaming data to local files.
+
+## Supported formats
+
+- `csv`
+- `parquet`
+- `json` (written as JSON Lines / NDJSON, `.jsonl`)
+- `avro`
+- `orc`
+
+## Common `WITH` options
+
+- `connector='filesystem'`
+- `type='sink'`
+- `format='csv'|'parquet'|'json'|'avro'|'orc'`
+- `path='/path/to/output'` (or `sink.path`)
+- `parquet.compression` (only for `parquet`)
+
diff --git a/docs/streaming-sql/Sink/iceberg-sink-zh.md b/docs/streaming-sql/Sink/iceberg-sink-zh.md
new file mode 100644
index 00000000..09667059
--- /dev/null
+++ b/docs/streaming-sql/Sink/iceberg-sink-zh.md
@@ -0,0 +1,31 @@
+# Iceberg Sink
+
+`iceberg` connector 对应 Iceberg 数据湖写出通道。
+
+## 支持格式
+
+- `csv`
+- `parquet`
+
+## 常用 WITH 参数
+
+- `connector='iceberg'`
+- `type='sink'`
+- `format='csv'|'parquet'`
+- `path='/data/iceberg/orders'` 或对象存储前缀
+- 可选 S3 参数：`s3.bucket` / `s3.region` / `s3.endpoint` / AKSK
+- `parquet.compression`（仅 `parquet`）
+
+## 示例（CREATE STREAMING TABLE）
+
+```sql
+CREATE STREAMING TABLE st_iceberg_parquet
+WITH (
+  connector='iceberg',
+  type='sink',
+  format='parquet',
+  path='/tmp/iceberg_orders',
+  'parquet.compression'='zstd'
+) AS
+SELECT * FROM src_kafka_orders;
+```
diff --git a/docs/streaming-sql/Sink/iceberg-sink.md b/docs/streaming-sql/Sink/iceberg-sink.md
new file mode 100644
index 00000000..0cafe549
--- /dev/null
+++ b/docs/streaming-sql/Sink/iceberg-sink.md
@@ -0,0 +1,18 @@
+# Iceberg Sink
+
+`iceberg` sink is the Iceberg Lakehouse write path.
+
+## Supported formats
+
+- `csv`
+- `parquet`
+
+## Common `WITH` options
+
+- `connector='iceberg'`
+- `type='sink'`
+- `format='csv'|'parquet'`
+- `path='/data/iceberg/orders'` or object-store prefix
+- Optional S3 options: `s3.bucket` / `s3.region` / `s3.endpoint` / access keys
+- `parquet.compression` (only for `parquet`)
+
diff --git a/docs/streaming-sql/Sink/kafka-sink-zh.md b/docs/streaming-sql/Sink/kafka-sink-zh.md
new file mode 100644
index 00000000..e6d3ecbd
--- /dev/null
+++ b/docs/streaming-sql/Sink/kafka-sink-zh.md
@@ -0,0 +1,32 @@
+# Kafka Sink
+
+`kafka` sink 用于将流数据写入 Kafka topic。
+
+## 支持格式
+
+- `json`
+- `raw_string`
+- `raw_bytes`
+
+## 常用 WITH 参数
+
+- `connector='kafka'`
+- `type='sink'`
+- `topic='topic_name'`
+- `bootstrap.servers='host:9092'`
+- `format='json'|'raw_string'|'raw_bytes'`
+- `sink.commit.mode='at-least-once'|'exactly-once'`（可选）
+
+## 示例（CREATE STREAMING TABLE）
+
+```sql
+CREATE STREAMING TABLE st_kafka_json
+WITH (
+  connector='kafka',
+  type='sink',
+  topic='topic_out',
+  'bootstrap.servers'='127.0.0.1:9092',
+  format='json'
+) AS
+SELECT * FROM src_kafka_orders;
+```
diff --git a/docs/streaming-sql/Sink/kafka-sink.md b/docs/streaming-sql/Sink/kafka-sink.md
new file mode 100644
index 00000000..43fbdb72
--- /dev/null
+++ b/docs/streaming-sql/Sink/kafka-sink.md
@@ -0,0 +1,33 @@
+# Kafka Sink
+
+`kafka` sink writes streaming output records to Kafka topics.
+
+## Supported formats
+
+- `json`
+- `raw_string`
+- `raw_bytes`
+
+## Common `WITH` options
+
+- `connector='kafka'`
+- `type='sink'`
+- `topic='topic_name'`
+- `bootstrap.servers='host:9092'`
+- `format='json'|'raw_string'|'raw_bytes'`
+- `sink.commit.mode='at-least-once'|'exactly-once'` (optional)
+
+## Example
+
+```sql
+CREATE STREAMING TABLE st_kafka_json
+WITH (
+  connector='kafka',
+  type='sink',
+  topic='topic_out',
+  'bootstrap.servers'='127.0.0.1:9092',
+  format='json'
+) AS
+SELECT * FROM src_kafka_orders;
+```
+
diff --git a/docs/streaming-sql/Sink/lancedb-sink-zh.md b/docs/streaming-sql/Sink/lancedb-sink-zh.md
new file mode 100644
index 00000000..f2dc4d0d
--- /dev/null
+++ b/docs/streaming-sql/Sink/lancedb-sink-zh.md
@@ -0,0 +1,28 @@
+# LanceDB Sink
+
+`lanceDB` 是 Lance 数据集专用写出 connector。
+
+## 支持格式
+
+- 仅支持 `lance`
+
+## 常用 WITH 参数
+
+- `connector='lanceDB'`
+- `type='sink'`
+- `format='lance'`
+- `path='/data/lance/orders'`（本地目录）或对象存储前缀
+- 可选 S3 参数：`s3.bucket` / `s3.region` / `s3.endpoint` / AKSK
+
+## 示例（CREATE STREAMING TABLE）
+
+```sql
+CREATE STREAMING TABLE st_lancedb
+WITH (
+  connector='lanceDB',
+  type='sink',
+  format='lance',
+  path='/tmp/lance_orders'
+) AS
+SELECT * FROM src_kafka_orders;
+```
diff --git a/docs/streaming-sql/Sink/lancedb-sink.md b/docs/streaming-sql/Sink/lancedb-sink.md
new file mode 100644
index 00000000..d1e7fa74
--- /dev/null
+++ b/docs/streaming-sql/Sink/lancedb-sink.md
@@ -0,0 +1,16 @@
+# LanceDB Sink
+
+`lanceDB` is a dedicated sink connector for Lance datasets.
+
+## Supported formats
+
+- `lance` only
+
+## Common `WITH` options
+
+- `connector='lanceDB'`
+- `type='sink'`
+- `format='lance'`
+- `path='/data/lance/orders'` (local directory) or object-store prefix
+- Optional S3 options: `s3.bucket` / `s3.region` / `s3.endpoint` / access keys
+
diff --git a/docs/streaming-sql/Sink/s3-sink-zh.md b/docs/streaming-sql/Sink/s3-sink-zh.md
new file mode 100644
index 00000000..34bdfc2f
--- /dev/null
+++ b/docs/streaming-sql/Sink/s3-sink-zh.md
@@ -0,0 +1,35 @@
+# S3 Sink
+
+`s3` 用于将流数据写入对象存储（AWS S3 / S3 兼容存储）。
+
+## 支持格式
+
+- `csv`
+- `parquet`
+
+## 常用 WITH 参数
+
+- `connector='s3'`
+- `type='sink'`
+- `format='csv'|'parquet'`
+- `path='prefix/path'`
+- `s3.bucket='your-bucket'`
+- `s3.region='us-east-1'`
+- `s3.endpoint='http://minio:9000'`（可选，S3 兼容）
+- `s3.access_key_id` / `s3.secret_access_key` / `s3.session_token`（可选）
+- `parquet.compression`（仅 `parquet`）
+
+## 示例（CREATE STREAMING TABLE）
+
+```sql
+CREATE STREAMING TABLE st_s3_csv
+WITH (
+  connector='s3',
+  type='sink',
+  format='csv',
+  path='streaming/orders',
+  's3.bucket'='fs-dev',
+  's3.region'='us-east-1'
+) AS
+SELECT * FROM src_kafka_orders;
+```
diff --git a/docs/streaming-sql/Sink/s3-sink.md b/docs/streaming-sql/Sink/s3-sink.md
new file mode 100644
index 00000000..c4b8f062
--- /dev/null
+++ b/docs/streaming-sql/Sink/s3-sink.md
@@ -0,0 +1,21 @@
+# S3 Sink
+
+`s3` sink writes streaming data to object storage (AWS S3 or S3-compatible services).
+
+## Supported formats
+
+- `csv`
+- `parquet`
+
+## Common `WITH` options
+
+- `connector='s3'`
+- `type='sink'`
+- `format='csv'|'parquet'`
+- `path='prefix/path'`
+- `s3.bucket='your-bucket'`
+- `s3.region='us-east-1'`
+- `s3.endpoint='http://minio:9000'` (optional, for S3-compatible storage)
+- `s3.access_key_id` / `s3.secret_access_key` / `s3.session_token` (optional)
+- `parquet.compression` (only for `parquet`)
+
diff --git a/docs/streaming-sql/Source/README-zh.md b/docs/streaming-sql/Source/README-zh.md
new file mode 100644
index 00000000..e38a763f
--- /dev/null
+++ b/docs/streaming-sql/Source/README-zh.md
@@ -0,0 +1,7 @@
+# Streaming SQL Source 文档
+
+当前 Streaming SQL Source 仅支持 Kafka。通常先创建 Source 表，再被 `CREATE STREAMING TABLE ... AS SELECT ...` 引用。
+
+## 文档列表
+
+- [Kafka Source](kafka-source-zh.md)
diff --git a/docs/streaming-sql/Source/README.md b/docs/streaming-sql/Source/README.md
new file mode 100644
index 00000000..a84d3437
--- /dev/null
+++ b/docs/streaming-sql/Source/README.md
@@ -0,0 +1,6 @@
+# Streaming SQL Source Docs
+
+## Supported sources
+
+- [Kafka Source](kafka-source.md)
+
diff --git a/docs/streaming-sql/Source/kafka-source-zh.md b/docs/streaming-sql/Source/kafka-source-zh.md
new file mode 100644
index 00000000..e3da7597
--- /dev/null
+++ b/docs/streaming-sql/Source/kafka-source-zh.md
@@ -0,0 +1,36 @@
+# Kafka Source
+
+`kafka` source 用于从 Kafka topic 持续消费数据。
+
+## 支持格式
+
+- `json`
+- `raw_string`
+- `raw_bytes`
+
+## 常用 WITH 参数
+
+- `connector='kafka'`
+- `type='source'`（默认）
+- `topic='topic_in'`
+- `bootstrap.servers='host:9092'`
+- `group.id='consumer_group'`
+- `format='json'|'raw_string'|'raw_bytes'`
+- `scan.startup.mode='earliest'|'latest'|'group-offsets'`（可选）
+
+## 示例
+
+```sql
+CREATE TABLE src_kafka_json (
+  user_id BIGINT,
+  event STRING,
+  ts TIMESTAMP
+) WITH (
+  connector='kafka',
+  type='source',
+  topic='topic_in',
+  'bootstrap.servers'='127.0.0.1:9092',
+  'group.id'='fs_demo',
+  format='json'
+);
+```
diff --git a/docs/streaming-sql/Source/kafka-source.md b/docs/streaming-sql/Source/kafka-source.md
new file mode 100644
index 00000000..505a6f3e
--- /dev/null
+++ b/docs/streaming-sql/Source/kafka-source.md
@@ -0,0 +1,19 @@
+# Kafka Source
+
+`kafka` source is used to ingest streaming records from Kafka topics.
+
+## Supported formats
+
+- `json`
+- `raw_string`
+- `raw_bytes`
+
+## Common `WITH` options
+
+- `connector='kafka'`
+- `topic='topic_name'`
+- `bootstrap.servers='host:9092'`
+- `format='json'|'raw_string'|'raw_bytes'`
+- `scan.startup.mode='earliest'|'latest'` (optional)
+- `group.id='consumer_group'` (optional)
+
diff --git a/protocol/proto/function_stream_graph.proto b/protocol/proto/function_stream_graph.proto
index 48b68a2f..18903e4f 100644
--- a/protocol/proto/function_stream_graph.proto
+++ b/protocol/proto/function_stream_graph.proto
@@ -70,7 +70,11 @@ message ConnectorOp {
   oneof config {
     KafkaSourceConfig kafka_source = 6;
     KafkaSinkConfig kafka_sink = 7;
-    GenericConnectorConfig generic = 8;
+    FilesystemSinkConfig filesystem_sink = 9;
+    DeltaSinkConfig delta_sink = 10;
+    IcebergSinkConfig iceberg_sink = 11;
+    S3SinkConfig s3_sink = 12;
+    LanceDbSinkConfig lancedb_sink = 13;
   }
 }
 
@@ -103,9 +107,55 @@ message KafkaSinkConfig {
   optional string value_subject = 9;
 }
 
-// Fallback for non-Kafka connectors that are not yet strongly typed.
-message GenericConnectorConfig {
-  map<string, string> properties = 1;
+message FilesystemSinkConfig {
+  string path = 1;
+  SinkFormatProto format = 2;
+  optional ParquetCompressionProto parquet_compression = 3;
+  map<string, string> extra_properties = 100;
+  map<string, string> runtime_properties = 101;
+}
+
+message DeltaSinkConfig {
+  string path = 1;
+  SinkFormatProto format = 2;
+  optional ParquetCompressionProto parquet_compression = 3;
+  map<string, string> extra_properties = 100;
+  map<string, string> runtime_properties = 101;
+}
+
+message IcebergSinkConfig {
+  string path = 1;
+  SinkFormatProto format = 2;
+  optional ParquetCompressionProto parquet_compression = 3;
+  map<string, string> extra_properties = 100;
+  map<string, string> runtime_properties = 101;
+}
+
+message S3SinkConfig {
+  string path = 1;
+  SinkFormatProto format = 2;
+  string bucket = 3;
+  string region = 4;
+  optional string endpoint = 5;
+  optional string access_key_id = 6;
+  optional string secret_access_key = 7;
+  optional string session_token = 8;
+  optional ParquetCompressionProto parquet_compression = 9;
+  map<string, string> extra_properties = 100;
+  map<string, string> runtime_properties = 101;
+}
+
+message LanceDbSinkConfig {
+  string path = 1;
+  SinkFormatProto format = 2;
+  optional string s3_bucket = 3;
+  optional string s3_region = 4;
+  optional string s3_endpoint = 5;
+  optional string s3_access_key_id = 6;
+  optional string s3_secret_access_key = 7;
+  optional string s3_session_token = 8;
+  map<string, string> extra_properties = 100;
+  map<string, string> runtime_properties = 101;
 }
 
 // ─────────────────────── Kafka Auth ───────────────────────
@@ -189,6 +239,26 @@ enum KafkaSinkCommitMode {
   KAFKA_SINK_EXACTLY_ONCE = 1;
 }
 
+enum SinkFormatProto {
+  SINK_FORMAT_UNSPECIFIED = 0;
+  SINK_FORMAT_CSV = 1;
+  SINK_FORMAT_JSONL = 2;
+  SINK_FORMAT_AVRO = 3;
+  SINK_FORMAT_PARQUET = 4;
+  SINK_FORMAT_ORC = 5;
+  SINK_FORMAT_LANCE = 6;
+}
+
+enum ParquetCompressionProto {
+  PARQUET_COMPRESSION_UNSPECIFIED = 0;
+  PARQUET_COMPRESSION_UNCOMPRESSED = 1;
+  PARQUET_COMPRESSION_SNAPPY = 2;
+  PARQUET_COMPRESSION_GZIP = 3;
+  PARQUET_COMPRESSION_ZSTD = 4;
+  PARQUET_COMPRESSION_LZ4 = 5;
+  PARQUET_COMPRESSION_LZ4_RAW = 6;
+}
+
 message ValuePlanOperator {
   string name = 1;
   bytes physical_plan = 2;
diff --git a/protocol/proto/storage.proto b/protocol/proto/storage.proto
index fd021727..719d0f63 100644
--- a/protocol/proto/storage.proto
+++ b/protocol/proto/storage.proto
@@ -58,9 +58,12 @@ message KafkaSourceSubtaskCheckpoint {
   repeated KafkaPartitionOffset partitions = 4;
 }
 
-// Generic source checkpoint payload envelope (enum-like via oneof).
-message SourceCheckpointPayload {
-  oneof checkpoint {
+// Aggregated, epoch-aligned per-source checkpoint info persisted by the coordinator
+// (one entry per source subtask). The envelope decouples the catalog/meta-store from
+// any specific source implementation so new source types can be added by registering a
+// new `CheckpointAggregator` in the runtime without touching storage schema layouts.
+message SourceCheckpointInfo {
+  oneof info {
     KafkaSourceSubtaskCheckpoint kafka = 1;
   }
 }
@@ -81,10 +84,10 @@ message StreamingTableDefinition {
   // Updated by JobManager after all operators ACK. Used for crash recovery.
   uint64 latest_checkpoint_epoch = 6;
 
-  // Kafka source per-subtask offsets at the same committed epoch as `latest_checkpoint_epoch`.
-  // Populated by the runtime coordinator from source checkpoint ACKs. Optional `.bin` files under
-  // the job state dir may exist only for local recovery materialization from this field.
-  repeated KafkaSourceSubtaskCheckpoint kafka_source_checkpoints = 7;
+  // Source-agnostic per-subtask checkpoint entries aligned with `latest_checkpoint_epoch`.
+  // Populated by the runtime CheckpointCoordinator via the configured aggregator registry;
+  // storage no longer knows the internal layout of any particular source type.
+  repeated SourceCheckpointInfo source_checkpoints = 7;
 }
 
 // =============================================================================
diff --git a/src/config/global_config.rs b/src/config/global_config.rs
index dcfbcf5c..3c6eab67 100644
--- a/src/config/global_config.rs
+++ b/src/config/global_config.rs
@@ -25,6 +25,8 @@ pub const DEFAULT_STREAMING_RUNTIME_MEMORY_BYTES: u64 = 10 * 1024 * 1024;
 
 /// Default for [`StreamingConfig::operator_state_store_memory_bytes`] when unset. **5 MiB** per stateful operator cap.
 pub const DEFAULT_OPERATOR_STATE_STORE_MEMORY_BYTES: u64 = 5 * 1024 * 1024;
+/// Default sink in-memory buffer reservation. **64 MiB**.
+pub const DEFAULT_SINK_BUFFER_MEMORY_BYTES: u64 = 64 * 1024 * 1024;
 
 #[derive(Debug, Clone, Serialize, Deserialize, Default)]
 pub struct StreamingConfig {
diff --git a/src/coordinator/dataset/show_catalog_tables_result.rs b/src/coordinator/dataset/show_catalog_tables_result.rs
index 9811ff82..87c5c3ee 100644
--- a/src/coordinator/dataset/show_catalog_tables_result.rs
+++ b/src/coordinator/dataset/show_catalog_tables_result.rs
@@ -17,7 +17,8 @@ use arrow_schema::{DataType, Field, Schema};
 use datafusion::arrow::datatypes::Schema as DfSchema;
 
 use super::DataSet;
-use crate::sql::schema::table::Table as CatalogTable;
+use crate::sql::schema::catalog::ExternalTable;
+use crate::sql::schema::table::CatalogEntity;
 use crate::sql::schema::{catalog_table_row_detail, schema_columns_one_line};
 
 #[derive(Clone, Debug)]
@@ -30,7 +31,7 @@ pub struct ShowCatalogTablesResult {
 }
 
 impl ShowCatalogTablesResult {
-    pub fn from_tables(tables: &[Arc<CatalogTable>]) -> Self {
+    pub fn from_tables(tables: &[Arc<CatalogEntity>]) -> Self {
         let mut names = Vec::with_capacity(tables.len());
         let mut kinds = Vec::with_capacity(tables.len());
         let mut column_counts = Vec::with_capacity(tables.len());
@@ -39,18 +40,19 @@ impl ShowCatalogTablesResult {
 
         for t in tables {
             let schema = match t.as_ref() {
-                CatalogTable::ConnectorTable(source) | CatalogTable::LookupTable(source) => {
-                    source.produce_physical_schema()
-                }
-                CatalogTable::TableFromQuery { .. } => DfSchema::new(t.get_fields()),
+                CatalogEntity::ExternalConnector(b) => b.as_ref().produce_physical_schema(),
+                CatalogEntity::ComputedTable { .. } => DfSchema::new(t.get_fields()),
             };
             let ncols = schema.fields().len() as i32;
             names.push(t.name().to_string());
             kinds.push(
                 match t.as_ref() {
-                    CatalogTable::ConnectorTable(_) => "SOURCE",
-                    CatalogTable::LookupTable(_) => "LOOKUP",
-                    CatalogTable::TableFromQuery { .. } => "QUERY",
+                    CatalogEntity::ExternalConnector(b) => match b.as_ref() {
+                        ExternalTable::Source(_) => "SOURCE",
+                        ExternalTable::Sink(_) => "SINK",
+                        ExternalTable::Lookup(_) => "LOOKUP",
+                    },
+                    CatalogEntity::ComputedTable { .. } => "QUERY",
                 }
                 .to_string(),
             );
diff --git a/src/coordinator/execution/executor.rs b/src/coordinator/execution/executor.rs
index 6fb03134..22c5c1b7 100644
--- a/src/coordinator/execution/executor.rs
+++ b/src/coordinator/execution/executor.rs
@@ -21,9 +21,9 @@ use crate::coordinator::dataset::{
     ShowFunctionsResult, ShowStreamingTablesResult, empty_record_batch,
 };
 use crate::coordinator::plan::{
-    CreateFunctionPlan, CreatePythonFunctionPlan, CreateTablePlan, CreateTablePlanBody,
-    DropFunctionPlan, DropStreamingTablePlan, DropTablePlan, LookupTablePlan, PlanNode,
-    PlanVisitor, PlanVisitorContext, PlanVisitorResult, ShowCatalogTablesPlan,
+    CompileErrorPlan, CreateFunctionPlan, CreatePythonFunctionPlan, CreateTablePlan,
+    CreateTablePlanBody, DropFunctionPlan, DropStreamingTablePlan, DropTablePlan, LookupTablePlan,
+    PlanNode, PlanVisitor, PlanVisitorContext, PlanVisitorResult, ShowCatalogTablesPlan,
     ShowCreateStreamingTablePlan, ShowCreateTablePlan, ShowFunctionsPlan, ShowStreamingTablesPlan,
     StartFunctionPlan, StopFunctionPlan, StreamingTable, StreamingTableConnectorPlan,
 };
@@ -34,8 +34,9 @@ use crate::coordinator::streaming_table_options::{
 use crate::runtime::streaming::job::JobManager;
 use crate::runtime::streaming::protocol::control::StopMode;
 use crate::runtime::wasm::taskexecutor::TaskManager;
+use crate::sql::schema::catalog::ExternalTable;
 use crate::sql::schema::show_create_catalog_table;
-use crate::sql::schema::table::Table as CatalogTable;
+use crate::sql::schema::table::CatalogEntity;
 use crate::storage::stream_catalog::CatalogManager;
 
 #[derive(Error, Debug)]
@@ -89,6 +90,14 @@ impl Executor {
 }
 
 impl PlanVisitor for Executor {
+    fn visit_compile_error_plan(
+        &self,
+        plan: &CompileErrorPlan,
+        _context: &PlanVisitorContext,
+    ) -> PlanVisitorResult {
+        PlanVisitorResult::Execute(Err(ExecuteError::Validation(plan.message.clone())))
+    }
+
     fn visit_create_function(
         &self,
         plan: &CreateFunctionPlan,
@@ -275,13 +284,17 @@ impl PlanVisitor for Executor {
     ) -> PlanVisitorResult {
         let execute = || -> Result<ExecuteResult, ExecuteError> {
             let (table_name, if_not_exists, catalog_table) = match &plan.body {
-                CreateTablePlanBody::ConnectorSource {
-                    source_table,
+                CreateTablePlanBody::External {
+                    table,
                     if_not_exists,
                 } => {
-                    let table_name = source_table.name().to_string();
-                    let table_instance =
-                        CatalogTable::ConnectorTable(source_table.as_ref().clone());
+                    if matches!(table.as_ref(), ExternalTable::Sink(_)) {
+                        return Err(ExecuteError::Internal(
+                            "`CREATE TABLE` cannot produce a Sink; use `CREATE STREAMING TABLE ... AS SELECT`".into(),
+                        ));
+                    }
+                    let table_name = table.name().to_string();
+                    let table_instance = CatalogEntity::external(table.as_ref().clone());
                     (table_name, *if_not_exists, table_instance)
                 }
                 CreateTablePlanBody::DataFusion(_) => {
@@ -302,13 +315,12 @@ impl PlanVisitor for Executor {
                 .add_catalog_table(catalog_table)
                 .map_err(|e| {
                     ExecuteError::Internal(format!(
-                        "Failed to register connector source table '{}': {}",
-                        table_name, e
+                        "Failed to register external table '{table_name}': {e}"
                     ))
                 })?;
 
             Ok(ExecuteResult::ok(format!(
-                "Created connector source table '{table_name}'"
+                "Created external table '{table_name}'"
             )))
         };
 
@@ -356,6 +368,7 @@ impl PlanVisitor for Executor {
                     fs_program,
                     custom_interval,
                     None,
+                    vec![],
                 ))
             })
             .map_err(|e| ExecuteError::Internal(format!("Failed to submit streaming job: {e}")))?;
diff --git a/src/coordinator/plan/ast_utils.rs b/src/coordinator/plan/ast_utils.rs
new file mode 100644
index 00000000..d2432888
--- /dev/null
+++ b/src/coordinator/plan/ast_utils.rs
@@ -0,0 +1,91 @@
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+//! Pure AST inspection helpers.
+//!
+//! This module is deliberately stateless: it only knows how to pull data out
+//! of `sqlparser` AST nodes. Nothing here touches `StreamSchemaProvider`,
+//! connectors, Builders, or logical plans. Put any piece of code that would
+//! be "fine" to unit test on a naked AST here — and only here.
+
+use datafusion::common::{Result, plan_err};
+use datafusion::sql::sqlparser::ast::{Expr as SqlExpr, SqlOption, TableConstraint};
+
+use crate::sql::common::with_option_keys as opt;
+
+/// Namespace for AST extraction helpers.
+pub struct AstUtils;
+
+impl AstUtils {
+    /// Extract the single PRIMARY KEY column list from the constraint
+    /// clauses of a `CREATE TABLE`. Rejects multiple PRIMARY KEY declarations.
+    pub fn parse_primary_keys(constraints: &[TableConstraint]) -> Result<Vec<String>> {
+        let mut keys = None;
+        for constraint in constraints {
+            if let TableConstraint::PrimaryKey { columns, .. } = constraint {
+                if keys.is_some() {
+                    return plan_err!(
+                        "Constraint Violation: Multiple PRIMARY KEY constraints are forbidden"
+                    );
+                }
+                keys = Some(columns.iter().map(|ident| ident.value.clone()).collect());
+            }
+        }
+        Ok(keys.unwrap_or_default())
+    }
+
+    /// Extract the (at most one) `WATERMARK FOR col [AS expr]` clause from
+    /// the constraint list. The resulting tuple is `(column_name, opt_expr)`
+    /// so the caller can decide whether it is legal in its context.
+    pub fn parse_watermark_strategy(
+        constraints: &[TableConstraint],
+    ) -> Result<Option<(String, Option<SqlExpr>)>> {
+        let mut strategy = None;
+        for constraint in constraints {
+            if let TableConstraint::Watermark {
+                column_name,
+                watermark_expr,
+            } = constraint
+            {
+                if strategy.is_some() {
+                    return plan_err!(
+                        "Constraint Violation: Only a single WATERMARK FOR clause is permitted"
+                    );
+                }
+                strategy = Some((column_name.value.clone(), watermark_expr.clone()));
+            }
+        }
+        Ok(strategy)
+    }
+
+    /// True iff the WITH clause declares a `connector=` property. Used by
+    /// the router to decide whether to hand off to the external-table
+    /// compiler.
+    pub fn contains_connector_property(options: &[SqlOption]) -> bool {
+        options.iter().any(|o| match o {
+            SqlOption::KeyValue { key, .. } => key.value.eq_ignore_ascii_case(opt::CONNECTOR),
+            _ => false,
+        })
+    }
+
+    /// Peek at the declared `type` in the WITH clause without consuming it,
+    /// returning its lowercased value. Used by the router to split the
+    /// source- and sink-compile paths before either Builder runs.
+    pub fn peek_table_role(with_options: &[SqlOption]) -> Option<String> {
+        with_options.iter().find_map(|o| match o {
+            SqlOption::KeyValue { key, value } if key.value.eq_ignore_ascii_case(opt::TYPE) => {
+                Some(value.to_string().trim_matches('\'').to_ascii_lowercase())
+            }
+            _ => None,
+        })
+    }
+}
diff --git a/src/sql/schema/table_execution_unit.rs b/src/coordinator/plan/compile_error_plan.rs
similarity index 54%
rename from src/sql/schema/table_execution_unit.rs
rename to src/coordinator/plan/compile_error_plan.rs
index c23dda7a..52b6e481 100644
--- a/src/sql/schema/table_execution_unit.rs
+++ b/src/coordinator/plan/compile_error_plan.rs
@@ -10,24 +10,21 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
-use super::temporal_pipeline_config::TemporalPipelineConfig;
+use super::{PlanNode, PlanVisitor, PlanVisitorContext, PlanVisitorResult};
 
-#[derive(Debug, Clone)]
-pub struct EngineDescriptor {
-    pub engine_type: String,
-    pub raw_payload: String,
+#[derive(Debug)]
+pub struct CompileErrorPlan {
+    pub message: String,
 }
 
-#[derive(Debug, Clone, Copy, PartialEq, Eq, Hash)]
-pub enum SyncMode {
-    AppendOnly,
-    Incremental,
+impl CompileErrorPlan {
+    pub fn new(message: String) -> Self {
+        Self { message }
+    }
 }
 
-#[derive(Debug, Clone)]
-pub struct TableExecutionUnit {
-    pub label: String,
-    pub engine_meta: EngineDescriptor,
-    pub sync_mode: SyncMode,
-    pub temporal_offset: TemporalPipelineConfig,
+impl PlanNode for CompileErrorPlan {
+    fn accept(&self, visitor: &dyn PlanVisitor, context: &PlanVisitorContext) -> PlanVisitorResult {
+        visitor.visit_compile_error_plan(self, context)
+    }
 }
diff --git a/src/coordinator/plan/create_table_plan.rs b/src/coordinator/plan/create_table_plan.rs
index 11ae14a4..7bb47a53 100644
--- a/src/coordinator/plan/create_table_plan.rs
+++ b/src/coordinator/plan/create_table_plan.rs
@@ -12,16 +12,16 @@
 
 use datafusion::logical_expr::LogicalPlan;
 
-use crate::sql::schema::SourceTable;
+use crate::sql::schema::ExternalTable;
 
 use super::{PlanNode, PlanVisitor, PlanVisitorContext, PlanVisitorResult};
 
-/// Payload for [`CreateTablePlan`]: either a DataFusion DDL plan or a connector `CREATE TABLE` (no `AS SELECT`).
+/// DataFusion DDL, or `CREATE TABLE` with connector → [`ExternalTable`].
 #[derive(Debug, Clone)]
 pub enum CreateTablePlanBody {
     DataFusion(Box<LogicalPlan>),
-    ConnectorSource {
-        source_table: Box<SourceTable>,
+    External {
+        table: Box<ExternalTable>,
         if_not_exists: bool,
     },
 }
@@ -38,10 +38,11 @@ impl CreateTablePlan {
         }
     }
 
-    pub fn connector_source(source_table: SourceTable, if_not_exists: bool) -> Self {
+    /// [`ExternalTable`] from the DDL compiler.
+    pub fn external_table(table: ExternalTable, if_not_exists: bool) -> Self {
         Self {
-            body: CreateTablePlanBody::ConnectorSource {
-                source_table: Box::new(source_table),
+            body: CreateTablePlanBody::External {
+                table: Box::new(table),
                 if_not_exists,
             },
         }
diff --git a/src/coordinator/plan/ddl_compiler.rs b/src/coordinator/plan/ddl_compiler.rs
new file mode 100644
index 00000000..5c5af39e
--- /dev/null
+++ b/src/coordinator/plan/ddl_compiler.rs
@@ -0,0 +1,397 @@
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+//! Compiles `CREATE TABLE ... WITH (...)` into Source or Lookup tables.
+
+use std::time::Duration;
+
+use datafusion::arrow::datatypes::{DataType, Field, Schema};
+use datafusion::common::{DFSchema, Result, plan_datafusion_err, plan_err};
+use datafusion::error::DataFusionError;
+use datafusion::logical_expr::Expr;
+use datafusion::sql::TableReference;
+use datafusion::sql::planner::{PlannerContext, SqlToRel};
+use datafusion::sql::sqlparser::ast;
+use datafusion::sql::sqlparser::ast::CreateTable as SqlCreateTable;
+use datafusion_expr::ExprSchemable;
+use tracing::warn;
+
+use super::ast_utils::AstUtils;
+use crate::coordinator::tool::ConnectorOptions;
+use crate::sql::analysis::StreamSchemaProvider;
+use crate::sql::common::constants::{connection_table_role, connector_type, sql_field};
+use crate::sql::common::with_option_keys as opt;
+use crate::sql::common::{BadData, Format, Framing, JsonCompression, JsonFormat};
+use crate::sql::connector::registry::REGISTRY;
+use crate::sql::schema::ColumnDescriptor;
+use crate::sql::schema::catalog::{ExternalTable, LookupTable, SourceTable};
+use crate::sql::schema::data_encoding_format::DataEncodingFormat;
+use crate::sql::schema::table_role::{apply_adapter_specific_rules, validate_adapter_availability};
+use crate::sql::schema::temporal_pipeline_config::TemporalPipelineConfig;
+
+pub struct DdlCompiler<'a> {
+    schema_provider: &'a StreamSchemaProvider,
+}
+
+impl<'a> DdlCompiler<'a> {
+    pub fn new(schema_provider: &'a StreamSchemaProvider) -> Self {
+        Self { schema_provider }
+    }
+
+    pub fn compile(
+        &self,
+        stmt: &SqlCreateTable,
+        declared_role: Option<&str>,
+    ) -> Result<ExternalTable> {
+        Self::assert_ddl_flags(stmt)?;
+
+        match declared_role {
+            Some(connection_table_role::LOOKUP) => self.compile_lookup(stmt),
+            Some(connection_table_role::SOURCE) | None => self.compile_source(stmt),
+            Some(connection_table_role::SINK) => plan_err!(
+                "`CREATE TABLE ... WITH (type='sink')` is not supported; use `CREATE STREAMING TABLE ... AS SELECT`"
+            ),
+            Some(other) => {
+                plan_err!("Invalid connection type '{other}' — expected 'source' or 'lookup'")
+            }
+        }
+    }
+
+    fn compile_source(&self, stmt: &SqlCreateTable) -> Result<ExternalTable> {
+        let target_name = stmt.name.to_string();
+        let description = stmt
+            .comment
+            .clone()
+            .map(|c| c.to_string())
+            .unwrap_or_default();
+
+        let mut columns = self.extract_columns(stmt)?;
+
+        let mut options = ConnectorOptions::new(&stmt.with_options, &None)?;
+        let adapter_type = Self::extract_adapter(&mut options)?;
+        Self::assert_connector_match(&mut options, &adapter_type)?;
+        Self::absorb_type_option(&mut options, connection_table_role::SOURCE)?;
+
+        validate_adapter_availability(&adapter_type)?;
+
+        let pk_constraints = AstUtils::parse_primary_keys(&stmt.constraints)?;
+        let catalog_with_options = options.snapshot_for_catalog();
+
+        let format = Format::from_opts(&mut options)?;
+        Self::assert_format_compatibility(&format, &adapter_type)?;
+        let _framing = Framing::from_opts(&mut options)?;
+        let bad_data = BadData::from_opts(&mut options)?;
+
+        let encoding = DataEncodingFormat::from_format(format.as_ref());
+
+        columns = apply_adapter_specific_rules(&adapter_type, columns);
+        columns = encoding.apply_envelope(columns)?;
+
+        if encoding.supports_delta_updates() && pk_constraints.is_empty() {
+            return plan_err!("CDC source requires at least one PRIMARY KEY field");
+        }
+
+        let watermark = AstUtils::parse_watermark_strategy(&stmt.constraints)?;
+        let mut temporal_config = resolve_source_watermark(
+            &target_name,
+            &mut columns,
+            watermark,
+            &mut options,
+            self.schema_provider,
+        )?;
+
+        let idle_from_micros = options
+            .pull_opt_i64(opt::IDLE_MICROS)?
+            .filter(|t| *t > 0)
+            .map(|t| Duration::from_micros(t as u64));
+        let idle_from_duration = options.pull_opt_duration(opt::IDLE_TIME)?;
+        temporal_config.liveness_timeout = idle_from_micros.or(idle_from_duration);
+
+        let provider = REGISTRY.get_source(&adapter_type)?;
+        let connector_config = provider.build_source_config(&mut options, &format, bad_data)?;
+
+        Self::assert_options_fully_consumed(
+            &options,
+            connection_table_role::SOURCE,
+            &adapter_type,
+        )?;
+
+        Ok(ExternalTable::Source(SourceTable {
+            table_identifier: target_name,
+            adapter_type,
+            schema_specs: columns,
+            connector_config,
+            temporal_config,
+            key_constraints: pk_constraints,
+            payload_format: Some(encoding),
+            connection_format: format,
+            description,
+            catalog_with_options,
+            registry_id: None,
+            inferred_fields: None,
+        }))
+    }
+
+    fn compile_lookup(&self, stmt: &SqlCreateTable) -> Result<ExternalTable> {
+        let target_name = stmt.name.to_string();
+
+        if AstUtils::parse_watermark_strategy(&stmt.constraints)?.is_some() {
+            return plan_err!(
+                "Syntax Error: WATERMARK FOR cannot be defined on a Lookup table (`{}`)",
+                target_name
+            );
+        }
+
+        let description = stmt
+            .comment
+            .clone()
+            .map(|c| c.to_string())
+            .unwrap_or_default();
+
+        let columns = self.extract_columns(stmt)?;
+
+        let mut options = ConnectorOptions::new(&stmt.with_options, &None)?;
+        let adapter_type = Self::extract_adapter(&mut options)?;
+        Self::assert_connector_match(&mut options, &adapter_type)?;
+        Self::absorb_type_option(&mut options, connection_table_role::LOOKUP)?;
+
+        validate_adapter_availability(&adapter_type)?;
+
+        let pk_constraints = AstUtils::parse_primary_keys(&stmt.constraints)?;
+        let catalog_with_options = options.snapshot_for_catalog();
+
+        let connection_format = Format::from_opts(&mut options)?;
+        Self::assert_format_compatibility(&connection_format, &adapter_type)?;
+        let bad_data = BadData::from_opts(&mut options)?;
+
+        let lookup_cache_max_bytes = options.pull_opt_u64(opt::LOOKUP_CACHE_MAX_BYTES)?;
+        let lookup_cache_ttl = options.pull_opt_duration(opt::LOOKUP_CACHE_TTL)?;
+
+        let provider = REGISTRY.get_source(&adapter_type)?;
+        let connector_config =
+            provider.build_source_config(&mut options, &connection_format, bad_data)?;
+
+        Self::assert_options_fully_consumed(
+            &options,
+            connection_table_role::LOOKUP,
+            &adapter_type,
+        )?;
+
+        Ok(ExternalTable::Lookup(LookupTable {
+            table_identifier: target_name,
+            adapter_type,
+            schema_specs: columns,
+            connector_config,
+            key_constraints: pk_constraints,
+            lookup_cache_max_bytes,
+            lookup_cache_ttl,
+            connection_format,
+            description,
+            catalog_with_options,
+            registry_id: None,
+            inferred_fields: None,
+        }))
+    }
+
+    fn extract_adapter(options: &mut ConnectorOptions) -> Result<String> {
+        options.pull_opt_str(opt::CONNECTOR)?.ok_or_else(|| {
+            plan_datafusion_err!(
+                "Configuration Error: Missing required property '{}' in WITH clause",
+                opt::CONNECTOR
+            )
+        })
+    }
+
+    fn extract_columns(&self, stmt: &SqlCreateTable) -> Result<Vec<ColumnDescriptor>> {
+        let schema_compiler = datafusion::sql::planner::SqlToRel::new(self.schema_provider);
+        let arrow_schema = schema_compiler.build_schema(stmt.columns.clone())?;
+        Ok(arrow_schema
+            .fields()
+            .iter()
+            .map(|f| ColumnDescriptor::from((**f).clone()))
+            .collect())
+    }
+
+    fn assert_ddl_flags(stmt: &SqlCreateTable) -> Result<()> {
+        if stmt.query.is_some() {
+            return plan_err!(
+                "Syntax Error: CREATE TABLE ... AS SELECT combined with WITH ('connector'=...) is invalid. Use CREATE STREAMING TABLE instead."
+            );
+        }
+        if stmt.or_replace {
+            return plan_err!(
+                "Syntax Error: OR REPLACE is not supported for external connector tables."
+            );
+        }
+        if stmt.temporary {
+            return plan_err!(
+                "Syntax Error: TEMPORARY is not supported for external connector tables."
+            );
+        }
+        if stmt.external {
+            return plan_err!(
+                "Syntax Error: EXTERNAL keyword is redundant and unsupported for connector configurations."
+            );
+        }
+        Ok(())
+    }
+
+    fn assert_connector_match(options: &mut ConnectorOptions, connector_name: &str) -> Result<()> {
+        if let Some(c) = options.pull_opt_str(opt::CONNECTOR)?
+            && c != connector_name
+        {
+            return plan_err!(
+                "WITH option `connector` is '{c}' but table uses connector '{connector_name}'"
+            );
+        }
+        Ok(())
+    }
+
+    fn absorb_type_option(options: &mut ConnectorOptions, expected_role: &str) -> Result<()> {
+        let Some(raw) = options.pull_opt_str(opt::TYPE)? else {
+            return Ok(());
+        };
+        if !raw.eq_ignore_ascii_case(expected_role) {
+            return plan_err!(
+                "Role mismatch: WITH option 'type' = '{raw}' is incompatible with the compiled role '{expected_role}'"
+            );
+        }
+        Ok(())
+    }
+
+    fn assert_format_compatibility(format: &Option<Format>, adapter_type: &str) -> Result<()> {
+        if let Some(Format::Json(JsonFormat { compression, .. })) = format
+            && !matches!(compression, JsonCompression::Uncompressed)
+            && adapter_type != connector_type::FILESYSTEM
+        {
+            return plan_err!("'json.compression' is only supported for the filesystem connector");
+        }
+        Ok(())
+    }
+
+    fn assert_options_fully_consumed(
+        options: &ConnectorOptions,
+        role: &str,
+        adapter_type: &str,
+    ) -> Result<()> {
+        if !options.is_empty() {
+            let unknown_keys: Vec<String> = options.keys().cloned().collect();
+            return plan_err!(
+                "Unknown options for {role} connector '{adapter_type}': {unknown_keys:?}"
+            );
+        }
+        Ok(())
+    }
+}
+
+fn resolve_source_watermark(
+    table_identifier: &str,
+    columns: &mut Vec<ColumnDescriptor>,
+    watermark: Option<(String, Option<ast::Expr>)>,
+    options: &mut ConnectorOptions,
+    schema_provider: &StreamSchemaProvider,
+) -> Result<TemporalPipelineConfig> {
+    let mut config = TemporalPipelineConfig::default();
+
+    if let Some(event_time_field) = options.pull_opt_field(opt::EVENT_TIME_FIELD)? {
+        warn!("`event_time_field` WITH option is deprecated; use WATERMARK FOR syntax");
+        config.event_column = Some(event_time_field);
+    }
+    if let Some(watermark_field) = options.pull_opt_field(opt::WATERMARK_FIELD)? {
+        warn!("`watermark_field` WITH option is deprecated; use WATERMARK FOR syntax");
+        config.watermark_strategy_column = Some(watermark_field);
+    }
+
+    let Some((time_field, watermark_expr)) = watermark else {
+        return Ok(config);
+    };
+
+    let declared_field = columns
+        .iter()
+        .find(|c| c.arrow_field().name().as_str() == time_field.as_str())
+        .ok_or_else(|| {
+            plan_datafusion_err!(
+                "WATERMARK FOR field `{}` does not exist in table",
+                time_field
+            )
+        })?;
+
+    if !matches!(
+        declared_field.arrow_field().data_type(),
+        DataType::Timestamp(_, None)
+    ) {
+        return plan_err!(
+            "WATERMARK FOR field `{time_field}` has type {}, but expected TIMESTAMP",
+            declared_field.arrow_field().data_type()
+        );
+    }
+
+    for col in columns.iter_mut() {
+        if col.arrow_field().name().as_str() == time_field.as_str() {
+            col.set_nullable(false);
+            break;
+        }
+    }
+
+    config.event_column = Some(time_field.clone());
+
+    match watermark_expr {
+        Some(expr) => {
+            let table_ref = TableReference::bare(table_identifier.to_string());
+            let physical_schema = Schema::new(
+                columns
+                    .iter()
+                    .filter(|c| !c.is_computed())
+                    .map(|c| c.arrow_field().clone())
+                    .collect::<Vec<_>>(),
+            );
+            let df_schema = DFSchema::try_from_qualified_schema(table_ref, &physical_schema)?;
+
+            let logical_expr =
+                plan_generating_expr(&expr, &df_schema, schema_provider).map_err(|e| {
+                    DataFusionError::Plan(format!("could not plan watermark expression: {e}"))
+                })?;
+
+            let (data_type, _nullable) = logical_expr.data_type_and_nullable(&df_schema)?;
+            if !matches!(data_type, DataType::Timestamp(_, _)) {
+                return plan_err!(
+                    "the type of the WATERMARK FOR expression must be TIMESTAMP, but was {data_type}"
+                );
+            }
+
+            columns.push(ColumnDescriptor::new_computed(
+                Field::new(
+                    sql_field::COMPUTED_WATERMARK,
+                    logical_expr.get_type(&df_schema)?,
+                    false,
+                ),
+                logical_expr,
+            ));
+            config.watermark_strategy_column = Some(sql_field::COMPUTED_WATERMARK.to_string());
+        }
+        None => {
+            config.watermark_strategy_column = Some(time_field);
+        }
+    }
+
+    Ok(config)
+}
+
+fn plan_generating_expr(
+    ast_expr: &ast::Expr,
+    df_schema: &DFSchema,
+    schema_provider: &StreamSchemaProvider,
+) -> Result<Expr> {
+    let planner = SqlToRel::new(schema_provider);
+    let mut ctx = PlannerContext::new();
+    planner.sql_to_expr(ast_expr.clone(), df_schema, &mut ctx)
+}
diff --git a/src/coordinator/plan/logical_plan_visitor.rs b/src/coordinator/plan/logical_plan_visitor.rs
index d49d0314..cb22d24e 100644
--- a/src/coordinator/plan/logical_plan_visitor.rs
+++ b/src/coordinator/plan/logical_plan_visitor.rs
@@ -10,26 +10,16 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
-use std::sync::Arc;
-
-use datafusion::common::{Result, plan_datafusion_err, plan_err};
-use datafusion::execution::SessionStateBuilder;
-use datafusion::sql::sqlparser::ast::{
-    CreateTable as SqlCreateTable, Expr as SqlExpr, ObjectType, SqlOption,
-    Statement as DFStatement, TableConstraint,
-};
-use datafusion_common::TableReference;
-use datafusion_execution::config::SessionConfig;
-use datafusion_expr::{Expr, Extension, LogicalPlan, col};
+use datafusion::sql::sqlparser::ast::{ObjectType, Statement as DFStatement};
 use sqlparser::ast::Statement;
 use tracing::debug;
 
 use crate::coordinator::analyze::analysis::Analysis;
 use crate::coordinator::plan::{
-    CreateFunctionPlan, CreatePythonFunctionPlan, CreateTablePlan, DropFunctionPlan,
-    DropStreamingTablePlan, DropTablePlan, PlanNode, ShowCatalogTablesPlan,
+    CompileErrorPlan, CreateFunctionPlan, CreatePythonFunctionPlan, CreateTablePlan,
+    DropFunctionPlan, DropStreamingTablePlan, DropTablePlan, PlanNode, ShowCatalogTablesPlan,
     ShowCreateStreamingTablePlan, ShowCreateTablePlan, ShowFunctionsPlan, ShowStreamingTablesPlan,
-    StartFunctionPlan, StopFunctionPlan, StreamingTable,
+    StartFunctionPlan, StopFunctionPlan,
 };
 use crate::coordinator::statement::{
     CreateFunction, CreatePythonFunction, CreateTable, DropFunction, DropStreamingTableStatement,
@@ -37,17 +27,11 @@ use crate::coordinator::statement::{
     ShowFunctions, ShowStreamingTables, StartFunction, StatementVisitor, StatementVisitorContext,
     StatementVisitorResult, StopFunction, StreamingTableStatement,
 };
-use crate::coordinator::tool::ConnectorOptions;
-use crate::sql::analysis::{StreamSchemaProvider, maybe_add_key_extension_to_sink, rewrite_sinks};
-use crate::sql::common::with_option_keys as opt;
-use crate::sql::functions::{is_json_union, serialize_outgoing_json};
-use crate::sql::logical_node::logical::{LogicalProgram, ProgramConfig};
-use crate::sql::logical_node::sink::StreamEgressNode;
-use crate::sql::logical_planner::optimizers::{ChainingOptimizer, produce_optimized_plan};
-use crate::sql::logical_planner::planner::PlanToGraphVisitor;
-use crate::sql::rewrite_plan;
-use crate::sql::schema::source_table::SourceTable;
-use crate::sql::schema::{ColumnDescriptor, ConnectionType, Table};
+use crate::sql::analysis::StreamSchemaProvider;
+
+use super::ast_utils::AstUtils;
+use super::ddl_compiler::DdlCompiler;
+use super::streaming_compiler::StreamingCompiler;
 
 #[derive(Clone)]
 pub struct LogicalPlanVisitor {
@@ -65,280 +49,11 @@ impl LogicalPlanVisitor {
 
         match stmt.accept(self, &context) {
             StatementVisitorResult::Plan(plan) => plan,
-            _ => panic!("Fatal: LogicalPlanVisitor must yield a PlanNode variant"),
-        }
-    }
-
-    pub fn build_streaming_table(
-        schema_provider: &StreamSchemaProvider,
-        stmt: &StreamingTableStatement,
-    ) -> Result<StreamingTable> {
-        Self::new(schema_provider.clone()).compile_streaming_sink(stmt)
-    }
-
-    fn compile_streaming_sink(&self, stmt: &StreamingTableStatement) -> Result<StreamingTable> {
-        let DFStatement::CreateStreamingTable {
-            name,
-            with_options,
-            comment,
-            query,
-        } = &stmt.statement
-        else {
-            return plan_err!("Statement mismatch: Expected CREATE STREAMING TABLE AST node");
-        };
-
-        let sink_table_name = name.to_string();
-        debug!(
-            "Initiating streaming sink compilation for identifier: {}",
-            sink_table_name
-        );
-
-        let mut sink_properties = ConnectorOptions::new(with_options, &None)?;
-        let connector_type = sink_properties
-            .pull_opt_str(opt::CONNECTOR)?
-            .ok_or_else(|| {
-                plan_datafusion_err!(
-                    "Validation Error: Streaming table '{}' requires the '{}' property",
-                    sink_table_name,
-                    opt::CONNECTOR
-                )
-            })?;
-
-        let partition_keys = Self::extract_partitioning_keys(&mut sink_properties)?;
-
-        let sink_description = comment
-            .as_deref()
-            .map(str::trim)
-            .filter(|s| !s.is_empty())
-            .map(str::to_string)
-            .unwrap_or_else(|| format!("sink `{}` ({connector_type})", sink_table_name));
-
-        let mut query_logical_plan = rewrite_plan(
-            produce_optimized_plan(&Statement::Query(query.clone()), &self.schema_provider)?,
-            &self.schema_provider,
-        )?;
-
-        if query_logical_plan
-            .schema()
-            .fields()
-            .iter()
-            .any(|f| is_json_union(f.data_type()))
-        {
-            query_logical_plan =
-                serialize_outgoing_json(&self.schema_provider, Arc::new(query_logical_plan));
-        }
-
-        let output_schema_fields = query_logical_plan
-            .schema()
-            .fields()
-            .iter()
-            .map(|f| ColumnDescriptor::from((**f).clone()))
-            .collect::<Vec<_>>();
-
-        let mut sink_definition = SourceTable::from_options(
-            &sink_table_name,
-            &connector_type,
-            false,
-            output_schema_fields,
-            vec![],
-            None,
-            &mut sink_properties,
-            None,
-            &self.schema_provider,
-            Some(ConnectionType::Sink),
-            sink_description,
-        )?;
-        sink_definition.partition_exprs = Arc::new(partition_keys);
-
-        let output_schema = query_logical_plan.schema().clone();
-        let sink_plan_node = StreamEgressNode::try_new(
-            TableReference::bare(sink_table_name.clone()),
-            Table::ConnectorTable(sink_definition.clone()),
-            output_schema,
-            query_logical_plan,
-        )?;
-
-        let mut rewritten_plans = rewrite_sinks(vec![maybe_add_key_extension_to_sink(
-            LogicalPlan::Extension(Extension {
-                node: Arc::new(sink_plan_node),
-            }),
-        )?])?;
-
-        let final_logical_plan = rewritten_plans.remove(0);
-
-        let validated_program = self.validate_graph_topology(&final_logical_plan)?;
-
-        let streaming_with_options: Option<std::collections::HashMap<String, String>> =
-            if with_options.is_empty() {
-                None
-            } else {
-                let map: std::collections::HashMap<String, String> = with_options
-                    .iter()
-                    .filter_map(|opt| match opt {
-                        SqlOption::KeyValue { key, value } => Some((
-                            key.value.clone(),
-                            value.to_string().trim_matches('\'').to_string(),
-                        )),
-                        _ => None,
-                    })
-                    .collect();
-                if map.is_empty() { None } else { Some(map) }
-            };
-
-        Ok(StreamingTable {
-            name: sink_table_name,
-            comment: comment.clone(),
-            program: validated_program,
-            with_options: streaming_with_options,
-        })
-    }
-
-    fn validate_graph_topology(&self, logical_plan: &LogicalPlan) -> Result<LogicalProgram> {
-        let mut session_config = SessionConfig::new();
-        let opts = session_config.options_mut();
-        opts.optimizer.enable_round_robin_repartition = false;
-        opts.optimizer.repartition_aggregations = false;
-        opts.optimizer.repartition_windows = false;
-        opts.optimizer.repartition_sorts = false;
-        opts.optimizer.repartition_joins = false;
-        opts.execution.target_partitions = 1;
-
-        let session_state = SessionStateBuilder::new()
-            .with_config(session_config)
-            .with_default_features()
-            .with_physical_optimizer_rules(vec![])
-            .build();
-
-        let mut graph_compiler = PlanToGraphVisitor::new(&self.schema_provider, &session_state);
-        graph_compiler.add_plan(logical_plan.clone())?;
-
-        let mut executable_program =
-            LogicalProgram::new(graph_compiler.into_graph(), ProgramConfig::default());
-        executable_program.optimize(&ChainingOptimizer {});
-
-        Ok(executable_program)
-    }
-
-    fn extract_partitioning_keys(options: &mut ConnectorOptions) -> Result<Option<Vec<Expr>>> {
-        options
-            .pull_opt_str(opt::PARTITION_BY)?
-            .map(|raw_cols| raw_cols.split(',').map(|c| col(c.trim())).collect())
-            .map(Ok)
-            .transpose()
-    }
-
-    fn contains_connector_property(options: &[SqlOption]) -> bool {
-        options.iter().any(|opt| match opt {
-            SqlOption::KeyValue { key, .. } => key.value.eq_ignore_ascii_case(opt::CONNECTOR),
-            _ => false,
-        })
-    }
-
-    fn parse_primary_keys(constraints: &[TableConstraint]) -> Result<Vec<String>> {
-        let mut keys = None;
-        for constraint in constraints {
-            if let TableConstraint::PrimaryKey { columns, .. } = constraint {
-                if keys.is_some() {
-                    return plan_err!(
-                        "Constraint Violation: Multiple PRIMARY KEY constraints are forbidden"
-                    );
-                }
-                keys = Some(columns.iter().map(|ident| ident.value.clone()).collect());
-            }
-        }
-        Ok(keys.unwrap_or_default())
-    }
-
-    fn parse_watermark_strategy(
-        constraints: &[TableConstraint],
-    ) -> Result<Option<(String, Option<SqlExpr>)>> {
-        let mut strategy = None;
-        for constraint in constraints {
-            if let TableConstraint::Watermark {
-                column_name,
-                watermark_expr,
-            } = constraint
-            {
-                if strategy.is_some() {
-                    return plan_err!(
-                        "Constraint Violation: Only a single WATERMARK FOR clause is permitted"
-                    );
-                }
-                strategy = Some((column_name.value.clone(), watermark_expr.clone()));
-            }
+            _ => Box::new(CompileErrorPlan::new(
+                "LogicalPlanVisitor did not yield a PlanNode variant for the given statement"
+                    .to_string(),
+            )),
         }
-        Ok(strategy)
-    }
-
-    fn compile_connector_source_plan(&self, stmt: &SqlCreateTable) -> Result<CreateTablePlan> {
-        if stmt.query.is_some() {
-            return plan_err!(
-                "Syntax Error: CREATE TABLE ... AS SELECT combined with WITH ('connector'=...) is invalid. Use CREATE STREAMING TABLE instead."
-            );
-        }
-        if stmt.or_replace {
-            return plan_err!(
-                "Syntax Error: OR REPLACE is not supported for external connector tables."
-            );
-        }
-        if stmt.temporary {
-            return plan_err!(
-                "Syntax Error: TEMPORARY is not supported for external connector tables."
-            );
-        }
-        if stmt.external {
-            return plan_err!(
-                "Syntax Error: EXTERNAL keyword is redundant and unsupported for connector configurations."
-            );
-        }
-
-        let target_name = stmt.name.to_string();
-        let table_description = stmt
-            .comment
-            .clone()
-            .map(|c| c.to_string())
-            .unwrap_or_default();
-
-        let schema_compiler = datafusion::sql::planner::SqlToRel::new(&self.schema_provider);
-        let arrow_schema = schema_compiler.build_schema(stmt.columns.clone())?;
-
-        let schema_descriptors = arrow_schema
-            .fields()
-            .iter()
-            .map(|f| ColumnDescriptor::from((**f).clone()))
-            .collect::<Vec<_>>();
-
-        let mut connector_options = ConnectorOptions::new(&stmt.with_options, &None)?;
-        let adapter_type = connector_options
-            .pull_opt_str(opt::CONNECTOR)?
-            .ok_or_else(|| {
-                plan_datafusion_err!(
-                    "Configuration Error: Missing required property '{}' in WITH clause",
-                    opt::CONNECTOR
-                )
-            })?;
-
-        let pk_constraints = Self::parse_primary_keys(&stmt.constraints)?;
-        let watermark_strategy = Self::parse_watermark_strategy(&stmt.constraints)?;
-
-        let source_definition = SourceTable::from_options(
-            &target_name,
-            &adapter_type,
-            false,
-            schema_descriptors,
-            pk_constraints,
-            watermark_strategy,
-            &mut connector_options,
-            None,
-            &self.schema_provider,
-            Some(ConnectionType::Source),
-            table_description,
-        )?;
-
-        Ok(CreateTablePlan::connector_source(
-            source_definition,
-            stmt.if_not_exists,
-        ))
     }
 }
 
@@ -422,16 +137,18 @@ impl StatementVisitor for LogicalPlanVisitor {
     ) -> StatementVisitorResult {
         if let Statement::CreateTable(ast_node) = &stmt.statement
             && ast_node.query.is_none()
-            && Self::contains_connector_property(&ast_node.with_options)
+            && AstUtils::contains_connector_property(&ast_node.with_options)
         {
-            let execution_plan =
-                self.compile_connector_source_plan(ast_node)
-                    .unwrap_or_else(|err| {
-                        panic!(
-                            "Fatal Compiler Error: Connector source resolution failed - {err:#}"
-                        );
-                    });
-            return StatementVisitorResult::Plan(Box::new(execution_plan));
+            let declared_role = AstUtils::peek_table_role(&ast_node.with_options);
+            let compiler = DdlCompiler::new(&self.schema_provider);
+            return match compiler.compile(ast_node, declared_role.as_deref()) {
+                Ok(external_table) => StatementVisitorResult::Plan(Box::new(
+                    CreateTablePlan::external_table(external_table, ast_node.if_not_exists),
+                )),
+                Err(err) => StatementVisitorResult::Plan(Box::new(CompileErrorPlan::new(format!(
+                    "Ingest table resolution failed - {err:#}"
+                )))),
+            };
         }
 
         let schema_compiler = datafusion::sql::planner::SqlToRel::new(&self.schema_provider);
@@ -443,7 +160,9 @@ impl StatementVisitor for LogicalPlanVisitor {
                 );
                 StatementVisitorResult::Plan(Box::new(CreateTablePlan::new(logical_plan)))
             }
-            Err(err) => panic!("Fatal Compiler Error: Logical plan translation failed - {err}"),
+            Err(err) => StatementVisitorResult::Plan(Box::new(CompileErrorPlan::new(format!(
+                "Logical plan translation failed - {err}"
+            )))),
         }
     }
 
@@ -452,10 +171,13 @@ impl StatementVisitor for LogicalPlanVisitor {
         stmt: &StreamingTableStatement,
         _ctx: &StatementVisitorContext,
     ) -> StatementVisitorResult {
-        let execution_plan = self.compile_streaming_sink(stmt).unwrap_or_else(|err| {
-            panic!("Fatal Compiler Error: Streaming sink compilation aborted - {err}");
-        });
-        StatementVisitorResult::Plan(Box::new(execution_plan))
+        let compiler = StreamingCompiler::new(&self.schema_provider);
+        match compiler.compile(stmt) {
+            Ok(execution_plan) => StatementVisitorResult::Plan(Box::new(execution_plan)),
+            Err(err) => StatementVisitorResult::Plan(Box::new(CompileErrorPlan::new(format!(
+                "Streaming sink compilation aborted - {err}"
+            )))),
+        }
     }
 
     fn visit_drop_table_statement(
@@ -470,16 +192,20 @@ impl StatementVisitor for LogicalPlanVisitor {
             ..
         } = &stmt.statement
         else {
-            panic!("Fatal Compiler Error: AST mismatch on DropTableStatement");
+            return StatementVisitorResult::Plan(Box::new(CompileErrorPlan::new(
+                "AST mismatch: expected DROP statement for DropTableStatement".to_string(),
+            )));
         };
 
         if *object_type != ObjectType::Table {
-            panic!("Fatal Compiler Error: Drop target must be of type TABLE");
+            return StatementVisitorResult::Plan(Box::new(CompileErrorPlan::new(format!(
+                "Drop target must be of type TABLE, got {object_type:?}"
+            ))));
         }
         if names.len() != 1 {
-            panic!(
-                "Fatal Compiler Error: Bulk drop operations are not supported. Specify exactly one table."
-            );
+            return StatementVisitorResult::Plan(Box::new(CompileErrorPlan::new(
+                "Bulk drop operations are not supported. Specify exactly one table.".to_string(),
+            )));
         }
 
         StatementVisitorResult::Plan(Box::new(DropTablePlan::new(
diff --git a/src/coordinator/plan/lookup_table_plan.rs b/src/coordinator/plan/lookup_table_plan.rs
index 65103b61..9d1fea93 100644
--- a/src/coordinator/plan/lookup_table_plan.rs
+++ b/src/coordinator/plan/lookup_table_plan.rs
@@ -10,14 +10,13 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
-use crate::sql::schema::source_table::SourceTable;
+use crate::sql::schema::LookupTable;
 
 use super::{PlanNode, PlanVisitor, PlanVisitorContext, PlanVisitorResult};
 
-/// Plan node that exposes a lookup table config as a logical plan input.
 #[derive(Debug)]
 pub struct LookupTablePlan {
-    pub table: SourceTable,
+    pub table: LookupTable,
 }
 
 impl PlanNode for LookupTablePlan {
diff --git a/src/coordinator/plan/mod.rs b/src/coordinator/plan/mod.rs
index 8166d444..95978467 100644
--- a/src/coordinator/plan/mod.rs
+++ b/src/coordinator/plan/mod.rs
@@ -10,9 +10,12 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
+mod ast_utils;
+mod compile_error_plan;
 mod create_function_plan;
 mod create_python_function_plan;
 mod create_table_plan;
+mod ddl_compiler;
 mod drop_function_plan;
 mod drop_streaming_table_plan;
 mod drop_table_plan;
@@ -26,10 +29,12 @@ mod show_functions_plan;
 mod show_streaming_tables_plan;
 mod start_function_plan;
 mod stop_function_plan;
+mod streaming_compiler;
 mod streaming_table_connector_plan;
 mod streaming_table_plan;
 mod visitor;
 
+pub use compile_error_plan::CompileErrorPlan;
 pub use create_function_plan::CreateFunctionPlan;
 pub use create_python_function_plan::CreatePythonFunctionPlan;
 pub use create_table_plan::{CreateTablePlan, CreateTablePlanBody};
diff --git a/src/coordinator/plan/streaming_compiler.rs b/src/coordinator/plan/streaming_compiler.rs
new file mode 100644
index 00000000..dbdd4dec
--- /dev/null
+++ b/src/coordinator/plan/streaming_compiler.rs
@@ -0,0 +1,234 @@
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+//! `CREATE STREAMING TABLE ... AS SELECT` → [`SinkTable`], egress, [`LogicalProgram`].
+
+use std::collections::HashMap;
+use std::sync::Arc;
+
+use datafusion::common::{Result, plan_datafusion_err, plan_err};
+use datafusion::execution::SessionStateBuilder;
+use datafusion::sql::sqlparser::ast::{SqlOption, Statement as DFStatement};
+use datafusion_common::TableReference;
+use datafusion_execution::config::SessionConfig;
+use datafusion_expr::{Expr, Extension, LogicalPlan, col};
+use sqlparser::ast::Statement;
+use tracing::debug;
+
+use super::StreamingTable;
+use crate::coordinator::statement::StreamingTableStatement;
+use crate::coordinator::tool::ConnectorOptions;
+use crate::sql::analysis::{StreamSchemaProvider, maybe_add_key_extension_to_sink, rewrite_sinks};
+use crate::sql::common::constants::connector_type;
+use crate::sql::common::with_option_keys as opt;
+use crate::sql::common::{Format, JsonCompression, JsonFormat};
+use crate::sql::connector::registry::REGISTRY;
+use crate::sql::connector::sink::runtime_config::SinkRuntimeConfig;
+use crate::sql::functions::{is_json_union, serialize_outgoing_json};
+use crate::sql::logical_node::logical::{LogicalProgram, ProgramConfig};
+use crate::sql::logical_node::sink::StreamEgressNode;
+use crate::sql::logical_planner::optimizers::{ChainingOptimizer, produce_optimized_plan};
+use crate::sql::logical_planner::planner::PlanToGraphVisitor;
+use crate::sql::rewrite_plan;
+use crate::sql::schema::ColumnDescriptor;
+use crate::sql::schema::catalog::{ExternalTable, SinkTable};
+use crate::sql::schema::table::CatalogEntity;
+use crate::sql::schema::table_role::validate_adapter_availability;
+
+pub struct StreamingCompiler<'a> {
+    schema_provider: &'a StreamSchemaProvider,
+}
+
+impl<'a> StreamingCompiler<'a> {
+    pub fn new(schema_provider: &'a StreamSchemaProvider) -> Self {
+        Self { schema_provider }
+    }
+
+    pub fn compile(&self, stmt: &StreamingTableStatement) -> Result<StreamingTable> {
+        let DFStatement::CreateStreamingTable {
+            name,
+            with_options,
+            comment,
+            query,
+        } = &stmt.statement
+        else {
+            return plan_err!("Statement mismatch: Expected CREATE STREAMING TABLE AST node");
+        };
+
+        let sink_table_name = name.to_string();
+        debug!(
+            "Initiating streaming sink compilation for identifier: {}",
+            sink_table_name
+        );
+
+        let mut sink_properties = ConnectorOptions::new(with_options, &None)?;
+        let adapter_type = sink_properties
+            .pull_opt_str(opt::CONNECTOR)?
+            .ok_or_else(|| {
+                plan_datafusion_err!(
+                    "Validation Error: Streaming table '{}' requires the '{}' property",
+                    sink_table_name,
+                    opt::CONNECTOR
+                )
+            })?;
+        validate_adapter_availability(&adapter_type)?;
+
+        let partition_keys = Self::extract_partitioning_keys(&mut sink_properties)?;
+        let catalog_with_options = sink_properties.snapshot_for_catalog();
+
+        let connection_format = Format::from_opts(&mut sink_properties)?;
+        Self::assert_format_compatibility(&connection_format, &adapter_type)?;
+
+        let sink_description = comment
+            .as_deref()
+            .map(str::trim)
+            .filter(|s| !s.is_empty())
+            .map(str::to_string)
+            .unwrap_or_else(|| format!("sink `{}` ({adapter_type})", sink_table_name));
+
+        let mut query_logical_plan = rewrite_plan(
+            produce_optimized_plan(&Statement::Query(query.clone()), self.schema_provider)?,
+            self.schema_provider,
+        )?;
+
+        if query_logical_plan
+            .schema()
+            .fields()
+            .iter()
+            .any(|f| is_json_union(f.data_type()))
+        {
+            query_logical_plan =
+                serialize_outgoing_json(self.schema_provider, Arc::new(query_logical_plan));
+        }
+
+        let output_schema_fields: Vec<ColumnDescriptor> = query_logical_plan
+            .schema()
+            .fields()
+            .iter()
+            .map(|f| ColumnDescriptor::from((**f).clone()))
+            .collect();
+
+        let runtime_props =
+            SinkRuntimeConfig::extract_from_options(&mut sink_properties)?.to_runtime_properties();
+        let provider = REGISTRY.get_sink(&adapter_type)?;
+        let connector_config =
+            provider.build_sink_config(&mut sink_properties, &connection_format, &runtime_props)?;
+
+        if !sink_properties.is_empty() {
+            let unknown_keys: Vec<String> = sink_properties.keys().cloned().collect();
+            return plan_err!(
+                "Unknown options for streaming sink connector '{adapter_type}': {unknown_keys:?}"
+            );
+        }
+
+        let sink_table = SinkTable {
+            table_identifier: sink_table_name.clone(),
+            adapter_type,
+            schema_specs: output_schema_fields,
+            connector_config,
+            partition_exprs: Arc::new(partition_keys),
+            key_constraints: Vec::new(),
+            connection_format,
+            description: sink_description,
+            catalog_with_options,
+        };
+
+        let output_schema = query_logical_plan.schema().clone();
+        let sink_plan_node = StreamEgressNode::try_new(
+            TableReference::bare(sink_table_name.clone()),
+            CatalogEntity::external(ExternalTable::Sink(sink_table)),
+            output_schema,
+            query_logical_plan,
+        )?;
+
+        let mut rewritten_plans = rewrite_sinks(vec![maybe_add_key_extension_to_sink(
+            LogicalPlan::Extension(Extension {
+                node: Arc::new(sink_plan_node),
+            }),
+        )?])?;
+
+        let final_logical_plan = rewritten_plans.remove(0);
+        let validated_program = self.validate_graph_topology(&final_logical_plan)?;
+
+        Ok(StreamingTable {
+            name: sink_table_name,
+            comment: comment.clone(),
+            program: validated_program,
+            with_options: Self::echo_with_options(with_options),
+        })
+    }
+
+    /// Compile the final logical plan into the executable streaming program,
+    /// disabling DataFusion's batch-oriented repartition heuristics.
+    fn validate_graph_topology(&self, logical_plan: &LogicalPlan) -> Result<LogicalProgram> {
+        let mut session_config = SessionConfig::new();
+        let opts = session_config.options_mut();
+        opts.optimizer.enable_round_robin_repartition = false;
+        opts.optimizer.repartition_aggregations = false;
+        opts.optimizer.repartition_windows = false;
+        opts.optimizer.repartition_sorts = false;
+        opts.optimizer.repartition_joins = false;
+        opts.execution.target_partitions = 1;
+
+        let session_state = SessionStateBuilder::new()
+            .with_config(session_config)
+            .with_default_features()
+            .with_physical_optimizer_rules(vec![])
+            .build();
+
+        let mut graph_compiler = PlanToGraphVisitor::new(self.schema_provider, &session_state);
+        graph_compiler.add_plan(logical_plan.clone())?;
+
+        let mut executable_program =
+            LogicalProgram::new(graph_compiler.into_graph(), ProgramConfig::default());
+        executable_program.optimize(&ChainingOptimizer {});
+
+        Ok(executable_program)
+    }
+
+    fn extract_partitioning_keys(options: &mut ConnectorOptions) -> Result<Option<Vec<Expr>>> {
+        options
+            .pull_opt_str(opt::PARTITION_BY)?
+            .map(|raw_cols| raw_cols.split(',').map(|c| col(c.trim())).collect())
+            .map(Ok)
+            .transpose()
+    }
+
+    fn assert_format_compatibility(format: &Option<Format>, adapter_type: &str) -> Result<()> {
+        if let Some(Format::Json(JsonFormat { compression, .. })) = format
+            && !matches!(compression, JsonCompression::Uncompressed)
+            && adapter_type != connector_type::FILESYSTEM
+        {
+            return plan_err!("'json.compression' is only supported for the filesystem connector");
+        }
+        Ok(())
+    }
+
+    /// Snapshot the original WITH options so the streaming-table catalog
+    /// can later reproduce the exact DDL for `SHOW CREATE STREAMING TABLE`.
+    fn echo_with_options(with_options: &[SqlOption]) -> Option<HashMap<String, String>> {
+        if with_options.is_empty() {
+            return None;
+        }
+        let map: HashMap<String, String> = with_options
+            .iter()
+            .filter_map(|o| match o {
+                SqlOption::KeyValue { key, value } => Some((
+                    key.value.clone(),
+                    value.to_string().trim_matches('\'').to_string(),
+                )),
+                _ => None,
+            })
+            .collect();
+        if map.is_empty() { None } else { Some(map) }
+    }
+}
diff --git a/src/coordinator/plan/streaming_table_connector_plan.rs b/src/coordinator/plan/streaming_table_connector_plan.rs
index 214e2e15..105f9bd7 100644
--- a/src/coordinator/plan/streaming_table_connector_plan.rs
+++ b/src/coordinator/plan/streaming_table_connector_plan.rs
@@ -10,14 +10,13 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
-use crate::sql::schema::source_table::SourceTable;
+use crate::sql::schema::SinkTable;
 
 use super::{PlanNode, PlanVisitor, PlanVisitorContext, PlanVisitorResult};
 
-/// Plan node that exposes a connector table config as a logical plan input.
 #[derive(Debug)]
 pub struct StreamingTableConnectorPlan {
-    pub table: SourceTable,
+    pub table: SinkTable,
 }
 
 impl PlanNode for StreamingTableConnectorPlan {
diff --git a/src/coordinator/plan/visitor.rs b/src/coordinator/plan/visitor.rs
index a94d761f..5c566b74 100644
--- a/src/coordinator/plan/visitor.rs
+++ b/src/coordinator/plan/visitor.rs
@@ -11,10 +11,11 @@
 // limitations under the License.
 
 use super::{
-    CreateFunctionPlan, CreatePythonFunctionPlan, CreateTablePlan, DropFunctionPlan,
-    DropStreamingTablePlan, DropTablePlan, LookupTablePlan, ShowCatalogTablesPlan,
-    ShowCreateStreamingTablePlan, ShowCreateTablePlan, ShowFunctionsPlan, ShowStreamingTablesPlan,
-    StartFunctionPlan, StopFunctionPlan, StreamingTable, StreamingTableConnectorPlan,
+    CompileErrorPlan, CreateFunctionPlan, CreatePythonFunctionPlan, CreateTablePlan,
+    DropFunctionPlan, DropStreamingTablePlan, DropTablePlan, LookupTablePlan,
+    ShowCatalogTablesPlan, ShowCreateStreamingTablePlan, ShowCreateTablePlan, ShowFunctionsPlan,
+    ShowStreamingTablesPlan, StartFunctionPlan, StopFunctionPlan, StreamingTable,
+    StreamingTableConnectorPlan,
 };
 
 /// Context passed to PlanVisitor methods
@@ -51,6 +52,12 @@ pub enum PlanVisitorResult {
 }
 
 pub trait PlanVisitor {
+    fn visit_compile_error_plan(
+        &self,
+        plan: &CompileErrorPlan,
+        context: &PlanVisitorContext,
+    ) -> PlanVisitorResult;
+
     fn visit_create_function(
         &self,
         plan: &CreateFunctionPlan,
diff --git a/src/coordinator/streaming_table_options.rs b/src/coordinator/streaming_table_options.rs
index 51e020b0..4d9db7b8 100644
--- a/src/coordinator/streaming_table_options.rs
+++ b/src/coordinator/streaming_table_options.rs
@@ -12,6 +12,8 @@
 
 use std::collections::HashMap;
 
+use crate::sql::common::with_option_keys as opt;
+
 fn parse_positive_u64(raw: &str) -> Option<u64> {
     let t = raw.trim().trim_matches('\'');
     t.parse::<u64>().ok().filter(|&v| v > 0)
@@ -23,12 +25,12 @@ fn parse_positive_u32(raw: &str) -> Option<u32> {
 }
 
 pub fn parse_checkpoint_interval_ms(opts: Option<&HashMap<String, String>>) -> Option<u64> {
-    opts.and_then(|m| m.get("checkpoint.interval"))
+    opts.and_then(|m| m.get(opt::CHECKPOINT_INTERVAL_MS))
         .and_then(|s| parse_positive_u64(s))
 }
 
 pub fn parse_pipeline_parallelism(opts: Option<&HashMap<String, String>>) -> Option<u32> {
-    opts.and_then(|m| m.get("parallelism"))
+    opts.and_then(|m| m.get(opt::PIPELINE_PARALLELISM))
         .and_then(|s| parse_positive_u32(s))
 }
 
@@ -39,8 +41,8 @@ mod tests {
     #[test]
     fn parses_checkpoint_and_parallelism() {
         let mut m = HashMap::new();
-        m.insert("checkpoint.interval".to_string(), "30000".to_string());
-        m.insert("parallelism".to_string(), "2".to_string());
+        m.insert(opt::CHECKPOINT_INTERVAL_MS.to_string(), "30000".to_string());
+        m.insert(opt::PIPELINE_PARALLELISM.to_string(), "2".to_string());
         assert_eq!(parse_checkpoint_interval_ms(Some(&m)), Some(30_000));
         assert_eq!(parse_pipeline_parallelism(Some(&m)), Some(2));
     }
diff --git a/src/runtime/streaming/api/context.rs b/src/runtime/streaming/api/context.rs
index 8b778502..f2557e7a 100644
--- a/src/runtime/streaming/api/context.rs
+++ b/src/runtime/streaming/api/context.rs
@@ -16,7 +16,7 @@ use std::time::{Duration, SystemTime};
 
 use anyhow::{Context, Result, anyhow};
 use arrow_array::RecordBatch;
-use protocol::storage::SourceCheckpointPayload;
+use protocol::storage::SourceCheckpointInfo;
 use tokio::sync::mpsc;
 
 use crate::runtime::memory::{MemoryBlock, MemoryPool, get_array_memory_size};
@@ -78,7 +78,7 @@ pub struct TaskContext {
     /// Last globally-committed safe epoch for crash recovery.
     safe_epoch: u64,
 
-    /// When set, pipelines report checkpoint completion (and optional Kafka offsets) to the job coordinator.
+    /// When set, pipelines report checkpoint completion to the job coordinator.
     checkpoint_ack_tx: Option<mpsc::Sender<JobMasterEvent>>,
 }
 
@@ -128,17 +128,13 @@ impl TaskContext {
     }
 
     /// Notify the job checkpoint coordinator that this pipeline has finished the barrier for `epoch`.
-    pub async fn send_checkpoint_ack(
-        &self,
-        epoch: u64,
-        source_payloads: Vec<SourceCheckpointPayload>,
-    ) {
+    pub async fn send_checkpoint_ack(&self, epoch: u64, source_infos: Vec<SourceCheckpointInfo>) {
         if let Some(tx) = &self.checkpoint_ack_tx {
             let _ = tx
                 .send(JobMasterEvent::CheckpointAck {
                     pipeline_id: self.pipeline_id,
                     epoch,
-                    source_payloads,
+                    source_infos,
                 })
                 .await;
         }
diff --git a/src/runtime/streaming/api/operator.rs b/src/runtime/streaming/api/operator.rs
index 8eb9e8c4..fc75e475 100644
--- a/src/runtime/streaming/api/operator.rs
+++ b/src/runtime/streaming/api/operator.rs
@@ -66,7 +66,7 @@ pub trait Operator: Send + 'static {
     /// `commit_transaction` on the producer stashed during [`Self::snapshot_state`].
     async fn commit_checkpoint(
         &mut self,
-        epoch: u32,
+        epoch: u64,
         _ctx: &mut TaskContext,
     ) -> anyhow::Result<()> {
         let _ = epoch;
@@ -76,7 +76,7 @@ pub trait Operator: Send + 'static {
     /// Global checkpoint **rollback** when phase 2 must not commit (e.g. catalog persist failed).
     ///
     /// Default is no-op. Transactional Kafka sink overrides with `abort_transaction` on the stashed producer.
-    async fn abort_checkpoint(&mut self, epoch: u32, _ctx: &mut TaskContext) -> anyhow::Result<()> {
+    async fn abort_checkpoint(&mut self, epoch: u64, _ctx: &mut TaskContext) -> anyhow::Result<()> {
         let _ = epoch;
         Ok(())
     }
diff --git a/src/runtime/streaming/api/source.rs b/src/runtime/streaming/api/source.rs
index 26851eb2..9c531f2c 100644
--- a/src/runtime/streaming/api/source.rs
+++ b/src/runtime/streaming/api/source.rs
@@ -15,7 +15,7 @@ use crate::sql::common::{CheckpointBarrier, Watermark};
 use arrow_array::RecordBatch;
 use async_trait::async_trait;
 use protocol::storage::{
-    KafkaSourceSubtaskCheckpoint, SourceCheckpointPayload, source_checkpoint_payload,
+    KafkaSourceSubtaskCheckpoint, SourceCheckpointInfo, source_checkpoint_info,
 };
 
 #[derive(Debug, Clone, Copy, PartialEq, Eq, Default)]
@@ -34,17 +34,19 @@ pub enum SourceEvent {
     EndOfStream,
 }
 
-/// Optional metadata returned when a source completes a checkpoint barrier snapshot.
+/// Checkpoint metadata produced by a source subtask during a barrier snapshot.
+/// Sources fill this directly with [`SourceCheckpointInfo`] — the coordinator collects
+/// and persists these entries without any further translation step.
 #[derive(Debug, Default, Clone)]
 pub struct SourceCheckpointReport {
-    pub payloads: Vec<SourceCheckpointPayload>,
+    pub infos: Vec<SourceCheckpointInfo>,
 }
 
 impl SourceCheckpointReport {
     pub fn from_kafka_checkpoint(kafka: KafkaSourceSubtaskCheckpoint) -> Self {
         Self {
-            payloads: vec![SourceCheckpointPayload {
-                checkpoint: Some(source_checkpoint_payload::Checkpoint::Kafka(kafka)),
+            infos: vec![SourceCheckpointInfo {
+                info: Some(source_checkpoint_info::Info::Kafka(kafka)),
             }],
         }
     }
@@ -54,6 +56,11 @@ impl SourceCheckpointReport {
 pub trait SourceOperator: Send + 'static {
     fn name(&self) -> &str;
 
+    /// Inject persisted checkpoint records before the source is started.
+    /// Called by the engine after the operator is constructed and before [`Self::on_start`].
+    /// Default implementation is a no-op; sources with stateful recovery override this.
+    fn set_recovery_checkpoint(&mut self, _infos: Vec<SourceCheckpointInfo>) {}
+
     async fn on_start(&mut self, _ctx: &mut TaskContext) -> anyhow::Result<()> {
         Ok(())
     }
@@ -74,7 +81,7 @@ pub trait SourceOperator: Send + 'static {
     /// Kafka source keeps the default: offsets are reported at the barrier in [`Self::snapshot_state`].
     async fn commit_checkpoint(
         &mut self,
-        epoch: u32,
+        epoch: u64,
         _ctx: &mut TaskContext,
     ) -> anyhow::Result<()> {
         let _ = epoch;
@@ -82,7 +89,7 @@ pub trait SourceOperator: Send + 'static {
     }
 
     /// Same rollback hook as [`super::operator::Operator::abort_checkpoint`].
-    async fn abort_checkpoint(&mut self, epoch: u32, _ctx: &mut TaskContext) -> anyhow::Result<()> {
+    async fn abort_checkpoint(&mut self, epoch: u64, _ctx: &mut TaskContext) -> anyhow::Result<()> {
         let _ = epoch;
         Ok(())
     }
diff --git a/src/runtime/streaming/execution/pipeline.rs b/src/runtime/streaming/execution/pipeline.rs
index 91309a48..7c2ca17a 100644
--- a/src/runtime/streaming/execution/pipeline.rs
+++ b/src/runtime/streaming/execution/pipeline.rs
@@ -110,7 +110,7 @@ impl Pipeline {
                                         }
                                     }
                                     AlignmentStatus::Complete => {
-                                        let epoch = barrier.epoch as u64;
+                                        let epoch = barrier.epoch;
                                         self.chain_head
                                             .process_event(
                                                 idx,
diff --git a/src/runtime/streaming/execution/source_driver.rs b/src/runtime/streaming/execution/source_driver.rs
index b4e7d327..0118c4ee 100644
--- a/src/runtime/streaming/execution/source_driver.rs
+++ b/src/runtime/streaming/execution/source_driver.rs
@@ -161,7 +161,7 @@ impl SourceDriver {
                 let b: CheckpointBarrier = barrier.clone().into();
                 let report = self.operator.snapshot_state(b, &mut self.ctx).await?;
                 self.dispatch_event(StreamEvent::Barrier(b)).await?;
-                pending_source_checkpoint = Some((b.epoch as u64, report));
+                pending_source_checkpoint = Some((b.epoch, report));
             }
             ControlCommand::Commit { epoch } => {
                 self.operator
@@ -186,7 +186,7 @@ impl SourceDriver {
         }
 
         if let Some((epoch, report)) = pending_source_checkpoint {
-            self.ctx.send_checkpoint_ack(epoch, report.payloads).await;
+            self.ctx.send_checkpoint_ack(epoch, report.infos).await;
         }
 
         Ok(stop)
diff --git a/src/runtime/streaming/execution/tracker/barrier_aligner.rs b/src/runtime/streaming/execution/tracker/barrier_aligner.rs
index 4f954a7d..dbc30cd1 100644
--- a/src/runtime/streaming/execution/tracker/barrier_aligner.rs
+++ b/src/runtime/streaming/execution/tracker/barrier_aligner.rs
@@ -23,7 +23,7 @@ pub enum AlignmentStatus {
 #[derive(Debug)]
 pub struct BarrierAligner {
     input_count: usize,
-    current_epoch: Option<u32>,
+    current_epoch: Option<u64>,
     reached_inputs: HashSet<usize>,
 }
 
diff --git a/src/runtime/streaming/factory/connector/delta.rs b/src/runtime/streaming/factory/connector/delta.rs
new file mode 100644
index 00000000..726f87ef
--- /dev/null
+++ b/src/runtime/streaming/factory/connector/delta.rs
@@ -0,0 +1,82 @@
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+use std::collections::HashMap;
+use std::sync::Arc;
+
+use anyhow::{Context, Result, bail};
+use prost::Message;
+use protocol::function_stream_graph::ConnectorOp;
+use protocol::function_stream_graph::connector_op::Config;
+
+use crate::runtime::streaming::api::operator::ConstructedOperator;
+use crate::runtime::streaming::factory::connector::sink_props_codec::{
+    apply_common_sink_fields, normalized_props, parse_sink_memory_bytes,
+};
+use crate::runtime::streaming::factory::global::Registry;
+use crate::runtime::streaming::factory::operator_constructor::OperatorConstructor;
+use crate::runtime::streaming::operators::sink::delta::{DeltaFormat, DeltaSinkOperator};
+use crate::runtime::streaming::operators::sink::filesystem::compression_from_str;
+use crate::sql::common::constants::connection_format_value;
+use crate::sql::common::with_option_keys as opt;
+
+pub struct DeltaSinkDispatcher;
+
+impl OperatorConstructor for DeltaSinkDispatcher {
+    fn with_config(&self, payload: &[u8], _registry: Arc<Registry>) -> Result<ConstructedOperator> {
+        let op = ConnectorOp::decode(payload).context("failed to decode connector op")?;
+        let props = match op.config {
+            Some(Config::DeltaSink(cfg)) => delta_props(cfg),
+            _ => bail!("delta sink expects DeltaSinkConfig"),
+        };
+        let props = normalized_props(props);
+
+        let format = props
+            .get(opt::FORMAT)
+            .map(String::as_str)
+            .unwrap_or(connection_format_value::PARQUET)
+            .to_ascii_lowercase();
+        let path = props
+            .get(opt::PATH)
+            .cloned()
+            .or_else(|| props.get(opt::SINK_PATH).cloned())
+            .unwrap_or_else(|| ".".to_string());
+        let compression =
+            compression_from_str(props.get(opt::PARQUET_COMPRESSION).map(String::as_str))?;
+        let sink_memory_bytes = parse_sink_memory_bytes(&props)?;
+        let format = match format.as_str() {
+            connection_format_value::CSV => DeltaFormat::Csv,
+            connection_format_value::PARQUET => DeltaFormat::Parquet,
+            connection_format_value::JSON => DeltaFormat::JsonL,
+            connection_format_value::AVRO => DeltaFormat::Avro,
+            connection_format_value::ORC => DeltaFormat::Orc,
+            other => bail!("unsupported delta sink format '{other}'"),
+        };
+        Ok(ConstructedOperator::Operator(Box::new(
+            DeltaSinkOperator::try_new(
+                op.name,
+                path,
+                format,
+                compression,
+                sink_memory_bytes,
+                props,
+            )?,
+        )))
+    }
+}
+
+fn delta_props(cfg: protocol::function_stream_graph::DeltaSinkConfig) -> HashMap<String, String> {
+    let mut props = cfg.extra_properties;
+    props.extend(cfg.runtime_properties);
+    apply_common_sink_fields(&mut props, cfg.path, cfg.format, cfg.parquet_compression);
+    props
+}
diff --git a/src/runtime/streaming/factory/connector/dispatchers.rs b/src/runtime/streaming/factory/connector/dispatchers.rs
index 430a49f9..7d626600 100644
--- a/src/runtime/streaming/factory/connector/dispatchers.rs
+++ b/src/runtime/streaming/factory/connector/dispatchers.rs
@@ -12,19 +12,30 @@
 
 use std::sync::Arc;
 
-use anyhow::Result;
+use anyhow::{Context, Result, bail};
+use prost::Message;
+use protocol::function_stream_graph::ConnectorOp;
 
 use crate::runtime::streaming::api::operator::ConstructedOperator;
 use crate::runtime::streaming::factory::global::Registry;
 use crate::runtime::streaming::factory::operator_constructor::OperatorConstructor;
+use crate::sql::common::constants::connector_type;
 
-use super::kafka::KafkaConnectorDispatcher;
+use super::{
+    DeltaSinkDispatcher, FilesystemSinkDispatcher, IcebergSinkDispatcher, LanceDbSinkDispatcher,
+    S3SinkDispatcher, kafka::KafkaConnectorDispatcher,
+};
 
 pub struct ConnectorSourceDispatcher;
 
 impl OperatorConstructor for ConnectorSourceDispatcher {
     fn with_config(&self, config: &[u8], registry: Arc<Registry>) -> Result<ConstructedOperator> {
-        KafkaConnectorDispatcher.with_config(config, registry)
+        let op = ConnectorOp::decode(config)
+            .context("failed decoding connector op for source dispatch")?;
+        match op.connector.to_ascii_lowercase().as_str() {
+            connector_type::KAFKA => KafkaConnectorDispatcher.with_config(config, registry),
+            _ => bail!("unsupported source connector '{}'", op.connector),
+        }
     }
 }
 
@@ -32,6 +43,16 @@ pub struct ConnectorSinkDispatcher;
 
 impl OperatorConstructor for ConnectorSinkDispatcher {
     fn with_config(&self, config: &[u8], registry: Arc<Registry>) -> Result<ConstructedOperator> {
-        KafkaConnectorDispatcher.with_config(config, registry)
+        let op = ConnectorOp::decode(config)
+            .context("failed decoding connector op for sink dispatch")?;
+        match op.connector.to_ascii_lowercase().as_str() {
+            connector_type::KAFKA => KafkaConnectorDispatcher.with_config(config, registry),
+            connector_type::FILESYSTEM => FilesystemSinkDispatcher.with_config(config, registry),
+            connector_type::DELTA => DeltaSinkDispatcher.with_config(config, registry),
+            connector_type::ICEBERG => IcebergSinkDispatcher.with_config(config, registry),
+            connector_type::S3 => S3SinkDispatcher.with_config(config, registry),
+            "lancedb" => LanceDbSinkDispatcher.with_config(config, registry),
+            _ => bail!("unsupported sink connector '{}'", op.connector),
+        }
     }
 }
diff --git a/src/runtime/streaming/factory/connector/filesystem.rs b/src/runtime/streaming/factory/connector/filesystem.rs
new file mode 100644
index 00000000..94101407
--- /dev/null
+++ b/src/runtime/streaming/factory/connector/filesystem.rs
@@ -0,0 +1,78 @@
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+use std::collections::HashMap;
+use std::sync::Arc;
+
+use anyhow::{Context, Result, bail};
+use prost::Message;
+use protocol::function_stream_graph::ConnectorOp;
+use protocol::function_stream_graph::connector_op::Config;
+
+use crate::runtime::streaming::api::operator::ConstructedOperator;
+use crate::runtime::streaming::factory::connector::sink_props_codec::{
+    apply_common_sink_fields, normalized_props, parse_sink_memory_bytes,
+};
+use crate::runtime::streaming::factory::global::Registry;
+use crate::runtime::streaming::factory::operator_constructor::OperatorConstructor;
+use crate::runtime::streaming::operators::sink::filesystem::{
+    FilesystemFormat, FilesystemSinkOperator, compression_from_str,
+};
+use crate::sql::common::constants::connection_format_value;
+use crate::sql::common::with_option_keys as opt;
+
+pub struct FilesystemSinkDispatcher;
+
+impl OperatorConstructor for FilesystemSinkDispatcher {
+    fn with_config(&self, payload: &[u8], _registry: Arc<Registry>) -> Result<ConstructedOperator> {
+        let op = ConnectorOp::decode(payload).context("failed to decode connector op")?;
+        let props = match op.config {
+            Some(Config::FilesystemSink(cfg)) => filesystem_props(cfg),
+            _ => bail!("filesystem sink expects FilesystemSinkConfig"),
+        };
+        let props = normalized_props(props);
+
+        let format = props
+            .get(opt::FORMAT)
+            .map(String::as_str)
+            .unwrap_or(connection_format_value::PARQUET)
+            .to_ascii_lowercase();
+        let path = props
+            .get(opt::PATH)
+            .cloned()
+            .or_else(|| props.get(opt::SINK_PATH).cloned())
+            .unwrap_or_else(|| ".".to_string());
+        let compression =
+            compression_from_str(props.get(opt::PARQUET_COMPRESSION).map(String::as_str))?;
+        let sink_memory_bytes = parse_sink_memory_bytes(&props)?;
+        let format = match format.as_str() {
+            connection_format_value::CSV => FilesystemFormat::Csv,
+            connection_format_value::PARQUET => FilesystemFormat::Parquet,
+            connection_format_value::JSON => FilesystemFormat::JsonL,
+            connection_format_value::AVRO => FilesystemFormat::Avro,
+            connection_format_value::ORC => FilesystemFormat::Orc,
+            other => bail!("unsupported filesystem sink format '{other}'"),
+        };
+        Ok(ConstructedOperator::Operator(Box::new(
+            FilesystemSinkOperator::try_new(op.name, path, format, compression, sink_memory_bytes)?,
+        )))
+    }
+}
+
+fn filesystem_props(
+    cfg: protocol::function_stream_graph::FilesystemSinkConfig,
+) -> HashMap<String, String> {
+    let mut props = cfg.extra_properties;
+    props.extend(cfg.runtime_properties);
+    apply_common_sink_fields(&mut props, cfg.path, cfg.format, cfg.parquet_compression);
+    props
+}
diff --git a/src/runtime/streaming/factory/connector/iceberg.rs b/src/runtime/streaming/factory/connector/iceberg.rs
new file mode 100644
index 00000000..58e0809f
--- /dev/null
+++ b/src/runtime/streaming/factory/connector/iceberg.rs
@@ -0,0 +1,81 @@
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+use std::collections::HashMap;
+use std::sync::Arc;
+
+use anyhow::{Context, Result, bail};
+use prost::Message;
+use protocol::function_stream_graph::ConnectorOp;
+use protocol::function_stream_graph::connector_op::Config;
+
+use crate::runtime::streaming::api::operator::ConstructedOperator;
+use crate::runtime::streaming::factory::connector::sink_props_codec::{
+    apply_common_sink_fields, normalized_props, parse_sink_memory_bytes,
+};
+use crate::runtime::streaming::factory::global::Registry;
+use crate::runtime::streaming::factory::operator_constructor::OperatorConstructor;
+use crate::runtime::streaming::operators::sink::filesystem::compression_from_str;
+use crate::runtime::streaming::operators::sink::iceberg::{IcebergFormat, IcebergSinkOperator};
+use crate::sql::common::constants::connection_format_value;
+use crate::sql::common::with_option_keys as opt;
+
+pub struct IcebergSinkDispatcher;
+
+impl OperatorConstructor for IcebergSinkDispatcher {
+    fn with_config(&self, payload: &[u8], _registry: Arc<Registry>) -> Result<ConstructedOperator> {
+        let op = ConnectorOp::decode(payload).context("failed to decode connector op")?;
+        let props = match op.config {
+            Some(Config::IcebergSink(cfg)) => iceberg_props(cfg),
+            _ => bail!("iceberg sink expects IcebergSinkConfig"),
+        };
+        let props = normalized_props(props);
+
+        let format = props
+            .get(opt::FORMAT)
+            .map(String::as_str)
+            .unwrap_or(connection_format_value::PARQUET)
+            .to_ascii_lowercase();
+        let path = props
+            .get(opt::PATH)
+            .cloned()
+            .or_else(|| props.get(opt::SINK_PATH).cloned())
+            .unwrap_or_else(|| ".".to_string());
+        let compression =
+            compression_from_str(props.get(opt::PARQUET_COMPRESSION).map(String::as_str))?;
+        let sink_memory_bytes = parse_sink_memory_bytes(&props)?;
+        let format = match format.as_str() {
+            connection_format_value::CSV => IcebergFormat::Csv,
+            connection_format_value::PARQUET => IcebergFormat::Parquet,
+            other => bail!("unsupported iceberg sink format '{other}'"),
+        };
+        Ok(ConstructedOperator::Operator(Box::new(
+            IcebergSinkOperator::try_new(
+                op.name,
+                path,
+                format,
+                compression,
+                sink_memory_bytes,
+                props,
+            )?,
+        )))
+    }
+}
+
+fn iceberg_props(
+    cfg: protocol::function_stream_graph::IcebergSinkConfig,
+) -> HashMap<String, String> {
+    let mut props = cfg.extra_properties;
+    props.extend(cfg.runtime_properties);
+    apply_common_sink_fields(&mut props, cfg.path, cfg.format, cfg.parquet_compression);
+    props
+}
diff --git a/src/runtime/streaming/factory/connector/kafka.rs b/src/runtime/streaming/factory/connector/kafka.rs
index 9d2f114d..17838e3e 100644
--- a/src/runtime/streaming/factory/connector/kafka.rs
+++ b/src/runtime/streaming/factory/connector/kafka.rs
@@ -136,8 +136,8 @@ impl OperatorConstructor for KafkaConnectorDispatcher {
                 Self::build_kafka_source(&op.name, cfg, fs_schema)
             }
             Some(Config::KafkaSink(ref cfg)) => Self::build_kafka_sink(&op.name, cfg, fs_schema),
-            Some(Config::Generic(_)) => bail!(
-                "ConnectorOp '{}': GenericConnectorConfig dispatch not yet implemented",
+            Some(_) => bail!(
+                "ConnectorOp '{}': received non-kafka connector config for Kafka dispatcher",
                 op.name
             ),
             None => bail!("ConnectorOp '{}' has no configuration payload", op.name),
diff --git a/src/runtime/streaming/factory/connector/lancedb.rs b/src/runtime/streaming/factory/connector/lancedb.rs
new file mode 100644
index 00000000..51f298f3
--- /dev/null
+++ b/src/runtime/streaming/factory/connector/lancedb.rs
@@ -0,0 +1,106 @@
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+use std::collections::HashMap;
+use std::sync::Arc;
+
+use anyhow::{Context, Result, bail};
+use prost::Message;
+use protocol::function_stream_graph::ConnectorOp;
+use protocol::function_stream_graph::connector_op::Config;
+
+use crate::runtime::streaming::api::operator::ConstructedOperator;
+use crate::runtime::streaming::factory::connector::sink_props_codec::{
+    apply_common_sink_fields, normalized_props,
+};
+use crate::runtime::streaming::factory::global::Registry;
+use crate::runtime::streaming::factory::operator_constructor::OperatorConstructor;
+use crate::runtime::streaming::operators::sink::lancedb::LanceDbSinkOperator;
+use crate::sql::common::constants::connection_format_value;
+use crate::sql::common::with_option_keys as opt;
+
+pub struct LanceDbSinkDispatcher;
+
+impl OperatorConstructor for LanceDbSinkDispatcher {
+    fn with_config(&self, payload: &[u8], _registry: Arc<Registry>) -> Result<ConstructedOperator> {
+        let op = ConnectorOp::decode(payload).context("failed to decode connector op")?;
+        let props = match op.config {
+            Some(Config::LancedbSink(cfg)) => lancedb_props(cfg),
+            _ => bail!("lanceDB connector expects LanceDbSinkConfig"),
+        };
+        let props = normalized_props(props);
+
+        let format = props
+            .get(opt::FORMAT)
+            .map(String::as_str)
+            .unwrap_or(connection_format_value::LANCE)
+            .to_ascii_lowercase();
+        if format != connection_format_value::LANCE {
+            bail!("lanceDB requires format='lance', got '{format}'");
+        }
+
+        let dataset_uri = resolve_lance_uri(&props)?;
+        let sink = LanceDbSinkOperator::new(op.name, dataset_uri);
+        Ok(ConstructedOperator::Operator(Box::new(sink)))
+    }
+}
+
+fn resolve_lance_uri(props: &HashMap<String, String>) -> Result<String> {
+    let path = props
+        .get(opt::PATH)
+        .cloned()
+        .or_else(|| props.get(opt::SINK_PATH).cloned())
+        .unwrap_or_else(|| ".".to_string());
+
+    // If path already contains a fully-qualified URI scheme, use it as-is.
+    if path.contains("://") {
+        return Ok(path);
+    }
+
+    if let Some(bucket) = props.get(opt::S3_BUCKET) {
+        let trimmed = path.trim_matches('/');
+        if trimmed.is_empty() {
+            Ok(format!("s3://{bucket}"))
+        } else {
+            Ok(format!("s3://{bucket}/{trimmed}"))
+        }
+    } else {
+        Ok(path)
+    }
+}
+
+fn lancedb_props(
+    cfg: protocol::function_stream_graph::LanceDbSinkConfig,
+) -> HashMap<String, String> {
+    let mut props = cfg.extra_properties;
+    props.extend(cfg.runtime_properties);
+    apply_common_sink_fields(&mut props, cfg.path, cfg.format, None);
+    if let Some(v) = cfg.s3_bucket {
+        props.insert(opt::S3_BUCKET.to_string(), v);
+    }
+    if let Some(v) = cfg.s3_region {
+        props.insert(opt::S3_REGION.to_string(), v);
+    }
+    if let Some(v) = cfg.s3_endpoint {
+        props.insert(opt::S3_ENDPOINT.to_string(), v);
+    }
+    if let Some(v) = cfg.s3_access_key_id {
+        props.insert(opt::S3_ACCESS_KEY_ID.to_string(), v);
+    }
+    if let Some(v) = cfg.s3_secret_access_key {
+        props.insert(opt::S3_SECRET_ACCESS_KEY.to_string(), v);
+    }
+    if let Some(v) = cfg.s3_session_token {
+        props.insert(opt::S3_SESSION_TOKEN.to_string(), v);
+    }
+    props
+}
diff --git a/src/runtime/streaming/factory/connector/mod.rs b/src/runtime/streaming/factory/connector/mod.rs
index 381de89c..d1a6d7f3 100644
--- a/src/runtime/streaming/factory/connector/mod.rs
+++ b/src/runtime/streaming/factory/connector/mod.rs
@@ -10,8 +10,19 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
+pub mod delta;
 mod dispatchers;
+pub mod filesystem;
+pub mod iceberg;
 pub mod kafka;
+pub mod lancedb;
+pub mod s3;
+pub mod sink_props_codec;
 
+pub use delta::DeltaSinkDispatcher;
 pub use dispatchers::{ConnectorSinkDispatcher, ConnectorSourceDispatcher};
+pub use filesystem::FilesystemSinkDispatcher;
+pub use iceberg::IcebergSinkDispatcher;
 pub use kafka::KafkaConnectorDispatcher;
+pub use lancedb::LanceDbSinkDispatcher;
+pub use s3::S3SinkDispatcher;
diff --git a/src/runtime/streaming/factory/connector/s3.rs b/src/runtime/streaming/factory/connector/s3.rs
new file mode 100644
index 00000000..4b67fb9e
--- /dev/null
+++ b/src/runtime/streaming/factory/connector/s3.rs
@@ -0,0 +1,90 @@
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+use std::collections::HashMap;
+use std::sync::Arc;
+
+use anyhow::{Context, Result, bail};
+use prost::Message;
+use protocol::function_stream_graph::ConnectorOp;
+use protocol::function_stream_graph::connector_op::Config;
+
+use crate::runtime::streaming::api::operator::ConstructedOperator;
+use crate::runtime::streaming::factory::connector::sink_props_codec::{
+    apply_common_sink_fields, normalized_props,
+};
+use crate::runtime::streaming::factory::global::Registry;
+use crate::runtime::streaming::factory::operator_constructor::OperatorConstructor;
+use crate::runtime::streaming::operators::sink::s3::{
+    S3Format, S3SinkOperator, compression_from_str,
+};
+use crate::sql::common::constants::connection_format_value;
+use crate::sql::common::with_option_keys as opt;
+
+pub struct S3SinkDispatcher;
+
+impl OperatorConstructor for S3SinkDispatcher {
+    fn with_config(&self, payload: &[u8], _registry: Arc<Registry>) -> Result<ConstructedOperator> {
+        let op = ConnectorOp::decode(payload).context("failed to decode connector op")?;
+        let props = match op.config {
+            Some(Config::S3Sink(cfg)) => s3_props(cfg),
+            _ => bail!("s3 sink expects S3SinkConfig"),
+        };
+        let props = normalized_props(props);
+
+        let format = props
+            .get(opt::FORMAT)
+            .map(String::as_str)
+            .unwrap_or(connection_format_value::PARQUET)
+            .to_ascii_lowercase();
+        let path = props
+            .get(opt::PATH)
+            .cloned()
+            .or_else(|| props.get(opt::SINK_PATH).cloned())
+            .unwrap_or_default();
+        let compression =
+            compression_from_str(props.get(opt::PARQUET_COMPRESSION).map(String::as_str))?;
+        let format = match format.as_str() {
+            connection_format_value::CSV => S3Format::Csv,
+            connection_format_value::PARQUET => S3Format::Parquet,
+            other => bail!("unsupported s3 sink format '{other}'"),
+        };
+        Ok(ConstructedOperator::Operator(Box::new(
+            S3SinkOperator::try_new(op.name, path, format, compression, props)?,
+        )))
+    }
+}
+
+fn s3_props(cfg: protocol::function_stream_graph::S3SinkConfig) -> HashMap<String, String> {
+    let mut props = cfg.extra_properties;
+    props.extend(cfg.runtime_properties);
+    apply_common_sink_fields(&mut props, cfg.path, cfg.format, cfg.parquet_compression);
+    if !cfg.bucket.is_empty() {
+        props.insert(opt::S3_BUCKET.to_string(), cfg.bucket);
+    }
+    if !cfg.region.is_empty() {
+        props.insert(opt::S3_REGION.to_string(), cfg.region);
+    }
+    if let Some(v) = cfg.endpoint {
+        props.insert(opt::S3_ENDPOINT.to_string(), v);
+    }
+    if let Some(v) = cfg.access_key_id {
+        props.insert(opt::S3_ACCESS_KEY_ID.to_string(), v);
+    }
+    if let Some(v) = cfg.secret_access_key {
+        props.insert(opt::S3_SECRET_ACCESS_KEY.to_string(), v);
+    }
+    if let Some(v) = cfg.session_token {
+        props.insert(opt::S3_SESSION_TOKEN.to_string(), v);
+    }
+    props
+}
diff --git a/src/runtime/streaming/factory/connector/sink_props_codec.rs b/src/runtime/streaming/factory/connector/sink_props_codec.rs
new file mode 100644
index 00000000..a4128f24
--- /dev/null
+++ b/src/runtime/streaming/factory/connector/sink_props_codec.rs
@@ -0,0 +1,97 @@
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+use std::collections::HashMap;
+
+use anyhow::{Context, Result, bail};
+use protocol::function_stream_graph::{ParquetCompressionProto, SinkFormatProto};
+
+use crate::config::global_config::DEFAULT_SINK_BUFFER_MEMORY_BYTES;
+use crate::sql::common::constants::{connection_format_value, parquet_compression_value};
+use crate::sql::common::with_option_keys as opt;
+
+pub fn normalized_props(raw: HashMap<String, String>) -> HashMap<String, String> {
+    raw.into_iter()
+        .map(|(k, v)| (k.to_ascii_lowercase(), normalize_value(&v)))
+        .collect()
+}
+
+pub fn apply_common_sink_fields(
+    props: &mut HashMap<String, String>,
+    path: String,
+    format: i32,
+    parquet_compression: Option<i32>,
+) {
+    if !path.is_empty() {
+        props.insert(opt::PATH.to_string(), path);
+    }
+    if let Some(fmt) = SinkFormatProto::try_from(format)
+        .ok()
+        .and_then(sink_format_as_str)
+    {
+        props.insert(opt::FORMAT.to_string(), fmt.to_string());
+    }
+    if let Some(comp) = parquet_compression
+        .and_then(|c| ParquetCompressionProto::try_from(c).ok())
+        .and_then(parquet_compression_as_str)
+    {
+        props.insert(opt::PARQUET_COMPRESSION.to_string(), comp.to_string());
+    }
+}
+
+pub fn sink_format_as_str(v: SinkFormatProto) -> Option<&'static str> {
+    use SinkFormatProto as F;
+    match v {
+        F::SinkFormatCsv => Some(connection_format_value::CSV),
+        F::SinkFormatJsonl => Some(connection_format_value::JSONL),
+        F::SinkFormatAvro => Some(connection_format_value::AVRO),
+        F::SinkFormatParquet => Some(connection_format_value::PARQUET),
+        F::SinkFormatOrc => Some(connection_format_value::ORC),
+        F::SinkFormatLance => Some(connection_format_value::LANCE),
+        F::SinkFormatUnspecified => None,
+    }
+}
+
+pub fn parquet_compression_as_str(v: ParquetCompressionProto) -> Option<&'static str> {
+    use ParquetCompressionProto as C;
+    match v {
+        C::ParquetCompressionUncompressed => Some(parquet_compression_value::UNCOMPRESSED),
+        C::ParquetCompressionSnappy => Some(parquet_compression_value::SNAPPY),
+        C::ParquetCompressionGzip => Some(parquet_compression_value::GZIP),
+        C::ParquetCompressionZstd => Some(parquet_compression_value::ZSTD),
+        C::ParquetCompressionLz4 => Some(parquet_compression_value::LZ4),
+        C::ParquetCompressionLz4Raw => Some(parquet_compression_value::LZ4_RAW),
+        C::ParquetCompressionUnspecified => None,
+    }
+}
+
+pub fn parse_sink_memory_bytes(props: &HashMap<String, String>) -> Result<u64> {
+    let parsed = match props.get(opt::SINK_MEMORY_BYTES) {
+        Some(raw) => raw
+            .parse::<u64>()
+            .with_context(|| format!("invalid '{}' value '{}'", opt::SINK_MEMORY_BYTES, raw))?,
+        None => DEFAULT_SINK_BUFFER_MEMORY_BYTES,
+    };
+    if parsed == 0 {
+        bail!("'{}' must be > 0", opt::SINK_MEMORY_BYTES);
+    }
+    Ok(parsed)
+}
+
+fn normalize_value(v: &str) -> String {
+    let s = v.trim();
+    if (s.starts_with('\'') && s.ends_with('\'')) || (s.starts_with('"') && s.ends_with('"')) {
+        s[1..s.len() - 1].to_string()
+    } else {
+        s.to_string()
+    }
+}
diff --git a/src/runtime/streaming/format/encoder.rs b/src/runtime/streaming/format/encoder.rs
new file mode 100644
index 00000000..198b06b2
--- /dev/null
+++ b/src/runtime/streaming/format/encoder.rs
@@ -0,0 +1,427 @@
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+use std::io::Cursor;
+
+use anyhow::{Context, Result, bail};
+use apache_avro::types::Value as AvroValue;
+use apache_avro::{Codec as AvroCodec, Schema as AvroSchema, Writer as AvroWriter};
+use arrow::csv::WriterBuilder as CsvWriterBuilder;
+use arrow::json::LineDelimitedWriter;
+use arrow_array::{
+    Array, BinaryArray, BooleanArray, Float32Array, Float64Array, Int8Array, Int16Array,
+    Int32Array, Int64Array, LargeBinaryArray, LargeStringArray, RecordBatch, StringArray,
+    TimestampMicrosecondArray, TimestampMillisecondArray, TimestampNanosecondArray,
+    TimestampSecondArray, UInt8Array, UInt16Array, UInt32Array, UInt64Array,
+};
+use arrow_ipc::writer::FileWriter as ArrowIpcFileWriter;
+use arrow_schema::{DataType, Field, TimeUnit};
+use parquet::arrow::ArrowWriter;
+use parquet::basic::Compression;
+use parquet::file::properties::WriterProperties;
+use serde_json::{Map as JsonMap, Value as JsonValue};
+
+/// Pure in-memory format encoder for sink payload generation.
+/// It only converts `RecordBatch` into bytes and does not perform any I/O.
+pub struct FormatEncoder;
+
+impl FormatEncoder {
+    /// Encode batches into CSV bytes.
+    pub fn encode_csv(batches: &[RecordBatch]) -> Result<Vec<u8>> {
+        if batches.is_empty() {
+            return Ok(Vec::new());
+        }
+
+        let mut out = Vec::new();
+        let mut writer = CsvWriterBuilder::new().with_header(true).build(&mut out);
+        for batch in batches {
+            writer
+                .write(batch)
+                .context("failed to encode record batch to CSV")?;
+        }
+        drop(writer);
+        Ok(out)
+    }
+
+    /// Encode batches into Parquet bytes.
+    pub fn encode_parquet(batches: &[RecordBatch], compression: Compression) -> Result<Vec<u8>> {
+        if batches.is_empty() {
+            return Ok(Vec::new());
+        }
+
+        let schema = batches[0].schema();
+        let props = WriterProperties::builder()
+            .set_compression(compression)
+            .build();
+        let mut cursor = Cursor::new(Vec::new());
+        let mut writer = ArrowWriter::try_new(&mut cursor, schema, Some(props))
+            .context("failed to init parquet writer")?;
+
+        for batch in batches {
+            writer
+                .write(batch)
+                .context("failed to encode record batch to parquet")?;
+        }
+        writer.close().context("failed to finalize parquet")?;
+        Ok(cursor.into_inner())
+    }
+
+    /// Encode batches into NDJSON (JSON Lines) bytes.
+    pub fn encode_jsonl(batches: &[RecordBatch]) -> Result<Vec<u8>> {
+        if batches.is_empty() {
+            return Ok(Vec::new());
+        }
+
+        let mut out = Vec::new();
+        let mut writer = LineDelimitedWriter::new(&mut out);
+        for batch in batches {
+            writer
+                .write(batch)
+                .context("failed to encode record batch to JSONL")?;
+        }
+        writer.finish().context("failed to finalize JSONL stream")?;
+        Ok(out)
+    }
+
+    pub fn encode_avro(batches: &[RecordBatch]) -> Result<Vec<u8>> {
+        if batches.is_empty() {
+            return Ok(Vec::new());
+        }
+
+        let schema_json = build_avro_schema_json(&batches[0])?;
+        let avro_schema =
+            AvroSchema::parse_str(&schema_json).context("failed to parse generated avro schema")?;
+        let mut writer = AvroWriter::with_codec(&avro_schema, Vec::new(), AvroCodec::Null);
+
+        for batch in batches {
+            let accessors = build_column_accessors(batch)?;
+            let schema = batch.schema();
+            let fields = schema.fields();
+            let col_names: Vec<String> = fields.iter().map(|f| f.name().clone()).collect();
+
+            for row_idx in 0..batch.num_rows() {
+                let mut row_records = Vec::with_capacity(accessors.len());
+                for (col_idx, accessor) in accessors.iter().enumerate() {
+                    let nullable = fields[col_idx].is_nullable();
+                    let val = accessor.avro_value_at(row_idx, nullable)?;
+                    row_records.push((col_names[col_idx].clone(), val));
+                }
+                writer
+                    .append(AvroValue::Record(row_records))
+                    .map_err(|e| anyhow::anyhow!("failed to append row into avro writer: {e}"))?;
+            }
+        }
+
+        writer.flush().context("failed to flush avro writer")?;
+        writer
+            .into_inner()
+            .context("failed to finalize avro container bytes")
+    }
+
+    pub fn encode_orc(batches: &[RecordBatch]) -> Result<Vec<u8>> {
+        if batches.is_empty() {
+            return Ok(Vec::new());
+        }
+
+        let schema = batches[0].schema();
+        let mut out = Cursor::new(Vec::new());
+        let mut writer = ArrowIpcFileWriter::try_new(&mut out, &schema)
+            .context("failed to init ORC-compatible file writer")?;
+
+        for batch in batches {
+            writer
+                .write(batch)
+                .context("failed to encode record batch into ORC-compatible payload")?;
+        }
+
+        writer
+            .finish()
+            .context("failed to finalize ORC-compatible payload")?;
+        Ok(out.into_inner())
+    }
+}
+
+fn build_avro_schema_json(batch: &RecordBatch) -> Result<String> {
+    let fields = batch
+        .schema()
+        .fields()
+        .iter()
+        .map(|field| {
+            let avro_type = avro_type_for_field(field)?;
+            let field_type = if field.is_nullable() {
+                JsonValue::Array(vec![JsonValue::String("null".to_string()), avro_type])
+            } else {
+                avro_type
+            };
+            Ok(JsonValue::Object(JsonMap::from_iter([
+                (
+                    "name".to_string(),
+                    JsonValue::String(field.name().to_string()),
+                ),
+                ("type".to_string(), field_type),
+                ("default".to_string(), JsonValue::Null),
+            ])))
+        })
+        .collect::<Result<Vec<_>>>()?;
+
+    let schema = JsonValue::Object(JsonMap::from_iter([
+        ("type".to_string(), JsonValue::String("record".to_string())),
+        (
+            "name".to_string(),
+            JsonValue::String("FunctionStreamRecord".to_string()),
+        ),
+        ("fields".to_string(), JsonValue::Array(fields)),
+    ]));
+
+    serde_json::to_string(&schema).context("failed to serialize avro schema json")
+}
+
+fn avro_type_for_field(field: &Field) -> Result<JsonValue> {
+    let ty = match field.data_type() {
+        DataType::Boolean => JsonValue::String("boolean".to_string()),
+        DataType::Int8 | DataType::Int16 | DataType::Int32 => JsonValue::String("int".to_string()),
+        DataType::UInt8
+        | DataType::UInt16
+        | DataType::UInt32
+        | DataType::Int64
+        | DataType::UInt64 => JsonValue::String("long".to_string()),
+        DataType::Float32 => JsonValue::String("float".to_string()),
+        DataType::Float64 => JsonValue::String("double".to_string()),
+        DataType::Utf8 | DataType::LargeUtf8 => JsonValue::String("string".to_string()),
+        DataType::Binary | DataType::LargeBinary => JsonValue::String("bytes".to_string()),
+        DataType::Date32 => JsonValue::Object(JsonMap::from_iter([
+            ("type".to_string(), JsonValue::String("int".to_string())),
+            (
+                "logicalType".to_string(),
+                JsonValue::String("date".to_string()),
+            ),
+        ])),
+        DataType::Timestamp(unit, _) => {
+            let logical = match unit {
+                TimeUnit::Second | TimeUnit::Millisecond => "timestamp-millis",
+                TimeUnit::Microsecond | TimeUnit::Nanosecond => "timestamp-micros",
+            };
+            JsonValue::Object(JsonMap::from_iter([
+                ("type".to_string(), JsonValue::String("long".to_string())),
+                (
+                    "logicalType".to_string(),
+                    JsonValue::String(logical.to_string()),
+                ),
+            ]))
+        }
+        other => bail!("unsupported data type for avro encoding: {other:?}"),
+    };
+    Ok(ty)
+}
+
+/// Downcasts each column once per batch; row iteration only matches on this enum.
+enum ColumnAccessor<'a> {
+    Boolean(&'a BooleanArray),
+    Int8(&'a Int8Array),
+    Int16(&'a Int16Array),
+    Int32(&'a Int32Array),
+    Int64(&'a Int64Array),
+    UInt8(&'a UInt8Array),
+    UInt16(&'a UInt16Array),
+    UInt32(&'a UInt32Array),
+    UInt64(&'a UInt64Array),
+    Float32(&'a Float32Array),
+    Float64(&'a Float64Array),
+    Utf8(&'a StringArray),
+    LargeUtf8(&'a LargeStringArray),
+    Binary(&'a BinaryArray),
+    LargeBinary(&'a LargeBinaryArray),
+    Date32(&'a Int32Array),
+    TimestampSec(&'a TimestampSecondArray),
+    TimestampMs(&'a TimestampMillisecondArray),
+    TimestampUs(&'a TimestampMicrosecondArray),
+    TimestampNs(&'a TimestampNanosecondArray),
+}
+
+impl<'a> ColumnAccessor<'a> {
+    fn avro_value_at(&self, row: usize, nullable: bool) -> Result<AvroValue> {
+        let is_null = match self {
+            Self::Boolean(a) => a.is_null(row),
+            Self::Int8(a) => a.is_null(row),
+            Self::Int16(a) => a.is_null(row),
+            Self::Int32(a) => a.is_null(row),
+            Self::Int64(a) => a.is_null(row),
+            Self::UInt8(a) => a.is_null(row),
+            Self::UInt16(a) => a.is_null(row),
+            Self::UInt32(a) => a.is_null(row),
+            Self::UInt64(a) => a.is_null(row),
+            Self::Float32(a) => a.is_null(row),
+            Self::Float64(a) => a.is_null(row),
+            Self::Utf8(a) => a.is_null(row),
+            Self::LargeUtf8(a) => a.is_null(row),
+            Self::Binary(a) => a.is_null(row),
+            Self::LargeBinary(a) => a.is_null(row),
+            Self::Date32(a) => a.is_null(row),
+            Self::TimestampSec(a) => a.is_null(row),
+            Self::TimestampMs(a) => a.is_null(row),
+            Self::TimestampUs(a) => a.is_null(row),
+            Self::TimestampNs(a) => a.is_null(row),
+        };
+
+        if is_null {
+            if !nullable {
+                bail!("null value in non-nullable avro field at row {row}");
+            }
+            return Ok(AvroValue::Union(0, Box::new(AvroValue::Null)));
+        }
+
+        let raw = match self {
+            Self::Boolean(a) => AvroValue::Boolean(a.value(row)),
+            Self::Int8(a) => AvroValue::Int(i32::from(a.value(row))),
+            Self::Int16(a) => AvroValue::Int(i32::from(a.value(row))),
+            Self::Int32(a) => AvroValue::Int(a.value(row)),
+            Self::Int64(a) => AvroValue::Long(a.value(row)),
+            Self::UInt8(a) => AvroValue::Int(i32::from(a.value(row))),
+            Self::UInt16(a) => AvroValue::Int(i32::from(a.value(row))),
+            Self::UInt32(a) => AvroValue::Long(i64::from(a.value(row))),
+            Self::UInt64(a) => {
+                let v = a.value(row);
+                AvroValue::Long(i64::try_from(v).with_context(|| {
+                    format!("UInt64 value {v} does not fit Avro long for row {row}")
+                })?)
+            }
+            Self::Float32(a) => {
+                let v = a.value(row);
+                if !v.is_finite() {
+                    bail!("non-finite f32 at row {row}: {v}");
+                }
+                AvroValue::Float(v)
+            }
+            Self::Float64(a) => {
+                let v = a.value(row);
+                if !v.is_finite() {
+                    bail!("non-finite f64 at row {row}: {v}");
+                }
+                AvroValue::Double(v)
+            }
+            Self::Utf8(a) => AvroValue::String(a.value(row).to_string()),
+            Self::LargeUtf8(a) => AvroValue::String(a.value(row).to_string()),
+            Self::Binary(a) => AvroValue::Bytes(a.value(row).to_vec()),
+            Self::LargeBinary(a) => AvroValue::Bytes(a.value(row).to_vec()),
+            Self::Date32(a) => AvroValue::Int(a.value(row)),
+            Self::TimestampSec(a) => AvroValue::Long(a.value(row).saturating_mul(1000)),
+            Self::TimestampMs(a) => AvroValue::Long(a.value(row)),
+            Self::TimestampUs(a) => AvroValue::Long(a.value(row)),
+            Self::TimestampNs(a) => AvroValue::Long(a.value(row) / 1000),
+        };
+
+        Ok(if nullable {
+            AvroValue::Union(1, Box::new(raw))
+        } else {
+            raw
+        })
+    }
+}
+
+fn build_column_accessors(batch: &RecordBatch) -> Result<Vec<ColumnAccessor<'_>>> {
+    let mut accessors = Vec::with_capacity(batch.num_columns());
+    for col in batch.columns() {
+        let accessor = match col.data_type() {
+            DataType::Boolean => ColumnAccessor::Boolean(
+                col.as_any()
+                    .downcast_ref()
+                    .context("expected BooleanArray")?,
+            ),
+            DataType::Int8 => {
+                ColumnAccessor::Int8(col.as_any().downcast_ref().context("expected Int8Array")?)
+            }
+            DataType::Int16 => {
+                ColumnAccessor::Int16(col.as_any().downcast_ref().context("expected Int16Array")?)
+            }
+            DataType::Int32 => {
+                ColumnAccessor::Int32(col.as_any().downcast_ref().context("expected Int32Array")?)
+            }
+            DataType::Int64 => {
+                ColumnAccessor::Int64(col.as_any().downcast_ref().context("expected Int64Array")?)
+            }
+            DataType::UInt8 => {
+                ColumnAccessor::UInt8(col.as_any().downcast_ref().context("expected UInt8Array")?)
+            }
+            DataType::UInt16 => ColumnAccessor::UInt16(
+                col.as_any()
+                    .downcast_ref()
+                    .context("expected UInt16Array")?,
+            ),
+            DataType::UInt32 => ColumnAccessor::UInt32(
+                col.as_any()
+                    .downcast_ref()
+                    .context("expected UInt32Array")?,
+            ),
+            DataType::UInt64 => ColumnAccessor::UInt64(
+                col.as_any()
+                    .downcast_ref()
+                    .context("expected UInt64Array")?,
+            ),
+            DataType::Float32 => ColumnAccessor::Float32(
+                col.as_any()
+                    .downcast_ref()
+                    .context("expected Float32Array")?,
+            ),
+            DataType::Float64 => ColumnAccessor::Float64(
+                col.as_any()
+                    .downcast_ref()
+                    .context("expected Float64Array")?,
+            ),
+            DataType::Utf8 => ColumnAccessor::Utf8(
+                col.as_any()
+                    .downcast_ref()
+                    .context("expected StringArray")?,
+            ),
+            DataType::LargeUtf8 => ColumnAccessor::LargeUtf8(
+                col.as_any()
+                    .downcast_ref()
+                    .context("expected LargeStringArray")?,
+            ),
+            DataType::Binary => ColumnAccessor::Binary(
+                col.as_any()
+                    .downcast_ref()
+                    .context("expected BinaryArray")?,
+            ),
+            DataType::LargeBinary => ColumnAccessor::LargeBinary(
+                col.as_any()
+                    .downcast_ref()
+                    .context("expected LargeBinaryArray")?,
+            ),
+            DataType::Date32 => {
+                ColumnAccessor::Date32(col.as_any().downcast_ref().context("expected Int32Array")?)
+            }
+            DataType::Timestamp(TimeUnit::Second, _) => ColumnAccessor::TimestampSec(
+                col.as_any()
+                    .downcast_ref()
+                    .context("expected TimestampSecondArray")?,
+            ),
+            DataType::Timestamp(TimeUnit::Millisecond, _) => ColumnAccessor::TimestampMs(
+                col.as_any()
+                    .downcast_ref()
+                    .context("expected TimestampMillisecondArray")?,
+            ),
+            DataType::Timestamp(TimeUnit::Microsecond, _) => ColumnAccessor::TimestampUs(
+                col.as_any()
+                    .downcast_ref()
+                    .context("expected TimestampMicrosecondArray")?,
+            ),
+            DataType::Timestamp(TimeUnit::Nanosecond, _) => ColumnAccessor::TimestampNs(
+                col.as_any()
+                    .downcast_ref()
+                    .context("expected TimestampNanosecondArray")?,
+            ),
+            other => bail!("unsupported data type for avro column accessor: {other:?}"),
+        };
+        accessors.push(accessor);
+    }
+    Ok(accessors)
+}
diff --git a/src/runtime/streaming/format/mod.rs b/src/runtime/streaming/format/mod.rs
index d5e63a9d..3112a1fe 100644
--- a/src/runtime/streaming/format/mod.rs
+++ b/src/runtime/streaming/format/mod.rs
@@ -12,6 +12,7 @@
 
 pub mod config;
 pub mod deserializer;
+pub mod encoder;
 pub mod json_encoder;
 pub mod serializer;
 
diff --git a/src/runtime/streaming/job/job_manager.rs b/src/runtime/streaming/job/job_manager.rs
index a9bc546f..e4b9916b 100644
--- a/src/runtime/streaming/job/job_manager.rs
+++ b/src/runtime/streaming/job/job_manager.rs
@@ -24,9 +24,7 @@ use tokio_stream::wrappers::ReceiverStream;
 use tracing::{error, info, warn};
 
 use protocol::function_stream_graph::{ChainedOperator, FsProgram};
-use protocol::storage::{
-    KafkaSourceSubtaskCheckpoint, SourceCheckpointPayload, source_checkpoint_payload,
-};
+use protocol::storage::{SourceCheckpointInfo, source_checkpoint_info};
 
 use crate::config::{
     DEFAULT_CHECKPOINT_INTERVAL_MS, DEFAULT_OPERATOR_STATE_STORE_MEMORY_BYTES,
@@ -157,7 +155,6 @@ struct CheckpointCoordinatorConfig {
     expected_pipeline_ids: HashSet<u32>,
     interval_ms: u64,
     start_epoch: u64,
-    job_state_dir: PathBuf,
     timeout: Duration,
 }
 
@@ -170,25 +167,6 @@ impl PipelineRunner {
     }
 }
 
-fn decode_kafka_checkpoints_from_source_payloads(
-    payloads: Vec<SourceCheckpointPayload>,
-    epoch: u64,
-) -> Vec<KafkaSourceSubtaskCheckpoint> {
-    let mut out = Vec::new();
-    for p in payloads {
-        match p.checkpoint {
-            Some(source_checkpoint_payload::Checkpoint::Kafka(mut cp)) => {
-                if cp.checkpoint_epoch != epoch {
-                    cp.checkpoint_epoch = epoch;
-                }
-                out.push(cp);
-            }
-            None => warn!("Skip empty source checkpoint payload"),
-        }
-    }
-    out
-}
-
 impl JobManager {
     pub fn new(
         operator_factory: Arc<OperatorFactory>,
@@ -255,7 +233,7 @@ impl JobManager {
         self.state_config.pipeline_parallelism
     }
 
-    /// Per-job state directory (Kafka offset snapshots, operator state roots, etc.).
+    /// Per-job state directory (source offset snapshots, operator state roots, etc.).
     #[inline]
     pub fn job_state_directory(&self, job_id: &str) -> PathBuf {
         self.state_base_dir.join(job_id)
@@ -267,6 +245,7 @@ impl JobManager {
         program: FsProgram,
         custom_checkpoint_interval_ms: Option<u64>,
         recovery_epoch: Option<u64>,
+        source_checkpoint_infos: Vec<SourceCheckpointInfo>,
     ) -> Result<String> {
         let mut edge_manager = EdgeManager::build(&program.nodes, &program.edges);
         let mut pipelines = HashMap::with_capacity(program.nodes.len());
@@ -295,6 +274,7 @@ impl JobManager {
                     &job_state_dir,
                     job_master_tx.clone(),
                     safe_epoch,
+                    &source_checkpoint_infos,
                 )
                 .with_context(|| {
                     format!(
@@ -322,7 +302,6 @@ impl JobManager {
             expected_pipeline_ids,
             interval_ms,
             start_epoch: safe_epoch + 1,
-            job_state_dir: job_state_dir.clone(),
             timeout: Duration::from_millis(interval_ms.max(1) * 3),
         });
 
@@ -530,6 +509,7 @@ impl JobManager {
         job_state_dir: &Path,
         job_master_tx: mpsc::Sender<JobMasterEvent>,
         recovery_epoch: u64,
+        source_checkpoint_infos: &[SourceCheckpointInfo],
     ) -> Result<(PhysicalPipeline, bool)> {
         let (raw_inboxes, raw_outboxes) =
             edge_manager.take_endpoints(pipeline_id).with_context(|| {
@@ -610,7 +590,20 @@ impl JobManager {
             Some(job_master_tx.clone()),
         );
 
-        let runner = if let Some(source) = chain.source {
+        let runner = if let Some(mut source) = chain.source {
+            // Filter checkpoint records for this pipeline and inject into the source operator
+            // so it can restore partition offsets in on_start without touching TaskContext.
+            let pipeline_checkpoint_infos: Vec<SourceCheckpointInfo> = source_checkpoint_infos
+                .iter()
+                .filter(|info| match &info.info {
+                    Some(source_checkpoint_info::Info::Kafka(cp)) => cp.pipeline_id == pipeline_id,
+                    None => false,
+                })
+                .cloned()
+                .collect();
+            if !pipeline_checkpoint_infos.is_empty() {
+                source.set_recovery_checkpoint(pipeline_checkpoint_infos);
+            }
             let chain_head = ChainBuilder::build(chain.operators);
             PipelineRunner::Source(SourceDriver::new(source, chain_head, ctx, control_rx))
         } else {
@@ -733,7 +726,6 @@ impl JobManager {
                 expected_pipeline_ids,
                 interval_ms,
                 start_epoch,
-                job_state_dir,
                 timeout,
             } = cfg;
             if interval_ms == 0 {
@@ -749,7 +741,7 @@ impl JobManager {
                 epoch: u64,
                 missing_acks: HashSet<u32>,
                 start_time: Instant,
-                source_reports: Vec<SourceCheckpointPayload>,
+                source_infos: Vec<SourceCheckpointInfo>,
             }
             let mut active_checkpoint: Option<PendingCheckpoint> = None;
 
@@ -768,15 +760,15 @@ impl JobManager {
                             JobMasterEvent::CheckpointAck {
                                 pipeline_id,
                                 epoch,
-                                source_payloads,
+                                source_infos,
                             } => {
                                 if let Some(pending) = &mut active_checkpoint {
                                     if pending.epoch != epoch {
                                         continue;
                                     }
                                     pending.missing_acks.remove(&pipeline_id);
-                                    if !source_payloads.is_empty() {
-                                        pending.source_reports.extend(source_payloads);
+                                    if !source_infos.is_empty() {
+                                        pending.source_infos.extend(source_infos);
                                     }
 
                                     if pending.missing_acks.is_empty() {
@@ -786,16 +778,13 @@ impl JobManager {
                                         );
 
                                         let completed = active_checkpoint.take().expect("active checkpoint exists");
-                                        let kf = decode_kafka_checkpoints_from_source_payloads(completed.source_reports, epoch);
-                                        let epoch_u32 = u32::try_from(epoch).unwrap_or(u32::MAX);
 
                                         let mut catalog_ok = true;
                                         if let Some(catalog) = CatalogManager::try_global() {
                                             if let Err(e) = catalog.commit_job_checkpoint(
                                                 &job_id,
                                                 epoch,
-                                                &job_state_dir,
-                                                kf,
+                                                completed.source_infos,
                                             ) {
                                                 catalog_ok = false;
                                                 error!(
@@ -812,9 +801,9 @@ impl JobManager {
                                         }
 
                                         let phase2 = if catalog_ok {
-                                            ControlCommand::Commit { epoch: epoch_u32 }
+                                            ControlCommand::Commit { epoch }
                                         } else {
-                                            ControlCommand::AbortCheckpoint { epoch: epoch_u32 }
+                                            ControlCommand::AbortCheckpoint { epoch }
                                         };
                                         broadcast_cmd(phase2);
                                     }
@@ -828,9 +817,7 @@ impl JobManager {
                                         job_id = %job_id, epoch = epoch, pipeline_id = pipeline_id,
                                         reason = %reason, "Checkpoint FAILED!"
                                     );
-                                    broadcast_cmd(ControlCommand::AbortCheckpoint {
-                                        epoch: u32::try_from(epoch).unwrap_or(u32::MAX),
-                                    });
+                                    broadcast_cmd(ControlCommand::AbortCheckpoint { epoch });
                                     active_checkpoint = None;
                                 }
                             }
@@ -846,7 +833,7 @@ impl JobManager {
                                     "Checkpoint timed out; aborting active epoch"
                                 );
                                 broadcast_cmd(ControlCommand::AbortCheckpoint {
-                                    epoch: u32::try_from(pending.epoch).unwrap_or(u32::MAX),
+                                    epoch: pending.epoch,
                                 });
                             } else {
                                 continue;
@@ -862,7 +849,7 @@ impl JobManager {
                         info!(job_id = %job_id, epoch = current_epoch, "Triggering global Checkpoint Barrier.");
 
                         let barrier = CheckpointBarrier {
-                            epoch: current_epoch as u32,
+                            epoch: current_epoch,
                             min_epoch: 0,
                             timestamp: std::time::SystemTime::now(),
                             then_stop: false,
@@ -871,7 +858,7 @@ impl JobManager {
                             epoch: current_epoch,
                             missing_acks: expected_pipeline_ids.clone(),
                             start_time: Instant::now(),
-                            source_reports: Vec::new(),
+                            source_infos: Vec::new(),
                         });
 
                         for tx in &source_control_txs {
diff --git a/src/runtime/streaming/operators/grouping/incremental_aggregate.rs b/src/runtime/streaming/operators/grouping/incremental_aggregate.rs
index a2325e7c..ff997c11 100644
--- a/src/runtime/streaming/operators/grouping/incremental_aggregate.rs
+++ b/src/runtime/streaming/operators/grouping/incremental_aggregate.rs
@@ -939,7 +939,7 @@ impl Operator for IncrementalAggregatingFunc {
 
         // Flush to Parquet
         store
-            .prepare_checkpoint_epoch(barrier.epoch as u64)
+            .prepare_checkpoint_epoch(barrier.epoch)
             .map_err(|e| anyhow!("Snapshot failed: {e}"))?;
 
         info!(
@@ -952,11 +952,11 @@ impl Operator for IncrementalAggregatingFunc {
         Ok(())
     }
 
-    async fn commit_checkpoint(&mut self, epoch: u32, _ctx: &mut TaskContext) -> Result<()> {
+    async fn commit_checkpoint(&mut self, epoch: u64, _ctx: &mut TaskContext) -> Result<()> {
         self.state_store
             .as_ref()
             .expect("state store not initialized")
-            .commit_checkpoint_epoch(epoch as u64)
+            .commit_checkpoint_epoch(epoch)
             .map_err(|e| anyhow!("Commit checkpoint failed: {e}"))?;
         Ok(())
     }
diff --git a/src/runtime/streaming/operators/joins/join_instance.rs b/src/runtime/streaming/operators/joins/join_instance.rs
index 098e5a73..5c04c1fb 100644
--- a/src/runtime/streaming/operators/joins/join_instance.rs
+++ b/src/runtime/streaming/operators/joins/join_instance.rs
@@ -360,16 +360,16 @@ impl Operator for InstantJoinOperator {
         self.state_store
             .as_ref()
             .unwrap()
-            .prepare_checkpoint_epoch(barrier.epoch as u64)
+            .prepare_checkpoint_epoch(barrier.epoch)
             .map_err(|e| anyhow!("Snapshot failed: {e}"))?;
         Ok(())
     }
 
-    async fn commit_checkpoint(&mut self, epoch: u32, _ctx: &mut TaskContext) -> Result<()> {
+    async fn commit_checkpoint(&mut self, epoch: u64, _ctx: &mut TaskContext) -> Result<()> {
         self.state_store
             .as_ref()
             .unwrap()
-            .commit_checkpoint_epoch(epoch as u64)
+            .commit_checkpoint_epoch(epoch)
             .map_err(|e| anyhow!("Commit checkpoint failed: {e}"))?;
         Ok(())
     }
diff --git a/src/runtime/streaming/operators/joins/join_with_expiration.rs b/src/runtime/streaming/operators/joins/join_with_expiration.rs
index 6a2a240c..5ed8dfa3 100644
--- a/src/runtime/streaming/operators/joins/join_with_expiration.rs
+++ b/src/runtime/streaming/operators/joins/join_with_expiration.rs
@@ -314,18 +314,18 @@ impl Operator for JoinWithExpirationOperator {
             .expect("State store not initialized");
 
         store
-            .prepare_checkpoint_epoch(barrier.epoch as u64)
+            .prepare_checkpoint_epoch(barrier.epoch)
             .map_err(|e| anyhow!("Snapshot failed: {e}"))?;
 
         info!(epoch = barrier.epoch, "Join Operator snapshotted state.");
         Ok(())
     }
 
-    async fn commit_checkpoint(&mut self, epoch: u32, _ctx: &mut TaskContext) -> Result<()> {
+    async fn commit_checkpoint(&mut self, epoch: u64, _ctx: &mut TaskContext) -> Result<()> {
         self.state_store
             .as_ref()
             .expect("State store not initialized")
-            .commit_checkpoint_epoch(epoch as u64)
+            .commit_checkpoint_epoch(epoch)
             .map_err(|e| anyhow!("Commit checkpoint failed: {e}"))?;
         Ok(())
     }
diff --git a/src/runtime/streaming/operators/sink/delta/mod.rs b/src/runtime/streaming/operators/sink/delta/mod.rs
new file mode 100644
index 00000000..4df6b3b5
--- /dev/null
+++ b/src/runtime/streaming/operators/sink/delta/mod.rs
@@ -0,0 +1,273 @@
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+use std::collections::HashMap;
+use std::fs::create_dir_all;
+use std::path::PathBuf;
+use std::sync::Arc;
+
+use anyhow::{Context, Result};
+use arrow_array::RecordBatch;
+use async_trait::async_trait;
+use bytes::Bytes;
+use object_store::aws::AmazonS3Builder;
+use object_store::path::Path as ObjectStorePath;
+use object_store::{ObjectStore, PutPayload};
+use parquet::basic::Compression;
+use tokio::io::AsyncWriteExt;
+use tracing::{debug, info, warn};
+
+use crate::runtime::memory::{MemoryBlock, try_global_memory_pool};
+use crate::runtime::streaming::StreamOutput;
+use crate::runtime::streaming::api::context::TaskContext;
+use crate::runtime::streaming::api::operator::{Collector, Operator};
+use crate::runtime::streaming::format::encoder::FormatEncoder;
+use crate::sql::common::constants::factory_operator_name;
+use crate::sql::common::with_option_keys as opt;
+use crate::sql::common::{CheckpointBarrier, Watermark};
+
+/// Flush early when buffered batches exceed this size.
+const DEFAULT_MAX_BUFFER_BYTES: usize = 64 * 1024 * 1024;
+
+enum DeltaDestination {
+    Local(PathBuf),
+    S3 {
+        prefix: String,
+        client: Arc<dyn ObjectStore>,
+    },
+}
+
+pub struct DeltaSinkOperator {
+    table_name: String,
+    destination: DeltaDestination,
+    parquet_compression: Compression,
+    pending: Vec<RecordBatch>,
+    pending_bytes: usize,
+    sink_memory_block: Option<Arc<MemoryBlock>>,
+    early_flush_threshold_bytes: usize,
+    file_counter: u64,
+    format: DeltaFormat,
+}
+
+#[derive(Debug, Clone, Copy)]
+pub enum DeltaFormat {
+    Csv,
+    Parquet,
+    JsonL,
+    Avro,
+    Orc,
+}
+
+impl DeltaSinkOperator {
+    pub fn try_new(
+        table_name: String,
+        path: String,
+        format: DeltaFormat,
+        parquet_compression: Compression,
+        sink_memory_bytes: u64,
+        options: HashMap<String, String>,
+    ) -> Result<Self> {
+        let destination = if let Some(bucket) = options.get(opt::S3_BUCKET) {
+            let region = options
+                .get(opt::S3_REGION)
+                .cloned()
+                .unwrap_or_else(|| "us-east-1".to_string());
+            let mut builder = AmazonS3Builder::new()
+                .with_bucket_name(bucket.clone())
+                .with_region(region);
+            if let Some(endpoint) = options.get(opt::S3_ENDPOINT) {
+                builder = builder.with_endpoint(endpoint);
+                if endpoint.to_ascii_lowercase().starts_with("http://") {
+                    builder = builder.with_allow_http(true);
+                }
+            }
+            if let Some(v) = options.get(opt::S3_ACCESS_KEY_ID) {
+                builder = builder.with_access_key_id(v);
+            }
+            if let Some(v) = options.get(opt::S3_SECRET_ACCESS_KEY) {
+                builder = builder.with_secret_access_key(v);
+            }
+            if let Some(v) = options.get(opt::S3_SESSION_TOKEN) {
+                builder = builder.with_token(v);
+            }
+            let client = builder
+                .build()
+                .context("failed to build s3 client for delta sink")?;
+            DeltaDestination::S3 {
+                prefix: path.trim_matches('/').to_string(),
+                client: Arc::new(client),
+            }
+        } else {
+            let root = PathBuf::from(path.clone());
+            create_dir_all(&root)
+                .with_context(|| format!("failed to create delta sink dir {}", root.display()))?;
+            DeltaDestination::Local(root)
+        };
+
+        let mut sink_memory_block = None;
+        let reserve_bytes = usize::try_from(sink_memory_bytes).unwrap_or(DEFAULT_MAX_BUFFER_BYTES);
+        let mut early_flush_threshold_bytes = reserve_bytes;
+        if let Ok(pool) = try_global_memory_pool()
+            && let Ok(block) = pool.try_request_block(reserve_bytes as u64)
+        {
+            early_flush_threshold_bytes = ((block.capacity() as usize) * 8) / 10;
+            sink_memory_block = Some(block);
+        }
+
+        Ok(Self {
+            table_name,
+            destination,
+            parquet_compression,
+            pending: Vec::new(),
+            pending_bytes: 0,
+            sink_memory_block,
+            early_flush_threshold_bytes,
+            file_counter: 0,
+            format,
+        })
+    }
+
+    async fn flush_epoch(&mut self, epoch: u64, subtask_idx: usize) -> Result<()> {
+        if self.pending.is_empty() {
+            return Ok(());
+        }
+
+        let batches = std::mem::take(&mut self.pending);
+        let format = self.format;
+        let compression = self.parquet_compression;
+        let bytes = tokio::task::spawn_blocking(move || -> Result<Vec<u8>> {
+            match format {
+                DeltaFormat::Csv => FormatEncoder::encode_csv(&batches),
+                DeltaFormat::Parquet => FormatEncoder::encode_parquet(&batches, compression),
+                DeltaFormat::JsonL => FormatEncoder::encode_jsonl(&batches),
+                DeltaFormat::Avro => FormatEncoder::encode_avro(&batches),
+                DeltaFormat::Orc => FormatEncoder::encode_orc(&batches),
+            }
+        })
+        .await
+        .context("tokio blocking task panicked during serialization")??;
+
+        self.file_counter += 1;
+        let file_name = format!(
+            "delta-part-{:05}-epoch-{:010}-{:06}.{}",
+            subtask_idx,
+            epoch,
+            self.file_counter,
+            match self.format {
+                DeltaFormat::Csv => "csv",
+                DeltaFormat::Parquet => "parquet",
+                DeltaFormat::JsonL => "jsonl",
+                DeltaFormat::Avro => "avro",
+                DeltaFormat::Orc => "orc",
+            }
+        );
+        match &self.destination {
+            DeltaDestination::Local(root) => {
+                let out = root.join(file_name);
+                let mut f = tokio::fs::File::create(&out).await.with_context(|| {
+                    format!("failed creating delta sink file {}", out.display())
+                })?;
+                f.write_all(&bytes)
+                    .await
+                    .with_context(|| format!("failed writing delta sink file {}", out.display()))?;
+            }
+            DeltaDestination::S3 { prefix, client } => {
+                let key = if prefix.is_empty() {
+                    file_name
+                } else {
+                    format!("{prefix}/{file_name}")
+                };
+                client
+                    .put(
+                        &ObjectStorePath::from(key),
+                        PutPayload::from(Bytes::from(bytes)),
+                    )
+                    .await
+                    .context("failed writing object to s3")?;
+            }
+        }
+        self.pending_bytes = 0;
+        Ok(())
+    }
+}
+
+#[async_trait]
+impl Operator for DeltaSinkOperator {
+    fn name(&self) -> &str {
+        factory_operator_name::CONNECTOR_SINK
+    }
+
+    async fn on_start(&mut self, _ctx: &mut TaskContext) -> Result<()> {
+        let reserved_block_bytes = self
+            .sink_memory_block
+            .as_ref()
+            .map(|b| b.capacity())
+            .unwrap_or(0);
+        info!(
+            table = %self.table_name,
+            format = ?self.format,
+            reserved_block_bytes,
+            early_flush_threshold_bytes = self.early_flush_threshold_bytes,
+            "Starting delta sink operator"
+        );
+        Ok(())
+    }
+
+    async fn process_data(
+        &mut self,
+        _input_idx: usize,
+        batch: RecordBatch,
+        ctx: &mut TaskContext,
+        _collector: &mut dyn Collector,
+    ) -> Result<()> {
+        let batch_size = batch.get_array_memory_size();
+        self.pending.push(batch);
+        self.pending_bytes += batch_size;
+
+        if self.pending_bytes > self.early_flush_threshold_bytes {
+            debug!(
+                table = %self.table_name,
+                bytes = self.pending_bytes,
+                "memory watermark reached, triggering early flush"
+            );
+            self.flush_epoch(0, ctx.subtask_index as usize).await?;
+        }
+        Ok(())
+    }
+
+    async fn process_watermark(
+        &mut self,
+        _watermark: Watermark,
+        _ctx: &mut TaskContext,
+        _collector: &mut dyn Collector,
+    ) -> Result<()> {
+        Ok(())
+    }
+
+    async fn snapshot_state(
+        &mut self,
+        barrier: CheckpointBarrier,
+        ctx: &mut TaskContext,
+    ) -> Result<()> {
+        self.flush_epoch(barrier.epoch, ctx.subtask_index as usize)
+            .await
+    }
+
+    async fn on_close(&mut self, ctx: &mut TaskContext) -> Result<Vec<StreamOutput>> {
+        if !self.pending.is_empty() {
+            warn!(table = %self.table_name, "flushing remaining delta sink batches on close");
+            self.flush_epoch(u64::MAX, ctx.subtask_index as usize)
+                .await?;
+        }
+        Ok(vec![])
+    }
+}
diff --git a/src/runtime/streaming/operators/sink/filesystem/mod.rs b/src/runtime/streaming/operators/sink/filesystem/mod.rs
new file mode 100644
index 00000000..a865a752
--- /dev/null
+++ b/src/runtime/streaming/operators/sink/filesystem/mod.rs
@@ -0,0 +1,233 @@
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+use std::fs::create_dir_all;
+use std::path::PathBuf;
+
+use anyhow::{Context, Result, bail};
+use arrow_array::RecordBatch;
+use async_trait::async_trait;
+use parquet::basic::Compression;
+use tokio::io::AsyncWriteExt;
+use tracing::{debug, info, warn};
+
+use crate::runtime::memory::{MemoryBlock, try_global_memory_pool};
+use crate::runtime::streaming::StreamOutput;
+use crate::runtime::streaming::api::context::TaskContext;
+use crate::runtime::streaming::api::operator::{Collector, Operator};
+use crate::runtime::streaming::format::encoder::FormatEncoder;
+use crate::sql::common::constants::factory_operator_name;
+use crate::sql::common::{CheckpointBarrier, Watermark};
+
+const DEFAULT_MAX_BUFFER_BYTES: usize = 64 * 1024 * 1024;
+
+#[derive(Debug, Clone, Copy)]
+pub enum FilesystemFormat {
+    Csv,
+    Parquet,
+    JsonL,
+    Avro,
+    Orc,
+}
+
+pub struct FilesystemSinkOperator {
+    table_name: String,
+    output_dir: PathBuf,
+    format: FilesystemFormat,
+    parquet_compression: Compression,
+    pending: Vec<RecordBatch>,
+    pending_bytes: usize,
+    sink_memory_block: Option<std::sync::Arc<MemoryBlock>>,
+    early_flush_threshold_bytes: usize,
+    file_counter: u64,
+}
+
+impl FilesystemSinkOperator {
+    pub fn try_new(
+        table_name: String,
+        output_dir: String,
+        format: FilesystemFormat,
+        parquet_compression: Compression,
+        sink_memory_bytes: u64,
+    ) -> Result<Self> {
+        let output_dir_path = PathBuf::from(&output_dir);
+        create_dir_all(&output_dir_path).with_context(|| {
+            format!(
+                "failed to create filesystem sink directory {}",
+                output_dir_path.display()
+            )
+        })?;
+
+        let mut sink_memory_block = None;
+        let reserve_bytes = usize::try_from(sink_memory_bytes).unwrap_or(DEFAULT_MAX_BUFFER_BYTES);
+        let mut early_flush_threshold_bytes = reserve_bytes;
+        if let Ok(pool) = try_global_memory_pool()
+            && let Ok(block) = pool.try_request_block(reserve_bytes as u64)
+        {
+            early_flush_threshold_bytes = ((block.capacity() as usize) * 8) / 10;
+            sink_memory_block = Some(block);
+        }
+
+        Ok(Self {
+            table_name,
+            output_dir: output_dir_path,
+            format,
+            parquet_compression,
+            pending: Vec::new(),
+            pending_bytes: 0,
+            sink_memory_block,
+            early_flush_threshold_bytes,
+            file_counter: 0,
+        })
+    }
+
+    fn extension(&self) -> &'static str {
+        match self.format {
+            FilesystemFormat::Csv => "csv",
+            FilesystemFormat::Parquet => "parquet",
+            FilesystemFormat::JsonL => "jsonl",
+            FilesystemFormat::Avro => "avro",
+            FilesystemFormat::Orc => "orc",
+        }
+    }
+
+    async fn flush_file_epoch(&mut self, epoch: u64, subtask_idx: usize) -> Result<()> {
+        if self.pending.is_empty() {
+            return Ok(());
+        }
+
+        let batches = std::mem::take(&mut self.pending);
+        let format = self.format;
+        let compression = self.parquet_compression;
+        let bytes = tokio::task::spawn_blocking(move || -> Result<Vec<u8>> {
+            match format {
+                FilesystemFormat::Csv => FormatEncoder::encode_csv(&batches),
+                FilesystemFormat::Parquet => FormatEncoder::encode_parquet(&batches, compression),
+                FilesystemFormat::JsonL => FormatEncoder::encode_jsonl(&batches),
+                FilesystemFormat::Avro => FormatEncoder::encode_avro(&batches),
+                FilesystemFormat::Orc => FormatEncoder::encode_orc(&batches),
+            }
+        })
+        .await
+        .context("tokio blocking task panicked during serialization")??;
+
+        if bytes.is_empty() {
+            self.pending_bytes = 0;
+            return Ok(());
+        }
+
+        self.file_counter += 1;
+        let file_name = format!(
+            "part-{:05}-epoch-{:010}-{:06}.{}",
+            subtask_idx,
+            epoch,
+            self.file_counter,
+            self.extension()
+        );
+        let output_file = self.output_dir.join(file_name);
+        let mut f = tokio::fs::File::create(&output_file)
+            .await
+            .with_context(|| format!("failed creating sink file {}", output_file.display()))?;
+        f.write_all(&bytes)
+            .await
+            .with_context(|| format!("failed writing sink file {}", output_file.display()))?;
+        self.pending_bytes = 0;
+        Ok(())
+    }
+}
+
+#[async_trait]
+impl Operator for FilesystemSinkOperator {
+    fn name(&self) -> &str {
+        factory_operator_name::CONNECTOR_SINK
+    }
+
+    async fn on_start(&mut self, _ctx: &mut TaskContext) -> Result<()> {
+        let reserved_block_bytes = self
+            .sink_memory_block
+            .as_ref()
+            .map(|b| b.capacity())
+            .unwrap_or(0);
+        info!(
+            table = %self.table_name,
+            path = %self.output_dir.display(),
+            format = ?self.format,
+            reserved_block_bytes,
+            early_flush_threshold_bytes = self.early_flush_threshold_bytes,
+            "Starting filesystem sink operator"
+        );
+        Ok(())
+    }
+
+    async fn process_data(
+        &mut self,
+        _input_idx: usize,
+        batch: RecordBatch,
+        ctx: &mut TaskContext,
+        _collector: &mut dyn Collector,
+    ) -> Result<()> {
+        let batch_size = batch.get_array_memory_size();
+        self.pending.push(batch);
+        self.pending_bytes += batch_size;
+        if self.pending_bytes > self.early_flush_threshold_bytes {
+            debug!(
+                table = %self.table_name,
+                bytes = self.pending_bytes,
+                "memory watermark reached, triggering early flush to filesystem"
+            );
+            self.flush_file_epoch(0, ctx.subtask_index as usize).await?;
+        }
+        Ok(())
+    }
+
+    async fn process_watermark(
+        &mut self,
+        _watermark: Watermark,
+        _ctx: &mut TaskContext,
+        _collector: &mut dyn Collector,
+    ) -> Result<()> {
+        Ok(())
+    }
+
+    async fn snapshot_state(
+        &mut self,
+        barrier: CheckpointBarrier,
+        ctx: &mut TaskContext,
+    ) -> Result<()> {
+        self.flush_file_epoch(barrier.epoch, ctx.subtask_index as usize)
+            .await
+    }
+
+    async fn on_close(&mut self, ctx: &mut TaskContext) -> Result<Vec<StreamOutput>> {
+        if !self.pending.is_empty() {
+            warn!(
+                table = %self.table_name,
+                "flushing remaining filesystem sink batches on close"
+            );
+            self.flush_file_epoch(u64::MAX, ctx.subtask_index as usize)
+                .await?;
+        }
+        Ok(vec![])
+    }
+}
+
+pub fn compression_from_str(v: Option<&str>) -> Result<Compression> {
+    match v.unwrap_or("zstd").to_ascii_lowercase().as_str() {
+        "uncompressed" => Ok(Compression::UNCOMPRESSED),
+        "snappy" => Ok(Compression::SNAPPY),
+        "gzip" => Ok(Compression::GZIP(Default::default())),
+        "zstd" => Ok(Compression::ZSTD(Default::default())),
+        "lz4" => Ok(Compression::LZ4),
+        "lz4_raw" => Ok(Compression::LZ4_RAW),
+        other => bail!("unsupported parquet compression '{other}'"),
+    }
+}
diff --git a/src/runtime/streaming/operators/sink/iceberg/mod.rs b/src/runtime/streaming/operators/sink/iceberg/mod.rs
new file mode 100644
index 00000000..b6c17414
--- /dev/null
+++ b/src/runtime/streaming/operators/sink/iceberg/mod.rs
@@ -0,0 +1,271 @@
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+use std::collections::HashMap;
+use std::fs::create_dir_all;
+use std::path::PathBuf;
+use std::sync::Arc;
+
+use anyhow::{Context, Result};
+use arrow_array::RecordBatch;
+use async_trait::async_trait;
+use bytes::Bytes;
+use object_store::aws::AmazonS3Builder;
+use object_store::path::Path as ObjectStorePath;
+use object_store::{ObjectStore, PutPayload};
+use parquet::basic::Compression;
+use tokio::io::AsyncWriteExt;
+use tracing::{debug, info, warn};
+
+use crate::runtime::memory::{MemoryBlock, try_global_memory_pool};
+use crate::runtime::streaming::StreamOutput;
+use crate::runtime::streaming::api::context::TaskContext;
+use crate::runtime::streaming::api::operator::{Collector, Operator};
+use crate::runtime::streaming::format::encoder::FormatEncoder;
+use crate::sql::common::constants::factory_operator_name;
+use crate::sql::common::with_option_keys as opt;
+use crate::sql::common::{CheckpointBarrier, Watermark};
+
+const DEFAULT_MAX_BUFFER_BYTES: usize = 64 * 1024 * 1024;
+
+enum IcebergDestination {
+    Local(PathBuf),
+    S3 {
+        prefix: String,
+        client: Arc<dyn ObjectStore>,
+    },
+}
+
+pub struct IcebergSinkOperator {
+    table_name: String,
+    destination: IcebergDestination,
+    parquet_compression: Compression,
+    pending: Vec<RecordBatch>,
+    pending_bytes: usize,
+    sink_memory_block: Option<Arc<MemoryBlock>>,
+    early_flush_threshold_bytes: usize,
+    file_counter: u64,
+    format: IcebergFormat,
+}
+
+#[derive(Debug, Clone, Copy)]
+pub enum IcebergFormat {
+    Csv,
+    Parquet,
+}
+
+impl IcebergSinkOperator {
+    pub fn try_new(
+        table_name: String,
+        path: String,
+        format: IcebergFormat,
+        parquet_compression: Compression,
+        sink_memory_bytes: u64,
+        options: HashMap<String, String>,
+    ) -> Result<Self> {
+        let destination = if let Some(bucket) = options.get(opt::S3_BUCKET) {
+            let region = options
+                .get(opt::S3_REGION)
+                .cloned()
+                .unwrap_or_else(|| "us-east-1".to_string());
+            let mut builder = AmazonS3Builder::new()
+                .with_bucket_name(bucket.clone())
+                .with_region(region);
+            if let Some(endpoint) = options.get(opt::S3_ENDPOINT) {
+                builder = builder.with_endpoint(endpoint);
+                if endpoint.to_ascii_lowercase().starts_with("http://") {
+                    builder = builder.with_allow_http(true);
+                }
+            }
+            if let Some(v) = options.get(opt::S3_ACCESS_KEY_ID) {
+                builder = builder.with_access_key_id(v);
+            }
+            if let Some(v) = options.get(opt::S3_SECRET_ACCESS_KEY) {
+                builder = builder.with_secret_access_key(v);
+            }
+            if let Some(v) = options.get(opt::S3_SESSION_TOKEN) {
+                builder = builder.with_token(v);
+            }
+            let client = builder
+                .build()
+                .context("failed to build s3 client for iceberg sink")?;
+            IcebergDestination::S3 {
+                prefix: path.trim_matches('/').to_string(),
+                client: Arc::new(client),
+            }
+        } else {
+            let root = PathBuf::from(path.clone());
+            create_dir_all(&root)
+                .with_context(|| format!("failed to create iceberg sink dir {}", root.display()))?;
+            IcebergDestination::Local(root)
+        };
+
+        let mut sink_memory_block = None;
+        let reserve_bytes = usize::try_from(sink_memory_bytes).unwrap_or(DEFAULT_MAX_BUFFER_BYTES);
+        let mut early_flush_threshold_bytes = reserve_bytes;
+        if let Ok(pool) = try_global_memory_pool()
+            && let Ok(block) = pool.try_request_block(reserve_bytes as u64)
+        {
+            early_flush_threshold_bytes = ((block.capacity() as usize) * 8) / 10;
+            sink_memory_block = Some(block);
+        }
+
+        Ok(Self {
+            table_name,
+            destination,
+            parquet_compression,
+            pending: Vec::new(),
+            pending_bytes: 0,
+            sink_memory_block,
+            early_flush_threshold_bytes,
+            file_counter: 0,
+            format,
+        })
+    }
+
+    async fn flush_epoch(&mut self, epoch: u64, subtask_idx: usize) -> Result<()> {
+        if self.pending.is_empty() {
+            return Ok(());
+        }
+
+        let batches = std::mem::take(&mut self.pending);
+        let format = self.format;
+        let compression = self.parquet_compression;
+        let bytes = tokio::task::spawn_blocking(move || -> Result<Vec<u8>> {
+            match format {
+                IcebergFormat::Csv => FormatEncoder::encode_csv(&batches),
+                IcebergFormat::Parquet => FormatEncoder::encode_parquet(&batches, compression),
+            }
+        })
+        .await
+        .context("tokio blocking task panicked during serialization")??;
+
+        if bytes.is_empty() {
+            self.pending_bytes = 0;
+            return Ok(());
+        }
+
+        self.file_counter += 1;
+        let file_name = format!(
+            "iceberg-part-{:05}-epoch-{:010}-{:06}.{}",
+            subtask_idx,
+            epoch,
+            self.file_counter,
+            match self.format {
+                IcebergFormat::Csv => "csv",
+                IcebergFormat::Parquet => "parquet",
+            }
+        );
+        match &self.destination {
+            IcebergDestination::Local(root) => {
+                let out = root.join(file_name);
+                let mut f = tokio::fs::File::create(&out).await.with_context(|| {
+                    format!("failed creating iceberg sink file {}", out.display())
+                })?;
+                f.write_all(&bytes).await.with_context(|| {
+                    format!("failed writing iceberg sink file {}", out.display())
+                })?;
+            }
+            IcebergDestination::S3 { prefix, client } => {
+                let key = if prefix.is_empty() {
+                    file_name
+                } else {
+                    format!("{prefix}/{file_name}")
+                };
+                client
+                    .put(
+                        &ObjectStorePath::from(key),
+                        PutPayload::from(Bytes::from(bytes)),
+                    )
+                    .await
+                    .context("failed writing iceberg data object to s3")?;
+            }
+        }
+        self.pending_bytes = 0;
+        Ok(())
+    }
+}
+
+#[async_trait]
+impl Operator for IcebergSinkOperator {
+    fn name(&self) -> &str {
+        factory_operator_name::CONNECTOR_SINK
+    }
+
+    async fn on_start(&mut self, _ctx: &mut TaskContext) -> Result<()> {
+        let reserved_block_bytes = self
+            .sink_memory_block
+            .as_ref()
+            .map(|b| b.capacity())
+            .unwrap_or(0);
+        info!(
+            table = %self.table_name,
+            format = ?self.format,
+            reserved_block_bytes,
+            early_flush_threshold_bytes = self.early_flush_threshold_bytes,
+            "Starting iceberg sink operator"
+        );
+        Ok(())
+    }
+
+    async fn process_data(
+        &mut self,
+        _input_idx: usize,
+        batch: RecordBatch,
+        ctx: &mut TaskContext,
+        _collector: &mut dyn Collector,
+    ) -> Result<()> {
+        let batch_size = batch.get_array_memory_size();
+        self.pending.push(batch);
+        self.pending_bytes += batch_size;
+        if self.pending_bytes > self.early_flush_threshold_bytes {
+            debug!(
+                table = %self.table_name,
+                bytes = self.pending_bytes,
+                threshold = self.early_flush_threshold_bytes,
+                "memory watermark reached, triggering early flush for iceberg sink"
+            );
+            self.flush_epoch(0, ctx.subtask_index as usize).await?;
+        }
+        Ok(())
+    }
+
+    async fn process_watermark(
+        &mut self,
+        _watermark: Watermark,
+        _ctx: &mut TaskContext,
+        _collector: &mut dyn Collector,
+    ) -> Result<()> {
+        Ok(())
+    }
+
+    async fn snapshot_state(
+        &mut self,
+        barrier: CheckpointBarrier,
+        ctx: &mut TaskContext,
+    ) -> Result<()> {
+        self.flush_epoch(barrier.epoch, ctx.subtask_index as usize)
+            .await
+    }
+
+    async fn on_close(&mut self, ctx: &mut TaskContext) -> Result<Vec<StreamOutput>> {
+        if !self.pending.is_empty() {
+            warn!(
+                table = %self.table_name,
+                "flushing remaining iceberg sink batches on close"
+            );
+            self.flush_epoch(u64::MAX, ctx.subtask_index as usize)
+                .await?;
+        }
+        Ok(vec![])
+    }
+}
diff --git a/src/runtime/streaming/operators/sink/kafka/mod.rs b/src/runtime/streaming/operators/sink/kafka/mod.rs
index a9c4b50e..b30bc572 100644
--- a/src/runtime/streaming/operators/sink/kafka/mod.rs
+++ b/src/runtime/streaming/operators/sink/kafka/mod.rs
@@ -336,7 +336,7 @@ impl Operator for KafkaSinkOperator {
         Ok(())
     }
 
-    async fn commit_checkpoint(&mut self, epoch: u32, _ctx: &mut TaskContext) -> Result<()> {
+    async fn commit_checkpoint(&mut self, epoch: u64, _ctx: &mut TaskContext) -> Result<()> {
         if matches!(self.consistency_mode, ConsistencyMode::AtLeastOnce) {
             return Ok(());
         }
@@ -380,7 +380,7 @@ impl Operator for KafkaSinkOperator {
         Ok(())
     }
 
-    async fn abort_checkpoint(&mut self, epoch: u32, _ctx: &mut TaskContext) -> Result<()> {
+    async fn abort_checkpoint(&mut self, epoch: u64, _ctx: &mut TaskContext) -> Result<()> {
         if matches!(self.consistency_mode, ConsistencyMode::AtLeastOnce) {
             return Ok(());
         }
diff --git a/src/runtime/streaming/operators/sink/lancedb/mod.rs b/src/runtime/streaming/operators/sink/lancedb/mod.rs
new file mode 100644
index 00000000..f8a082e1
--- /dev/null
+++ b/src/runtime/streaming/operators/sink/lancedb/mod.rs
@@ -0,0 +1,158 @@
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+use std::io::Cursor;
+
+use anyhow::{Context, Result};
+use arrow_array::RecordBatch;
+use arrow_array_lance::RecordBatchIterator as LanceBatchIterator;
+use async_trait::async_trait;
+use lance::Dataset;
+use lance::dataset::{WriteMode, WriteParams};
+use tracing::{info, warn};
+
+use crate::runtime::streaming::StreamOutput;
+use crate::runtime::streaming::api::context::TaskContext;
+use crate::runtime::streaming::api::operator::{Collector, Operator};
+use crate::sql::common::constants::factory_operator_name;
+use crate::sql::common::{CheckpointBarrier, Watermark};
+
+pub struct LanceDbSinkOperator {
+    table_name: String,
+    dataset_uri: String,
+    pending: Vec<RecordBatch>,
+    initialized: bool,
+}
+
+impl LanceDbSinkOperator {
+    pub fn new(table_name: String, dataset_uri: String) -> Self {
+        Self {
+            table_name,
+            dataset_uri,
+            pending: Vec::new(),
+            initialized: false,
+        }
+    }
+
+    fn to_lance_batches(
+        &self,
+        batches: &[RecordBatch],
+    ) -> Result<Vec<arrow_array_lance::RecordBatch>> {
+        let schema = batches
+            .first()
+            .map(|b| b.schema())
+            .context("lanceDB sink requires at least one record batch")?;
+
+        let mut ipc_payload = Vec::<u8>::new();
+        {
+            let mut writer = arrow_ipc::writer::FileWriter::try_new(&mut ipc_payload, &schema)
+                .context("failed to build ipc writer for lanceDB conversion")?;
+            for batch in batches {
+                writer
+                    .write(batch)
+                    .context("failed writing ipc payload for lanceDB conversion")?;
+            }
+            writer
+                .finish()
+                .context("failed finishing ipc payload for lanceDB conversion")?;
+        }
+
+        let reader = arrow_ipc_lance::reader::FileReader::try_new(Cursor::new(ipc_payload), None)
+            .context("failed reading lance-compatible ipc payload")?;
+        reader
+            .collect::<std::result::Result<Vec<_>, _>>()
+            .context("failed converting batches into lance-compatible batches")
+    }
+
+    async fn flush_epoch(&mut self) -> Result<()> {
+        if self.pending.is_empty() {
+            return Ok(());
+        }
+
+        let lance_batches = self.to_lance_batches(&self.pending)?;
+        let schema = lance_batches
+            .first()
+            .map(|b| b.schema())
+            .context("lanceDB sink produced no converted batches")?;
+        let reader = LanceBatchIterator::new(lance_batches.into_iter().map(Ok), schema);
+        let params = WriteParams {
+            mode: if self.initialized {
+                WriteMode::Append
+            } else {
+                WriteMode::Create
+            },
+            ..Default::default()
+        };
+        Dataset::write(reader, &self.dataset_uri, Some(params))
+            .await
+            .with_context(|| format!("failed writing lance dataset '{}'", self.dataset_uri))?;
+
+        self.initialized = true;
+        self.pending.clear();
+        Ok(())
+    }
+}
+
+#[async_trait]
+impl Operator for LanceDbSinkOperator {
+    fn name(&self) -> &str {
+        factory_operator_name::CONNECTOR_SINK
+    }
+
+    async fn on_start(&mut self, _ctx: &mut TaskContext) -> Result<()> {
+        info!(
+            table = %self.table_name,
+            dataset = %self.dataset_uri,
+            "Starting lanceDB sink operator"
+        );
+        Ok(())
+    }
+
+    async fn process_data(
+        &mut self,
+        _input_idx: usize,
+        batch: RecordBatch,
+        _ctx: &mut TaskContext,
+        _collector: &mut dyn Collector,
+    ) -> Result<()> {
+        self.pending.push(batch);
+        Ok(())
+    }
+
+    async fn process_watermark(
+        &mut self,
+        _watermark: Watermark,
+        _ctx: &mut TaskContext,
+        _collector: &mut dyn Collector,
+    ) -> Result<()> {
+        Ok(())
+    }
+
+    async fn snapshot_state(
+        &mut self,
+        _barrier: CheckpointBarrier,
+        _ctx: &mut TaskContext,
+    ) -> Result<()> {
+        self.flush_epoch().await
+    }
+
+    async fn on_close(&mut self, _ctx: &mut TaskContext) -> Result<Vec<StreamOutput>> {
+        if !self.pending.is_empty() {
+            warn!(
+                table = %self.table_name,
+                "flushing remaining lanceDB sink batches on close"
+            );
+            self.flush_epoch().await?;
+        }
+        Ok(vec![])
+    }
+}
diff --git a/src/runtime/streaming/operators/sink/mod.rs b/src/runtime/streaming/operators/sink/mod.rs
index b9574391..ff893d57 100644
--- a/src/runtime/streaming/operators/sink/mod.rs
+++ b/src/runtime/streaming/operators/sink/mod.rs
@@ -10,4 +10,9 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
+pub mod delta;
+pub mod filesystem;
+pub mod iceberg;
 pub mod kafka;
+pub mod lancedb;
+pub mod s3;
diff --git a/src/runtime/streaming/operators/sink/s3/mod.rs b/src/runtime/streaming/operators/sink/s3/mod.rs
new file mode 100644
index 00000000..715b5b86
--- /dev/null
+++ b/src/runtime/streaming/operators/sink/s3/mod.rs
@@ -0,0 +1,259 @@
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+use std::collections::HashMap;
+use std::io::Cursor;
+
+use anyhow::{Context, Result, bail};
+use arrow::csv::WriterBuilder as CsvWriterBuilder;
+use arrow_array::RecordBatch;
+use async_trait::async_trait;
+use bytes::Bytes;
+use object_store::aws::AmazonS3Builder;
+use object_store::path::Path as ObjectStorePath;
+use object_store::{ObjectStore, PutPayload};
+use parquet::arrow::ArrowWriter;
+use parquet::basic::Compression;
+use parquet::file::properties::WriterProperties;
+use tracing::{info, warn};
+
+use crate::runtime::streaming::StreamOutput;
+use crate::runtime::streaming::api::context::TaskContext;
+use crate::runtime::streaming::api::operator::{Collector, Operator};
+use crate::sql::common::constants::factory_operator_name;
+use crate::sql::common::with_option_keys as opt;
+use crate::sql::common::{CheckpointBarrier, Watermark};
+
+#[derive(Debug, Clone, Copy)]
+pub enum S3Format {
+    Csv,
+    Parquet,
+}
+
+pub struct S3SinkOperator {
+    table_name: String,
+    bucket: String,
+    prefix: String,
+    format: S3Format,
+    parquet_compression: Compression,
+    client: Box<dyn ObjectStore>,
+    pending: Vec<RecordBatch>,
+    file_counter: u64,
+}
+
+impl S3SinkOperator {
+    pub fn try_new(
+        table_name: String,
+        path: String,
+        format: S3Format,
+        parquet_compression: Compression,
+        s3_options: HashMap<String, String>,
+    ) -> Result<Self> {
+        let bucket = s3_options
+            .get(opt::S3_BUCKET)
+            .cloned()
+            .context("s3 sink requires 's3.bucket'")?;
+        let region = s3_options
+            .get(opt::S3_REGION)
+            .cloned()
+            .unwrap_or_else(|| "us-east-1".to_string());
+
+        let mut builder = AmazonS3Builder::new()
+            .with_bucket_name(bucket.clone())
+            .with_region(region);
+        if let Some(endpoint) = s3_options.get(opt::S3_ENDPOINT) {
+            builder = builder.with_endpoint(endpoint);
+            if endpoint.to_ascii_lowercase().starts_with("http://") {
+                builder = builder.with_allow_http(true);
+            }
+        }
+        if let Some(v) = s3_options.get(opt::S3_ACCESS_KEY_ID) {
+            builder = builder.with_access_key_id(v);
+        }
+        if let Some(v) = s3_options.get(opt::S3_SECRET_ACCESS_KEY) {
+            builder = builder.with_secret_access_key(v);
+        }
+        if let Some(v) = s3_options.get(opt::S3_SESSION_TOKEN) {
+            builder = builder.with_token(v);
+        }
+        let client = builder
+            .build()
+            .context("failed to build s3 object-store client")?;
+
+        let prefix = path.trim_matches('/').to_string();
+
+        Ok(Self {
+            table_name,
+            bucket,
+            prefix,
+            format,
+            parquet_compression,
+            client: Box::new(client),
+            pending: Vec::new(),
+            file_counter: 0,
+        })
+    }
+
+    fn extension(&self) -> &'static str {
+        match self.format {
+            S3Format::Csv => "csv",
+            S3Format::Parquet => "parquet",
+        }
+    }
+
+    fn serialize_csv(&self) -> Result<Vec<u8>> {
+        let mut out = Vec::new();
+        let mut writer = CsvWriterBuilder::new().with_header(true).build(&mut out);
+        for batch in &self.pending {
+            writer.write(batch).context("failed writing csv batch")?;
+        }
+        drop(writer);
+        Ok(out)
+    }
+
+    fn serialize_parquet(&self) -> Result<Vec<u8>> {
+        let schema = self
+            .pending
+            .first()
+            .map(|b| b.schema())
+            .context("parquet serialization requires at least one record batch")?;
+        let props = WriterProperties::builder()
+            .set_compression(self.parquet_compression)
+            .build();
+        let mut cursor = Cursor::new(Vec::<u8>::new());
+        let mut writer = ArrowWriter::try_new(&mut cursor, schema, Some(props))
+            .context("failed to initialize parquet writer")?;
+        for batch in &self.pending {
+            writer
+                .write(batch)
+                .context("failed writing parquet batch")?;
+        }
+        writer.close().context("failed to close parquet writer")?;
+        Ok(cursor.into_inner())
+    }
+
+    async fn flush_epoch(&mut self, epoch: u64, subtask_idx: usize, bytes: Vec<u8>) -> Result<()> {
+        if self.pending.is_empty() {
+            return Ok(());
+        }
+        if bytes.is_empty() {
+            self.pending.clear();
+            return Ok(());
+        }
+
+        self.file_counter += 1;
+        let file_name = format!(
+            "part-{:05}-epoch-{:010}-{:06}.{}",
+            subtask_idx,
+            epoch,
+            self.file_counter,
+            self.extension()
+        );
+        let key = if self.prefix.is_empty() {
+            file_name
+        } else {
+            format!("{}/{}", self.prefix, file_name)
+        };
+        self.client
+            .put(
+                &ObjectStorePath::from(key),
+                PutPayload::from(Bytes::from(bytes)),
+            )
+            .await
+            .context("failed writing object to s3")?;
+
+        self.pending.clear();
+        Ok(())
+    }
+}
+
+#[async_trait]
+impl Operator for S3SinkOperator {
+    fn name(&self) -> &str {
+        factory_operator_name::CONNECTOR_SINK
+    }
+
+    async fn on_start(&mut self, _ctx: &mut TaskContext) -> Result<()> {
+        info!(
+            table = %self.table_name,
+            bucket = %self.bucket,
+            prefix = %self.prefix,
+            format = ?self.format,
+            "Starting s3 sink operator"
+        );
+        Ok(())
+    }
+
+    async fn process_data(
+        &mut self,
+        _input_idx: usize,
+        batch: RecordBatch,
+        _ctx: &mut TaskContext,
+        _collector: &mut dyn Collector,
+    ) -> Result<()> {
+        self.pending.push(batch);
+        Ok(())
+    }
+
+    async fn process_watermark(
+        &mut self,
+        _watermark: Watermark,
+        _ctx: &mut TaskContext,
+        _collector: &mut dyn Collector,
+    ) -> Result<()> {
+        Ok(())
+    }
+
+    async fn snapshot_state(
+        &mut self,
+        barrier: CheckpointBarrier,
+        ctx: &mut TaskContext,
+    ) -> Result<()> {
+        if self.pending.is_empty() {
+            return Ok(());
+        }
+        let bytes = match self.format {
+            S3Format::Csv => self.serialize_csv()?,
+            S3Format::Parquet => self.serialize_parquet()?,
+        };
+        self.flush_epoch(barrier.epoch, ctx.subtask_index as usize, bytes)
+            .await
+    }
+
+    async fn on_close(&mut self, ctx: &mut TaskContext) -> Result<Vec<StreamOutput>> {
+        if !self.pending.is_empty() {
+            warn!(
+                table = %self.table_name,
+                "flushing remaining s3 sink batches on close"
+            );
+            let bytes = match self.format {
+                S3Format::Csv => self.serialize_csv()?,
+                S3Format::Parquet => self.serialize_parquet()?,
+            };
+            self.flush_epoch(0, ctx.subtask_index as usize, bytes)
+                .await?;
+        }
+        Ok(vec![])
+    }
+}
+
+pub fn compression_from_str(v: Option<&str>) -> Result<Compression> {
+    match v.unwrap_or("zstd").to_ascii_lowercase().as_str() {
+        "uncompressed" => Ok(Compression::UNCOMPRESSED),
+        "snappy" => Ok(Compression::SNAPPY),
+        "gzip" => Ok(Compression::GZIP(Default::default())),
+        "zstd" => Ok(Compression::ZSTD(Default::default())),
+        "lz4" => Ok(Compression::LZ4),
+        "lz4_raw" => Ok(Compression::LZ4_RAW),
+        other => bail!("unsupported parquet compression '{other}'"),
+    }
+}
diff --git a/src/runtime/streaming/operators/source/kafka/mod.rs b/src/runtime/streaming/operators/source/kafka/mod.rs
index 9f5b84ad..75edb968 100644
--- a/src/runtime/streaming/operators/source/kafka/mod.rs
+++ b/src/runtime/streaming/operators/source/kafka/mod.rs
@@ -10,21 +10,24 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
-//! Kafka source checkpointing: `enable.auto.commit=false`, offsets captured at the checkpoint barrier
-//! and reported to the job coordinator for catalog persistence; restart rewinds from that snapshot.
+//! Kafka source checkpointing: `enable.auto.commit=false`, offsets captured at the checkpoint
+//! barrier and reported to the job coordinator for catalog persistence; on restart the catalog
+//! records are injected directly into [`TaskContext::source_checkpoint_infos`] — no intermediate
+//! on-disk snapshot files are used.
 
 use anyhow::{Context as _, Result, anyhow};
 use arrow_array::RecordBatch;
 use arrow_schema::SchemaRef;
 use async_trait::async_trait;
-use bincode::{Decode, Encode};
 use governor::{DefaultDirectRateLimiter, Quota, RateLimiter as GovernorRateLimiter};
-use protocol::storage::{KafkaPartitionOffset, KafkaSourceSubtaskCheckpoint};
+use protocol::storage::{
+    KafkaPartitionOffset, KafkaSourceSubtaskCheckpoint, SourceCheckpointInfo,
+    source_checkpoint_info,
+};
 use rdkafka::consumer::{CommitMode, Consumer, StreamConsumer};
 use rdkafka::{ClientConfig, Message as KMessage, Offset, TopicPartitionList};
 use std::collections::HashMap;
 use std::num::NonZeroU32;
-use std::path::PathBuf;
 use std::time::{Duration, Instant};
 use tracing::{debug, error, info, warn};
 
@@ -35,80 +38,6 @@ use crate::runtime::streaming::api::source::{
 use crate::runtime::streaming::format::{BadDataPolicy, DataDeserializer, Format};
 use crate::sql::common::fs_schema::FieldValueType;
 use crate::sql::common::{CheckpointBarrier, MetadataField};
-// ============================================================================
-// ============================================================================
-
-#[derive(Copy, Clone, Debug, Encode, Decode, PartialEq, PartialOrd)]
-pub struct KafkaState {
-    pub partition: i32,
-    pub offset: i64,
-}
-
-/// Last committed partition offsets for this source subtask, tied to a checkpoint epoch.
-/// Materialized into a `.bin` under the job state dir from catalog before restart; see
-/// [`TaskContext::latest_safe_epoch`] and `StreamingTableDefinition` in `storage.proto`.
-#[derive(Debug, Encode, Decode)]
-pub(crate) struct KafkaSourceSavedOffsets {
-    /// Same numbering as [`CheckpointBarrier::epoch`] / catalog `latest_checkpoint_epoch` (as u64).
-    pub(crate) epoch: u64,
-    pub(crate) partitions: Vec<KafkaState>,
-}
-
-pub(crate) fn encode_kafka_offset_snapshot(saved: &KafkaSourceSavedOffsets) -> Result<Vec<u8>> {
-    bincode::encode_to_vec(saved, bincode::config::standard())
-        .map_err(|e| anyhow!("bincode encode Kafka offset snapshot: {e}"))
-}
-
-pub(crate) fn decode_kafka_offset_snapshot(bytes: &[u8]) -> Result<KafkaSourceSavedOffsets> {
-    let (saved, _) = bincode::decode_from_slice(bytes, bincode::config::standard())
-        .map_err(|e| anyhow!("bincode decode Kafka offset snapshot: {e}"))?;
-    Ok(saved)
-}
-
-pub(crate) fn kafka_snapshot_path(
-    job_dir: &std::path::Path,
-    pipeline_id: u32,
-    subtask_index: u32,
-) -> PathBuf {
-    job_dir.join(format!(
-        "kafka_source_offsets_pipe{}_sub{}.bin",
-        pipeline_id, subtask_index
-    ))
-}
-
-fn kafka_offsets_snapshot_path(ctx: &TaskContext) -> PathBuf {
-    kafka_snapshot_path(&ctx.state_dir, ctx.pipeline_id, ctx.subtask_index)
-}
-
-fn load_saved_offsets_if_recovering(ctx: &TaskContext) -> Option<KafkaSourceSavedOffsets> {
-    let safe = ctx.latest_safe_epoch();
-    if safe == 0 {
-        return None;
-    }
-    let path = kafka_offsets_snapshot_path(ctx);
-    let bytes = std::fs::read(&path).ok()?;
-    let saved = match decode_kafka_offset_snapshot(&bytes) {
-        Ok(v) => v,
-        Err(e) => {
-            warn!(
-                path = %path.display(),
-                error = %e,
-                "Failed to decode Kafka offset snapshot"
-            );
-            return None;
-        }
-    };
-    if saved.epoch > safe {
-        warn!(
-            path = %path.display(),
-            saved_epoch = saved.epoch,
-            safe_epoch = safe,
-            "Ignoring Kafka offset snapshot newer than catalog safe epoch"
-        );
-        return None;
-    }
-    Some(saved)
-}
 
 pub trait BatchDeserializer: Send + 'static {
     fn deserialize_slice(
@@ -220,8 +149,10 @@ pub struct KafkaSourceOperator {
 
     current_offsets: HashMap<i32, i64>,
     is_empty_assignment: bool,
-
     last_flush_time: Instant,
+
+    /// Checkpoint records injected before `on_start`; consumed once to restore partition offsets.
+    recovery_checkpoint_infos: Vec<SourceCheckpointInfo>,
 }
 
 impl KafkaSourceOperator {
@@ -252,14 +183,44 @@ impl KafkaSourceOperator {
             current_offsets: HashMap::new(),
             is_empty_assignment: false,
             last_flush_time: Instant::now(),
+            recovery_checkpoint_infos: vec![],
         }
     }
 
-    async fn init_and_assign_consumer(
-        &mut self,
-        ctx: &mut TaskContext,
-        saved_offsets: Option<KafkaSourceSavedOffsets>,
-    ) -> Result<()> {
+    fn load_recovery_offsets(&mut self, ctx: &TaskContext) -> (bool, HashMap<i32, i64>) {
+        if ctx.latest_safe_epoch() == 0 || self.recovery_checkpoint_infos.is_empty() {
+            return (false, HashMap::new());
+        }
+        let cp = self.recovery_checkpoint_infos.iter().find_map(|info| {
+            if let Some(source_checkpoint_info::Info::Kafka(cp)) = &info.info
+                && cp.subtask_index == ctx.subtask_index
+            {
+                return Some(cp);
+            }
+            None
+        });
+        match cp {
+            Some(cp) => {
+                info!(
+                    job_id = %ctx.job_id,
+                    pipeline_id = ctx.pipeline_id,
+                    subtask = ctx.subtask_index,
+                    epoch = cp.checkpoint_epoch,
+                    partitions = cp.partitions.len(),
+                    "Restoring Kafka source offsets from catalog checkpoint"
+                );
+                let map = cp
+                    .partitions
+                    .iter()
+                    .map(|p| (p.partition, p.offset))
+                    .collect();
+                (true, map)
+            }
+            None => (false, HashMap::new()),
+        }
+    }
+
+    async fn init_and_assign_consumer(&mut self, ctx: &mut TaskContext) -> Result<()> {
         info!("Creating kafka consumer for {}", self.bootstrap_servers);
         let mut client_config = ClientConfig::new();
 
@@ -282,24 +243,7 @@ impl KafkaSourceOperator {
             .set("group.id", &group_id)
             .create()?;
 
-        let (has_state, state_map) = if let Some(saved) = saved_offsets {
-            info!(
-                job_id = %ctx.job_id,
-                pipeline_id = ctx.pipeline_id,
-                subtask = ctx.subtask_index,
-                epoch = saved.epoch,
-                safe_epoch = ctx.latest_safe_epoch(),
-                partitions = saved.partitions.len(),
-                "Restoring Kafka source offsets from materialized checkpoint snapshot"
-            );
-            let mut m = HashMap::with_capacity(saved.partitions.len());
-            for s in saved.partitions {
-                m.insert(s.partition, s);
-            }
-            (true, m)
-        } else {
-            (false, HashMap::new())
-        };
+        let (has_state, saved_offsets_map) = self.load_recovery_offsets(ctx);
 
         let metadata = consumer
             .fetch_metadata(Some(&self.topic), Duration::from_secs(30))
@@ -317,10 +261,10 @@ impl KafkaSourceOperator {
 
         for p in partitions {
             if p.id().rem_euclid(pmax) == ctx.subtask_index as i32 {
-                // `current_offsets` / snapshot store last consumed offset; resume at next offset.
-                let offset = state_map
+                // saved_offsets_map stores last consumed offset; resume at next offset.
+                let offset = saved_offsets_map
                     .get(&p.id())
-                    .map(|s| Offset::Offset(s.offset.saturating_add(1)))
+                    .map(|&last| Offset::Offset(last.saturating_add(1)))
                     .unwrap_or_else(|| {
                         if has_state {
                             Offset::Beginning
@@ -357,9 +301,12 @@ impl SourceOperator for KafkaSourceOperator {
         &self.topic
     }
 
+    fn set_recovery_checkpoint(&mut self, infos: Vec<SourceCheckpointInfo>) {
+        self.recovery_checkpoint_infos = infos;
+    }
+
     async fn on_start(&mut self, ctx: &mut TaskContext) -> Result<()> {
-        let saved = load_saved_offsets_if_recovering(ctx);
-        self.init_and_assign_consumer(ctx, saved).await?;
+        self.init_and_assign_consumer(ctx).await?;
         self.rate_limiter = Some(GovernorRateLimiter::direct(Quota::per_second(
             self.messages_per_second,
         )));
@@ -479,7 +426,7 @@ impl SourceOperator for KafkaSourceOperator {
             warn!("Failed to commit async offset to Kafka Broker: {:?}", e);
         }
 
-        let epoch = u64::from(barrier.epoch);
+        let epoch = barrier.epoch;
         if self.current_offsets.is_empty() {
             return Ok(SourceCheckpointReport::default());
         }
diff --git a/src/runtime/streaming/operators/windows/session_aggregating_window.rs b/src/runtime/streaming/operators/windows/session_aggregating_window.rs
index 2da2c285..2056cdd9 100644
--- a/src/runtime/streaming/operators/windows/session_aggregating_window.rs
+++ b/src/runtime/streaming/operators/windows/session_aggregating_window.rs
@@ -871,7 +871,7 @@ impl Operator for SessionWindowOperator {
         self.state_store
             .as_ref()
             .expect("State store not initialized")
-            .prepare_checkpoint_epoch(barrier.epoch as u64)
+            .prepare_checkpoint_epoch(barrier.epoch)
             .map_err(|e| anyhow!("Snapshot failed: {e}"))?;
 
         info!(
@@ -881,11 +881,11 @@ impl Operator for SessionWindowOperator {
         Ok(())
     }
 
-    async fn commit_checkpoint(&mut self, epoch: u32, _ctx: &mut TaskContext) -> Result<()> {
+    async fn commit_checkpoint(&mut self, epoch: u64, _ctx: &mut TaskContext) -> Result<()> {
         self.state_store
             .as_ref()
             .expect("State store not initialized")
-            .commit_checkpoint_epoch(epoch as u64)
+            .commit_checkpoint_epoch(epoch)
             .map_err(|e| anyhow!("Commit checkpoint failed: {e}"))?;
         Ok(())
     }
diff --git a/src/runtime/streaming/operators/windows/sliding_aggregating_window.rs b/src/runtime/streaming/operators/windows/sliding_aggregating_window.rs
index 3516e950..f18b3b14 100644
--- a/src/runtime/streaming/operators/windows/sliding_aggregating_window.rs
+++ b/src/runtime/streaming/operators/windows/sliding_aggregating_window.rs
@@ -613,16 +613,16 @@ impl Operator for SlidingWindowOperator {
         self.state_store
             .as_ref()
             .expect("State store not initialized")
-            .prepare_checkpoint_epoch(barrier.epoch as u64)
+            .prepare_checkpoint_epoch(barrier.epoch)
             .map_err(|e| anyhow!("Snapshot failed: {e}"))?;
         Ok(())
     }
 
-    async fn commit_checkpoint(&mut self, epoch: u32, _ctx: &mut TaskContext) -> Result<()> {
+    async fn commit_checkpoint(&mut self, epoch: u64, _ctx: &mut TaskContext) -> Result<()> {
         self.state_store
             .as_ref()
             .expect("State store not initialized")
-            .commit_checkpoint_epoch(epoch as u64)
+            .commit_checkpoint_epoch(epoch)
             .map_err(|e| anyhow!("Commit checkpoint failed: {e}"))?;
         Ok(())
     }
diff --git a/src/runtime/streaming/operators/windows/tumbling_aggregating_window.rs b/src/runtime/streaming/operators/windows/tumbling_aggregating_window.rs
index 6b6b6029..5c805625 100644
--- a/src/runtime/streaming/operators/windows/tumbling_aggregating_window.rs
+++ b/src/runtime/streaming/operators/windows/tumbling_aggregating_window.rs
@@ -393,16 +393,16 @@ impl Operator for TumblingWindowOperator {
         self.state_store
             .as_ref()
             .expect("State store not initialized")
-            .prepare_checkpoint_epoch(barrier.epoch as u64)
+            .prepare_checkpoint_epoch(barrier.epoch)
             .map_err(|e| anyhow!("Snapshot failed: {e}"))?;
         Ok(())
     }
 
-    async fn commit_checkpoint(&mut self, epoch: u32, _ctx: &mut TaskContext) -> Result<()> {
+    async fn commit_checkpoint(&mut self, epoch: u64, _ctx: &mut TaskContext) -> Result<()> {
         self.state_store
             .as_ref()
             .expect("State store not initialized")
-            .commit_checkpoint_epoch(epoch as u64)
+            .commit_checkpoint_epoch(epoch)
             .map_err(|e| anyhow!("Commit checkpoint failed: {e}"))?;
         Ok(())
     }
diff --git a/src/runtime/streaming/operators/windows/window_function.rs b/src/runtime/streaming/operators/windows/window_function.rs
index 1249233e..37815b78 100644
--- a/src/runtime/streaming/operators/windows/window_function.rs
+++ b/src/runtime/streaming/operators/windows/window_function.rs
@@ -269,16 +269,16 @@ impl Operator for WindowFunctionOperator {
         self.state_store
             .as_ref()
             .expect("State store not initialized")
-            .prepare_checkpoint_epoch(barrier.epoch as u64)
+            .prepare_checkpoint_epoch(barrier.epoch)
             .map_err(|e| anyhow!("Snapshot failed: {e}"))?;
         Ok(())
     }
 
-    async fn commit_checkpoint(&mut self, epoch: u32, _ctx: &mut TaskContext) -> Result<()> {
+    async fn commit_checkpoint(&mut self, epoch: u64, _ctx: &mut TaskContext) -> Result<()> {
         self.state_store
             .as_ref()
             .expect("State store not initialized")
-            .commit_checkpoint_epoch(epoch as u64)
+            .commit_checkpoint_epoch(epoch)
             .map_err(|e| anyhow!("Commit checkpoint failed: {e}"))?;
         Ok(())
     }
diff --git a/src/runtime/streaming/protocol/control.rs b/src/runtime/streaming/protocol/control.rs
index 6d0bc492..70ab71be 100644
--- a/src/runtime/streaming/protocol/control.rs
+++ b/src/runtime/streaming/protocol/control.rs
@@ -11,15 +11,15 @@
 // limitations under the License.
 
 use super::event::CheckpointBarrier;
-use protocol::storage::SourceCheckpointPayload;
+use protocol::storage::SourceCheckpointInfo;
 use serde::{Deserialize, Serialize};
 use std::time::Duration;
 use tokio::sync::mpsc::{self, Receiver, Sender};
 
 #[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
 pub struct CheckpointBarrierWire {
-    pub epoch: u32,
-    pub min_epoch: u32,
+    pub epoch: u64,
+    pub min_epoch: u64,
     pub timestamp_secs: u64,
     pub timestamp_subsec_nanos: u32,
     pub then_stop: bool,
@@ -62,11 +62,11 @@ pub enum ControlCommand {
     DropState,
     /// Phase 2 of checkpoint 2PC: metadata durable; transactional Kafka sink should `commit_transaction`.
     Commit {
-        epoch: u32,
+        epoch: u64,
     },
     /// Roll back pre-committed transactional Kafka writes when checkpoint metadata commit failed or barrier declined.
     AbortCheckpoint {
-        epoch: u32,
+        epoch: u64,
     },
     UpdateConfig {
         config_json: String,
@@ -99,8 +99,8 @@ pub enum JobMasterEvent {
     CheckpointAck {
         pipeline_id: u32,
         epoch: u64,
-        /// Source protocol checkpoint payloads (enum-style oneof envelope).
-        source_payloads: Vec<SourceCheckpointPayload>,
+        /// Per-subtask checkpoint records produced directly by the source during snapshot.
+        source_infos: Vec<SourceCheckpointInfo>,
     },
     CheckpointDecline {
         pipeline_id: u32,
diff --git a/src/runtime/streaming/protocol/event.rs b/src/runtime/streaming/protocol/event.rs
index 21be6852..093d99ba 100644
--- a/src/runtime/streaming/protocol/event.rs
+++ b/src/runtime/streaming/protocol/event.rs
@@ -27,8 +27,8 @@ pub enum Watermark {
 
 #[derive(Debug, Copy, Clone, PartialEq, Eq, Encode, Decode, Serialize, Deserialize)]
 pub struct CheckpointBarrier {
-    pub epoch: u32,
-    pub min_epoch: u32,
+    pub epoch: u64,
+    pub min_epoch: u64,
     pub timestamp: SystemTime,
     pub then_stop: bool,
 }
diff --git a/src/sql/analysis/source_metadata_visitor.rs b/src/sql/analysis/source_metadata_visitor.rs
index 55350301..b37d3a1b 100644
--- a/src/sql/analysis/source_metadata_visitor.rs
+++ b/src/sql/analysis/source_metadata_visitor.rs
@@ -51,7 +51,11 @@ impl<'a> SourceMetadataVisitor<'a> {
 
         let table = self.schema_provider.get_catalog_table(&table_name)?;
         match table {
-            crate::sql::schema::table::Table::ConnectorTable(t) => t.registry_id,
+            crate::sql::schema::table::CatalogEntity::ExternalConnector(b) => match b.as_ref() {
+                crate::sql::schema::catalog::ExternalTable::Source(t) => t.registry_id,
+                crate::sql::schema::catalog::ExternalTable::Lookup(t) => t.registry_id,
+                _ => None,
+            },
             _ => None,
         }
     }
diff --git a/src/sql/analysis/source_rewriter.rs b/src/sql/analysis/source_rewriter.rs
index 0bd15e85..620ea336 100644
--- a/src/sql/analysis/source_rewriter.rs
+++ b/src/sql/analysis/source_rewriter.rs
@@ -27,8 +27,8 @@ use crate::sql::logical_node::table_source::StreamIngestionNode;
 use crate::sql::logical_node::watermark_node::EventTimeWatermarkNode;
 use crate::sql::schema::ColumnDescriptor;
 use crate::sql::schema::StreamSchemaProvider;
-use crate::sql::schema::source_table::SourceTable;
-use crate::sql::schema::table::Table;
+use crate::sql::schema::catalog::{ExternalTable, SourceTable};
+use crate::sql::schema::table::CatalogEntity;
 use crate::sql::types::TIMESTAMP_FIELD;
 
 /// Rewrites table scans: projections are lifted out of scans into a dedicated projection node
@@ -285,15 +285,21 @@ impl TreeNodeRewriter for SourceRewriter<'_> {
             .ok_or_else(|| DataFusionError::Plan(format!("Table {table_name} not found")))?;
 
         match table {
-            Table::ConnectorTable(table) => self.mutate_connector_table(&table_scan, table),
-            Table::LookupTable(_table) => {
-                // TODO: implement LookupSource extension
-                plan_err!("Lookup tables are not yet supported")
-            }
-            Table::TableFromQuery {
+            CatalogEntity::ExternalConnector(b) => match b.as_ref() {
+                ExternalTable::Source(source) => self.mutate_connector_table(&table_scan, source),
+                ExternalTable::Lookup(_) => {
+                    // TODO: implement LookupSource extension
+                    plan_err!("Lookup tables are not yet supported")
+                }
+                ExternalTable::Sink(sink) => plan_err!(
+                    "Cannot SELECT from sink table '{}' (sinks are write-only)",
+                    sink.name()
+                ),
+            },
+            CatalogEntity::ComputedTable {
                 name: _,
                 logical_plan,
-            } => self.mutate_table_from_query(&table_scan, logical_plan),
+            } => self.mutate_table_from_query(&table_scan, logical_plan.as_ref()),
         }
     }
 }
diff --git a/src/sql/api/mod.rs b/src/sql/api/mod.rs
index cdc119b7..9fc6b23f 100644
--- a/src/sql/api/mod.rs
+++ b/src/sql/api/mod.rs
@@ -25,8 +25,6 @@ pub mod var_str;
 
 use serde::{Deserialize, Serialize};
 
-pub use connections::ConnectionProfile;
-
 #[derive(Serialize, Deserialize, Clone, Debug)]
 #[serde(rename_all = "camelCase")]
 pub struct PaginatedCollection<T> {
diff --git a/src/sql/common/connector_options.rs b/src/sql/common/connector_options.rs
index e2e306b6..0702d945 100644
--- a/src/sql/common/connector_options.rs
+++ b/src/sql/common/connector_options.rs
@@ -106,6 +106,21 @@ impl ConnectorOptions {
         }
     }
 
+    pub fn peek_opt_str(&self, name: &str) -> DFResult<Option<String>> {
+        match self.options.get(name) {
+            Some(Expr::Value(ValueWithSpan {
+                value: SqlValue::SingleQuotedString(s),
+                span: _,
+            })) => Ok(Some(s.clone())),
+            Some(e) => Err(plan_datafusion_err!(
+                "expected with option '{}' to be a single-quoted string, but it was `{:?}`",
+                name,
+                e
+            )),
+            None => Ok(None),
+        }
+    }
+
     pub fn pull_str(&mut self, name: &str) -> DFResult<String> {
         self.pull_opt_str(name)?
             .ok_or_else(|| plan_datafusion_err!("required option '{}' not set", name))
diff --git a/src/sql/common/constants.rs b/src/sql/common/constants.rs
index 19fdbcb3..8cdb68e3 100644
--- a/src/sql/common/constants.rs
+++ b/src/sql/common/constants.rs
@@ -139,6 +139,11 @@ pub mod interval_duration_unit {
 
 pub mod connection_format_value {
     pub const JSON: &str = "json";
+    pub const CSV: &str = "csv";
+    pub const JSONL: &str = "jsonl";
+    pub const NDJSON: &str = "ndjson";
+    pub const LANCE: &str = "lance";
+    pub const ORC: &str = "orc";
     pub const DEBEZIUM_JSON: &str = "debezium_json";
     pub const AVRO: &str = "avro";
     pub const PARQUET: &str = "parquet";
@@ -246,6 +251,8 @@ pub mod connector_type {
     pub const FILESYSTEM: &str = "filesystem";
     pub const DELTA: &str = "delta";
     pub const ICEBERG: &str = "iceberg";
+    pub const LANCE_DB: &str = "lanceDB";
+    pub const S3: &str = "s3";
     pub const PULSAR: &str = "pulsar";
     pub const NATS: &str = "nats";
     pub const REDIS: &str = "redis";
@@ -264,7 +271,14 @@ pub mod connection_table_role {
     pub const LOOKUP: &str = "lookup";
 }
 
-pub const SUPPORTED_CONNECTOR_ADAPTERS: &[&str] = &[connector_type::KAFKA];
+pub const SUPPORTED_CONNECTOR_ADAPTERS: &[&str] = &[
+    connector_type::KAFKA,
+    connector_type::FILESYSTEM,
+    connector_type::S3,
+    connector_type::DELTA,
+    connector_type::ICEBERG,
+    connector_type::LANCE_DB,
+];
 
 pub mod kafka_with_value {
     pub const SCAN_LATEST: &str = "latest";
diff --git a/src/sql/common/format_from_opts.rs b/src/sql/common/format_from_opts.rs
index 276235c1..ffd29572 100644
--- a/src/sql/common/format_from_opts.rs
+++ b/src/sql/common/format_from_opts.rs
@@ -19,9 +19,9 @@ use datafusion::common::{Result as DFResult, plan_datafusion_err, plan_err};
 use super::connector_options::ConnectorOptions;
 use super::constants::{bad_data_value, connection_format_value, framing_method_value};
 use super::formats::{
-    AvroFormat, BadData, DecimalEncoding, Format, Framing, JsonCompression, JsonFormat,
-    NewlineDelimitedFraming, ParquetCompression, ParquetFormat, ProtobufFormat, RawBytesFormat,
-    RawStringFormat, TimestampFormat,
+    AvroFormat, BadData, CsvFormat, DecimalEncoding, Format, Framing, JsonCompression, JsonFormat,
+    LanceFormat, NewlineDelimitedFraming, ParquetCompression, ParquetFormat, ProtobufFormat,
+    RawBytesFormat, RawStringFormat, TimestampFormat,
 };
 use super::with_option_keys as opt;
 
@@ -61,12 +61,14 @@ impl JsonFormat {
 
 impl Format {
     pub fn from_opts(opts: &mut ConnectorOptions) -> DFResult<Option<Self>> {
-        let Some(name) = opts.pull_opt_str(opt::FORMAT)? else {
+        let Some(name) = opts.peek_opt_str(opt::FORMAT)? else {
             return Ok(None);
         };
         let n = name.to_lowercase();
         match n.as_str() {
             connection_format_value::JSON => Ok(Some(Format::Json(JsonFormat::from_opts(opts)?))),
+            connection_format_value::CSV => Ok(Some(Format::Csv(CsvFormat {}))),
+            connection_format_value::LANCE => Ok(Some(Format::Lance(LanceFormat {}))),
             connection_format_value::DEBEZIUM_JSON => {
                 let mut j = JsonFormat::from_opts(opts)?;
                 j.debezium = true;
diff --git a/src/sql/common/formats.rs b/src/sql/common/formats.rs
index aad3ce18..a47d93cf 100644
--- a/src/sql/common/formats.rs
+++ b/src/sql/common/formats.rs
@@ -115,6 +115,14 @@ pub struct RawStringFormat {}
 #[serde(rename_all = "snake_case")]
 pub struct RawBytesFormat {}
 
+#[derive(Clone, Debug, Serialize, Deserialize, PartialEq, Eq, Hash, PartialOrd)]
+#[serde(rename_all = "snake_case")]
+pub struct CsvFormat {}
+
+#[derive(Clone, Debug, Serialize, Deserialize, PartialEq, Eq, Hash, PartialOrd)]
+#[serde(rename_all = "snake_case")]
+pub struct LanceFormat {}
+
 #[derive(Clone, Debug, Serialize, Deserialize, PartialEq, Eq, Hash, PartialOrd)]
 #[serde(rename_all = "snake_case")]
 pub struct AvroFormat {
@@ -199,6 +207,8 @@ pub struct ProtobufFormat {
 #[serde(rename_all = "snake_case", tag = "type")]
 pub enum Format {
     Json(JsonFormat),
+    Csv(CsvFormat),
+    Lance(LanceFormat),
     Avro(AvroFormat),
     Protobuf(ProtobufFormat),
     Parquet(ParquetFormat),
@@ -216,6 +226,8 @@ impl Format {
     pub fn name(&self) -> &'static str {
         match self {
             Format::Json(_) => connection_format_value::JSON,
+            Format::Csv(_) => connection_format_value::CSV,
+            Format::Lance(_) => connection_format_value::LANCE,
             Format::Avro(_) => connection_format_value::AVRO,
             Format::Protobuf(_) => connection_format_value::PROTOBUF,
             Format::Parquet(_) => connection_format_value::PARQUET,
diff --git a/src/sql/common/with_option_keys.rs b/src/sql/common/with_option_keys.rs
index b998d1eb..21bfa691 100644
--- a/src/sql/common/with_option_keys.rs
+++ b/src/sql/common/with_option_keys.rs
@@ -16,6 +16,8 @@ pub const FORMAT: &str = "format";
 pub const DEFAULT_FORMAT_VALUE: &str = "json";
 pub const BAD_DATA: &str = "bad_data";
 pub const PARTITION_BY: &str = "partition_by";
+pub const PATH: &str = "path";
+pub const SINK_PATH: &str = "sink.path";
 
 pub const EVENT_TIME_FIELD: &str = "event_time_field";
 pub const WATERMARK_FIELD: &str = "watermark_field";
@@ -71,6 +73,15 @@ pub const AVRO_SCHEMA_ID: &str = "avro.schema_id";
 pub const PARQUET_COMPRESSION: &str = "parquet.compression";
 pub const PARQUET_ROW_GROUP_BYTES: &str = "parquet.row_group_bytes";
 
+// ── S3 ────────────────────────────────────────────────────────────────────
+
+pub const S3_BUCKET: &str = "s3.bucket";
+pub const S3_REGION: &str = "s3.region";
+pub const S3_ENDPOINT: &str = "s3.endpoint";
+pub const S3_ACCESS_KEY_ID: &str = "s3.access_key_id";
+pub const S3_SECRET_ACCESS_KEY: &str = "s3.secret_access_key";
+pub const S3_SESSION_TOKEN: &str = "s3.session_token";
+
 // ── Protobuf ────────────────────────────────────────────────────────────────
 
 pub const PROTOBUF_INTO_UNSTRUCTURED_JSON: &str = "protobuf.into_unstructured_json";
@@ -84,3 +95,11 @@ pub const FRAMING_METHOD: &str = "framing.method";
 pub const FRAMING_MAX_LINE_LENGTH: &str = "framing.max_line_length";
 
 pub const FORMAT_DEBEZIUM_FLAG: &str = "format.debezium";
+
+// ── Streaming runtime common options ───────────────────────────────────────
+
+pub const CHECKPOINT_INTERVAL_MS: &str = "checkpoint.interval.ms";
+pub const PIPELINE_PARALLELISM: &str = "pipeline.parallelism";
+pub const KEY_BY_PARALLELISM: &str = "key_by.parallelism";
+pub const OPERATOR_MEMORY_BYTES: &str = "operator.memory.bytes";
+pub const SINK_MEMORY_BYTES: &str = "sink.memory.bytes";
diff --git a/src/sql/connector/config.rs b/src/sql/connector/config.rs
new file mode 100644
index 00000000..c3eaf00f
--- /dev/null
+++ b/src/sql/connector/config.rs
@@ -0,0 +1,91 @@
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+use protocol::function_stream_graph::{
+    DeltaSinkConfig, FilesystemSinkConfig, IcebergSinkConfig, KafkaSinkConfig, KafkaSourceConfig,
+    LanceDbSinkConfig, S3SinkConfig, connector_op,
+};
+
+#[derive(Debug, Clone)]
+pub enum ConnectorConfig {
+    KafkaSource(KafkaSourceConfig),
+    KafkaSink(KafkaSinkConfig),
+    FilesystemSink(FilesystemSinkConfig),
+    DeltaSink(DeltaSinkConfig),
+    IcebergSink(IcebergSinkConfig),
+    S3Sink(S3SinkConfig),
+    LanceDbSink(LanceDbSinkConfig),
+}
+
+impl ConnectorConfig {
+    pub fn to_proto_config(&self) -> connector_op::Config {
+        match self {
+            ConnectorConfig::KafkaSource(cfg) => connector_op::Config::KafkaSource(cfg.clone()),
+            ConnectorConfig::KafkaSink(cfg) => connector_op::Config::KafkaSink(cfg.clone()),
+            ConnectorConfig::FilesystemSink(cfg) => {
+                connector_op::Config::FilesystemSink(cfg.clone())
+            }
+            ConnectorConfig::DeltaSink(cfg) => connector_op::Config::DeltaSink(cfg.clone()),
+            ConnectorConfig::IcebergSink(cfg) => connector_op::Config::IcebergSink(cfg.clone()),
+            ConnectorConfig::S3Sink(cfg) => connector_op::Config::S3Sink(cfg.clone()),
+            ConnectorConfig::LanceDbSink(cfg) => connector_op::Config::LancedbSink(cfg.clone()),
+        }
+    }
+}
+
+impl PartialEq for ConnectorConfig {
+    fn eq(&self, other: &Self) -> bool {
+        use prost::Message;
+        match (self, other) {
+            (ConnectorConfig::KafkaSource(a), ConnectorConfig::KafkaSource(b)) => {
+                a.encode_to_vec() == b.encode_to_vec()
+            }
+            (ConnectorConfig::KafkaSink(a), ConnectorConfig::KafkaSink(b)) => {
+                a.encode_to_vec() == b.encode_to_vec()
+            }
+            (ConnectorConfig::FilesystemSink(a), ConnectorConfig::FilesystemSink(b)) => {
+                a.encode_to_vec() == b.encode_to_vec()
+            }
+            (ConnectorConfig::DeltaSink(a), ConnectorConfig::DeltaSink(b)) => {
+                a.encode_to_vec() == b.encode_to_vec()
+            }
+            (ConnectorConfig::IcebergSink(a), ConnectorConfig::IcebergSink(b)) => {
+                a.encode_to_vec() == b.encode_to_vec()
+            }
+            (ConnectorConfig::S3Sink(a), ConnectorConfig::S3Sink(b)) => {
+                a.encode_to_vec() == b.encode_to_vec()
+            }
+            (ConnectorConfig::LanceDbSink(a), ConnectorConfig::LanceDbSink(b)) => {
+                a.encode_to_vec() == b.encode_to_vec()
+            }
+            _ => false,
+        }
+    }
+}
+
+impl Eq for ConnectorConfig {}
+
+impl std::hash::Hash for ConnectorConfig {
+    fn hash<H: std::hash::Hasher>(&self, state: &mut H) {
+        use prost::Message;
+        std::mem::discriminant(self).hash(state);
+        match self {
+            ConnectorConfig::KafkaSource(cfg) => cfg.encode_to_vec().hash(state),
+            ConnectorConfig::KafkaSink(cfg) => cfg.encode_to_vec().hash(state),
+            ConnectorConfig::FilesystemSink(cfg) => cfg.encode_to_vec().hash(state),
+            ConnectorConfig::DeltaSink(cfg) => cfg.encode_to_vec().hash(state),
+            ConnectorConfig::IcebergSink(cfg) => cfg.encode_to_vec().hash(state),
+            ConnectorConfig::S3Sink(cfg) => cfg.encode_to_vec().hash(state),
+            ConnectorConfig::LanceDbSink(cfg) => cfg.encode_to_vec().hash(state),
+        }
+    }
+}
diff --git a/src/sql/connector/factory.rs b/src/sql/connector/factory.rs
new file mode 100644
index 00000000..8c37a15a
--- /dev/null
+++ b/src/sql/connector/factory.rs
@@ -0,0 +1,67 @@
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+use std::collections::HashMap;
+
+use datafusion::arrow::datatypes::Schema;
+use datafusion::common::Result;
+
+use super::config::ConnectorConfig;
+use super::registry::REGISTRY;
+use super::sink::runtime_config::SinkRuntimeConfig;
+use crate::sql::common::connector_options::ConnectorOptions;
+use crate::sql::common::formats::{BadData, Format};
+use crate::sql::schema::table_role::TableRole;
+
+pub fn build_connector_config(
+    connector_name: &str,
+    role: TableRole,
+    options: &mut ConnectorOptions,
+    format: &Option<Format>,
+    bad_data: BadData,
+) -> Result<ConnectorConfig> {
+    let runtime_opts_map = options.snapshot_for_catalog().into_iter().collect();
+    let runtime_props =
+        SinkRuntimeConfig::from_options_map(&runtime_opts_map)?.to_runtime_properties();
+    match role {
+        TableRole::Ingestion | TableRole::Reference => REGISTRY
+            .get_source(connector_name)?
+            .build_source_config(options, format, bad_data),
+        TableRole::Egress => {
+            REGISTRY
+                .get_sink(connector_name)?
+                .build_sink_config(options, format, &runtime_props)
+        }
+    }
+}
+
+pub fn build_connector_config_from_options(
+    connector_name: &str,
+    role: TableRole,
+    options: &mut ConnectorOptions,
+    format: &Option<Format>,
+    bad_data: BadData,
+) -> Result<ConnectorConfig> {
+    build_connector_config(connector_name, role, options, format, bad_data)
+}
+
+pub fn build_connector_config_from_catalog(
+    connector_name: &str,
+    role: TableRole,
+    opts: HashMap<String, String>,
+    _physical_schema: &Schema,
+) -> Result<ConnectorConfig> {
+    let mut options = ConnectorOptions::from_flat_string_map(opts)?;
+    let format = Format::from_opts(&mut options)?;
+    let bad_data = BadData::from_opts(&mut options)?;
+    build_connector_config(connector_name, role, &mut options, &format, bad_data)
+}
diff --git a/src/sql/connector/mod.rs b/src/sql/connector/mod.rs
new file mode 100644
index 00000000..f477c976
--- /dev/null
+++ b/src/sql/connector/mod.rs
@@ -0,0 +1,18 @@
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+pub mod config;
+pub mod factory;
+pub mod provider;
+pub mod registry;
+pub mod sink;
+pub mod source;
diff --git a/src/sql/connector/provider.rs b/src/sql/connector/provider.rs
new file mode 100644
index 00000000..8875ee0c
--- /dev/null
+++ b/src/sql/connector/provider.rs
@@ -0,0 +1,52 @@
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+use datafusion::common::{DataFusionError, Result};
+
+use super::config::ConnectorConfig;
+use super::sink::runtime_config::SinkRuntimeProperties;
+use crate::sql::common::connector_options::ConnectorOptions;
+use crate::sql::common::formats::{BadData, Format};
+
+pub trait SourceProvider: Send + Sync {
+    fn name(&self) -> &'static str;
+
+    fn build_source_config(
+        &self,
+        options: &mut ConnectorOptions,
+        format: &Option<Format>,
+        bad_data: BadData,
+    ) -> Result<ConnectorConfig>;
+}
+
+pub trait SinkProvider: Send + Sync {
+    fn name(&self) -> &'static str;
+    fn build_sink_config(
+        &self,
+        options: &mut ConnectorOptions,
+        format: &Option<Format>,
+        runtime_props: &SinkRuntimeProperties,
+    ) -> Result<ConnectorConfig>;
+}
+
+pub fn require_option(
+    options: &mut ConnectorOptions,
+    key: &str,
+    connector_name: &str,
+) -> Result<String> {
+    options.pull_opt_str(key)?.ok_or_else(|| {
+        DataFusionError::Plan(format!(
+            "Connector '{}' requires option '{}' to be set",
+            connector_name, key
+        ))
+    })
+}
diff --git a/src/sql/connector/registry.rs b/src/sql/connector/registry.rs
new file mode 100644
index 00000000..4a8a8c1c
--- /dev/null
+++ b/src/sql/connector/registry.rs
@@ -0,0 +1,86 @@
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+use std::collections::HashMap;
+use std::sync::{Arc, LazyLock};
+
+use datafusion::common::{DataFusionError, Result};
+
+use super::provider::{SinkProvider, SourceProvider};
+use super::sink::delta::DeltaSinkConnector;
+use super::sink::filesystem::FilesystemSinkConnector;
+use super::sink::iceberg::IcebergSinkConnector;
+use super::sink::kafka::KafkaSinkConnector;
+use super::sink::lancedb::LanceDbSinkConnector;
+use super::sink::s3::S3SinkConnector;
+use super::source::kafka::KafkaSourceConnector;
+
+pub struct ConnectorRegistry {
+    sources: HashMap<String, Arc<dyn SourceProvider>>,
+    sinks: HashMap<String, Arc<dyn SinkProvider>>,
+}
+
+impl ConnectorRegistry {
+    fn new() -> Self {
+        let mut registry = Self {
+            sources: HashMap::new(),
+            sinks: HashMap::new(),
+        };
+
+        registry.register_source(Arc::new(KafkaSourceConnector));
+
+        registry.register_sink(Arc::new(KafkaSinkConnector));
+        registry.register_sink(Arc::new(S3SinkConnector));
+        registry.register_sink(Arc::new(FilesystemSinkConnector));
+        registry.register_sink(Arc::new(DeltaSinkConnector));
+        registry.register_sink(Arc::new(IcebergSinkConnector));
+        registry.register_sink(Arc::new(LanceDbSinkConnector));
+
+        registry
+    }
+
+    pub fn register_source(&mut self, provider: Arc<dyn SourceProvider>) {
+        self.sources
+            .insert(provider.name().to_ascii_lowercase(), provider);
+    }
+
+    pub fn register_sink(&mut self, provider: Arc<dyn SinkProvider>) {
+        self.sinks
+            .insert(provider.name().to_ascii_lowercase(), provider);
+    }
+
+    pub fn get_source(&self, connector_name: &str) -> Result<Arc<dyn SourceProvider>> {
+        self.sources
+            .get(&connector_name.to_ascii_lowercase())
+            .cloned()
+            .ok_or_else(|| {
+                DataFusionError::Plan(format!(
+                    "Connector '{}' is not registered or does not support being used as a SOURCE",
+                    connector_name
+                ))
+            })
+    }
+
+    pub fn get_sink(&self, connector_name: &str) -> Result<Arc<dyn SinkProvider>> {
+        self.sinks
+            .get(&connector_name.to_ascii_lowercase())
+            .cloned()
+            .ok_or_else(|| {
+                DataFusionError::Plan(format!(
+                    "Connector '{}' is not registered or does not support being used as a SINK",
+                    connector_name
+                ))
+            })
+    }
+}
+
+pub static REGISTRY: LazyLock<ConnectorRegistry> = LazyLock::new(ConnectorRegistry::new);
diff --git a/src/sql/connector/sink/delta.rs b/src/sql/connector/sink/delta.rs
new file mode 100644
index 00000000..cd86660d
--- /dev/null
+++ b/src/sql/connector/sink/delta.rs
@@ -0,0 +1,60 @@
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+use datafusion::common::Result;
+use protocol::function_stream_graph::{DeltaSinkConfig, SinkFormatProto};
+
+use crate::sql::common::Format;
+use crate::sql::common::connector_options::ConnectorOptions;
+use crate::sql::common::constants::connector_type;
+use crate::sql::connector::config::ConnectorConfig;
+use crate::sql::connector::provider::SinkProvider;
+use crate::sql::connector::sink::runtime_config::SinkRuntimeProperties;
+use crate::sql::connector::sink::utils::SinkUtils;
+
+pub struct DeltaSinkConnector;
+
+impl SinkProvider for DeltaSinkConnector {
+    fn name(&self) -> &'static str {
+        connector_type::DELTA
+    }
+
+    fn build_sink_config(
+        &self,
+        options: &mut ConnectorOptions,
+        format: &Option<Format>,
+        runtime_props: &SinkRuntimeProperties,
+    ) -> Result<ConnectorConfig> {
+        let path = SinkUtils::require_path(options)?;
+        let parquet_compression = SinkUtils::extract_parquet_compression(options)?;
+        let format_proto = SinkUtils::resolve_sink_format(
+            format,
+            self.name(),
+            &[
+                SinkFormatProto::SinkFormatCsv,
+                SinkFormatProto::SinkFormatJsonl,
+                SinkFormatProto::SinkFormatAvro,
+                SinkFormatProto::SinkFormatParquet,
+                SinkFormatProto::SinkFormatOrc,
+            ],
+        )?;
+        let extra_properties = options.drain_remaining_string_values()?;
+
+        Ok(ConnectorConfig::DeltaSink(DeltaSinkConfig {
+            path,
+            format: format_proto,
+            parquet_compression,
+            extra_properties,
+            runtime_properties: runtime_props.clone(),
+        }))
+    }
+}
diff --git a/src/sql/connector/sink/filesystem.rs b/src/sql/connector/sink/filesystem.rs
new file mode 100644
index 00000000..224b1805
--- /dev/null
+++ b/src/sql/connector/sink/filesystem.rs
@@ -0,0 +1,60 @@
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+use datafusion::common::Result;
+use protocol::function_stream_graph::{FilesystemSinkConfig, SinkFormatProto};
+
+use crate::sql::common::Format;
+use crate::sql::common::connector_options::ConnectorOptions;
+use crate::sql::common::constants::connector_type;
+use crate::sql::connector::config::ConnectorConfig;
+use crate::sql::connector::provider::SinkProvider;
+use crate::sql::connector::sink::runtime_config::SinkRuntimeProperties;
+use crate::sql::connector::sink::utils::SinkUtils;
+
+pub struct FilesystemSinkConnector;
+
+impl SinkProvider for FilesystemSinkConnector {
+    fn name(&self) -> &'static str {
+        connector_type::FILESYSTEM
+    }
+
+    fn build_sink_config(
+        &self,
+        options: &mut ConnectorOptions,
+        format: &Option<Format>,
+        runtime_props: &SinkRuntimeProperties,
+    ) -> Result<ConnectorConfig> {
+        let path = SinkUtils::require_path(options)?;
+        let parquet_compression = SinkUtils::extract_parquet_compression(options)?;
+        let format_proto = SinkUtils::resolve_sink_format(
+            format,
+            self.name(),
+            &[
+                SinkFormatProto::SinkFormatCsv,
+                SinkFormatProto::SinkFormatJsonl,
+                SinkFormatProto::SinkFormatAvro,
+                SinkFormatProto::SinkFormatParquet,
+                SinkFormatProto::SinkFormatOrc,
+            ],
+        )?;
+        let extra_properties = options.drain_remaining_string_values()?;
+
+        Ok(ConnectorConfig::FilesystemSink(FilesystemSinkConfig {
+            path,
+            format: format_proto,
+            parquet_compression,
+            extra_properties,
+            runtime_properties: runtime_props.clone(),
+        }))
+    }
+}
diff --git a/src/sql/connector/sink/iceberg.rs b/src/sql/connector/sink/iceberg.rs
new file mode 100644
index 00000000..12f0d378
--- /dev/null
+++ b/src/sql/connector/sink/iceberg.rs
@@ -0,0 +1,57 @@
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+use datafusion::common::Result;
+use protocol::function_stream_graph::{IcebergSinkConfig, SinkFormatProto};
+
+use crate::sql::common::Format;
+use crate::sql::common::connector_options::ConnectorOptions;
+use crate::sql::common::constants::connector_type;
+use crate::sql::connector::config::ConnectorConfig;
+use crate::sql::connector::provider::SinkProvider;
+use crate::sql::connector::sink::runtime_config::SinkRuntimeProperties;
+use crate::sql::connector::sink::utils::SinkUtils;
+
+pub struct IcebergSinkConnector;
+
+impl SinkProvider for IcebergSinkConnector {
+    fn name(&self) -> &'static str {
+        connector_type::ICEBERG
+    }
+
+    fn build_sink_config(
+        &self,
+        options: &mut ConnectorOptions,
+        format: &Option<Format>,
+        runtime_props: &SinkRuntimeProperties,
+    ) -> Result<ConnectorConfig> {
+        let path = SinkUtils::require_path(options)?;
+        let parquet_compression = SinkUtils::extract_parquet_compression(options)?;
+        let format_proto = SinkUtils::resolve_sink_format(
+            format,
+            self.name(),
+            &[
+                SinkFormatProto::SinkFormatCsv,
+                SinkFormatProto::SinkFormatParquet,
+            ],
+        )?;
+        let extra_properties = options.drain_remaining_string_values()?;
+
+        Ok(ConnectorConfig::IcebergSink(IcebergSinkConfig {
+            path,
+            format: format_proto,
+            parquet_compression,
+            extra_properties,
+            runtime_properties: runtime_props.clone(),
+        }))
+    }
+}
diff --git a/src/sql/connector/sink/kafka.rs b/src/sql/connector/sink/kafka.rs
new file mode 100644
index 00000000..a6fd115c
--- /dev/null
+++ b/src/sql/connector/sink/kafka.rs
@@ -0,0 +1,159 @@
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+use datafusion::common::{Result, plan_datafusion_err, plan_err};
+use protocol::function_stream_graph::{
+    DecimalEncodingProto, FormatConfig, JsonFormatConfig, KafkaAuthConfig, KafkaAuthNone,
+    KafkaSinkCommitMode, KafkaSinkConfig, RawBytesFormatConfig, RawStringFormatConfig,
+    TimestampFormatProto, format_config, kafka_auth_config,
+};
+
+use crate::sql::common::connector_options::ConnectorOptions;
+use crate::sql::common::constants::{connector_type, kafka_with_value};
+use crate::sql::common::formats::{
+    DecimalEncoding as SqlDecimalEncoding, Format as SqlFormat,
+    TimestampFormat as SqlTimestampFormat,
+};
+use crate::sql::common::with_option_keys as opt;
+use crate::sql::connector::config::ConnectorConfig;
+use crate::sql::connector::provider::SinkProvider;
+use crate::sql::connector::sink::runtime_config::SinkRuntimeProperties;
+
+pub struct KafkaSinkConnector;
+
+impl KafkaSinkConnector {
+    fn sql_format_to_proto(fmt: &SqlFormat) -> Result<FormatConfig> {
+        match fmt {
+            SqlFormat::Json(j) => Ok(FormatConfig {
+                format: Some(format_config::Format::Json(JsonFormatConfig {
+                    timestamp_format: match j.timestamp_format {
+                        SqlTimestampFormat::RFC3339 => {
+                            TimestampFormatProto::TimestampRfc3339 as i32
+                        }
+                        SqlTimestampFormat::UnixMillis => {
+                            TimestampFormatProto::TimestampUnixMillis as i32
+                        }
+                    },
+                    decimal_encoding: match j.decimal_encoding {
+                        SqlDecimalEncoding::Number => DecimalEncodingProto::DecimalNumber as i32,
+                        SqlDecimalEncoding::String => DecimalEncodingProto::DecimalString as i32,
+                        SqlDecimalEncoding::Bytes => DecimalEncodingProto::DecimalBytes as i32,
+                    },
+                    include_schema: j.include_schema,
+                    confluent_schema_registry: j.confluent_schema_registry,
+                    schema_id: j.schema_id,
+                    debezium: j.debezium,
+                    unstructured: j.unstructured,
+                })),
+            }),
+            SqlFormat::RawString(_) => Ok(FormatConfig {
+                format: Some(format_config::Format::RawString(RawStringFormatConfig {})),
+            }),
+            SqlFormat::RawBytes(_) => Ok(FormatConfig {
+                format: Some(format_config::Format::RawBytes(RawBytesFormatConfig {})),
+            }),
+            other => plan_err!(
+                "Kafka sink connector: format '{}' is not supported",
+                other.name()
+            ),
+        }
+    }
+}
+
+impl SinkProvider for KafkaSinkConnector {
+    fn name(&self) -> &'static str {
+        connector_type::KAFKA
+    }
+
+    fn build_sink_config(
+        &self,
+        options: &mut ConnectorOptions,
+        format: &Option<SqlFormat>,
+        _runtime_props: &SinkRuntimeProperties,
+    ) -> Result<ConnectorConfig> {
+        let bootstrap_servers = match options.pull_opt_str(opt::KAFKA_BOOTSTRAP_SERVERS)? {
+            Some(s) => s,
+            None => options
+                .pull_opt_str(opt::KAFKA_BOOTSTRAP_SERVERS_LEGACY)?
+                .ok_or_else(|| {
+                    plan_datafusion_err!(
+                        "Kafka connector requires 'bootstrap.servers' in the WITH clause"
+                    )
+                })?,
+        };
+
+        let topic = options.pull_opt_str(opt::KAFKA_TOPIC)?.ok_or_else(|| {
+            plan_datafusion_err!("Kafka connector requires 'topic' in the WITH clause")
+        })?;
+
+        let sql_format = format.as_ref().ok_or_else(|| {
+            plan_datafusion_err!(
+                "Kafka sink requires 'format' in the WITH clause (e.g. format = 'json')"
+            )
+        })?;
+        let proto_format = Self::sql_format_to_proto(sql_format)?;
+
+        let value_subject = options.pull_opt_str(opt::KAFKA_VALUE_SUBJECT)?;
+
+        let commit_mode = match options
+            .pull_opt_str(opt::KAFKA_SINK_COMMIT_MODE)?
+            .as_deref()
+        {
+            Some(s)
+                if s == kafka_with_value::SINK_COMMIT_EXACTLY_ONCE_HYPHEN
+                    || s == kafka_with_value::SINK_COMMIT_EXACTLY_ONCE_UNDERSCORE =>
+            {
+                KafkaSinkCommitMode::KafkaSinkExactlyOnce as i32
+            }
+            Some(s)
+                if s == kafka_with_value::SINK_COMMIT_AT_LEAST_ONCE_HYPHEN
+                    || s == kafka_with_value::SINK_COMMIT_AT_LEAST_ONCE_UNDERSCORE =>
+            {
+                KafkaSinkCommitMode::KafkaSinkAtLeastOnce as i32
+            }
+            None => KafkaSinkCommitMode::KafkaSinkAtLeastOnce as i32,
+            Some(other) => return plan_err!("invalid sink.commit.mode '{other}'"),
+        };
+
+        let key_field = match options.pull_opt_str(opt::KAFKA_SINK_KEY_FIELD)? {
+            Some(s) => Some(s),
+            None => options.pull_opt_str(opt::KAFKA_KEY_FIELD_LEGACY)?,
+        };
+        let timestamp_field = match options.pull_opt_str(opt::KAFKA_SINK_TIMESTAMP_FIELD)? {
+            Some(s) => Some(s),
+            None => options.pull_opt_str(opt::KAFKA_TIMESTAMP_FIELD_LEGACY)?,
+        };
+
+        let _ = options.pull_opt_str(opt::TYPE)?;
+        let _ = options.pull_opt_str(opt::CONNECTOR)?;
+
+        let mut client_configs = options.drain_remaining_string_values()?;
+        client_configs.remove(opt::CHECKPOINT_INTERVAL_MS);
+        client_configs.remove(opt::PIPELINE_PARALLELISM);
+        client_configs.remove(opt::KEY_BY_PARALLELISM);
+        client_configs.remove(opt::FORMAT);
+
+        Ok(ConnectorConfig::KafkaSink(KafkaSinkConfig {
+            topic,
+            bootstrap_servers,
+            commit_mode,
+            key_field,
+            timestamp_field,
+            auth: Some(KafkaAuthConfig {
+                auth: Some(kafka_auth_config::Auth::None(KafkaAuthNone {})),
+            }),
+            client_configs,
+            format: Some(proto_format),
+            value_subject,
+        }))
+    }
+}
diff --git a/src/sql/connector/sink/lancedb.rs b/src/sql/connector/sink/lancedb.rs
new file mode 100644
index 00000000..aee79735
--- /dev/null
+++ b/src/sql/connector/sink/lancedb.rs
@@ -0,0 +1,61 @@
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+use datafusion::common::Result;
+use protocol::function_stream_graph::{LanceDbSinkConfig, SinkFormatProto};
+
+use crate::sql::common::Format;
+use crate::sql::common::connector_options::ConnectorOptions;
+use crate::sql::common::with_option_keys as opt;
+use crate::sql::connector::config::ConnectorConfig;
+use crate::sql::connector::provider::SinkProvider;
+use crate::sql::connector::sink::runtime_config::SinkRuntimeProperties;
+use crate::sql::connector::sink::utils::SinkUtils;
+
+pub struct LanceDbSinkConnector;
+
+impl SinkProvider for LanceDbSinkConnector {
+    fn name(&self) -> &'static str {
+        "lancedb"
+    }
+
+    fn build_sink_config(
+        &self,
+        options: &mut ConnectorOptions,
+        _format: &Option<Format>,
+        runtime_props: &SinkRuntimeProperties,
+    ) -> Result<ConnectorConfig> {
+        let path = SinkUtils::require_path(options)?;
+
+        let s3_bucket = options.pull_opt_str(opt::S3_BUCKET)?;
+        let s3_region = options.pull_opt_str(opt::S3_REGION)?;
+        let s3_endpoint = options.pull_opt_str(opt::S3_ENDPOINT)?;
+        let s3_access_key_id = options.pull_opt_str(opt::S3_ACCESS_KEY_ID)?;
+        let s3_secret_access_key = options.pull_opt_str(opt::S3_SECRET_ACCESS_KEY)?;
+        let s3_session_token = options.pull_opt_str(opt::S3_SESSION_TOKEN)?;
+
+        let extra_properties = options.drain_remaining_string_values()?;
+
+        Ok(ConnectorConfig::LanceDbSink(LanceDbSinkConfig {
+            path,
+            format: SinkFormatProto::SinkFormatLance as i32,
+            s3_bucket,
+            s3_region,
+            s3_endpoint,
+            s3_access_key_id,
+            s3_secret_access_key,
+            s3_session_token,
+            extra_properties,
+            runtime_properties: runtime_props.clone(),
+        }))
+    }
+}
diff --git a/src/sql/connector/sink/mod.rs b/src/sql/connector/sink/mod.rs
new file mode 100644
index 00000000..b7d645ca
--- /dev/null
+++ b/src/sql/connector/sink/mod.rs
@@ -0,0 +1,20 @@
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+pub mod delta;
+pub mod filesystem;
+pub mod iceberg;
+pub mod kafka;
+pub mod lancedb;
+pub mod runtime_config;
+pub mod s3;
+pub mod utils;
diff --git a/src/sql/connector/sink/runtime_config.rs b/src/sql/connector/sink/runtime_config.rs
new file mode 100644
index 00000000..e0ffaeee
--- /dev/null
+++ b/src/sql/connector/sink/runtime_config.rs
@@ -0,0 +1,137 @@
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+use std::collections::HashMap;
+
+use datafusion::common::{DataFusionError, Result, plan_err};
+
+use crate::config::global_config::{
+    DEFAULT_OPERATOR_STATE_STORE_MEMORY_BYTES, DEFAULT_SINK_BUFFER_MEMORY_BYTES,
+};
+use crate::config::streaming_job::DEFAULT_CHECKPOINT_INTERVAL_MS;
+use crate::sql::common::connector_options::ConnectorOptions;
+use crate::sql::common::with_option_keys as opt;
+
+#[derive(Debug, Clone, PartialEq, Eq, Default)]
+pub struct SinkRuntimeConfig {
+    pub pipeline_parallelism: Option<u32>,
+    pub key_by_parallelism: Option<u32>,
+    pub checkpoint_interval_ms: u64,
+    pub operator_memory_bytes: u64,
+    pub sink_memory_bytes: u64,
+}
+
+pub type SinkRuntimeProperties = HashMap<String, String>;
+
+impl SinkRuntimeConfig {
+    pub fn extract_from_options(options: &mut ConnectorOptions) -> Result<Self> {
+        let pipeline_parallelism = options
+            .pull_opt_u64(opt::PIPELINE_PARALLELISM)?
+            .map(|v| v as u32);
+        let key_by_parallelism = options
+            .pull_opt_u64(opt::KEY_BY_PARALLELISM)?
+            .map(|v| v as u32);
+        let checkpoint_interval_ms = options
+            .pull_opt_u64(opt::CHECKPOINT_INTERVAL_MS)?
+            .unwrap_or(DEFAULT_CHECKPOINT_INTERVAL_MS);
+        let operator_memory_bytes = options
+            .pull_opt_u64(opt::OPERATOR_MEMORY_BYTES)?
+            .unwrap_or(DEFAULT_OPERATOR_STATE_STORE_MEMORY_BYTES);
+        let sink_memory_bytes = options
+            .pull_opt_u64(opt::SINK_MEMORY_BYTES)?
+            .unwrap_or(DEFAULT_SINK_BUFFER_MEMORY_BYTES);
+        Ok(Self {
+            pipeline_parallelism,
+            key_by_parallelism,
+            checkpoint_interval_ms,
+            operator_memory_bytes,
+            sink_memory_bytes,
+        })
+    }
+
+    pub fn from_options_map(opts: &HashMap<String, String>) -> Result<Self> {
+        let pipeline_parallelism = parse_opt_u32(opts, opt::PIPELINE_PARALLELISM)?;
+        let key_by_parallelism = parse_opt_u32(opts, opt::KEY_BY_PARALLELISM)?;
+        let checkpoint_interval_ms = parse_opt_u64(opts, opt::CHECKPOINT_INTERVAL_MS)?
+            .unwrap_or(DEFAULT_CHECKPOINT_INTERVAL_MS);
+        let operator_memory_bytes = parse_opt_u64(opts, opt::OPERATOR_MEMORY_BYTES)?
+            .unwrap_or(DEFAULT_OPERATOR_STATE_STORE_MEMORY_BYTES);
+        let sink_memory_bytes = parse_opt_u64(opts, opt::SINK_MEMORY_BYTES)?
+            .unwrap_or(DEFAULT_SINK_BUFFER_MEMORY_BYTES);
+        Ok(Self {
+            pipeline_parallelism,
+            key_by_parallelism,
+            checkpoint_interval_ms,
+            operator_memory_bytes,
+            sink_memory_bytes,
+        })
+    }
+
+    pub fn to_runtime_properties(&self) -> HashMap<String, String> {
+        let mut out = HashMap::new();
+        if let Some(v) = self.pipeline_parallelism {
+            out.insert(opt::PIPELINE_PARALLELISM.to_string(), v.to_string());
+        }
+        if let Some(v) = self.key_by_parallelism {
+            out.insert(opt::KEY_BY_PARALLELISM.to_string(), v.to_string());
+        }
+        out.insert(
+            opt::CHECKPOINT_INTERVAL_MS.to_string(),
+            self.checkpoint_interval_ms.to_string(),
+        );
+        out.insert(
+            opt::OPERATOR_MEMORY_BYTES.to_string(),
+            self.operator_memory_bytes.to_string(),
+        );
+        out.insert(
+            opt::SINK_MEMORY_BYTES.to_string(),
+            self.sink_memory_bytes.to_string(),
+        );
+        out
+    }
+}
+
+fn parse_opt_u32(opts: &HashMap<String, String>, key: &str) -> Result<Option<u32>> {
+    let Some(raw) = opts.get(key) else {
+        return Ok(None);
+    };
+    let normalized = normalize_numeric_option(raw);
+    let parsed = normalized.parse::<u32>().map_err(|_| {
+        DataFusionError::Plan(format!(
+            "WITH option '{key}' expects unsigned integer, got '{raw}'"
+        ))
+    })?;
+    if parsed == 0 {
+        return plan_err!("WITH option '{key}' must be > 0");
+    }
+    Ok(Some(parsed))
+}
+
+fn parse_opt_u64(opts: &HashMap<String, String>, key: &str) -> Result<Option<u64>> {
+    let Some(raw) = opts.get(key) else {
+        return Ok(None);
+    };
+    let normalized = normalize_numeric_option(raw);
+    let parsed = normalized.parse::<u64>().map_err(|_| {
+        DataFusionError::Plan(format!(
+            "WITH option '{key}' expects unsigned integer, got '{raw}'"
+        ))
+    })?;
+    if parsed == 0 {
+        return plan_err!("WITH option '{key}' must be > 0");
+    }
+    Ok(Some(parsed))
+}
+
+fn normalize_numeric_option(raw: &str) -> &str {
+    raw.trim().trim_matches('\'').trim_matches('"').trim()
+}
diff --git a/src/sql/connector/sink/s3.rs b/src/sql/connector/sink/s3.rs
new file mode 100644
index 00000000..5d04ce46
--- /dev/null
+++ b/src/sql/connector/sink/s3.rs
@@ -0,0 +1,75 @@
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+use datafusion::common::Result;
+use protocol::function_stream_graph::{S3SinkConfig, SinkFormatProto};
+
+use crate::sql::common::Format;
+use crate::sql::common::connector_options::ConnectorOptions;
+use crate::sql::common::constants::connector_type;
+use crate::sql::common::with_option_keys as opt;
+use crate::sql::connector::config::ConnectorConfig;
+use crate::sql::connector::provider::SinkProvider;
+use crate::sql::connector::sink::runtime_config::SinkRuntimeProperties;
+use crate::sql::connector::sink::utils::SinkUtils;
+
+pub struct S3SinkConnector;
+
+impl SinkProvider for S3SinkConnector {
+    fn name(&self) -> &'static str {
+        connector_type::S3
+    }
+
+    fn build_sink_config(
+        &self,
+        options: &mut ConnectorOptions,
+        format: &Option<Format>,
+        runtime_props: &SinkRuntimeProperties,
+    ) -> Result<ConnectorConfig> {
+        let path = SinkUtils::require_path(options)?;
+
+        let format_proto = SinkUtils::resolve_sink_format(
+            format,
+            self.name(),
+            &[
+                SinkFormatProto::SinkFormatCsv,
+                SinkFormatProto::SinkFormatParquet,
+            ],
+        )?;
+
+        let bucket = SinkUtils::require_str(options, opt::S3_BUCKET, self.name())?;
+        let region = options
+            .pull_opt_str(opt::S3_REGION)?
+            .unwrap_or_else(|| "us-east-1".to_string());
+        let endpoint = options.pull_opt_str(opt::S3_ENDPOINT)?;
+        let access_key_id = options.pull_opt_str(opt::S3_ACCESS_KEY_ID)?;
+        let secret_access_key = options.pull_opt_str(opt::S3_SECRET_ACCESS_KEY)?;
+        let session_token = options.pull_opt_str(opt::S3_SESSION_TOKEN)?;
+
+        let parquet_compression = SinkUtils::extract_parquet_compression(options)?;
+        let extra_properties = options.drain_remaining_string_values()?;
+
+        Ok(ConnectorConfig::S3Sink(S3SinkConfig {
+            path,
+            format: format_proto,
+            bucket,
+            region,
+            endpoint,
+            access_key_id,
+            secret_access_key,
+            session_token,
+            parquet_compression,
+            extra_properties,
+            runtime_properties: runtime_props.clone(),
+        }))
+    }
+}
diff --git a/src/sql/connector/sink/utils.rs b/src/sql/connector/sink/utils.rs
new file mode 100644
index 00000000..e61cd870
--- /dev/null
+++ b/src/sql/connector/sink/utils.rs
@@ -0,0 +1,91 @@
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+use datafusion::common::{DataFusionError, Result, plan_err};
+use protocol::function_stream_graph::{ParquetCompressionProto, SinkFormatProto};
+
+use crate::sql::common::Format;
+use crate::sql::common::connector_options::ConnectorOptions;
+use crate::sql::common::constants::parquet_compression_value;
+use crate::sql::common::with_option_keys as opt;
+
+pub struct SinkUtils;
+
+impl SinkUtils {
+    pub fn require_path(options: &mut ConnectorOptions) -> Result<String> {
+        if let Some(v) = options.pull_opt_str(opt::PATH)? {
+            return Ok(v);
+        }
+        if let Some(v) = options.pull_opt_str(opt::SINK_PATH)? {
+            return Ok(v);
+        }
+        plan_err!("Missing required WITH option 'path' (or 'sink.path')")
+    }
+
+    pub fn extract_parquet_compression(options: &mut ConnectorOptions) -> Result<Option<i32>> {
+        let Some(v) = options.pull_opt_str(opt::PARQUET_COMPRESSION)? else {
+            return Ok(None);
+        };
+        let parsed = match v.to_ascii_lowercase().as_str() {
+            parquet_compression_value::UNCOMPRESSED => {
+                ParquetCompressionProto::ParquetCompressionUncompressed
+            }
+            parquet_compression_value::SNAPPY => ParquetCompressionProto::ParquetCompressionSnappy,
+            parquet_compression_value::GZIP => ParquetCompressionProto::ParquetCompressionGzip,
+            parquet_compression_value::ZSTD => ParquetCompressionProto::ParquetCompressionZstd,
+            parquet_compression_value::LZ4 => ParquetCompressionProto::ParquetCompressionLz4,
+            parquet_compression_value::LZ4_RAW => ParquetCompressionProto::ParquetCompressionLz4Raw,
+            other => return plan_err!("Unsupported parquet.compression '{other}'"),
+        };
+        Ok(Some(parsed as i32))
+    }
+
+    pub fn require_str(
+        options: &mut ConnectorOptions,
+        key: &str,
+        connector: &str,
+    ) -> Result<String> {
+        options.pull_opt_str(key)?.ok_or_else(|| {
+            DataFusionError::Plan(format!(
+                "Connector '{connector}' requires WITH option '{key}'"
+            ))
+        })
+    }
+
+    pub fn resolve_sink_format(
+        format: &Option<Format>,
+        connector_name: &str,
+        supported_formats: &[SinkFormatProto],
+    ) -> Result<i32> {
+        let proto_format = match format {
+            Some(Format::Csv(_)) => SinkFormatProto::SinkFormatCsv,
+            Some(Format::Json(_)) => SinkFormatProto::SinkFormatJsonl,
+            Some(Format::Avro(_)) => SinkFormatProto::SinkFormatAvro,
+            Some(Format::Parquet(_)) => SinkFormatProto::SinkFormatParquet,
+            Some(Format::Lance(_)) => SinkFormatProto::SinkFormatLance,
+            Some(f) => {
+                return plan_err!("Format '{f:?}' cannot be mapped to a sink format");
+            }
+            None => {
+                return plan_err!("Connector '{connector_name}' requires a format to be specified");
+            }
+        };
+
+        if !supported_formats.contains(&proto_format) {
+            return plan_err!(
+                "Format {proto_format:?} is not supported by connector '{connector_name}'"
+            );
+        }
+
+        Ok(proto_format as i32)
+    }
+}
diff --git a/src/sql/connector/source/kafka.rs b/src/sql/connector/source/kafka.rs
new file mode 100644
index 00000000..0bc220d9
--- /dev/null
+++ b/src/sql/connector/source/kafka.rs
@@ -0,0 +1,185 @@
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+use datafusion::common::{Result, plan_datafusion_err, plan_err};
+use protocol::function_stream_graph::{
+    BadDataPolicy, DecimalEncodingProto, FormatConfig, JsonFormatConfig, KafkaAuthConfig,
+    KafkaAuthNone, KafkaOffsetMode, KafkaReadMode, KafkaSourceConfig, RawBytesFormatConfig,
+    RawStringFormatConfig, TimestampFormatProto, format_config, kafka_auth_config,
+};
+
+use crate::sql::common::connector_options::ConnectorOptions;
+use crate::sql::common::constants::{connector_type, kafka_with_value};
+use crate::sql::common::formats::{
+    BadData, DecimalEncoding as SqlDecimalEncoding, Format as SqlFormat,
+    TimestampFormat as SqlTimestampFormat,
+};
+use crate::sql::common::with_option_keys as opt;
+use crate::sql::connector::config::ConnectorConfig;
+use crate::sql::connector::provider::SourceProvider;
+
+pub struct KafkaSourceConnector;
+
+impl KafkaSourceConnector {
+    fn sql_format_to_proto(fmt: &SqlFormat) -> Result<FormatConfig> {
+        match fmt {
+            SqlFormat::Json(j) => Ok(FormatConfig {
+                format: Some(format_config::Format::Json(JsonFormatConfig {
+                    timestamp_format: match j.timestamp_format {
+                        SqlTimestampFormat::RFC3339 => {
+                            TimestampFormatProto::TimestampRfc3339 as i32
+                        }
+                        SqlTimestampFormat::UnixMillis => {
+                            TimestampFormatProto::TimestampUnixMillis as i32
+                        }
+                    },
+                    decimal_encoding: match j.decimal_encoding {
+                        SqlDecimalEncoding::Number => DecimalEncodingProto::DecimalNumber as i32,
+                        SqlDecimalEncoding::String => DecimalEncodingProto::DecimalString as i32,
+                        SqlDecimalEncoding::Bytes => DecimalEncodingProto::DecimalBytes as i32,
+                    },
+                    include_schema: j.include_schema,
+                    confluent_schema_registry: j.confluent_schema_registry,
+                    schema_id: j.schema_id,
+                    debezium: j.debezium,
+                    unstructured: j.unstructured,
+                })),
+            }),
+            SqlFormat::RawString(_) => Ok(FormatConfig {
+                format: Some(format_config::Format::RawString(RawStringFormatConfig {})),
+            }),
+            SqlFormat::RawBytes(_) => Ok(FormatConfig {
+                format: Some(format_config::Format::RawBytes(RawBytesFormatConfig {})),
+            }),
+            other => plan_err!(
+                "Kafka source connector: format '{}' is not supported",
+                other.name()
+            ),
+        }
+    }
+
+    fn bad_data_to_proto(bad: &BadData) -> i32 {
+        match bad {
+            BadData::Fail {} => BadDataPolicy::BadDataFail as i32,
+            BadData::Drop {} => BadDataPolicy::BadDataDrop as i32,
+        }
+    }
+}
+
+impl SourceProvider for KafkaSourceConnector {
+    fn name(&self) -> &'static str {
+        connector_type::KAFKA
+    }
+
+    fn build_source_config(
+        &self,
+        options: &mut ConnectorOptions,
+        format: &Option<SqlFormat>,
+        bad_data: BadData,
+    ) -> Result<ConnectorConfig> {
+        let bootstrap_servers = match options.pull_opt_str(opt::KAFKA_BOOTSTRAP_SERVERS)? {
+            Some(s) => s,
+            None => options
+                .pull_opt_str(opt::KAFKA_BOOTSTRAP_SERVERS_LEGACY)?
+                .ok_or_else(|| {
+                    plan_datafusion_err!(
+                        "Kafka connector requires 'bootstrap.servers' in the WITH clause"
+                    )
+                })?,
+        };
+
+        let topic = options.pull_opt_str(opt::KAFKA_TOPIC)?.ok_or_else(|| {
+            plan_datafusion_err!("Kafka connector requires 'topic' in the WITH clause")
+        })?;
+
+        let sql_format = format.as_ref().ok_or_else(|| {
+            plan_datafusion_err!(
+                "Kafka source requires 'format' in the WITH clause (e.g. format = 'json')"
+            )
+        })?;
+        let proto_format = Self::sql_format_to_proto(sql_format)?;
+
+        let rate_limit = options
+            .pull_opt_u64(opt::KAFKA_RATE_LIMIT_MESSAGES_PER_SECOND)?
+            .map(|v| v.clamp(1, u32::MAX as u64) as u32)
+            .unwrap_or(0);
+
+        let value_subject = options.pull_opt_str(opt::KAFKA_VALUE_SUBJECT)?;
+
+        let offset_mode = match options
+            .pull_opt_str(opt::KAFKA_SCAN_STARTUP_MODE)?
+            .as_deref()
+        {
+            Some(s) if s == kafka_with_value::SCAN_LATEST => {
+                KafkaOffsetMode::KafkaOffsetLatest as i32
+            }
+            Some(s) if s == kafka_with_value::SCAN_EARLIEST => {
+                KafkaOffsetMode::KafkaOffsetEarliest as i32
+            }
+            Some(s)
+                if s == kafka_with_value::SCAN_GROUP_OFFSETS
+                    || s == kafka_with_value::SCAN_GROUP =>
+            {
+                KafkaOffsetMode::KafkaOffsetGroup as i32
+            }
+            None => KafkaOffsetMode::KafkaOffsetGroup as i32,
+            Some(other) => {
+                return plan_err!(
+                    "invalid scan.startup.mode '{other}'; expected latest, earliest, or group-offsets"
+                );
+            }
+        };
+
+        let read_mode = match options.pull_opt_str(opt::KAFKA_ISOLATION_LEVEL)?.as_deref() {
+            Some(s) if s == kafka_with_value::ISOLATION_READ_COMMITTED => {
+                KafkaReadMode::KafkaReadCommitted as i32
+            }
+            Some(s) if s == kafka_with_value::ISOLATION_READ_UNCOMMITTED => {
+                KafkaReadMode::KafkaReadUncommitted as i32
+            }
+            None => KafkaReadMode::KafkaReadDefault as i32,
+            Some(other) => return plan_err!("invalid isolation.level '{other}'"),
+        };
+
+        let group_id = match options.pull_opt_str(opt::KAFKA_GROUP_ID)? {
+            Some(s) => Some(s),
+            None => options.pull_opt_str(opt::KAFKA_GROUP_ID_LEGACY)?,
+        };
+        let group_id_prefix = options.pull_opt_str(opt::KAFKA_GROUP_ID_PREFIX)?;
+
+        let _ = options.pull_opt_str(opt::TYPE)?;
+        let _ = options.pull_opt_str(opt::CONNECTOR)?;
+
+        let mut client_configs = options.drain_remaining_string_values()?;
+        client_configs.remove(opt::CHECKPOINT_INTERVAL_MS);
+        client_configs.remove(opt::PIPELINE_PARALLELISM);
+        client_configs.remove(opt::KEY_BY_PARALLELISM);
+        client_configs.remove(opt::FORMAT);
+
+        Ok(ConnectorConfig::KafkaSource(KafkaSourceConfig {
+            topic,
+            bootstrap_servers,
+            group_id,
+            group_id_prefix,
+            offset_mode,
+            read_mode,
+            auth: Some(KafkaAuthConfig {
+                auth: Some(kafka_auth_config::Auth::None(KafkaAuthNone {})),
+            }),
+            client_configs,
+            format: Some(proto_format),
+            bad_data_policy: Self::bad_data_to_proto(&bad_data),
+            rate_limit_msgs_per_sec: rate_limit,
+            value_subject,
+        }))
+    }
+}
diff --git a/src/sql/connector/source/mod.rs b/src/sql/connector/source/mod.rs
new file mode 100644
index 00000000..b9574391
--- /dev/null
+++ b/src/sql/connector/source/mod.rs
@@ -0,0 +1,13 @@
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+pub mod kafka;
diff --git a/src/sql/logical_node/lookup.rs b/src/sql/logical_node/lookup.rs
index c060ba82..d2817c85 100644
--- a/src/sql/logical_node/lookup.rs
+++ b/src/sql/logical_node/lookup.rs
@@ -10,7 +10,6 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
-use std::collections::HashMap;
 use std::fmt::Formatter;
 use std::sync::Arc;
 
@@ -22,9 +21,7 @@ use datafusion_proto::physical_plan::to_proto::serialize_physical_expr;
 use prost::Message;
 
 use protocol::function_stream_graph;
-use protocol::function_stream_graph::{
-    ConnectorOp, GenericConnectorConfig, LookupJoinCondition, LookupJoinOperator,
-};
+use protocol::function_stream_graph::{ConnectorOp, LookupJoinCondition, LookupJoinOperator};
 
 use crate::multifield_partial_ord;
 use crate::sql::common::constants::extension_node;
@@ -32,7 +29,7 @@ use crate::sql::common::{FsSchema, FsSchemaRef};
 use crate::sql::logical_node::logical::{LogicalEdge, LogicalEdgeType, LogicalNode, OperatorName};
 use crate::sql::logical_node::{CompiledTopologyNode, StreamingOperatorBlueprint};
 use crate::sql::logical_planner::planner::{NamedNode, Planner};
-use crate::sql::schema::SourceTable;
+use crate::sql::schema::LookupTable;
 use crate::sql::schema::utils::add_timestamp_field_arrow;
 
 pub const DICTIONARY_SOURCE_NODE_NAME: &str = extension_node::REFERENCE_TABLE_SOURCE;
@@ -40,7 +37,7 @@ pub const STREAM_DICTIONARY_JOIN_NODE_NAME: &str = extension_node::STREAM_REFERE
 
 #[derive(Debug, Clone, PartialEq, Eq, Hash)]
 pub struct ReferenceTableSourceNode {
-    pub(crate) source_definition: SourceTable,
+    pub(crate) source_definition: LookupTable,
     pub(crate) resolved_schema: DFSchemaRef,
 }
 
@@ -85,7 +82,7 @@ impl UserDefinedLogicalNodeCore for ReferenceTableSourceNode {
 pub struct StreamReferenceJoinNode {
     pub(crate) upstream_stream_plan: LogicalPlan,
     pub(crate) output_schema: DFSchemaRef,
-    pub(crate) external_dictionary: SourceTable,
+    pub(crate) external_dictionary: LookupTable,
     pub(crate) equijoin_conditions: Vec<(Expr, Column)>,
     pub(crate) post_join_filter: Option<Expr>,
     pub(crate) namespace_alias: Option<TableReference>,
@@ -140,13 +137,6 @@ impl StreamReferenceJoinNode {
         let lookup_fs_schema =
             FsSchema::from_schema_unkeyed(add_timestamp_field_arrow(dictionary_physical_schema))?;
 
-        let properties: HashMap<String, String> = self
-            .external_dictionary
-            .catalog_with_options
-            .iter()
-            .map(|(k, v)| (k.clone(), v.clone()))
-            .collect();
-
         Ok(LookupJoinOperator {
             input_schema: Some(internal_input_schema.into()),
             lookup_schema: Some(lookup_fs_schema.clone().into()),
@@ -155,11 +145,7 @@ impl StreamReferenceJoinNode {
                 fs_schema: Some(lookup_fs_schema.into()),
                 name: self.external_dictionary.table_identifier.clone(),
                 description: self.external_dictionary.description.clone(),
-                config: Some(
-                    protocol::function_stream_graph::connector_op::Config::Generic(
-                        GenericConnectorConfig { properties },
-                    ),
-                ),
+                config: Some(self.external_dictionary.connector_config.to_proto_config()),
             }),
             key_exprs: self.compile_join_conditions(planner)?,
             join_type: self.map_api_join_type()?,
diff --git a/src/sql/logical_node/sink.rs b/src/sql/logical_node/sink.rs
index 2edf8f27..d767afe3 100644
--- a/src/sql/logical_node/sink.rs
+++ b/src/sql/logical_node/sink.rs
@@ -23,7 +23,8 @@ use crate::sql::common::{FsSchema, FsSchemaRef, UPDATING_META_FIELD};
 use crate::sql::logical_node::logical::{LogicalEdge, LogicalEdgeType, LogicalNode, OperatorName};
 use crate::sql::logical_node::{CompiledTopologyNode, StreamingOperatorBlueprint};
 use crate::sql::logical_planner::planner::{NamedNode, Planner};
-use crate::sql::schema::Table;
+use crate::sql::schema::CatalogEntity;
+use crate::sql::schema::catalog::ExternalTable;
 
 use super::debezium::PackDebeziumEnvelopeNode;
 use super::remote_table::RemoteTableBoundaryNode;
@@ -42,7 +43,7 @@ pub(crate) const STREAM_EGRESS_NODE_NAME: &str = extension_node::STREAM_EGRESS;
 #[derive(Debug, Clone, PartialEq, Eq, Hash)]
 pub(crate) struct StreamEgressNode {
     pub(crate) target_identifier: TableReference,
-    pub(crate) destination_table: Table,
+    pub(crate) destination_table: CatalogEntity,
     pub(crate) egress_schema: DFSchemaRef,
     upstream_plans: Arc<Vec<LogicalPlan>>,
 }
@@ -52,7 +53,7 @@ multifield_partial_ord!(StreamEgressNode, target_identifier, upstream_plans);
 impl StreamEgressNode {
     pub fn try_new(
         target_identifier: TableReference,
-        destination_table: Table,
+        destination_table: CatalogEntity,
         initial_schema: DFSchemaRef,
         upstream_plan: LogicalPlan,
     ) -> Result<Self> {
@@ -72,43 +73,61 @@ impl StreamEgressNode {
     fn apply_cdc_transformations(
         plan: LogicalPlan,
         schema: DFSchemaRef,
-        destination: &Table,
+        destination: &CatalogEntity,
     ) -> Result<(LogicalPlan, DFSchemaRef)> {
         let is_upstream_updating = plan
             .schema()
             .has_column_with_unqualified_name(UPDATING_META_FIELD);
 
         match destination {
-            Table::ConnectorTable(connector) => {
-                let is_sink_updating = connector.is_updating();
-
-                match (is_upstream_updating, is_sink_updating) {
-                    (_, true) => {
-                        let debezium_encoder = PackDebeziumEnvelopeNode::try_new(plan)?;
-                        let wrapped_plan = LogicalPlan::Extension(Extension {
-                            node: Arc::new(debezium_encoder),
-                        });
-                        let new_schema = wrapped_plan.schema().clone();
-
-                        Ok((wrapped_plan, new_schema))
+            CatalogEntity::ExternalConnector(b) => match b.as_ref() {
+                ExternalTable::Sink(sink) => {
+                    let is_sink_updating = sink.is_updating();
+
+                    match (is_upstream_updating, is_sink_updating) {
+                        (_, true) => {
+                            let debezium_encoder = PackDebeziumEnvelopeNode::try_new(plan)?;
+                            let wrapped_plan = LogicalPlan::Extension(Extension {
+                                node: Arc::new(debezium_encoder),
+                            });
+                            let new_schema = wrapped_plan.schema().clone();
+
+                            Ok((wrapped_plan, new_schema))
+                        }
+                        (true, false) => {
+                            plan_err!(
+                                "Topology Mismatch: The upstream is producing an updating stream (CDC), \
+                                 but the target sink '{}' is not configured to accept updates. \
+                                 Hint: set `format = 'debezium_json'` in the WITH clause.",
+                                sink.name()
+                            )
+                        }
+                        (false, false) => Ok((plan, schema)),
                     }
-                    (true, false) => {
-                        plan_err!(
-                            "Topology Mismatch: The upstream is producing an updating stream (CDC), \
-                             but the target sink '{}' is not configured to accept updates. \
-                             Hint: set `format = 'debezium_json'` in the WITH clause.",
-                            connector.name()
-                        )
+                }
+                ExternalTable::Source(source) => {
+                    let is_sink_updating = source.is_updating();
+                    match (is_upstream_updating, is_sink_updating) {
+                        (_, true) => {
+                            let debezium_encoder = PackDebeziumEnvelopeNode::try_new(plan)?;
+                            let wrapped_plan = LogicalPlan::Extension(Extension {
+                                node: Arc::new(debezium_encoder),
+                            });
+                            let new_schema = wrapped_plan.schema().clone();
+                            Ok((wrapped_plan, new_schema))
+                        }
+                        (true, false) => plan_err!(
+                            "Topology Mismatch: upstream produces CDC but target '{}' is a non-updating source table",
+                            source.name()
+                        ),
+                        (false, false) => Ok((plan, schema)),
                     }
-                    (false, false) => Ok((plan, schema)),
                 }
-            }
-            Table::LookupTable(..) => {
-                plan_err!(
+                ExternalTable::Lookup(_) => plan_err!(
                     "Topology Violation: A Lookup Table cannot be used as a streaming data sink."
-                )
-            }
-            Table::TableFromQuery { .. } => Ok((plan, schema)),
+                ),
+            },
+            CatalogEntity::ComputedTable { .. } => Ok((plan, schema)),
         }
     }
 
diff --git a/src/sql/mod.rs b/src/sql/mod.rs
index 529c7a2d..f4a0eef6 100644
--- a/src/sql/mod.rs
+++ b/src/sql/mod.rs
@@ -14,6 +14,7 @@ pub mod api;
 pub mod common;
 
 pub mod analysis;
+pub mod connector;
 pub mod functions;
 pub mod logical_node;
 pub mod logical_planner;
diff --git a/src/sql/schema/catalog.rs b/src/sql/schema/catalog.rs
new file mode 100644
index 00000000..479df682
--- /dev/null
+++ b/src/sql/schema/catalog.rs
@@ -0,0 +1,609 @@
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+//! External connector catalog: [`ExternalTable`] as [`SourceTable`] | [`SinkTable`] | [`LookupTable`].
+
+use std::collections::BTreeMap;
+use std::sync::Arc;
+use std::time::Duration;
+
+use datafusion::arrow::datatypes::{DataType, Field, FieldRef, Schema};
+use datafusion::common::{Column, Result, plan_err};
+use datafusion::error::DataFusionError;
+use datafusion::logical_expr::Expr;
+use protocol::function_stream_graph::ConnectorOp;
+
+use super::column_descriptor::ColumnDescriptor;
+use super::data_encoding_format::DataEncodingFormat;
+use super::table::SqlSource;
+use super::temporal_pipeline_config::TemporalPipelineConfig;
+use crate::multifield_partial_ord;
+use crate::sql::common::constants::sql_field;
+use crate::sql::common::{Format, FsSchema};
+use crate::sql::connector::config::ConnectorConfig;
+use crate::sql::types::ProcessingMode;
+
+#[derive(Debug, Clone)]
+pub struct EngineDescriptor {
+    pub engine_type: String,
+    pub raw_payload: String,
+}
+
+#[derive(Debug, Clone, Copy, PartialEq, Eq, Hash)]
+pub enum SyncMode {
+    AppendOnly,
+    Incremental,
+}
+
+#[derive(Debug, Clone)]
+pub struct TableExecutionUnit {
+    pub label: String,
+    pub engine_meta: EngineDescriptor,
+    pub sync_mode: SyncMode,
+    pub temporal_offset: TemporalPipelineConfig,
+}
+
+/// The only legal shape an external-connector catalog row can take.
+#[derive(Debug, Clone, PartialEq, Eq, Hash)]
+pub enum ExternalTable {
+    Source(SourceTable),
+    Sink(SinkTable),
+    Lookup(LookupTable),
+}
+
+impl ExternalTable {
+    #[inline]
+    pub fn name(&self) -> &str {
+        match self {
+            ExternalTable::Source(t) => t.table_identifier.as_str(),
+            ExternalTable::Sink(t) => t.table_identifier.as_str(),
+            ExternalTable::Lookup(t) => t.table_identifier.as_str(),
+        }
+    }
+
+    #[inline]
+    pub fn adapter_type(&self) -> &str {
+        match self {
+            ExternalTable::Source(t) => t.adapter_type.as_str(),
+            ExternalTable::Sink(t) => t.adapter_type.as_str(),
+            ExternalTable::Lookup(t) => t.adapter_type.as_str(),
+        }
+    }
+
+    #[inline]
+    pub fn description(&self) -> &str {
+        match self {
+            ExternalTable::Source(t) => t.description.as_str(),
+            ExternalTable::Sink(t) => t.description.as_str(),
+            ExternalTable::Lookup(t) => t.description.as_str(),
+        }
+    }
+
+    #[inline]
+    pub fn schema_specs(&self) -> &[ColumnDescriptor] {
+        match self {
+            ExternalTable::Source(t) => &t.schema_specs,
+            ExternalTable::Sink(t) => &t.schema_specs,
+            ExternalTable::Lookup(t) => &t.schema_specs,
+        }
+    }
+
+    #[inline]
+    pub fn connector_config(&self) -> &ConnectorConfig {
+        match self {
+            ExternalTable::Source(t) => &t.connector_config,
+            ExternalTable::Sink(t) => &t.connector_config,
+            ExternalTable::Lookup(t) => &t.connector_config,
+        }
+    }
+
+    #[inline]
+    pub fn key_constraints(&self) -> &[String] {
+        match self {
+            ExternalTable::Source(t) => &t.key_constraints,
+            ExternalTable::Sink(t) => &t.key_constraints,
+            ExternalTable::Lookup(t) => &t.key_constraints,
+        }
+    }
+
+    #[inline]
+    pub fn connection_format(&self) -> Option<&Format> {
+        match self {
+            ExternalTable::Source(t) => t.connection_format.as_ref(),
+            ExternalTable::Sink(t) => t.connection_format.as_ref(),
+            ExternalTable::Lookup(t) => t.connection_format.as_ref(),
+        }
+    }
+
+    #[inline]
+    pub fn catalog_with_options(&self) -> &BTreeMap<String, String> {
+        match self {
+            ExternalTable::Source(t) => &t.catalog_with_options,
+            ExternalTable::Sink(t) => &t.catalog_with_options,
+            ExternalTable::Lookup(t) => &t.catalog_with_options,
+        }
+    }
+
+    pub fn produce_physical_schema(&self) -> Schema {
+        Schema::new(
+            self.schema_specs()
+                .iter()
+                .filter(|c| !c.is_computed())
+                .map(|c| c.arrow_field().clone())
+                .collect::<Vec<_>>(),
+        )
+    }
+
+    pub fn connector_op(&self) -> ConnectorOp {
+        let physical = self.produce_physical_schema();
+        let fields: Vec<Field> = physical
+            .fields()
+            .iter()
+            .map(|f| f.as_ref().clone())
+            .collect();
+        let fs_schema = FsSchema::from_fields(fields);
+
+        ConnectorOp {
+            connector: self.adapter_type().to_string(),
+            fs_schema: Some(fs_schema.into()),
+            name: self.name().to_string(),
+            description: self.description().to_string(),
+            config: Some(self.connector_config().to_proto_config()),
+        }
+    }
+
+    #[inline]
+    pub fn is_updating(&self) -> bool {
+        match self {
+            ExternalTable::Source(t) => t.is_updating(),
+            ExternalTable::Sink(t) => t
+                .connection_format
+                .as_ref()
+                .is_some_and(|f| f.is_updating()),
+            ExternalTable::Lookup(_) => false,
+        }
+    }
+
+    /// Variant-agnostic view of "persisted Arrow fields post-planning".
+    /// Only Source / Lookup track inferred schema — Sinks derive theirs from the upstream plan.
+    pub fn effective_fields(&self) -> Vec<FieldRef> {
+        match self {
+            ExternalTable::Source(t) => t.effective_fields(),
+            ExternalTable::Sink(t) => t.effective_fields(),
+            ExternalTable::Lookup(t) => t.effective_fields(),
+        }
+    }
+
+    #[inline]
+    pub fn as_source(&self) -> Option<&SourceTable> {
+        match self {
+            ExternalTable::Source(t) => Some(t),
+            _ => None,
+        }
+    }
+
+    #[inline]
+    pub fn as_sink(&self) -> Option<&SinkTable> {
+        match self {
+            ExternalTable::Sink(t) => Some(t),
+            _ => None,
+        }
+    }
+
+    #[inline]
+    pub fn as_lookup(&self) -> Option<&LookupTable> {
+        match self {
+            ExternalTable::Lookup(t) => Some(t),
+            _ => None,
+        }
+    }
+}
+
+/// Ingress external connector (`CREATE TABLE ... WITH (type='source', ...)`).
+#[derive(Debug, Clone, PartialEq, Eq, Hash)]
+pub struct SourceTable {
+    pub table_identifier: String,
+    pub adapter_type: String,
+    pub schema_specs: Vec<ColumnDescriptor>,
+    pub connector_config: ConnectorConfig,
+    pub temporal_config: TemporalPipelineConfig,
+    pub key_constraints: Vec<String>,
+    pub payload_format: Option<DataEncodingFormat>,
+    pub connection_format: Option<Format>,
+    pub description: String,
+    pub catalog_with_options: BTreeMap<String, String>,
+
+    // Planner / catalog; not in SQL text.
+    pub registry_id: Option<i64>,
+    pub inferred_fields: Option<Vec<FieldRef>>,
+}
+
+multifield_partial_ord!(
+    SourceTable,
+    registry_id,
+    adapter_type,
+    table_identifier,
+    description,
+    key_constraints,
+    connection_format,
+    catalog_with_options
+);
+
+impl SourceTable {
+    #[inline]
+    pub fn name(&self) -> &str {
+        self.table_identifier.as_str()
+    }
+
+    #[inline]
+    pub fn connector(&self) -> &str {
+        self.adapter_type.as_str()
+    }
+
+    pub fn event_time_field(&self) -> Option<&str> {
+        self.temporal_config.event_column.as_deref()
+    }
+
+    pub fn watermark_field(&self) -> Option<&str> {
+        self.temporal_config.watermark_strategy_column.as_deref()
+    }
+
+    /// Watermark column safe to persist to the stream catalog. Omits the
+    /// generated `__watermark` column — that is only resolvable at compile
+    /// time, the catalog round-trip cannot reconstruct it.
+    pub fn stream_catalog_watermark_field(&self) -> Option<String> {
+        self.temporal_config
+            .watermark_strategy_column
+            .as_deref()
+            .filter(|w| *w != sql_field::COMPUTED_WATERMARK)
+            .map(str::to_string)
+    }
+
+    #[inline]
+    pub fn catalog_with_options(&self) -> &BTreeMap<String, String> {
+        &self.catalog_with_options
+    }
+
+    pub fn idle_time(&self) -> Option<Duration> {
+        self.temporal_config.liveness_timeout
+    }
+
+    pub fn produce_physical_schema(&self) -> Schema {
+        Schema::new(
+            self.schema_specs
+                .iter()
+                .filter(|c| !c.is_computed())
+                .map(|c| c.arrow_field().clone())
+                .collect::<Vec<_>>(),
+        )
+    }
+
+    #[inline]
+    pub fn physical_schema(&self) -> Schema {
+        self.produce_physical_schema()
+    }
+
+    pub fn effective_fields(&self) -> Vec<FieldRef> {
+        self.inferred_fields.clone().unwrap_or_else(|| {
+            self.schema_specs
+                .iter()
+                .map(|c| Arc::new(c.arrow_field().clone()))
+                .collect()
+        })
+    }
+
+    pub fn convert_to_execution_unit(&self) -> Result<TableExecutionUnit> {
+        if self.is_cdc_enabled() && self.schema_specs.iter().any(|c| c.is_computed()) {
+            return plan_err!("CDC cannot be mixed with computed columns natively");
+        }
+
+        let mode = if self.is_cdc_enabled() {
+            SyncMode::Incremental
+        } else {
+            SyncMode::AppendOnly
+        };
+
+        Ok(TableExecutionUnit {
+            label: self.table_identifier.clone(),
+            engine_meta: EngineDescriptor {
+                engine_type: self.adapter_type.clone(),
+                raw_payload: String::new(),
+            },
+            sync_mode: mode,
+            temporal_offset: self.temporal_config.clone(),
+        })
+    }
+
+    #[inline]
+    pub fn to_execution_unit(&self) -> Result<TableExecutionUnit> {
+        self.convert_to_execution_unit()
+    }
+
+    fn is_cdc_enabled(&self) -> bool {
+        self.payload_format
+            .as_ref()
+            .is_some_and(|f| f.supports_delta_updates())
+    }
+
+    pub fn has_virtual_fields(&self) -> bool {
+        self.schema_specs.iter().any(|c| c.is_computed())
+    }
+
+    pub fn is_updating(&self) -> bool {
+        self.connection_format
+            .as_ref()
+            .is_some_and(|f| f.is_updating())
+            || self.payload_format == Some(DataEncodingFormat::DebeziumJson)
+    }
+
+    pub fn connector_op(&self) -> ConnectorOp {
+        let physical = self.produce_physical_schema();
+        let fields: Vec<Field> = physical
+            .fields()
+            .iter()
+            .map(|f| f.as_ref().clone())
+            .collect();
+        let fs_schema = FsSchema::from_fields(fields);
+
+        ConnectorOp {
+            connector: self.adapter_type.clone(),
+            fs_schema: Some(fs_schema.into()),
+            name: self.table_identifier.clone(),
+            description: self.description.clone(),
+            config: Some(self.connector_config.to_proto_config()),
+        }
+    }
+
+    pub fn processing_mode(&self) -> ProcessingMode {
+        if self.is_updating() {
+            ProcessingMode::Update
+        } else {
+            ProcessingMode::Append
+        }
+    }
+
+    pub fn timestamp_override(&self) -> Result<Option<Expr>> {
+        if let Some(field_name) = self.temporal_config.event_column.clone() {
+            if self.is_updating() {
+                return plan_err!("can't use event_time_field with update mode");
+            }
+            let _field = self.get_time_column(&field_name)?;
+            Ok(Some(Expr::Column(Column::from_name(field_name.as_str()))))
+        } else {
+            Ok(None)
+        }
+    }
+
+    fn get_time_column(&self, field_name: &str) -> Result<&ColumnDescriptor> {
+        self.schema_specs
+            .iter()
+            .find(|c| {
+                c.arrow_field().name() == field_name
+                    && matches!(c.arrow_field().data_type(), DataType::Timestamp(..))
+            })
+            .ok_or_else(|| {
+                DataFusionError::Plan(format!("field {field_name} not found or not a timestamp"))
+            })
+    }
+
+    pub fn watermark_column(&self) -> Result<Option<Expr>> {
+        if let Some(field_name) = self.temporal_config.watermark_strategy_column.clone() {
+            let _field = self.get_time_column(&field_name)?;
+            Ok(Some(Expr::Column(Column::from_name(field_name.as_str()))))
+        } else {
+            Ok(None)
+        }
+    }
+
+    pub fn as_sql_source(&self) -> Result<SourceOperator> {
+        if self.is_updating() && self.has_virtual_fields() {
+            return plan_err!("can't read from a source with virtual fields and update mode.");
+        }
+
+        let timestamp_override = self.timestamp_override()?;
+        let watermark_column = self.watermark_column()?;
+
+        let source = SqlSource {
+            id: self.registry_id,
+            struct_def: self
+                .schema_specs
+                .iter()
+                .filter(|c| !c.is_computed())
+                .map(|c| Arc::new(c.arrow_field().clone()))
+                .collect(),
+            config: self.connector_op(),
+            processing_mode: self.processing_mode(),
+            idle_time: self.temporal_config.liveness_timeout,
+        };
+
+        Ok(SourceOperator {
+            name: self.table_identifier.clone(),
+            source,
+            timestamp_override,
+            watermark_column,
+        })
+    }
+}
+
+/// Egress external connector, or the sink of `CREATE STREAMING TABLE ... AS SELECT`.
+#[derive(Debug, Clone, PartialEq, Eq, Hash)]
+pub struct SinkTable {
+    pub table_identifier: String,
+    pub adapter_type: String,
+    pub schema_specs: Vec<ColumnDescriptor>,
+    pub connector_config: ConnectorConfig,
+    pub partition_exprs: Arc<Option<Vec<Expr>>>,
+    pub key_constraints: Vec<String>,
+    pub connection_format: Option<Format>,
+    pub description: String,
+    pub catalog_with_options: BTreeMap<String, String>,
+}
+
+multifield_partial_ord!(
+    SinkTable,
+    adapter_type,
+    table_identifier,
+    description,
+    key_constraints,
+    connection_format,
+    catalog_with_options
+);
+
+impl SinkTable {
+    #[inline]
+    pub fn name(&self) -> &str {
+        self.table_identifier.as_str()
+    }
+
+    #[inline]
+    pub fn connector(&self) -> &str {
+        self.adapter_type.as_str()
+    }
+
+    #[inline]
+    pub fn catalog_with_options(&self) -> &BTreeMap<String, String> {
+        &self.catalog_with_options
+    }
+
+    pub fn produce_physical_schema(&self) -> Schema {
+        Schema::new(
+            self.schema_specs
+                .iter()
+                .filter(|c| !c.is_computed())
+                .map(|c| c.arrow_field().clone())
+                .collect::<Vec<_>>(),
+        )
+    }
+
+    pub fn effective_fields(&self) -> Vec<FieldRef> {
+        self.schema_specs
+            .iter()
+            .map(|c| Arc::new(c.arrow_field().clone()))
+            .collect()
+    }
+
+    pub fn is_updating(&self) -> bool {
+        self.connection_format
+            .as_ref()
+            .is_some_and(|f| f.is_updating())
+    }
+
+    pub fn connector_op(&self) -> ConnectorOp {
+        let physical = self.produce_physical_schema();
+        let fields: Vec<Field> = physical
+            .fields()
+            .iter()
+            .map(|f| f.as_ref().clone())
+            .collect();
+        let fs_schema = FsSchema::from_fields(fields);
+
+        ConnectorOp {
+            connector: self.adapter_type.clone(),
+            fs_schema: Some(fs_schema.into()),
+            name: self.table_identifier.clone(),
+            description: self.description.clone(),
+            config: Some(self.connector_config.to_proto_config()),
+        }
+    }
+}
+
+/// Lookup-join only; not a scan source.
+#[derive(Debug, Clone, PartialEq, Eq, Hash)]
+pub struct LookupTable {
+    pub table_identifier: String,
+    pub adapter_type: String,
+    pub schema_specs: Vec<ColumnDescriptor>,
+    pub connector_config: ConnectorConfig,
+    pub key_constraints: Vec<String>,
+    pub lookup_cache_max_bytes: Option<u64>,
+    pub lookup_cache_ttl: Option<Duration>,
+    pub connection_format: Option<Format>,
+    pub description: String,
+    pub catalog_with_options: BTreeMap<String, String>,
+
+    pub registry_id: Option<i64>,
+    pub inferred_fields: Option<Vec<FieldRef>>,
+}
+
+multifield_partial_ord!(
+    LookupTable,
+    registry_id,
+    adapter_type,
+    table_identifier,
+    description,
+    key_constraints,
+    connection_format,
+    catalog_with_options
+);
+
+impl LookupTable {
+    #[inline]
+    pub fn name(&self) -> &str {
+        self.table_identifier.as_str()
+    }
+
+    #[inline]
+    pub fn connector(&self) -> &str {
+        self.adapter_type.as_str()
+    }
+
+    #[inline]
+    pub fn catalog_with_options(&self) -> &BTreeMap<String, String> {
+        &self.catalog_with_options
+    }
+
+    pub fn produce_physical_schema(&self) -> Schema {
+        Schema::new(
+            self.schema_specs
+                .iter()
+                .filter(|c| !c.is_computed())
+                .map(|c| c.arrow_field().clone())
+                .collect::<Vec<_>>(),
+        )
+    }
+
+    pub fn effective_fields(&self) -> Vec<FieldRef> {
+        self.inferred_fields.clone().unwrap_or_else(|| {
+            self.schema_specs
+                .iter()
+                .map(|c| Arc::new(c.arrow_field().clone()))
+                .collect()
+        })
+    }
+
+    pub fn connector_op(&self) -> ConnectorOp {
+        let physical = self.produce_physical_schema();
+        let fields: Vec<Field> = physical
+            .fields()
+            .iter()
+            .map(|f| f.as_ref().clone())
+            .collect();
+        let fs_schema = FsSchema::from_fields(fields);
+
+        ConnectorOp {
+            connector: self.adapter_type.clone(),
+            fs_schema: Some(fs_schema.into()),
+            name: self.table_identifier.clone(),
+            description: self.description.clone(),
+            config: Some(self.connector_config.to_proto_config()),
+        }
+    }
+}
+
+/// [`SourceTable`] as an ingestion logical node input.
+#[derive(Debug, Clone)]
+pub struct SourceOperator {
+    pub name: String,
+    pub source: SqlSource,
+    pub timestamp_override: Option<Expr>,
+    pub watermark_column: Option<Expr>,
+}
diff --git a/src/sql/schema/catalog_ddl.rs b/src/sql/schema/catalog_ddl.rs
deleted file mode 100644
index 45936912..00000000
--- a/src/sql/schema/catalog_ddl.rs
+++ /dev/null
@@ -1,251 +0,0 @@
-// Licensed under the Apache License, Version 2.0 (the "License");
-// you may not use this file except in compliance with the License.
-// You may obtain a copy of the License at
-//
-//     http://www.apache.org/licenses/LICENSE-2.0
-//
-// Unless required by applicable law or agreed to in writing, software
-// distributed under the License is distributed on an "AS IS" BASIS,
-// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-// See the License for the specific language governing permissions and
-// limitations under the License.
-
-//! Best-effort SQL text for catalog introspection (`SHOW CREATE TABLE`).
-
-use std::collections::BTreeMap;
-
-use datafusion::arrow::datatypes::{DataType, TimeUnit};
-
-use super::schema_provider::StreamTable;
-use super::table::Table as CatalogTable;
-use crate::sql::logical_node::logical::LogicalProgram;
-
-fn data_type_sql(dt: &DataType) -> String {
-    match dt {
-        DataType::Null => "NULL".to_string(),
-        DataType::Boolean => "BOOLEAN".to_string(),
-        DataType::Int8 => "TINYINT".to_string(),
-        DataType::Int16 => "SMALLINT".to_string(),
-        DataType::Int32 => "INT".to_string(),
-        DataType::Int64 => "BIGINT".to_string(),
-        DataType::UInt8 => "TINYINT UNSIGNED".to_string(),
-        DataType::UInt16 => "SMALLINT UNSIGNED".to_string(),
-        DataType::UInt32 => "INT UNSIGNED".to_string(),
-        DataType::UInt64 => "BIGINT UNSIGNED".to_string(),
-        DataType::Float16 => "FLOAT".to_string(),
-        DataType::Float32 => "REAL".to_string(),
-        DataType::Float64 => "DOUBLE".to_string(),
-        DataType::Utf8 | DataType::LargeUtf8 | DataType::Utf8View => "VARCHAR".to_string(),
-        DataType::Binary | DataType::LargeBinary => "VARBINARY".to_string(),
-        DataType::Date32 => "DATE".to_string(),
-        DataType::Date64 => "DATE".to_string(),
-        DataType::Timestamp(unit, tz) => match (unit, tz) {
-            (TimeUnit::Second, None) => "TIMESTAMP(0)".to_string(),
-            (TimeUnit::Millisecond, None) => "TIMESTAMP(3)".to_string(),
-            (TimeUnit::Microsecond, None) => "TIMESTAMP(6)".to_string(),
-            (TimeUnit::Nanosecond, None) => "TIMESTAMP(9)".to_string(),
-            (_, Some(_)) => "TIMESTAMP WITH TIME ZONE".to_string(),
-        },
-        DataType::Decimal128(p, s) => format!("DECIMAL({p},{s})"),
-        DataType::Decimal256(p, s) => format!("DECIMAL({p},{s})"),
-        _ => dt.to_string(),
-    }
-}
-
-fn format_columns(schema: &datafusion::arrow::datatypes::Schema) -> Vec<String> {
-    schema
-        .fields()
-        .iter()
-        .map(|f| {
-            let null = if f.is_nullable() { "" } else { " NOT NULL" };
-            format!("  {} {}{}", f.name(), data_type_sql(f.data_type()), null)
-        })
-        .collect()
-}
-
-fn format_with_clause(opts: &BTreeMap<String, String>) -> String {
-    if opts.is_empty() {
-        return "WITH ('connector' = '...', 'format' = '...');\n/* Original WITH options are not persisted in the stream catalog. */\n"
-            .to_string();
-    }
-    let pairs: Vec<String> = opts
-        .iter()
-        .map(|(k, v)| {
-            let k_esc = k.replace('\'', "''");
-            let v_esc = v.replace('\'', "''");
-            format!("  '{k_esc}' = '{v_esc}'")
-        })
-        .collect();
-    format!("WITH (\n{}\n);\n", pairs.join(",\n"))
-}
-
-/// Single-line `col:TYPE` list for result grids.
-pub fn schema_columns_one_line(schema: &datafusion::arrow::datatypes::Schema) -> String {
-    schema
-        .fields()
-        .iter()
-        .map(|f| format!("{}:{}", f.name(), data_type_sql(f.data_type())))
-        .collect::<Vec<_>>()
-        .join(", ")
-}
-
-fn pipeline_summary_short(program: &LogicalProgram) -> String {
-    let mut parts: Vec<String> = Vec::new();
-    parts.push(format!("tasks={}", program.task_count()));
-    parts.push(format!("hash={}", program.get_hash()));
-    for nw in program.graph.node_weights() {
-        let chain = nw
-            .operator_chain
-            .operators
-            .iter()
-            .map(|o| format!("{}", o.operator_name))
-            .collect::<Vec<_>>()
-            .join("->");
-        parts.push(format!("n{}:{}", nw.node_id, chain));
-    }
-    parts.join(" | ")
-}
-
-/// Extra fields for `SHOW TABLES` result grid (pipeline summary; no full Graphviz).
-pub fn stream_table_row_detail(table: &StreamTable) -> String {
-    match table {
-        StreamTable::Source {
-            connector,
-            event_time_field,
-            watermark_field,
-            with_options,
-            ..
-        } => {
-            format!(
-                "connector={}, event_time={:?}, watermark={:?}, with_options={}",
-                connector,
-                event_time_field,
-                watermark_field,
-                with_options.len()
-            )
-        }
-        StreamTable::Sink { program, .. } => pipeline_summary_short(program),
-    }
-}
-
-fn pipeline_text(program: &LogicalProgram) -> String {
-    let mut lines: Vec<String> = Vec::new();
-    lines.push(format!("tasks_total: {}", program.task_count()));
-    lines.push(format!("program_hash: {}", program.get_hash()));
-    for nw in program.graph.node_weights() {
-        let chain = nw
-            .operator_chain
-            .operators
-            .iter()
-            .map(|o| format!("{}[{}]", o.operator_name, o.operator_id))
-            .collect::<Vec<_>>()
-            .join(" -> ");
-        lines.push(format!(
-            "node {} (parallelism={}): {chain}",
-            nw.node_id, nw.parallelism
-        ));
-    }
-    let dot = program.dot();
-    const MAX_DOT: usize = 12_000;
-    if dot.len() > MAX_DOT {
-        lines.push(format!(
-            "graphviz_dot_truncated:\n{}... [{} more bytes]",
-            &dot[..MAX_DOT],
-            dot.len() - MAX_DOT
-        ));
-    } else {
-        lines.push(format!("graphviz_dot:\n{dot}"));
-    }
-    lines.join("\n")
-}
-
-/// Human-readable `SHOW CREATE TABLE` text (sink `AS SELECT` is not stored).
-pub fn show_create_stream_table(table: &StreamTable) -> String {
-    match table {
-        StreamTable::Source {
-            name,
-            connector,
-            schema,
-            event_time_field,
-            watermark_field,
-            with_options,
-        } => {
-            let cols = format_columns(schema);
-            let mut ddl = format!("CREATE TABLE {name} (\n{}\n)", cols.join(",\n"));
-            if let Some(e) = event_time_field {
-                ddl.push_str(&format!("\n/* EVENT TIME COLUMN: {e} */\n"));
-            }
-            if let Some(w) = watermark_field {
-                ddl.push_str(&format!("/* WATERMARK: {w} */\n"));
-            }
-            let mut merged_opts = with_options.clone();
-            merged_opts
-                .entry("connector".to_string())
-                .or_insert_with(|| connector.clone());
-            ddl.push_str(&format_with_clause(&merged_opts));
-            ddl
-        }
-        StreamTable::Sink { name, program } => {
-            let schema = program.egress_arrow_schema().unwrap_or_else(|| {
-                std::sync::Arc::new(datafusion::arrow::datatypes::Schema::empty())
-            });
-            let cols = format_columns(&schema);
-            let mut ddl = format!(
-                "CREATE STREAMING TABLE {name}\nWITH ('connector' = '...') AS SELECT ...\n/* Sink WITH / AS SELECT text is not stored. Output schema:\n{}\n*/\n\n",
-                cols.join(",\n")
-            );
-            ddl.push_str("-- Resolved logical pipeline:\n");
-            ddl.push_str(&pipeline_text(program));
-            ddl.push('\n');
-            ddl
-        }
-    }
-}
-
-/// Extra fields for `SHOW TABLES` result grid for persisted catalog rows.
-pub fn catalog_table_row_detail(table: &CatalogTable) -> String {
-    match table {
-        CatalogTable::ConnectorTable(source) => format!(
-            "kind=connector, connector={}, event_time={:?}, watermark={:?}, with_options={}",
-            source.connector(),
-            source.event_time_field(),
-            source.temporal_config.watermark_strategy_column,
-            source.catalog_with_options().len()
-        ),
-        CatalogTable::LookupTable(source) => format!(
-            "kind=lookup, connector={}, event_time={:?}, watermark={:?}, with_options={}",
-            source.connector(),
-            source.event_time_field(),
-            source.temporal_config.watermark_strategy_column,
-            source.catalog_with_options().len()
-        ),
-        CatalogTable::TableFromQuery { .. } => "kind=query".to_string(),
-    }
-}
-
-/// Human-readable `SHOW CREATE TABLE` text for persisted catalog rows.
-pub fn show_create_catalog_table(table: &CatalogTable) -> String {
-    match table {
-        CatalogTable::ConnectorTable(source) | CatalogTable::LookupTable(source) => {
-            let schema = source.produce_physical_schema();
-            let cols = format_columns(&schema);
-            let mut ddl = format!("CREATE TABLE {} (\n{}\n)", source.name(), cols.join(",\n"));
-            if let Some(e) = source.event_time_field() {
-                ddl.push_str(&format!("\n/* EVENT TIME COLUMN: {e} */\n"));
-            }
-            if let Some(w) = source.temporal_config.watermark_strategy_column.as_deref() {
-                ddl.push_str(&format!("/* WATERMARK: {w} */\n"));
-            }
-            let mut opts = source.catalog_with_options().clone();
-            opts.entry("connector".to_string())
-                .or_insert_with(|| source.connector().to_string());
-            ddl.push_str(&format_with_clause(&opts));
-            ddl
-        }
-        CatalogTable::TableFromQuery { name, .. } => {
-            format!(
-                "CREATE TABLE {name} AS SELECT ...;\n/* logical query text is not persisted */\n"
-            )
-        }
-    }
-}
diff --git a/src/sql/schema/connector_config.rs b/src/sql/schema/connector_config.rs
deleted file mode 100644
index edb44bae..00000000
--- a/src/sql/schema/connector_config.rs
+++ /dev/null
@@ -1,72 +0,0 @@
-// Licensed under the Apache License, Version 2.0 (the "License");
-// you may not use this file except in compliance with the License.
-// You may obtain a copy of the License at
-//
-//     http://www.apache.org/licenses/LICENSE-2.0
-//
-// Unless required by applicable law or agreed to in writing, software
-// distributed under the License is distributed on an "AS IS" BASIS,
-// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-// See the License for the specific language governing permissions and
-// limitations under the License.
-
-use std::collections::HashMap;
-
-use protocol::function_stream_graph::{
-    GenericConnectorConfig, KafkaSinkConfig, KafkaSourceConfig, connector_op,
-};
-
-#[derive(Debug, Clone)]
-pub enum ConnectorConfig {
-    KafkaSource(KafkaSourceConfig),
-    KafkaSink(KafkaSinkConfig),
-    Generic(HashMap<String, String>),
-}
-
-impl ConnectorConfig {
-    pub fn to_proto_config(&self) -> connector_op::Config {
-        match self {
-            ConnectorConfig::KafkaSource(cfg) => connector_op::Config::KafkaSource(cfg.clone()),
-            ConnectorConfig::KafkaSink(cfg) => connector_op::Config::KafkaSink(cfg.clone()),
-            ConnectorConfig::Generic(props) => {
-                connector_op::Config::Generic(GenericConnectorConfig {
-                    properties: props.clone(),
-                })
-            }
-        }
-    }
-}
-
-impl PartialEq for ConnectorConfig {
-    fn eq(&self, other: &Self) -> bool {
-        use prost::Message;
-        match (self, other) {
-            (ConnectorConfig::KafkaSource(a), ConnectorConfig::KafkaSource(b)) => {
-                a.encode_to_vec() == b.encode_to_vec()
-            }
-            (ConnectorConfig::KafkaSink(a), ConnectorConfig::KafkaSink(b)) => {
-                a.encode_to_vec() == b.encode_to_vec()
-            }
-            (ConnectorConfig::Generic(a), ConnectorConfig::Generic(b)) => a == b,
-            _ => false,
-        }
-    }
-}
-
-impl Eq for ConnectorConfig {}
-
-impl std::hash::Hash for ConnectorConfig {
-    fn hash<H: std::hash::Hasher>(&self, state: &mut H) {
-        use prost::Message;
-        std::mem::discriminant(self).hash(state);
-        match self {
-            ConnectorConfig::KafkaSource(cfg) => cfg.encode_to_vec().hash(state),
-            ConnectorConfig::KafkaSink(cfg) => cfg.encode_to_vec().hash(state),
-            ConnectorConfig::Generic(m) => {
-                let mut pairs: Vec<_> = m.iter().collect();
-                pairs.sort_by_key(|(k, _)| (*k).clone());
-                pairs.hash(state);
-            }
-        }
-    }
-}
diff --git a/src/sql/schema/data_encoding_format.rs b/src/sql/schema/data_encoding_format.rs
index 1cd5c736..0b6f5e1d 100644
--- a/src/sql/schema/data_encoding_format.rs
+++ b/src/sql/schema/data_encoding_format.rs
@@ -10,78 +10,79 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
-use std::collections::HashMap;
-
 use datafusion::arrow::datatypes::{DataType, Field};
 use datafusion::common::{Result, plan_err};
 
 use super::column_descriptor::ColumnDescriptor;
 use crate::sql::common::Format;
-use crate::sql::common::constants::{cdc, connection_format_value, with_opt_bool_str};
-use crate::sql::common::with_option_keys as opt;
+use crate::sql::common::constants::cdc;
 
-/// High-level payload encoding (orthogonal to `Format` wire details in `ConnectionSchema`).
-#[derive(Debug, Clone, Copy, PartialEq, Eq, Hash)]
+#[derive(Debug, Clone, Copy, PartialEq, Eq, Hash, Default)]
 pub enum DataEncodingFormat {
+    #[default]
+    Raw,
     StandardJson,
     DebeziumJson,
     Avro,
     Parquet,
-    Raw,
+    Csv,
+    JsonL,
+    Orc,
+    Protobuf,
 }
 
 impl DataEncodingFormat {
-    pub fn extract_from_map(opts: &HashMap<String, String>) -> Result<Self> {
-        let format_str = opts
-            .get(opt::FORMAT)
-            .map(|s| s.as_str())
-            .unwrap_or(opt::DEFAULT_FORMAT_VALUE);
-        let is_debezium = opts
-            .get(opt::FORMAT_DEBEZIUM_FLAG)
-            .or_else(|| opts.get(opt::JSON_DEBEZIUM))
-            .map(|s| s == with_opt_bool_str::TRUE)
-            .unwrap_or(false);
-
-        match (format_str, is_debezium) {
-            (f, true) if f == connection_format_value::JSON => Ok(Self::DebeziumJson),
-            (f, _) if f == connection_format_value::DEBEZIUM_JSON => Ok(Self::DebeziumJson),
-            (f, false) if f == connection_format_value::JSON => Ok(Self::StandardJson),
-            (f, _) if f == connection_format_value::AVRO => Ok(Self::Avro),
-            (f, _) if f == connection_format_value::PARQUET => Ok(Self::Parquet),
-            _ => Ok(Self::Raw),
+    pub fn from_format(format: Option<&Format>) -> Self {
+        match format {
+            Some(Format::Json(j)) if j.debezium => Self::DebeziumJson,
+            Some(Format::Json(_)) => Self::StandardJson,
+            Some(Format::Avro(_)) => Self::Avro,
+            Some(Format::Parquet(_)) => Self::Parquet,
+            Some(Format::Csv(_)) => Self::Csv,
+            Some(Format::Protobuf(_)) => Self::Protobuf,
+            Some(Format::RawString(_)) | Some(Format::RawBytes(_)) | None => Self::Raw,
+            Some(_) => Self::Raw,
         }
     }
 
-    pub fn from_connection_format(format: &Format) -> Self {
-        match format {
-            Format::Json(j) if j.debezium => Self::DebeziumJson,
-            Format::Json(_) => Self::StandardJson,
-            Format::Avro(_) => Self::Avro,
-            Format::Parquet(_) => Self::Parquet,
-            Format::Protobuf(_) | Format::RawString(_) | Format::RawBytes(_) => Self::Raw,
-        }
+    pub fn is_cdc_format(&self) -> bool {
+        matches!(self, Self::DebeziumJson)
     }
 
+    #[inline]
     pub fn supports_delta_updates(&self) -> bool {
-        matches!(self, Self::DebeziumJson)
+        self.is_cdc_format()
     }
 
-    pub fn apply_envelope(self, columns: Vec<ColumnDescriptor>) -> Result<Vec<ColumnDescriptor>> {
-        if !self.supports_delta_updates() {
-            return Ok(columns);
+    pub fn apply_envelope(
+        &self,
+        logical_columns: Vec<ColumnDescriptor>,
+    ) -> Result<Vec<ColumnDescriptor>> {
+        if !self.is_cdc_format() {
+            return Ok(logical_columns);
         }
-        if columns.iter().any(|c| c.is_computed()) {
-            return plan_err!("Virtual fields are not supported with CDC envelope");
+
+        if logical_columns.is_empty() {
+            return Ok(logical_columns);
         }
-        if columns.is_empty() {
-            return Ok(columns);
+
+        if logical_columns.iter().any(|c| c.is_computed()) {
+            return plan_err!(
+                "Computed/Virtual columns are not supported directly inside a CDC source table; \
+                 define computed columns in a downstream VIEW or AS SELECT streaming query"
+            );
         }
-        let fields: Vec<Field> = columns.into_iter().map(|c| c.into_arrow_field()).collect();
-        let struct_type = DataType::Struct(fields.into());
+
+        let inner_fields: Vec<Field> = logical_columns
+            .into_iter()
+            .map(|c| c.into_arrow_field())
+            .collect();
+
+        let row_struct_type = DataType::Struct(inner_fields.into());
 
         Ok(vec![
-            ColumnDescriptor::new_physical(Field::new(cdc::BEFORE, struct_type.clone(), true)),
-            ColumnDescriptor::new_physical(Field::new(cdc::AFTER, struct_type.clone(), true)),
+            ColumnDescriptor::new_physical(Field::new(cdc::BEFORE, row_struct_type.clone(), true)),
+            ColumnDescriptor::new_physical(Field::new(cdc::AFTER, row_struct_type, true)),
             ColumnDescriptor::new_physical(Field::new(cdc::OP, DataType::Utf8, true)),
         ])
     }
diff --git a/src/sql/schema/introspection/ddl_formatter.rs b/src/sql/schema/introspection/ddl_formatter.rs
new file mode 100644
index 00000000..f4ce36e6
--- /dev/null
+++ b/src/sql/schema/introspection/ddl_formatter.rs
@@ -0,0 +1,156 @@
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+use std::collections::BTreeMap;
+use std::fmt::{self, Write};
+
+use datafusion::arrow::datatypes::{DataType, Schema, TimeUnit};
+
+use crate::sql::common::constants::sql_field;
+
+pub struct DdlBuilder<'a> {
+    table_name: &'a str,
+    schema: &'a Schema,
+    watermark_column: Option<&'a str>,
+    primary_keys: &'a [String],
+    options: BTreeMap<String, String>,
+}
+
+impl<'a> DdlBuilder<'a> {
+    pub fn new(table_name: &'a str, schema: &'a Schema) -> Self {
+        Self {
+            table_name,
+            schema,
+            watermark_column: None,
+            primary_keys: &[],
+            options: BTreeMap::new(),
+        }
+    }
+
+    pub fn with_watermark(mut self, watermark: Option<&'a str>) -> Self {
+        self.watermark_column = watermark;
+        self
+    }
+
+    pub fn with_primary_keys(mut self, keys: &'a [String]) -> Self {
+        self.primary_keys = keys;
+        self
+    }
+
+    pub fn with_options(
+        mut self,
+        opts: &BTreeMap<String, String>,
+        role: &str,
+        connector: &str,
+    ) -> Self {
+        self.options = opts.clone();
+        self.options
+            .entry("type".to_string())
+            .or_insert_with(|| role.to_string());
+        self.options
+            .entry("connector".to_string())
+            .or_insert_with(|| connector.to_string());
+        self
+    }
+}
+
+impl<'a> fmt::Display for DdlBuilder<'a> {
+    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
+        writeln!(f, "CREATE TABLE {} (", self.table_name)?;
+
+        let mut rows: Vec<String> = Vec::new();
+        for field in self.schema.fields() {
+            let null_constraint = if field.is_nullable() { "" } else { " NOT NULL" };
+            rows.push(format!(
+                "  {} {}{}",
+                field.name(),
+                format_data_type(field.data_type()),
+                null_constraint
+            ));
+        }
+
+        if let Some(wm) = self.watermark_column
+            && wm != sql_field::COMPUTED_WATERMARK
+        {
+            rows.push(format!("  WATERMARK FOR {wm}"));
+        }
+
+        if !self.primary_keys.is_empty() {
+            rows.push(format!("  PRIMARY KEY ({})", self.primary_keys.join(", ")));
+        }
+
+        writeln!(f, "{}", rows.join(",\n"))?;
+        write!(f, ")")?;
+
+        if !self.options.is_empty() {
+            writeln!(f)?;
+            writeln!(f, "WITH (")?;
+            let mut opt_lines: Vec<String> = Vec::with_capacity(self.options.len());
+            for (k, v) in &self.options {
+                let k_esc = k.replace('\'', "''");
+                let v_esc = v.replace('\'', "''");
+                opt_lines.push(format!("  '{k_esc}' = '{v_esc}'"));
+            }
+            write!(f, "{}\n);", opt_lines.join(",\n"))?;
+        } else {
+            write!(f, ";")?;
+        }
+
+        Ok(())
+    }
+}
+
+pub fn format_data_type(dt: &DataType) -> String {
+    match dt {
+        DataType::Null => "NULL".to_string(),
+        DataType::Boolean => "BOOLEAN".to_string(),
+        DataType::Int8 => "TINYINT".to_string(),
+        DataType::Int16 => "SMALLINT".to_string(),
+        DataType::Int32 => "INT".to_string(),
+        DataType::Int64 => "BIGINT".to_string(),
+        DataType::UInt8 => "TINYINT UNSIGNED".to_string(),
+        DataType::UInt16 => "SMALLINT UNSIGNED".to_string(),
+        DataType::UInt32 => "INT UNSIGNED".to_string(),
+        DataType::UInt64 => "BIGINT UNSIGNED".to_string(),
+        DataType::Float16 => "FLOAT".to_string(),
+        DataType::Float32 => "REAL".to_string(),
+        DataType::Float64 => "DOUBLE".to_string(),
+        DataType::Utf8 | DataType::LargeUtf8 | DataType::Utf8View => "VARCHAR".to_string(),
+        DataType::Binary | DataType::LargeBinary => "VARBINARY".to_string(),
+        DataType::Date32 | DataType::Date64 => "DATE".to_string(),
+        DataType::Timestamp(unit, tz) => match (unit, tz) {
+            (TimeUnit::Second, None) => "TIMESTAMP(0)".to_string(),
+            (TimeUnit::Millisecond, None) => "TIMESTAMP(3)".to_string(),
+            (TimeUnit::Microsecond, None) => "TIMESTAMP(6)".to_string(),
+            (TimeUnit::Nanosecond, None) => "TIMESTAMP(9)".to_string(),
+            (_, Some(_)) => "TIMESTAMP WITH TIME ZONE".to_string(),
+        },
+        DataType::Decimal128(p, s) | DataType::Decimal256(p, s) => format!("DECIMAL({p}, {s})"),
+        _ => dt.to_string(),
+    }
+}
+
+pub fn schema_columns_one_line(schema: &Schema) -> String {
+    let mut buf = String::new();
+    for (idx, field) in schema.fields().iter().enumerate() {
+        if idx > 0 {
+            buf.push_str(", ");
+        }
+        let _ = write!(
+            buf,
+            "{}:{}",
+            field.name(),
+            format_data_type(field.data_type())
+        );
+    }
+    buf
+}
diff --git a/src/sql/schema/introspection/mod.rs b/src/sql/schema/introspection/mod.rs
new file mode 100644
index 00000000..1ba9c816
--- /dev/null
+++ b/src/sql/schema/introspection/mod.rs
@@ -0,0 +1,21 @@
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+pub mod ddl_formatter;
+pub mod show_formatter;
+pub mod stream_formatter;
+
+#[allow(unused_imports)]
+pub use ddl_formatter::{DdlBuilder, format_data_type, schema_columns_one_line};
+pub use show_formatter::{catalog_table_row_detail, show_create_catalog_table};
+#[allow(unused_imports)]
+pub use stream_formatter::{show_create_stream_table, stream_table_row_detail};
diff --git a/src/sql/schema/introspection/show_formatter.rs b/src/sql/schema/introspection/show_formatter.rs
new file mode 100644
index 00000000..28a81ae9
--- /dev/null
+++ b/src/sql/schema/introspection/show_formatter.rs
@@ -0,0 +1,100 @@
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+use crate::sql::common::constants::connection_table_role;
+use crate::sql::schema::catalog::ExternalTable;
+use crate::sql::schema::table::CatalogEntity;
+
+use super::ddl_formatter::DdlBuilder;
+
+impl ExternalTable {
+    pub fn to_ddl_string(&self) -> String {
+        match self {
+            ExternalTable::Source(source) => {
+                let schema = source.produce_physical_schema();
+                DdlBuilder::new(&source.table_identifier, &schema)
+                    .with_watermark(source.temporal_config.watermark_strategy_column.as_deref())
+                    .with_primary_keys(&source.key_constraints)
+                    .with_options(
+                        &source.catalog_with_options,
+                        connection_table_role::SOURCE,
+                        &source.adapter_type,
+                    )
+                    .to_string()
+            }
+            ExternalTable::Sink(sink) => {
+                let schema = sink.produce_physical_schema();
+                DdlBuilder::new(&sink.table_identifier, &schema)
+                    .with_primary_keys(&sink.key_constraints)
+                    .with_options(
+                        &sink.catalog_with_options,
+                        connection_table_role::SINK,
+                        &sink.adapter_type,
+                    )
+                    .to_string()
+            }
+            ExternalTable::Lookup(lookup) => {
+                let schema = lookup.produce_physical_schema();
+                DdlBuilder::new(&lookup.table_identifier, &schema)
+                    .with_primary_keys(&lookup.key_constraints)
+                    .with_options(
+                        &lookup.catalog_with_options,
+                        connection_table_role::LOOKUP,
+                        &lookup.adapter_type,
+                    )
+                    .to_string()
+            }
+        }
+    }
+
+    pub fn to_row_detail(&self) -> String {
+        match self {
+            ExternalTable::Source(s) => format!(
+                "{{ kind: 'source', connector: '{}', watermark: '{}', options_count: {} }}",
+                s.adapter_type,
+                s.temporal_config
+                    .watermark_strategy_column
+                    .as_deref()
+                    .unwrap_or("none"),
+                s.catalog_with_options.len()
+            ),
+            ExternalTable::Sink(s) => format!(
+                "{{ kind: 'sink', connector: '{}', partitioned: {}, options_count: {} }}",
+                s.adapter_type,
+                s.partition_exprs.as_ref().is_some(),
+                s.catalog_with_options.len()
+            ),
+            ExternalTable::Lookup(s) => format!(
+                "{{ kind: 'lookup', connector: '{}', cache_ttl_secs: {}, options_count: {} }}",
+                s.adapter_type,
+                s.lookup_cache_ttl.map(|d| d.as_secs()).unwrap_or(0),
+                s.catalog_with_options.len()
+            ),
+        }
+    }
+}
+
+pub fn show_create_catalog_table(table: &CatalogEntity) -> String {
+    match table {
+        CatalogEntity::ExternalConnector(ext) => ext.to_ddl_string(),
+        CatalogEntity::ComputedTable { name, .. } => {
+            format!("-- Logical query view\nCREATE VIEW {name} AS SELECT ...;")
+        }
+    }
+}
+
+pub fn catalog_table_row_detail(table: &CatalogEntity) -> String {
+    match table {
+        CatalogEntity::ExternalConnector(ext) => ext.to_row_detail(),
+        CatalogEntity::ComputedTable { .. } => "{ kind: 'logical_view' }".to_string(),
+    }
+}
diff --git a/src/sql/schema/introspection/stream_formatter.rs b/src/sql/schema/introspection/stream_formatter.rs
new file mode 100644
index 00000000..ebb02330
--- /dev/null
+++ b/src/sql/schema/introspection/stream_formatter.rs
@@ -0,0 +1,120 @@
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+use std::sync::Arc;
+
+use datafusion::arrow::datatypes::Schema;
+
+use crate::sql::common::constants::connection_table_role;
+use crate::sql::logical_node::logical::LogicalProgram;
+use crate::sql::schema::schema_provider::StreamTable;
+
+use super::ddl_formatter::DdlBuilder;
+
+impl StreamTable {
+    pub fn to_ddl_string(&self) -> String {
+        match self {
+            StreamTable::Source {
+                name,
+                connector,
+                schema,
+                event_time_field: _,
+                watermark_field,
+                with_options,
+            } => DdlBuilder::new(name, schema)
+                .with_watermark(watermark_field.as_deref())
+                .with_options(with_options, connection_table_role::SOURCE, connector)
+                .to_string(),
+            StreamTable::Sink { name, program } => {
+                let schema: Arc<Schema> = program
+                    .egress_arrow_schema()
+                    .unwrap_or_else(|| Arc::new(Schema::empty()));
+
+                let mut ddl = format!("CREATE STREAMING TABLE {name} AS SELECT ...\n\n");
+                ddl.push_str("/* === SINK SCHEMA === */\n");
+                let schema_ddl = DdlBuilder::new(name, &schema).to_string();
+                ddl.push_str(&schema_ddl);
+                ddl.push_str("\n\n/* === STREAMING TOPOLOGY === */\n");
+                ddl.push_str(&format_pipeline(program));
+                ddl
+            }
+        }
+    }
+
+    pub fn to_row_detail(&self) -> String {
+        match self {
+            StreamTable::Source {
+                connector,
+                event_time_field,
+                watermark_field,
+                with_options,
+                ..
+            } => format!(
+                "{{ kind: 'stream_source', connector: '{}', event_time: '{}', watermark: '{}', options_count: {} }}",
+                connector,
+                event_time_field.as_deref().unwrap_or("none"),
+                watermark_field.as_deref().unwrap_or("none"),
+                with_options.len()
+            ),
+            StreamTable::Sink { program, .. } => format!(
+                "{{ kind: 'streaming_sink', tasks: {}, nodes: {} }}",
+                program.task_count(),
+                program.graph.node_count()
+            ),
+        }
+    }
+}
+
+pub fn show_create_stream_table(table: &StreamTable) -> String {
+    table.to_ddl_string()
+}
+
+pub fn stream_table_row_detail(table: &StreamTable) -> String {
+    table.to_row_detail()
+}
+
+fn format_pipeline(program: &LogicalProgram) -> String {
+    let mut lines: Vec<String> = Vec::new();
+    lines.push(format!("Pipeline Hash : {}", program.get_hash()));
+    lines.push(format!("Total Tasks   : {}", program.task_count()));
+    lines.push(format!("Node Count    : {}", program.graph.node_count()));
+    lines.push(String::from("Operator Chains:"));
+
+    for nw in program.graph.node_weights() {
+        let chain = nw
+            .operator_chain
+            .operators
+            .iter()
+            .map(|op| format!("{}[{}]", op.operator_name, op.operator_id))
+            .collect::<Vec<_>>()
+            .join(" -> ");
+
+        lines.push(format!(
+            "  Node {:<3} | Parallelism {:<3} | {}",
+            nw.node_id, nw.parallelism, chain
+        ));
+    }
+
+    let dot = program.dot();
+    const MAX_DOT: usize = 5_000;
+    if dot.len() > MAX_DOT {
+        lines.push(format!(
+            "\nGraphviz DOT (truncated, {} bytes omitted):\n{}...",
+            dot.len() - MAX_DOT,
+            &dot[..MAX_DOT]
+        ));
+    } else {
+        lines.push(format!("\nGraphviz DOT:\n{dot}"));
+    }
+
+    lines.join("\n")
+}
diff --git a/src/sql/schema/kafka_operator_config.rs b/src/sql/schema/kafka_operator_config.rs
deleted file mode 100644
index d87dda8f..00000000
--- a/src/sql/schema/kafka_operator_config.rs
+++ /dev/null
@@ -1,272 +0,0 @@
-// Licensed under the Apache License, Version 2.0 (the "License");
-// you may not use this file except in compliance with the License.
-//
-// Builds strongly-typed proto Kafka configs from SQL DDL WITH options.
-
-use std::collections::HashMap;
-
-use datafusion::arrow::datatypes::Schema;
-use datafusion::common::{Result as DFResult, plan_datafusion_err, plan_err};
-
-use protocol::function_stream_graph::connector_op::Config as ProtoConfig;
-use protocol::function_stream_graph::{
-    BadDataPolicy, DecimalEncodingProto, FormatConfig, JsonFormatConfig, KafkaAuthConfig,
-    KafkaAuthNone, KafkaOffsetMode, KafkaReadMode, KafkaSinkCommitMode, KafkaSinkConfig,
-    KafkaSourceConfig, RawBytesFormatConfig, RawStringFormatConfig, TimestampFormatProto,
-};
-
-use crate::sql::common::connector_options::ConnectorOptions;
-use crate::sql::common::constants::{connection_table_role, kafka_with_value};
-use crate::sql::common::formats::{
-    BadData, DecimalEncoding as SqlDecimalEncoding, Format as SqlFormat,
-    TimestampFormat as SqlTimestampFormat,
-};
-use crate::sql::common::with_option_keys as opt;
-use crate::sql::schema::table_role::TableRole;
-
-const STREAMING_JOB_OPTION_CHECKPOINT_INTERVAL: &str = "checkpoint.interval";
-const STREAMING_JOB_OPTION_PARALLELISM: &str = "parallelism";
-
-fn sql_format_to_proto(fmt: &SqlFormat) -> DFResult<FormatConfig> {
-    match fmt {
-        SqlFormat::Json(j) => Ok(FormatConfig {
-            format: Some(
-                protocol::function_stream_graph::format_config::Format::Json(JsonFormatConfig {
-                    timestamp_format: match j.timestamp_format {
-                        SqlTimestampFormat::RFC3339 => {
-                            TimestampFormatProto::TimestampRfc3339 as i32
-                        }
-                        SqlTimestampFormat::UnixMillis => {
-                            TimestampFormatProto::TimestampUnixMillis as i32
-                        }
-                    },
-                    decimal_encoding: match j.decimal_encoding {
-                        SqlDecimalEncoding::Number => DecimalEncodingProto::DecimalNumber as i32,
-                        SqlDecimalEncoding::String => DecimalEncodingProto::DecimalString as i32,
-                        SqlDecimalEncoding::Bytes => DecimalEncodingProto::DecimalBytes as i32,
-                    },
-                    include_schema: j.include_schema,
-                    confluent_schema_registry: j.confluent_schema_registry,
-                    schema_id: j.schema_id,
-                    debezium: j.debezium,
-                    unstructured: j.unstructured,
-                }),
-            ),
-        }),
-        SqlFormat::RawString(_) => Ok(FormatConfig {
-            format: Some(
-                protocol::function_stream_graph::format_config::Format::RawString(
-                    RawStringFormatConfig {},
-                ),
-            ),
-        }),
-        SqlFormat::RawBytes(_) => Ok(FormatConfig {
-            format: Some(
-                protocol::function_stream_graph::format_config::Format::RawBytes(
-                    RawBytesFormatConfig {},
-                ),
-            ),
-        }),
-        other => plan_err!(
-            "Kafka connector: format '{}' is not supported yet",
-            other.name()
-        ),
-    }
-}
-
-fn sql_bad_data_to_proto(bad: &BadData) -> i32 {
-    match bad {
-        BadData::Fail {} => BadDataPolicy::BadDataFail as i32,
-        BadData::Drop {} => BadDataPolicy::BadDataDrop as i32,
-    }
-}
-
-/// Build Kafka proto config from a flat string map (catalog rebuild path).
-pub fn build_kafka_proto_config_from_string_map(
-    map: HashMap<String, String>,
-    _physical_schema: &Schema,
-) -> DFResult<ProtoConfig> {
-    let mut options = ConnectorOptions::from_flat_string_map(map)?;
-    let format = crate::sql::common::formats::Format::from_opts(&mut options)
-        .map_err(|e| datafusion::error::DataFusionError::Plan(format!("invalid format: {e}")))?;
-    let bad_data = BadData::from_opts(&mut options).map_err(|e| {
-        datafusion::error::DataFusionError::Plan(format!("Invalid bad_data: '{e}'"))
-    })?;
-    let _framing = crate::sql::common::formats::Framing::from_opts(&mut options)
-        .map_err(|e| datafusion::error::DataFusionError::Plan(format!("invalid framing: '{e}'")))?;
-
-    let role = match options.pull_opt_str(opt::TYPE)?.as_deref() {
-        None | Some(connection_table_role::SOURCE) => TableRole::Ingestion,
-        Some(connection_table_role::SINK) => TableRole::Egress,
-        Some(connection_table_role::LOOKUP) => TableRole::Reference,
-        Some(other) => {
-            return plan_err!("invalid connection type '{other}' in WITH options");
-        }
-    };
-
-    build_kafka_proto_config(&mut options, role, &format, bad_data)
-}
-
-/// Core builder shared by SQL DDL and catalog reload paths.
-pub fn build_kafka_proto_config(
-    options: &mut ConnectorOptions,
-    role: TableRole,
-    format: &Option<SqlFormat>,
-    bad_data: BadData,
-) -> DFResult<ProtoConfig> {
-    let bootstrap_servers = match options.pull_opt_str(opt::KAFKA_BOOTSTRAP_SERVERS)? {
-        Some(s) => s,
-        None => options
-            .pull_opt_str(opt::KAFKA_BOOTSTRAP_SERVERS_LEGACY)?
-            .ok_or_else(|| {
-                plan_datafusion_err!(
-                    "Kafka connector requires 'bootstrap.servers' in the WITH clause"
-                )
-            })?,
-    };
-
-    let topic = options.pull_opt_str(opt::KAFKA_TOPIC)?.ok_or_else(|| {
-        plan_datafusion_err!("Kafka connector requires 'topic' in the WITH clause")
-    })?;
-
-    let sql_format = format.clone().ok_or_else(|| {
-        plan_datafusion_err!(
-            "Kafka connector requires 'format' in the WITH clause (e.g. format = 'json')"
-        )
-    })?;
-    let proto_format = sql_format_to_proto(&sql_format)?;
-
-    let rate_limit = options
-        .pull_opt_u64(opt::KAFKA_RATE_LIMIT_MESSAGES_PER_SECOND)?
-        .map(|v| v.clamp(1, u32::MAX as u64) as u32)
-        .unwrap_or(0);
-
-    let value_subject = options.pull_opt_str(opt::KAFKA_VALUE_SUBJECT)?;
-
-    let auth = Some(KafkaAuthConfig {
-        auth: Some(
-            protocol::function_stream_graph::kafka_auth_config::Auth::None(KafkaAuthNone {}),
-        ),
-    });
-
-    let _ = options.pull_opt_str(opt::TYPE)?;
-    let _ = options.pull_opt_str(opt::CONNECTOR)?;
-
-    match role {
-        TableRole::Ingestion => {
-            let offset_mode = match options
-                .pull_opt_str(opt::KAFKA_SCAN_STARTUP_MODE)?
-                .as_deref()
-            {
-                Some(s) if s == kafka_with_value::SCAN_LATEST => {
-                    KafkaOffsetMode::KafkaOffsetLatest as i32
-                }
-                Some(s) if s == kafka_with_value::SCAN_EARLIEST => {
-                    KafkaOffsetMode::KafkaOffsetEarliest as i32
-                }
-                Some(s)
-                    if s == kafka_with_value::SCAN_GROUP_OFFSETS
-                        || s == kafka_with_value::SCAN_GROUP =>
-                {
-                    KafkaOffsetMode::KafkaOffsetGroup as i32
-                }
-                None => KafkaOffsetMode::KafkaOffsetGroup as i32,
-                Some(other) => {
-                    return plan_err!(
-                        "invalid scan.startup.mode '{other}'; expected latest, earliest, or group-offsets"
-                    );
-                }
-            };
-
-            let read_mode = match options.pull_opt_str(opt::KAFKA_ISOLATION_LEVEL)?.as_deref() {
-                Some(s) if s == kafka_with_value::ISOLATION_READ_COMMITTED => {
-                    KafkaReadMode::KafkaReadCommitted as i32
-                }
-                Some(s) if s == kafka_with_value::ISOLATION_READ_UNCOMMITTED => {
-                    KafkaReadMode::KafkaReadUncommitted as i32
-                }
-                None => KafkaReadMode::KafkaReadDefault as i32,
-                Some(other) => {
-                    return plan_err!("invalid isolation.level '{other}'");
-                }
-            };
-
-            let group_id = match options.pull_opt_str(opt::KAFKA_GROUP_ID)? {
-                Some(s) => Some(s),
-                None => options.pull_opt_str(opt::KAFKA_GROUP_ID_LEGACY)?,
-            };
-            let group_id_prefix = options.pull_opt_str(opt::KAFKA_GROUP_ID_PREFIX)?;
-
-            let mut client_configs = options.drain_remaining_string_values()?;
-            // Streaming job-level options are parsed by planner/coordinator, not Kafka client.
-            client_configs.remove(STREAMING_JOB_OPTION_CHECKPOINT_INTERVAL);
-            client_configs.remove(STREAMING_JOB_OPTION_PARALLELISM);
-
-            Ok(ProtoConfig::KafkaSource(KafkaSourceConfig {
-                topic,
-                bootstrap_servers,
-                group_id,
-                group_id_prefix,
-                offset_mode,
-                read_mode,
-                auth,
-                client_configs,
-                format: Some(proto_format),
-                bad_data_policy: sql_bad_data_to_proto(&bad_data),
-                rate_limit_msgs_per_sec: rate_limit,
-                value_subject,
-            }))
-        }
-        TableRole::Egress => {
-            let commit_mode = match options
-                .pull_opt_str(opt::KAFKA_SINK_COMMIT_MODE)?
-                .as_deref()
-            {
-                Some(s)
-                    if s == kafka_with_value::SINK_COMMIT_EXACTLY_ONCE_HYPHEN
-                        || s == kafka_with_value::SINK_COMMIT_EXACTLY_ONCE_UNDERSCORE =>
-                {
-                    KafkaSinkCommitMode::KafkaSinkExactlyOnce as i32
-                }
-                None => KafkaSinkCommitMode::KafkaSinkAtLeastOnce as i32,
-                Some(s)
-                    if s == kafka_with_value::SINK_COMMIT_AT_LEAST_ONCE_HYPHEN
-                        || s == kafka_with_value::SINK_COMMIT_AT_LEAST_ONCE_UNDERSCORE =>
-                {
-                    KafkaSinkCommitMode::KafkaSinkAtLeastOnce as i32
-                }
-                Some(other) => {
-                    return plan_err!("invalid sink.commit.mode '{other}'");
-                }
-            };
-            let key_field = match options.pull_opt_str(opt::KAFKA_SINK_KEY_FIELD)? {
-                Some(s) => Some(s),
-                None => options.pull_opt_str(opt::KAFKA_KEY_FIELD_LEGACY)?,
-            };
-            let timestamp_field = match options.pull_opt_str(opt::KAFKA_SINK_TIMESTAMP_FIELD)? {
-                Some(s) => Some(s),
-                None => options.pull_opt_str(opt::KAFKA_TIMESTAMP_FIELD_LEGACY)?,
-            };
-
-            let mut client_configs = options.drain_remaining_string_values()?;
-            // Streaming job-level options are parsed by planner/coordinator, not Kafka client.
-            client_configs.remove(STREAMING_JOB_OPTION_CHECKPOINT_INTERVAL);
-            client_configs.remove(STREAMING_JOB_OPTION_PARALLELISM);
-
-            Ok(ProtoConfig::KafkaSink(KafkaSinkConfig {
-                topic,
-                bootstrap_servers,
-                commit_mode,
-                key_field,
-                timestamp_field,
-                auth,
-                client_configs,
-                format: Some(proto_format),
-                value_subject,
-            }))
-        }
-        TableRole::Reference => {
-            plan_err!("Kafka connector cannot be used as a lookup table in this path")
-        }
-    }
-}
diff --git a/src/sql/schema/mod.rs b/src/sql/schema/mod.rs
index be3cdda4..e11e4808 100644
--- a/src/sql/schema/mod.rs
+++ b/src/sql/schema/mod.rs
@@ -10,30 +10,21 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
-pub mod catalog_ddl;
+pub mod catalog;
 pub mod column_descriptor;
 pub mod connection_type;
-pub mod connector_config;
 pub mod data_encoding_format;
-pub mod kafka_operator_config;
-pub mod schema_context;
+pub mod introspection;
 pub mod schema_provider;
-pub mod source_table;
 pub mod table;
-pub mod table_execution_unit;
 pub mod table_role;
 pub mod temporal_pipeline_config;
 pub mod utils;
 
-pub use catalog_ddl::{
+pub use catalog::{ExternalTable, LookupTable, SinkTable, SourceTable};
+pub use column_descriptor::ColumnDescriptor;
+pub use introspection::{
     catalog_table_row_detail, schema_columns_one_line, show_create_catalog_table,
 };
-pub use column_descriptor::ColumnDescriptor;
-pub use connection_type::ConnectionType;
-pub use connector_config::ConnectorConfig;
-pub use source_table::SourceTable;
-
-/// Back-compat alias for [`SourceTable`].
-pub type ConnectorTable = SourceTable;
 pub use schema_provider::{ObjectName, StreamPlanningContext, StreamSchemaProvider, StreamTable};
-pub use table::Table;
+pub use table::CatalogEntity;
diff --git a/src/sql/schema/schema_context.rs b/src/sql/schema/schema_context.rs
deleted file mode 100644
index 851bf6af..00000000
--- a/src/sql/schema/schema_context.rs
+++ /dev/null
@@ -1,37 +0,0 @@
-// Licensed under the Apache License, Version 2.0 (the "License");
-// you may not use this file except in compliance with the License.
-// You may obtain a copy of the License at
-//
-//     http://www.apache.org/licenses/LICENSE-2.0
-//
-// Unless required by applicable law or agreed to in writing, software
-// distributed under the License is distributed on an "AS IS" BASIS,
-// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-// See the License for the specific language governing permissions and
-// limitations under the License.
-
-use datafusion::arrow::datatypes::{DataType, Schema};
-use datafusion::common::{DFSchema, Result};
-use datafusion::logical_expr::Expr;
-use datafusion_expr::ExprSchemable;
-
-pub trait SchemaContext {
-    fn resolve_expression(&self, expr: &Expr, schema: &Schema) -> Result<Expr>;
-    fn extract_datatype(&self, expr: &Expr, schema: &Schema) -> Result<DataType>;
-}
-
-/// [`SchemaContext`] backed by a [`DFSchema`] built from the physical Arrow schema.
-pub struct DfSchemaContext;
-
-impl SchemaContext for DfSchemaContext {
-    fn resolve_expression(&self, expr: &Expr, schema: &Schema) -> Result<Expr> {
-        let df = DFSchema::try_from(schema.clone())?;
-        let _ = expr.get_type(&df)?;
-        Ok(expr.clone())
-    }
-
-    fn extract_datatype(&self, expr: &Expr, schema: &Schema) -> Result<DataType> {
-        let df = DFSchema::try_from(schema.clone())?;
-        expr.get_type(&df)
-    }
-}
diff --git a/src/sql/schema/schema_provider.rs b/src/sql/schema/schema_provider.rs
index 26fd43e8..15cd58ee 100644
--- a/src/sql/schema/schema_provider.rs
+++ b/src/sql/schema/schema_provider.rs
@@ -14,7 +14,7 @@ use std::collections::{BTreeMap, HashMap, HashSet};
 use std::sync::Arc;
 
 use datafusion::arrow::datatypes::{self as datatypes, DataType, Field, Schema};
-use datafusion::common::{DataFusionError, Result};
+use datafusion::common::{DataFusionError, Result as DataFusionResult};
 use datafusion::datasource::{DefaultTableSource, TableProvider, TableType};
 use datafusion::execution::{FunctionRegistry, SessionStateDefaults};
 use datafusion::logical_expr::expr_rewriter::FunctionRewrite;
@@ -23,11 +23,13 @@ use datafusion::logical_expr::{AggregateUDF, Expr, ScalarUDF, TableSource, Windo
 use datafusion::optimizer::Analyzer;
 use datafusion::sql::TableReference;
 use datafusion::sql::planner::ContextProvider;
+use thiserror::Error;
+use tracing::{debug, error, info};
 use unicase::UniCase;
 
 use crate::sql::common::constants::{planning_placeholder_udf, window_fn};
 use crate::sql::logical_node::logical::{DylibUdfConfig, LogicalProgram};
-use crate::sql::schema::table::Table as CatalogTable;
+use crate::sql::schema::table::CatalogEntity;
 use crate::sql::schema::utils::window_arrow_struct;
 use crate::sql::types::{PlanningOptions, PlanningPlaceholderUdf, SqlConfig};
 
@@ -38,6 +40,25 @@ fn object_name(s: impl Into<String>) -> ObjectName {
     UniCase::new(s.into())
 }
 
+#[derive(Error, Debug)]
+pub enum PlanningError {
+    #[error("Catalog table not found: {0}")]
+    TableNotFound(String),
+    #[error("Planning init failed: {0}")]
+    InitError(String),
+    #[error("Engine error: {0}")]
+    Engine(#[from] DataFusionError),
+}
+
+impl From<PlanningError> for DataFusionError {
+    fn from(err: PlanningError) -> Self {
+        match err {
+            PlanningError::Engine(inner) => inner,
+            other => DataFusionError::Plan(other.to_string()),
+        }
+    }
+}
+
 #[derive(Clone, Debug)]
 pub enum StreamTable {
     Source {
@@ -46,7 +67,6 @@ pub enum StreamTable {
         schema: Arc<Schema>,
         event_time_field: Option<String>,
         watermark_field: Option<String>,
-        /// Persisted `WITH` options for `SHOW CREATE TABLE`.
         with_options: BTreeMap<String, String>,
     },
     Sink {
@@ -98,7 +118,7 @@ impl TableProvider for LogicalBatchInput {
         _projection: Option<&Vec<usize>>,
         _filters: &[Expr],
         _limit: Option<usize>,
-    ) -> Result<Arc<dyn datafusion::physical_plan::ExecutionPlan>> {
+    ) -> DataFusionResult<Arc<dyn datafusion::physical_plan::ExecutionPlan>> {
         Ok(Arc::new(crate::sql::physical::FsMemExec::new(
             self.table_name.clone(),
             Arc::clone(&self.schema),
@@ -117,7 +137,7 @@ pub struct FunctionCatalog {
 #[derive(Clone, Default)]
 pub struct TableCatalog {
     pub streams: HashMap<ObjectName, Arc<StreamTable>>,
-    pub catalogs: HashMap<ObjectName, Arc<CatalogTable>>,
+    pub catalogs: HashMap<ObjectName, Arc<CatalogEntity>>,
     pub source_defs: HashMap<String, String>,
 }
 
@@ -132,7 +152,6 @@ pub struct StreamPlanningContext {
     pub sql_config: SqlConfig,
 }
 
-/// Back-compat name for [`StreamPlanningContext`].
 pub type StreamSchemaProvider = StreamPlanningContext;
 
 impl StreamPlanningContext {
@@ -150,20 +169,25 @@ impl StreamPlanningContext {
         self.sql_config.key_by_parallelism
     }
 
-    /// Same registration order as the historical `StreamSchemaProvider::new` (placeholders, then DataFusion defaults).
+    pub fn try_new(config: SqlConfig) -> Result<Self, PlanningError> {
+        info!("Initializing StreamPlanningContext");
+        let mut builder = StreamPlanningContextBuilder::default();
+        builder
+            .with_streaming_extensions()?
+            .with_default_functions()?;
+        let mut ctx = builder.build();
+        ctx.sql_config = config;
+        Ok(ctx)
+    }
+
     pub fn new() -> Self {
-        let mut ctx = Self::builder()
-            .with_streaming_extensions()
-            .expect("streaming extensions")
-            .with_default_functions()
-            .expect("default functions")
-            .build();
-        ctx.sql_config = crate::sql::planning_runtime::sql_planning_snapshot();
-        ctx
+        let config = crate::sql::planning_runtime::sql_planning_snapshot();
+        Self::try_new(config).expect("StreamPlanningContext bootstrap")
     }
 
     pub fn register_stream_table(&mut self, table: StreamTable) {
         let key = object_name(table.name().to_string());
+        debug!(table = %key, "register stream table");
         self.tables.streams.insert(key, Arc::new(table));
     }
 
@@ -174,12 +198,13 @@ impl StreamPlanningContext {
             .cloned()
     }
 
-    pub fn register_catalog_table(&mut self, table: CatalogTable) {
+    pub fn register_catalog_table(&mut self, table: CatalogEntity) {
         let key = object_name(table.name().to_string());
+        debug!(table = %key, "register catalog table");
         self.tables.catalogs.insert(key, Arc::new(table));
     }
 
-    pub fn get_catalog_table(&self, table_name: impl AsRef<str>) -> Option<&CatalogTable> {
+    pub fn get_catalog_table(&self, table_name: impl AsRef<str>) -> Option<&CatalogEntity> {
         self.tables
             .catalogs
             .get(&object_name(table_name.as_ref().to_string()))
@@ -189,7 +214,7 @@ impl StreamPlanningContext {
     pub fn get_catalog_table_mut(
         &mut self,
         table_name: impl AsRef<str>,
-    ) -> Option<&mut CatalogTable> {
+    ) -> Option<&mut CatalogEntity> {
         self.tables
             .catalogs
             .get_mut(&object_name(table_name.as_ref().to_string()))
@@ -221,8 +246,7 @@ impl StreamPlanningContext {
         self.register_stream_table(table);
     }
 
-    /// Alias for [`Self::register_catalog_table`].
-    pub fn insert_catalog_table(&mut self, table: CatalogTable) {
+    pub fn insert_catalog_table(&mut self, table: CatalogEntity) {
         self.register_catalog_table(table);
     }
 
@@ -254,12 +278,15 @@ impl StreamPlanningContext {
 }
 
 impl ContextProvider for StreamPlanningContext {
-    fn get_table_source(&self, name: TableReference) -> Result<Arc<dyn TableSource>> {
-        let table = self
-            .get_stream_table(name.table())
-            .ok_or_else(|| DataFusionError::Plan(format!("Table {} not found", name)))?;
-
-        Ok(Self::create_table_source(name.to_string(), table.schema()))
+    fn get_table_source(&self, name: TableReference) -> DataFusionResult<Arc<dyn TableSource>> {
+        let name_str = name.table();
+        match self.get_stream_table(name_str) {
+            Some(table) => Ok(Self::create_table_source(name.to_string(), table.schema())),
+            None => {
+                error!(table = %name_str, "stream table lookup failed");
+                Err(DataFusionError::Plan(format!("Table {} not found", name)))
+            }
+        }
     }
 
     fn get_function_meta(&self, name: &str) -> Option<Arc<ScalarUDF>> {
@@ -304,7 +331,7 @@ impl FunctionRegistry for StreamPlanningContext {
         self.functions.scalars.keys().cloned().collect()
     }
 
-    fn udf(&self, name: &str) -> Result<Arc<ScalarUDF>> {
+    fn udf(&self, name: &str) -> DataFusionResult<Arc<ScalarUDF>> {
         self.functions
             .scalars
             .get(name)
@@ -312,7 +339,7 @@ impl FunctionRegistry for StreamPlanningContext {
             .ok_or_else(|| DataFusionError::Plan(format!("No UDF with name {name}")))
     }
 
-    fn udaf(&self, name: &str) -> Result<Arc<AggregateUDF>> {
+    fn udaf(&self, name: &str) -> DataFusionResult<Arc<AggregateUDF>> {
         self.functions
             .aggregates
             .get(name)
@@ -320,7 +347,7 @@ impl FunctionRegistry for StreamPlanningContext {
             .ok_or_else(|| DataFusionError::Plan(format!("No UDAF with name {name}")))
     }
 
-    fn udwf(&self, name: &str) -> Result<Arc<WindowUDF>> {
+    fn udwf(&self, name: &str) -> DataFusionResult<Arc<WindowUDF>> {
         self.functions
             .windows
             .get(name)
@@ -331,27 +358,33 @@ impl FunctionRegistry for StreamPlanningContext {
     fn register_function_rewrite(
         &mut self,
         rewrite: Arc<dyn FunctionRewrite + Send + Sync>,
-    ) -> Result<()> {
+    ) -> DataFusionResult<()> {
         self.analyzer.add_function_rewrite(rewrite);
         Ok(())
     }
 
-    fn register_udf(&mut self, udf: Arc<ScalarUDF>) -> Result<Option<Arc<ScalarUDF>>> {
+    fn register_udf(&mut self, udf: Arc<ScalarUDF>) -> DataFusionResult<Option<Arc<ScalarUDF>>> {
         Ok(self.functions.scalars.insert(udf.name().to_string(), udf))
     }
 
-    fn register_udaf(&mut self, udaf: Arc<AggregateUDF>) -> Result<Option<Arc<AggregateUDF>>> {
+    fn register_udaf(
+        &mut self,
+        udaf: Arc<AggregateUDF>,
+    ) -> DataFusionResult<Option<Arc<AggregateUDF>>> {
         Ok(self
             .functions
             .aggregates
             .insert(udaf.name().to_string(), udaf))
     }
 
-    fn register_udwf(&mut self, udwf: Arc<WindowUDF>) -> Result<Option<Arc<WindowUDF>>> {
+    fn register_udwf(&mut self, udwf: Arc<WindowUDF>) -> DataFusionResult<Option<Arc<WindowUDF>>> {
         Ok(self.functions.windows.insert(udwf.name().to_string(), udwf))
     }
 
-    fn register_expr_planner(&mut self, expr_planner: Arc<dyn ExprPlanner>) -> Result<()> {
+    fn register_expr_planner(
+        &mut self,
+        expr_planner: Arc<dyn ExprPlanner>,
+    ) -> DataFusionResult<()> {
         self.functions.planners.push(expr_planner);
         Ok(())
     }
@@ -371,7 +404,7 @@ impl StreamPlanningContextBuilder {
         Self::default()
     }
 
-    pub fn with_default_functions(mut self) -> Result<Self> {
+    pub fn with_default_functions(&mut self) -> Result<&mut Self, PlanningError> {
         for p in SessionStateDefaults::default_scalar_functions() {
             self.context.register_udf(p)?;
         }
@@ -387,7 +420,7 @@ impl StreamPlanningContextBuilder {
         Ok(self)
     }
 
-    pub fn with_streaming_extensions(mut self) -> Result<Self> {
+    pub fn with_streaming_extensions(&mut self) -> Result<&mut Self, PlanningError> {
         let extensions = vec![
             PlanningPlaceholderUdf::new_with_return(
                 window_fn::HOP,
diff --git a/src/sql/schema/source_table.rs b/src/sql/schema/source_table.rs
deleted file mode 100644
index 03eda9ac..00000000
--- a/src/sql/schema/source_table.rs
+++ /dev/null
@@ -1,592 +0,0 @@
-// Licensed under the Apache License, Version 2.0 (the "License");
-// you may not use this file except in compliance with the License.
-// You may obtain a copy of the License at
-//
-//     http://www.apache.org/licenses/LICENSE-2.0
-//
-// Unless required by applicable law or agreed to in writing, software
-// distributed under the License is distributed on an "AS IS" BASIS,
-// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-// See the License for the specific language governing permissions and
-// limitations under the License.
-
-use std::collections::{BTreeMap, HashMap};
-use std::sync::Arc;
-use std::time::Duration;
-
-use datafusion::arrow::datatypes::{DataType, Field, FieldRef, Schema};
-use datafusion::common::{Column, DFSchema, Result, plan_datafusion_err, plan_err};
-use datafusion::error::DataFusionError;
-use datafusion::logical_expr::Expr;
-use datafusion::sql::TableReference;
-use datafusion::sql::planner::{PlannerContext, SqlToRel};
-use datafusion::sql::sqlparser::ast;
-use datafusion_expr::ExprSchemable;
-use protocol::function_stream_graph::ConnectorOp;
-use tracing::warn;
-
-use super::StreamSchemaProvider;
-use super::column_descriptor::ColumnDescriptor;
-use super::connector_config::ConnectorConfig;
-use super::data_encoding_format::DataEncodingFormat;
-use super::schema_context::SchemaContext;
-use super::table_execution_unit::{EngineDescriptor, SyncMode, TableExecutionUnit};
-use super::table_role::{
-    TableRole, apply_adapter_specific_rules, deduce_role, serialize_backend_params,
-    validate_adapter_availability,
-};
-use super::temporal_pipeline_config::{
-    TemporalPipelineConfig, TemporalSpec, resolve_temporal_logic,
-};
-use crate::multifield_partial_ord;
-use crate::sql::api::ConnectionProfile;
-use crate::sql::common::connector_options::ConnectorOptions;
-use crate::sql::common::constants::{connection_table_role, connector_type, sql_field};
-use crate::sql::common::with_option_keys as opt;
-use crate::sql::common::{BadData, Format, Framing, FsSchema, JsonCompression, JsonFormat};
-use crate::sql::schema::ConnectionType;
-use crate::sql::schema::kafka_operator_config::build_kafka_proto_config;
-use crate::sql::schema::table::SqlSource;
-use crate::sql::types::ProcessingMode;
-
-/// Connector-backed catalog table (adapter / source-sink model).
-#[derive(Debug, Clone, PartialEq, Eq, Hash)]
-pub struct SourceTable {
-    pub registry_id: Option<i64>,
-    pub adapter_type: String,
-    pub table_identifier: String,
-    pub role: TableRole,
-    pub schema_specs: Vec<ColumnDescriptor>,
-    /// Strongly-typed connector runtime configuration — replaces the legacy `opaque_config: String`.
-    pub connector_config: ConnectorConfig,
-    pub temporal_config: TemporalPipelineConfig,
-    pub key_constraints: Vec<String>,
-    pub payload_format: Option<DataEncodingFormat>,
-    /// Wire [`Format`] when built from SQL `WITH` (updating mode, `ConnectionSchema`).
-    pub connection_format: Option<Format>,
-    pub description: String,
-    pub partition_exprs: Arc<Option<Vec<Expr>>>,
-    pub lookup_cache_max_bytes: Option<u64>,
-    pub lookup_cache_ttl: Option<Duration>,
-    pub inferred_fields: Option<Vec<FieldRef>>,
-    /// Original `WITH` options for catalog persistence / `SHOW CREATE TABLE`.
-    pub catalog_with_options: BTreeMap<String, String>,
-}
-
-multifield_partial_ord!(
-    SourceTable,
-    registry_id,
-    adapter_type,
-    table_identifier,
-    role,
-    description,
-    key_constraints,
-    connection_format,
-    catalog_with_options
-);
-
-impl SourceTable {
-    #[inline]
-    pub fn name(&self) -> &str {
-        self.table_identifier.as_str()
-    }
-
-    pub fn new(
-        table_identifier: impl Into<String>,
-        connector: impl Into<String>,
-        connection_type: ConnectionType,
-    ) -> Self {
-        Self {
-            registry_id: None,
-            adapter_type: connector.into(),
-            table_identifier: table_identifier.into(),
-            role: connection_type.into(),
-            schema_specs: Vec::new(),
-            connector_config: ConnectorConfig::Generic(HashMap::new()),
-            temporal_config: TemporalPipelineConfig::default(),
-            key_constraints: Vec::new(),
-            payload_format: None,
-            connection_format: None,
-            description: String::new(),
-            partition_exprs: Arc::new(None),
-            lookup_cache_max_bytes: None,
-            lookup_cache_ttl: None,
-            inferred_fields: None,
-            catalog_with_options: BTreeMap::new(),
-        }
-    }
-
-    #[inline]
-    pub fn connector(&self) -> &str {
-        self.adapter_type.as_str()
-    }
-
-    #[inline]
-    pub fn connection_type(&self) -> ConnectionType {
-        self.role.into()
-    }
-
-    pub fn event_time_field(&self) -> Option<&str> {
-        self.temporal_config.event_column.as_deref()
-    }
-
-    pub fn watermark_field(&self) -> Option<&str> {
-        self.temporal_config.watermark_strategy_column.as_deref()
-    }
-
-    /// Watermark column name safe to persist for [`StreamTable::Source`]. Omits the computed
-    /// [`sql_field::COMPUTED_WATERMARK`] column: stream catalog only stores Arrow physical fields,
-    /// so `__watermark` cannot be resolved when the table is planned from the catalog.
-    pub fn stream_catalog_watermark_field(&self) -> Option<String> {
-        self.temporal_config
-            .watermark_strategy_column
-            .as_deref()
-            .filter(|w| *w != sql_field::COMPUTED_WATERMARK)
-            .map(str::to_string)
-    }
-
-    #[inline]
-    pub fn catalog_with_options(&self) -> &BTreeMap<String, String> {
-        &self.catalog_with_options
-    }
-
-    pub fn idle_time(&self) -> Option<Duration> {
-        self.temporal_config.liveness_timeout
-    }
-
-    pub fn initialize_from_params(
-        id: &str,
-        adapter: &str,
-        raw_columns: Vec<ColumnDescriptor>,
-        pk_list: Vec<String>,
-        time_meta: Option<TemporalSpec>,
-        options: &mut HashMap<String, String>,
-        _schema_ctx: &dyn SchemaContext,
-    ) -> Result<Self> {
-        validate_adapter_availability(adapter)?;
-
-        let catalog_with_options: BTreeMap<String, String> = options
-            .iter()
-            .map(|(k, v)| (k.clone(), v.clone()))
-            .collect();
-
-        let encoding = DataEncodingFormat::extract_from_map(options)?;
-
-        let mut refined_columns = apply_adapter_specific_rules(adapter, raw_columns);
-        refined_columns = encoding.apply_envelope(refined_columns)?;
-
-        let temporal_settings = resolve_temporal_logic(&refined_columns, time_meta)?;
-        let _finalized_config = serialize_backend_params(adapter, options)?;
-        let role = deduce_role(options)?;
-
-        if role == TableRole::Ingestion && encoding.supports_delta_updates() && pk_list.is_empty() {
-            return plan_err!("CDC source requires at least one primary key");
-        }
-
-        Ok(Self {
-            registry_id: None,
-            adapter_type: adapter.to_string(),
-            table_identifier: id.to_string(),
-            role,
-            schema_specs: refined_columns,
-            connector_config: ConnectorConfig::Generic(
-                catalog_with_options.clone().into_iter().collect(),
-            ),
-            temporal_config: temporal_settings,
-            key_constraints: pk_list,
-            payload_format: Some(encoding),
-            connection_format: None,
-            description: String::new(),
-            partition_exprs: Arc::new(None),
-            lookup_cache_max_bytes: None,
-            lookup_cache_ttl: None,
-            inferred_fields: None,
-            catalog_with_options,
-        })
-    }
-
-    pub fn produce_physical_schema(&self) -> Schema {
-        Schema::new(
-            self.schema_specs
-                .iter()
-                .filter(|c| !c.is_computed())
-                .map(|c| c.arrow_field().clone())
-                .collect::<Vec<_>>(),
-        )
-    }
-
-    #[inline]
-    pub fn physical_schema(&self) -> Schema {
-        self.produce_physical_schema()
-    }
-
-    pub fn convert_to_execution_unit(&self) -> Result<TableExecutionUnit> {
-        if self.role == TableRole::Egress {
-            return plan_err!("Target [{}] is write-only", self.table_identifier);
-        }
-
-        if self.is_cdc_enabled() && self.schema_specs.iter().any(|c| c.is_computed()) {
-            return plan_err!("CDC cannot be mixed with computed columns natively");
-        }
-
-        let mode = if self.is_cdc_enabled() {
-            SyncMode::Incremental
-        } else {
-            SyncMode::AppendOnly
-        };
-
-        Ok(TableExecutionUnit {
-            label: self.table_identifier.clone(),
-            engine_meta: EngineDescriptor {
-                engine_type: self.adapter_type.clone(),
-                raw_payload: String::new(),
-            },
-            sync_mode: mode,
-            temporal_offset: self.temporal_config.clone(),
-        })
-    }
-
-    #[inline]
-    pub fn to_execution_unit(&self) -> Result<TableExecutionUnit> {
-        self.convert_to_execution_unit()
-    }
-
-    fn is_cdc_enabled(&self) -> bool {
-        self.payload_format
-            .as_ref()
-            .is_some_and(|f| f.supports_delta_updates())
-    }
-
-    #[allow(clippy::too_many_arguments)]
-    pub fn from_options(
-        table_identifier: &str,
-        connector_name: &str,
-        temporary: bool,
-        fields: Vec<ColumnDescriptor>,
-        primary_keys: Vec<String>,
-        watermark: Option<(String, Option<ast::Expr>)>,
-        options: &mut ConnectorOptions,
-        connection_profile: Option<&ConnectionProfile>,
-        schema_provider: &StreamSchemaProvider,
-        connection_type_override: Option<ConnectionType>,
-        description: String,
-    ) -> Result<Self> {
-        let _ = connection_profile;
-
-        let catalog_with_options = options.snapshot_for_catalog();
-
-        if let Some(c) = options.pull_opt_str(opt::CONNECTOR)?
-            && c != connector_name
-        {
-            return plan_err!(
-                "WITH option `connector` is '{c}' but table uses connector '{connector_name}'"
-            );
-        }
-
-        validate_adapter_availability(connector_name)?;
-
-        let mut columns = fields;
-        columns = apply_adapter_specific_rules(connector_name, columns);
-
-        let format = Format::from_opts(options)
-            .map_err(|e| DataFusionError::Plan(format!("invalid format: '{e}'")))?;
-
-        if let Some(Format::Json(JsonFormat { compression, .. })) = &format
-            && !matches!(compression, JsonCompression::Uncompressed)
-            && connector_name != connector_type::FILESYSTEM
-        {
-            return plan_err!("'json.compression' is only supported for the filesystem connector");
-        }
-
-        let _framing = Framing::from_opts(options)
-            .map_err(|e| DataFusionError::Plan(format!("invalid framing: '{e}'")))?;
-
-        if temporary
-            && let Some(t) = options.insert_str(opt::TYPE, connection_table_role::LOOKUP)?
-            && t != connection_table_role::LOOKUP
-        {
-            return plan_err!(
-                "Cannot have a temporary table with type '{t}'; temporary tables must be type 'lookup'"
-            );
-        }
-
-        let payload_format = format
-            .as_ref()
-            .map(DataEncodingFormat::from_connection_format);
-        let encoding = payload_format.unwrap_or(DataEncodingFormat::Raw);
-        columns = encoding.apply_envelope(columns)?;
-
-        let bad_data = BadData::from_opts(options)
-            .map_err(|e| DataFusionError::Plan(format!("Invalid bad_data: '{e}'")))?;
-
-        let role = if let Some(t) = connection_type_override {
-            t.into()
-        } else {
-            match options.pull_opt_str(opt::TYPE)?.as_deref() {
-                None | Some(connection_table_role::SOURCE) => TableRole::Ingestion,
-                Some(connection_table_role::SINK) => TableRole::Egress,
-                Some(connection_table_role::LOOKUP) => TableRole::Reference,
-                Some(other) => {
-                    return plan_err!("invalid connection type '{other}' in WITH options");
-                }
-            }
-        };
-
-        let mut table = SourceTable {
-            registry_id: None,
-            adapter_type: connector_name.to_string(),
-            table_identifier: table_identifier.to_string(),
-            role,
-            schema_specs: columns,
-            connector_config: ConnectorConfig::Generic(HashMap::new()),
-            temporal_config: TemporalPipelineConfig::default(),
-            key_constraints: Vec::new(),
-            payload_format,
-            connection_format: format.clone(),
-            description,
-            partition_exprs: Arc::new(None),
-            lookup_cache_max_bytes: None,
-            lookup_cache_ttl: None,
-            inferred_fields: None,
-            catalog_with_options,
-        };
-
-        if let Some(event_time_field) = options.pull_opt_field(opt::EVENT_TIME_FIELD)? {
-            warn!("`event_time_field` WITH option is deprecated; use WATERMARK FOR syntax");
-            table.temporal_config.event_column = Some(event_time_field);
-        }
-
-        if let Some(watermark_field) = options.pull_opt_field(opt::WATERMARK_FIELD)? {
-            warn!("`watermark_field` WITH option is deprecated; use WATERMARK FOR syntax");
-            table.temporal_config.watermark_strategy_column = Some(watermark_field);
-        }
-
-        if let Some((time_field, watermark_expr)) = watermark {
-            let field = table
-                .schema_specs
-                .iter()
-                .find(|c| c.arrow_field().name().as_str() == time_field.as_str())
-                .ok_or_else(|| {
-                    plan_datafusion_err!(
-                        "WATERMARK FOR field `{}` does not exist in table",
-                        time_field
-                    )
-                })?;
-
-            if !matches!(
-                field.arrow_field().data_type(),
-                DataType::Timestamp(_, None)
-            ) {
-                return plan_err!(
-                    "WATERMARK FOR field `{time_field}` has type {}, but expected TIMESTAMP",
-                    field.arrow_field().data_type()
-                );
-            }
-
-            for col in table.schema_specs.iter_mut() {
-                if col.arrow_field().name().as_str() == time_field.as_str() {
-                    col.set_nullable(false);
-                    break;
-                }
-            }
-
-            let table_ref = TableReference::bare(table.table_identifier.as_str());
-            let df_schema =
-                DFSchema::try_from_qualified_schema(table_ref, &table.produce_physical_schema())?;
-
-            table.temporal_config.event_column = Some(time_field.clone());
-
-            if let Some(expr) = watermark_expr {
-                let logical_expr = plan_generating_expr(&expr, &df_schema, schema_provider)
-                    .map_err(|e| {
-                        DataFusionError::Plan(format!("could not plan watermark expression: {e}"))
-                    })?;
-
-                let (data_type, _nullable) = logical_expr.data_type_and_nullable(&df_schema)?;
-                if !matches!(data_type, DataType::Timestamp(_, _)) {
-                    return plan_err!(
-                        "the type of the WATERMARK FOR expression must be TIMESTAMP, but was {data_type}"
-                    );
-                }
-
-                table.schema_specs.push(ColumnDescriptor::new_computed(
-                    Field::new(
-                        sql_field::COMPUTED_WATERMARK,
-                        logical_expr.get_type(&df_schema)?,
-                        false,
-                    ),
-                    logical_expr,
-                ));
-                table.temporal_config.watermark_strategy_column =
-                    Some(sql_field::COMPUTED_WATERMARK.to_string());
-            } else {
-                table.temporal_config.watermark_strategy_column = Some(time_field);
-            }
-        }
-
-        let idle_from_micros = options
-            .pull_opt_i64(opt::IDLE_MICROS)?
-            .filter(|t| *t > 0)
-            .map(|t| Duration::from_micros(t as u64));
-        let idle_from_duration = options.pull_opt_duration(opt::IDLE_TIME)?;
-        table.temporal_config.liveness_timeout = idle_from_micros.or(idle_from_duration);
-
-        table.lookup_cache_max_bytes = options.pull_opt_u64(opt::LOOKUP_CACHE_MAX_BYTES)?;
-
-        table.lookup_cache_ttl = options.pull_opt_duration(opt::LOOKUP_CACHE_TTL)?;
-
-        if connector_name.eq_ignore_ascii_case(connector_type::KAFKA) {
-            let proto_cfg = build_kafka_proto_config(options, role, &format, bad_data)?;
-            table.connector_config = match proto_cfg {
-                protocol::function_stream_graph::connector_op::Config::KafkaSource(cfg) => {
-                    ConnectorConfig::KafkaSource(cfg)
-                }
-                protocol::function_stream_graph::connector_op::Config::KafkaSink(cfg) => {
-                    ConnectorConfig::KafkaSink(cfg)
-                }
-                protocol::function_stream_graph::connector_op::Config::Generic(g) => {
-                    ConnectorConfig::Generic(g.properties)
-                }
-            };
-        } else {
-            let extra_opts = options.drain_remaining_string_values()?;
-            table.connector_config = ConnectorConfig::Generic(extra_opts);
-        }
-
-        if role == TableRole::Ingestion
-            && encoding.supports_delta_updates()
-            && primary_keys.is_empty()
-        {
-            return plan_err!("Debezium source must have at least one PRIMARY KEY field");
-        }
-
-        table.key_constraints = primary_keys;
-
-        Ok(table)
-    }
-
-    pub fn has_virtual_fields(&self) -> bool {
-        self.schema_specs.iter().any(|c| c.is_computed())
-    }
-
-    pub fn is_updating(&self) -> bool {
-        self.connection_format
-            .as_ref()
-            .is_some_and(|f| f.is_updating())
-            || self.payload_format == Some(DataEncodingFormat::DebeziumJson)
-    }
-
-    pub fn connector_op(&self) -> ConnectorOp {
-        let physical = self.produce_physical_schema();
-        let fields: Vec<Field> = physical
-            .fields()
-            .iter()
-            .map(|f| f.as_ref().clone())
-            .collect();
-        let fs_schema = FsSchema::from_fields(fields);
-
-        ConnectorOp {
-            connector: self.adapter_type.clone(),
-            fs_schema: Some(fs_schema.into()),
-            name: self.table_identifier.clone(),
-            description: self.description.clone(),
-            config: Some(self.connector_config.to_proto_config()),
-        }
-    }
-
-    pub fn processing_mode(&self) -> ProcessingMode {
-        if self.is_updating() {
-            ProcessingMode::Update
-        } else {
-            ProcessingMode::Append
-        }
-    }
-
-    pub fn timestamp_override(&self) -> Result<Option<Expr>> {
-        if let Some(field_name) = self.temporal_config.event_column.clone() {
-            if self.is_updating() {
-                return plan_err!("can't use event_time_field with update mode");
-            }
-            let _field = self.get_time_column(&field_name)?;
-            Ok(Some(Expr::Column(Column::from_name(field_name.as_str()))))
-        } else {
-            Ok(None)
-        }
-    }
-
-    fn get_time_column(&self, field_name: &str) -> Result<&ColumnDescriptor> {
-        self.schema_specs
-            .iter()
-            .find(|c| {
-                c.arrow_field().name() == field_name
-                    && matches!(c.arrow_field().data_type(), DataType::Timestamp(..))
-            })
-            .ok_or_else(|| {
-                DataFusionError::Plan(format!("field {field_name} not found or not a timestamp"))
-            })
-    }
-
-    pub fn watermark_column(&self) -> Result<Option<Expr>> {
-        if let Some(field_name) = self.temporal_config.watermark_strategy_column.clone() {
-            let _field = self.get_time_column(&field_name)?;
-            Ok(Some(Expr::Column(Column::from_name(field_name.as_str()))))
-        } else {
-            Ok(None)
-        }
-    }
-
-    pub fn as_sql_source(&self) -> Result<SourceOperator> {
-        match self.role {
-            TableRole::Ingestion => {}
-            TableRole::Egress | TableRole::Reference => {
-                return plan_err!("cannot read from sink");
-            }
-        };
-
-        if self.is_updating() && self.has_virtual_fields() {
-            return plan_err!("can't read from a source with virtual fields and update mode.");
-        }
-
-        let timestamp_override = self.timestamp_override()?;
-        let watermark_column = self.watermark_column()?;
-
-        let source = SqlSource {
-            id: self.registry_id,
-            struct_def: self
-                .schema_specs
-                .iter()
-                .filter(|c| !c.is_computed())
-                .map(|c| Arc::new(c.arrow_field().clone()))
-                .collect(),
-            config: self.connector_op(),
-            processing_mode: self.processing_mode(),
-            idle_time: self.temporal_config.liveness_timeout,
-        };
-
-        Ok(SourceOperator {
-            name: self.table_identifier.clone(),
-            source,
-            timestamp_override,
-            watermark_column,
-        })
-    }
-}
-
-/// Plan a SQL scalar expression against a table-qualified schema (e.g. watermark `AS` clause).
-fn plan_generating_expr(
-    ast: &ast::Expr,
-    df_schema: &DFSchema,
-    schema_provider: &StreamSchemaProvider,
-) -> Result<Expr> {
-    let planner = SqlToRel::new(schema_provider);
-    let mut ctx = PlannerContext::new();
-    planner.sql_to_expr(ast.clone(), df_schema, &mut ctx)
-}
-
-#[derive(Debug, Clone)]
-pub struct SourceOperator {
-    pub name: String,
-    pub source: SqlSource,
-    pub timestamp_override: Option<Expr>,
-    pub watermark_column: Option<Expr>,
-}
diff --git a/src/sql/schema/table.rs b/src/sql/schema/table.rs
index efa0c59f..6c001d9c 100644
--- a/src/sql/schema/table.rs
+++ b/src/sql/schema/table.rs
@@ -10,11 +10,11 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
-use super::source_table::SourceTable;
 use crate::sql::analysis::rewrite_plan;
 use crate::sql::logical_node::remote_table::RemoteTableBoundaryNode;
 use crate::sql::logical_planner::optimizers::produce_optimized_plan;
 use crate::sql::schema::StreamSchemaProvider;
+use crate::sql::schema::catalog::ExternalTable;
 use crate::sql::types::{ProcessingMode, QualifiedField};
 use datafusion::arrow::datatypes::FieldRef;
 use datafusion::common::{Result, plan_err};
@@ -24,23 +24,22 @@ use protocol::function_stream_graph::ConnectorOp;
 use std::sync::Arc;
 use std::time::Duration;
 
-/// Represents all table types in the FunctionStream SQL catalog.
-#[allow(clippy::enum_variant_names)]
 #[derive(Debug, Clone, PartialEq, Eq, Hash)]
-pub enum Table {
-    /// A lookup table backed by an external connector.
-    LookupTable(SourceTable),
-    /// A source/sink table backed by an external connector.
-    ConnectorTable(SourceTable),
-    /// A table defined by a query (CREATE VIEW / CREATE TABLE AS SELECT).
-    TableFromQuery {
+pub enum CatalogEntity {
+    /// Both payload variants are boxed so the enum is not padded to the largest field.
+    ExternalConnector(Box<ExternalTable>),
+    ComputedTable {
         name: String,
-        logical_plan: LogicalPlan,
+        logical_plan: Box<LogicalPlan>,
     },
 }
 
-impl Table {
-    /// Try to construct a Table from a CREATE TABLE or CREATE VIEW statement.
+impl CatalogEntity {
+    #[inline]
+    pub fn external(table: ExternalTable) -> Self {
+        Self::ExternalConnector(Box::new(table))
+    }
+
     pub fn try_from_statement(
         statement: &Statement,
         schema_provider: &StreamSchemaProvider,
@@ -69,11 +68,11 @@ impl Table {
                     resolved_schema: schema,
                     requires_materialization: true,
                 };
-                Ok(Some(Table::TableFromQuery {
+                Ok(Some(CatalogEntity::ComputedTable {
                     name: name.to_string(),
-                    logical_plan: LogicalPlan::Extension(Extension {
+                    logical_plan: Box::new(LogicalPlan::Extension(Extension {
                         node: Arc::new(remote),
-                    }),
+                    })),
                 }))
             }
             _ => Ok(None),
@@ -82,36 +81,25 @@ impl Table {
 
     pub fn name(&self) -> &str {
         match self {
-            Table::TableFromQuery { name, .. } => name.as_str(),
-            Table::ConnectorTable(c) | Table::LookupTable(c) => c.name(),
+            CatalogEntity::ComputedTable { name, .. } => name.as_str(),
+            CatalogEntity::ExternalConnector(e) => e.name(),
         }
     }
 
     pub fn get_fields(&self) -> Vec<FieldRef> {
         match self {
-            Table::ConnectorTable(SourceTable {
-                schema_specs,
-                inferred_fields,
-                ..
-            })
-            | Table::LookupTable(SourceTable {
-                schema_specs,
-                inferred_fields,
-                ..
-            }) => inferred_fields.clone().unwrap_or_else(|| {
-                schema_specs
-                    .iter()
-                    .map(|c| Arc::new(c.arrow_field().clone()))
-                    .collect()
-            }),
-            Table::TableFromQuery { logical_plan, .. } => {
+            CatalogEntity::ExternalConnector(e) => e.effective_fields(),
+            CatalogEntity::ComputedTable { logical_plan, .. } => {
                 logical_plan.schema().fields().iter().cloned().collect()
             }
         }
     }
 
     pub fn set_inferred_fields(&mut self, fields: Vec<QualifiedField>) -> Result<()> {
-        let Table::ConnectorTable(t) = self else {
+        let CatalogEntity::ExternalConnector(ext) = self else {
+            return Ok(());
+        };
+        let ExternalTable::Source(t) = ext.as_mut() else {
             return Ok(());
         };
 
@@ -139,14 +127,27 @@ impl Table {
 
     pub fn connector_op(&self) -> Result<ConnectorOp> {
         match self {
-            Table::ConnectorTable(c) | Table::LookupTable(c) => Ok(c.connector_op()),
-            Table::TableFromQuery { .. } => plan_err!("can't write to a query-defined table"),
+            CatalogEntity::ExternalConnector(e) => Ok(e.connector_op()),
+            CatalogEntity::ComputedTable { .. } => {
+                plan_err!("can't write to a query-defined table")
+            }
         }
     }
 
     pub fn partition_exprs(&self) -> Option<&Vec<datafusion::logical_expr::Expr>> {
+        let CatalogEntity::ExternalConnector(ext) = self else {
+            return None;
+        };
+        let ExternalTable::Sink(s) = ext.as_ref() else {
+            return None;
+        };
+        (*s.partition_exprs).as_ref()
+    }
+
+    #[inline]
+    pub fn as_external(&self) -> Option<&ExternalTable> {
         match self {
-            Table::ConnectorTable(c) => (*c.partition_exprs).as_ref(),
+            CatalogEntity::ExternalConnector(e) => Some(e.as_ref()),
             _ => None,
         }
     }
diff --git a/src/storage/mod.rs b/src/storage/mod.rs
index 823425d2..ec32bdfa 100644
--- a/src/storage/mod.rs
+++ b/src/storage/mod.rs
@@ -10,6 +10,44 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
+use std::sync::Arc;
+
+use anyhow::Context;
+
 pub mod state_backend;
 pub mod stream_catalog;
 pub mod task;
+
+/// Install the process-global [`stream_catalog::CatalogManager`] from configuration.
+/// In-memory when `config.stream_catalog.persist` is `false`, otherwise a durable
+/// [`stream_catalog::RocksDbMetaStore`] (default path: `{data_dir}/catalog.db`).
+pub fn initialize_stream_catalog(config: &crate::config::GlobalConfig) -> anyhow::Result<()> {
+    use stream_catalog::{CatalogManager, InMemoryMetaStore, MetaStore, RocksDbMetaStore};
+
+    let store: Arc<dyn MetaStore> = if !config.stream_catalog.persist {
+        Arc::new(InMemoryMetaStore::new())
+    } else {
+        let path = config
+            .stream_catalog
+            .db_path
+            .as_ref()
+            .map(|p| crate::config::resolve_path(p))
+            .unwrap_or_else(|| crate::config::get_data_dir().join("catalog.db"));
+
+        std::fs::create_dir_all(&path).with_context(|| {
+            format!(
+                "Failed to create stream catalog RocksDB directory {}",
+                path.display()
+            )
+        })?;
+
+        Arc::new(RocksDbMetaStore::open(&path).with_context(|| {
+            format!(
+                "Failed to open stream catalog RocksDB at {}",
+                path.display()
+            )
+        })?)
+    };
+
+    CatalogManager::init_global(store).context("Stream catalog (CatalogManager) global init failed")
+}
diff --git a/src/storage/stream_catalog/manager.rs b/src/storage/stream_catalog/manager.rs
index 3c9d561e..fa810e8d 100644
--- a/src/storage/stream_catalog/manager.rs
+++ b/src/storage/stream_catalog/manager.rs
@@ -10,24 +10,25 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
-use std::path::Path;
+use std::collections::BTreeMap;
 use std::sync::{Arc, OnceLock};
 
-use anyhow::{Context, anyhow, bail};
+use anyhow::{anyhow, bail};
 use datafusion::common::{Result as DFResult, internal_err, plan_err};
 use prost::Message;
 use protocol::function_stream_graph::FsProgram;
 use protocol::storage::{self as pb, table_definition};
 use tracing::{debug, info, warn};
 
-use crate::runtime::streaming::operators::source::kafka as kafka_snap;
 use unicase::UniCase;
 
 use crate::sql::common::constants::sql_field;
+use crate::sql::connector::config::ConnectorConfig;
+use crate::sql::schema::catalog::{ExternalTable, LookupTable, SourceTable};
 use crate::sql::schema::column_descriptor::ColumnDescriptor;
-use crate::sql::schema::connection_type::ConnectionType;
-use crate::sql::schema::source_table::SourceTable;
-use crate::sql::schema::table::Table as CatalogTable;
+use crate::sql::schema::table::CatalogEntity;
+use crate::sql::schema::table_role::TableRole;
+use crate::sql::schema::temporal_pipeline_config::TemporalPipelineConfig;
 use crate::sql::schema::{StreamPlanningContext, StreamTable};
 
 use super::codec::CatalogCodec;
@@ -36,91 +37,16 @@ use super::meta_store::MetaStore;
 const CATALOG_KEY_PREFIX: &str = "catalog:stream_table:";
 const STREAMING_JOB_KEY_PREFIX: &str = "streaming_job:";
 
-/// One persisted streaming job row from catalog (program + checkpoint metadata + Kafka offsets).
+/// One persisted streaming job row from catalog (program + checkpoint metadata).
 #[derive(Debug, Clone)]
 pub struct StoredStreamingJob {
     pub table_name: String,
     pub program: FsProgram,
     pub checkpoint_interval_ms: u64,
     pub latest_checkpoint_epoch: u64,
-    pub kafka_source_checkpoints: Vec<pb::KafkaSourceSubtaskCheckpoint>,
-}
-
-fn parse_kafka_offset_snapshot_filename(name: &str) -> Option<(u32, u32)> {
-    const PREFIX: &str = "kafka_source_offsets_pipe";
-    const SUFFIX: &str = ".bin";
-    if !name.starts_with(PREFIX) || !name.ends_with(SUFFIX) {
-        return None;
-    }
-    let mid = name.strip_prefix(PREFIX)?.strip_suffix(SUFFIX)?;
-    let (pipe, sub_part) = mid.split_once("_sub")?;
-    Some((pipe.parse().ok()?, sub_part.parse().ok()?))
-}
-
-/// Removes on-disk staging snapshots once their payload is committed into catalog (same epoch).
-fn cleanup_kafka_offset_snapshots_for_epoch(job_dir: &Path, epoch: u64) {
-    let Ok(rd) = std::fs::read_dir(job_dir) else {
-        return;
-    };
-    for ent in rd.flatten() {
-        let path = ent.path();
-        let name = ent.file_name().to_string_lossy().into_owned();
-        if parse_kafka_offset_snapshot_filename(&name).is_none() {
-            continue;
-        }
-        let Ok(bytes) = std::fs::read(&path) else {
-            continue;
-        };
-        let Ok(saved) = kafka_snap::decode_kafka_offset_snapshot(&bytes) else {
-            continue;
-        };
-        if saved.epoch == epoch && std::fs::remove_file(&path).is_err() {
-            debug!(path = %path.display(), "Could not remove staged Kafka offset snapshot (non-fatal)");
-        }
-    }
-}
-
-/// Writes catalog-stored Kafka checkpoints back to the job state dir before `submit_job` resumes sources.
-pub fn materialize_kafka_source_checkpoints_from_catalog(
-    job_dir: &Path,
-    checkpoints: &[pb::KafkaSourceSubtaskCheckpoint],
-) -> DFResult<()> {
-    if checkpoints.is_empty() {
-        return Ok(());
-    }
-    std::fs::create_dir_all(job_dir).map_err(|e| {
-        datafusion::common::DataFusionError::Execution(format!(
-            "create job state dir {}: {e}",
-            job_dir.display()
-        ))
-    })?;
-    for c in checkpoints {
-        let saved = kafka_snap::KafkaSourceSavedOffsets {
-            epoch: c.checkpoint_epoch,
-            partitions: c
-                .partitions
-                .iter()
-                .map(|p| kafka_snap::KafkaState {
-                    partition: p.partition,
-                    offset: p.offset,
-                })
-                .collect(),
-        };
-        let path = kafka_snap::kafka_snapshot_path(job_dir, c.pipeline_id, c.subtask_index);
-        let bytes = kafka_snap::encode_kafka_offset_snapshot(&saved).map_err(|e| {
-            datafusion::common::DataFusionError::Execution(format!(
-                "encode kafka snapshot for {}: {e}",
-                path.display()
-            ))
-        })?;
-        std::fs::write(&path, &bytes).map_err(|e| {
-            datafusion::common::DataFusionError::Execution(format!(
-                "write kafka snapshot {}: {e}",
-                path.display()
-            ))
-        })?;
-    }
-    Ok(())
+    /// Source-type-agnostic per-subtask checkpoint entries. Each entry is a
+    /// [`pb::SourceCheckpointInfo`] oneof envelope — the catalog does not inspect the payload.
+    pub source_checkpoints: Vec<pb::SourceCheckpointInfo>,
 }
 
 pub struct CatalogManager {
@@ -188,7 +114,7 @@ impl CatalogManager {
             comment: comment.to_string(),
             checkpoint_interval_ms,
             latest_checkpoint_epoch: 0,
-            kafka_source_checkpoints: vec![],
+            source_checkpoints: vec![],
         };
         let payload = def.encode_to_vec();
         let key = Self::build_streaming_job_key(table_name);
@@ -207,16 +133,14 @@ impl CatalogManager {
     /// Persist the globally-completed checkpoint epoch after all operators ACK.
     /// Only advances forward; stale epochs are silently ignored.
     ///
-    /// `kafka_source_checkpoints` is assembled by the job coordinator from source pipeline checkpoint
-    /// ACKs (in-memory); it is stored next to `latest_checkpoint_epoch` in the catalog.
-    ///
-    /// `job_state_dir` is only used to remove legacy on-disk staging snapshots for this epoch, if present.
+    /// `source_checkpoints` is the source-agnostic list assembled by the job coordinator via
+    /// [`CheckpointAggregatorRegistry::aggregate_all`]; it is stored atomically next to
+    /// `latest_checkpoint_epoch` via [`MetaStore::write_batch`].
     pub fn commit_job_checkpoint(
         &self,
         table_name: &str,
         epoch: u64,
-        job_state_dir: &Path,
-        kafka_source_checkpoints: Vec<pb::KafkaSourceSubtaskCheckpoint>,
+        source_checkpoints: Vec<pb::SourceCheckpointInfo>,
     ) -> DFResult<()> {
         let key = Self::build_streaming_job_key(table_name);
 
@@ -237,22 +161,21 @@ impl CatalogManager {
 
         if epoch > def.latest_checkpoint_epoch {
             def.latest_checkpoint_epoch = epoch;
-            def.kafka_source_checkpoints = kafka_source_checkpoints;
-            let new_payload = def.encode_to_vec();
-            self.store.put(&key, new_payload)?;
+            def.source_checkpoints = source_checkpoints;
+            self.store
+                .write_batch(vec![(key, Some(def.encode_to_vec()))])?;
             debug!(
                 table = %table_name,
                 epoch = epoch,
-                kafka_subtasks = def.kafka_source_checkpoints.len(),
-                "Checkpoint metadata committed to Catalog"
+                source_subtasks = def.source_checkpoints.len(),
+                "Checkpoint metadata committed to Catalog (write_batch)"
             );
-            cleanup_kafka_offset_snapshots_for_epoch(job_state_dir, epoch);
         }
 
         Ok(())
     }
 
-    /// Load all persisted streaming jobs (including Kafka offset checkpoints for restore).
+    /// Load all persisted streaming jobs (including source checkpoint data for restore).
     pub fn load_streaming_job_definitions(&self) -> DFResult<Vec<StoredStreamingJob>> {
         let records = self.store.scan_prefix(STREAMING_JOB_KEY_PREFIX)?;
         let mut out = Vec::with_capacity(records.len());
@@ -284,7 +207,7 @@ impl CatalogManager {
                 program,
                 checkpoint_interval_ms: def.checkpoint_interval_ms,
                 latest_checkpoint_epoch: def.latest_checkpoint_epoch,
-                kafka_source_checkpoints: def.kafka_source_checkpoints,
+                source_checkpoints: def.source_checkpoints,
             });
         }
         Ok(out)
@@ -294,7 +217,7 @@ impl CatalogManager {
     // Catalog table persistence (CREATE TABLE / DROP TABLE)
     // ========================================================================
 
-    pub fn add_catalog_table(&self, table: CatalogTable) -> DFResult<()> {
+    pub fn add_catalog_table(&self, table: CatalogEntity) -> DFResult<()> {
         let proto_def = self.encode_catalog_table(&table)?;
         let payload = proto_def.encode_to_vec();
         let key = Self::build_store_key(table.name());
@@ -332,36 +255,44 @@ impl CatalogManager {
         ctx.tables.catalogs = catalogs.clone();
 
         for (name, table) in catalogs {
-            let source = match table.as_ref() {
-                CatalogTable::ConnectorTable(s) | CatalogTable::LookupTable(s) => s,
-                CatalogTable::TableFromQuery { .. } => continue,
+            let stream = match table.as_ref() {
+                CatalogEntity::ExternalConnector(b) => match b.as_ref() {
+                    ExternalTable::Source(s) => Some(StreamTable::Source {
+                        name: s.name().to_string(),
+                        connector: s.connector().to_string(),
+                        schema: Arc::new(s.produce_physical_schema()),
+                        event_time_field: s.event_time_field().map(str::to_string),
+                        watermark_field: s.stream_catalog_watermark_field(),
+                        with_options: s.catalog_with_options().clone(),
+                    }),
+                    ExternalTable::Lookup(l) => Some(StreamTable::Source {
+                        name: l.name().to_string(),
+                        connector: l.connector().to_string(),
+                        schema: Arc::new(l.produce_physical_schema()),
+                        event_time_field: None,
+                        watermark_field: None,
+                        with_options: l.catalog_with_options().clone(),
+                    }),
+                    ExternalTable::Sink(_) => None,
+                },
+                CatalogEntity::ComputedTable { .. } => None,
             };
-
-            let schema = Arc::new(source.produce_physical_schema());
-            ctx.tables.streams.insert(
-                name,
-                Arc::new(StreamTable::Source {
-                    name: source.name().to_string(),
-                    connector: source.connector().to_string(),
-                    schema,
-                    event_time_field: source.event_time_field().map(str::to_string),
-                    watermark_field: source.stream_catalog_watermark_field(),
-                    with_options: source.catalog_with_options().clone(),
-                }),
-            );
+            if let Some(st) = stream {
+                ctx.tables.streams.insert(name, Arc::new(st));
+            }
         }
         ctx
     }
 
     /// All persisted catalog tables, sorted by table name.
-    pub fn list_catalog_tables(&self) -> DFResult<Vec<Arc<CatalogTable>>> {
-        let mut out: Vec<Arc<CatalogTable>> =
+    pub fn list_catalog_tables(&self) -> DFResult<Vec<Arc<CatalogEntity>>> {
+        let mut out: Vec<Arc<CatalogEntity>> =
             self.load_catalog_tables_map()?.into_values().collect();
         out.sort_by(|a, b| a.name().cmp(b.name()));
         Ok(out)
     }
 
-    pub fn get_catalog_table(&self, name: &str) -> DFResult<Option<Arc<CatalogTable>>> {
+    pub fn get_catalog_table(&self, name: &str) -> DFResult<Option<Arc<CatalogEntity>>> {
         let key = UniCase::new(name.to_string());
         Ok(self.load_catalog_tables_map()?.get(&key).cloned())
     }
@@ -376,17 +307,40 @@ impl CatalogManager {
                 watermark_field,
                 with_options,
             } => {
-                let mut source = SourceTable::new(name, connector, ConnectionType::Source);
-                source.schema_specs = schema
+                let schema_specs: Vec<ColumnDescriptor> = schema
                     .fields()
                     .iter()
                     .map(|f| ColumnDescriptor::new_physical((**f).clone()))
                     .collect();
-                source.inferred_fields = Some(schema.fields().iter().cloned().collect());
-                source.temporal_config.event_column = event_time_field;
-                source.temporal_config.watermark_strategy_column = watermark_field;
-                source.catalog_with_options = with_options;
-                self.add_catalog_table(CatalogTable::ConnectorTable(source))
+                let inferred_fields = Some(schema.fields().iter().cloned().collect());
+                let physical_schema = schema.as_ref().clone();
+
+                let connector_config = build_connector_config_for_role(
+                    &connector,
+                    TableRole::Ingestion,
+                    &with_options,
+                    &physical_schema,
+                )?;
+
+                let source = SourceTable {
+                    table_identifier: name,
+                    adapter_type: connector,
+                    schema_specs,
+                    connector_config,
+                    temporal_config: TemporalPipelineConfig {
+                        event_column: event_time_field,
+                        watermark_strategy_column: watermark_field,
+                        liveness_timeout: None,
+                    },
+                    key_constraints: Vec::new(),
+                    payload_format: None,
+                    connection_format: None,
+                    description: String::new(),
+                    catalog_with_options: with_options.into_iter().collect(),
+                    registry_id: None,
+                    inferred_fields,
+                };
+                self.add_catalog_table(CatalogEntity::external(ExternalTable::Source(source)))
             }
             StreamTable::Sink { name, .. } => plan_err!(
                 "Persisting streaming sink '{name}' in stream catalog is no longer supported"
@@ -406,19 +360,7 @@ impl CatalogManager {
         self.list_catalog_tables()
             .unwrap_or_default()
             .into_iter()
-            .filter_map(|t| match t.as_ref() {
-                CatalogTable::ConnectorTable(s) | CatalogTable::LookupTable(s) => {
-                    Some(Arc::new(StreamTable::Source {
-                        name: s.name().to_string(),
-                        connector: s.connector().to_string(),
-                        schema: Arc::new(s.produce_physical_schema()),
-                        event_time_field: s.event_time_field().map(str::to_string),
-                        watermark_field: s.stream_catalog_watermark_field(),
-                        with_options: s.catalog_with_options().clone(),
-                    }))
-                }
-                CatalogTable::TableFromQuery { .. } => None,
-            })
+            .filter_map(|t| external_to_stream_table(t.as_ref()).map(Arc::new))
             .collect()
     }
 
@@ -426,44 +368,54 @@ impl CatalogManager {
         self.get_catalog_table(name)
             .ok()
             .flatten()
-            .and_then(|t| match t.as_ref() {
-                CatalogTable::ConnectorTable(s) | CatalogTable::LookupTable(s) => {
-                    Some(Arc::new(StreamTable::Source {
-                        name: s.name().to_string(),
-                        connector: s.connector().to_string(),
-                        schema: Arc::new(s.produce_physical_schema()),
-                        event_time_field: s.event_time_field().map(str::to_string),
-                        watermark_field: s.stream_catalog_watermark_field(),
-                        with_options: s.catalog_with_options().clone(),
-                    }))
-                }
-                CatalogTable::TableFromQuery { .. } => None,
-            })
+            .and_then(|t| external_to_stream_table(t.as_ref()).map(Arc::new))
     }
 
-    fn encode_catalog_table(&self, table: &CatalogTable) -> DFResult<pb::TableDefinition> {
+    fn encode_catalog_table(&self, table: &CatalogEntity) -> DFResult<pb::TableDefinition> {
         let table_type = match table {
-            CatalogTable::ConnectorTable(source) | CatalogTable::LookupTable(source) => {
-                let mut opts = source.catalog_with_options().clone();
-                opts.entry("connector".to_string())
-                    .or_insert_with(|| source.connector().to_string());
-                let catalog_row = pb::CatalogSourceTable {
-                    arrow_schema_ipc: CatalogCodec::encode_schema(&Arc::new(
-                        source.produce_physical_schema(),
-                    ))?,
-                    event_time_field: source.event_time_field().map(str::to_string),
-                    watermark_field: source.stream_catalog_watermark_field(),
-                    with_options: opts.into_iter().collect(),
-                    connector: source.connector().to_string(),
-                    description: source.description.clone(),
-                };
-                if matches!(table, CatalogTable::LookupTable(_)) {
-                    table_definition::TableType::LookupTable(catalog_row)
-                } else {
+            CatalogEntity::ExternalConnector(b) => match b.as_ref() {
+                ExternalTable::Source(source) => {
+                    let mut opts: std::collections::HashMap<String, String> =
+                        source.catalog_with_options.clone().into_iter().collect();
+                    opts.entry("connector".to_string())
+                        .or_insert_with(|| source.connector().to_string());
+                    let catalog_row = pb::CatalogSourceTable {
+                        arrow_schema_ipc: CatalogCodec::encode_schema(&Arc::new(
+                            source.produce_physical_schema(),
+                        ))?,
+                        event_time_field: source.event_time_field().map(str::to_string),
+                        watermark_field: source.stream_catalog_watermark_field(),
+                        with_options: opts,
+                        connector: source.connector().to_string(),
+                        description: source.description.clone(),
+                    };
                     table_definition::TableType::ConnectorTable(catalog_row)
                 }
-            }
-            CatalogTable::TableFromQuery { name, .. } => {
+                ExternalTable::Lookup(lookup) => {
+                    let mut opts: std::collections::HashMap<String, String> =
+                        lookup.catalog_with_options.clone().into_iter().collect();
+                    opts.entry("connector".to_string())
+                        .or_insert_with(|| lookup.connector().to_string());
+                    let catalog_row = pb::CatalogSourceTable {
+                        arrow_schema_ipc: CatalogCodec::encode_schema(&Arc::new(
+                            lookup.produce_physical_schema(),
+                        ))?,
+                        event_time_field: None,
+                        watermark_field: None,
+                        with_options: opts,
+                        connector: lookup.connector().to_string(),
+                        description: lookup.description.clone(),
+                    };
+                    table_definition::TableType::LookupTable(catalog_row)
+                }
+                ExternalTable::Sink(sink) => {
+                    return plan_err!(
+                        "Persisting SINK table '{}' in stream catalog is not supported",
+                        sink.name()
+                    );
+                }
+            },
+            CatalogEntity::ComputedTable { name, .. } => {
                 return plan_err!(
                     "Persisting query-defined table '{}' is not supported by stream catalog storage",
                     name
@@ -483,7 +435,7 @@ impl CatalogManager {
         table_name: String,
         source_row: pb::CatalogSourceTable,
         as_lookup: bool,
-    ) -> DFResult<CatalogTable> {
+    ) -> DFResult<CatalogEntity> {
         let connector = if source_row.connector.is_empty() {
             source_row
                 .with_options
@@ -493,71 +445,75 @@ impl CatalogManager {
         } else {
             source_row.connector.clone()
         };
-        let mut source = SourceTable::new(
-            table_name,
-            connector,
-            if as_lookup {
-                ConnectionType::Lookup
-            } else {
-                ConnectionType::Source
-            },
-        );
+
         let schema = CatalogCodec::decode_schema(&source_row.arrow_schema_ipc)?;
-        source.schema_specs = schema
+        let schema_specs: Vec<ColumnDescriptor> = schema
             .fields()
             .iter()
             .map(|f| ColumnDescriptor::new_physical((**f).clone()))
             .collect();
-        source.inferred_fields = Some(schema.fields().iter().cloned().collect());
-        source.temporal_config.event_column = source_row.event_time_field;
-        source.temporal_config.watermark_strategy_column = source_row
-            .watermark_field
-            .filter(|w| w != sql_field::COMPUTED_WATERMARK);
-        source.catalog_with_options = source_row.with_options.into_iter().collect();
-        source.description = source_row.description;
-
-        // Rebuild strongly-typed ConnectorConfig from persisted WITH options.
-        if source.connector().eq_ignore_ascii_case("kafka") {
-            use crate::sql::schema::ConnectorConfig;
-            use crate::sql::schema::kafka_operator_config::build_kafka_proto_config_from_string_map;
-            let opts_map: std::collections::HashMap<String, String> = source
-                .catalog_with_options
-                .iter()
-                .map(|(k, v)| (k.clone(), v.clone()))
-                .collect();
-            let physical = source.produce_physical_schema();
-            if let Ok(proto_cfg) = build_kafka_proto_config_from_string_map(opts_map, &physical) {
-                source.connector_config = match proto_cfg {
-                    protocol::function_stream_graph::connector_op::Config::KafkaSource(cfg) => {
-                        ConnectorConfig::KafkaSource(cfg)
-                    }
-                    protocol::function_stream_graph::connector_op::Config::KafkaSink(cfg) => {
-                        ConnectorConfig::KafkaSink(cfg)
-                    }
-                    protocol::function_stream_graph::connector_op::Config::Generic(g) => {
-                        ConnectorConfig::Generic(g.properties)
-                    }
-                };
-            }
+        let inferred_fields = Some(schema.fields().iter().cloned().collect());
+        let physical_schema = schema.as_ref().clone();
+        let catalog_with_options: BTreeMap<String, String> =
+            source_row.with_options.clone().into_iter().collect();
+
+        let role = if as_lookup {
+            TableRole::Reference
         } else {
-            use crate::sql::schema::ConnectorConfig;
-            source.connector_config = ConnectorConfig::Generic(
-                source
-                    .catalog_with_options
-                    .iter()
-                    .map(|(k, v)| (k.clone(), v.clone()))
-                    .collect(),
-            );
-        }
+            TableRole::Ingestion
+        };
+        let connector_config = build_connector_config_for_role(
+            &connector,
+            role,
+            &source_row.with_options,
+            &physical_schema,
+        )?;
 
         if as_lookup {
-            Ok(CatalogTable::LookupTable(source))
+            Ok(CatalogEntity::external(ExternalTable::Lookup(
+                LookupTable {
+                    table_identifier: table_name,
+                    adapter_type: connector,
+                    schema_specs,
+                    connector_config,
+                    key_constraints: Vec::new(),
+                    lookup_cache_max_bytes: None,
+                    lookup_cache_ttl: None,
+                    connection_format: None,
+                    description: source_row.description,
+                    catalog_with_options,
+                    registry_id: None,
+                    inferred_fields,
+                },
+            )))
         } else {
-            Ok(CatalogTable::ConnectorTable(source))
+            let watermark_field = source_row
+                .watermark_field
+                .filter(|w| w != sql_field::COMPUTED_WATERMARK);
+            Ok(CatalogEntity::external(ExternalTable::Source(
+                SourceTable {
+                    table_identifier: table_name,
+                    adapter_type: connector,
+                    schema_specs,
+                    connector_config,
+                    temporal_config: TemporalPipelineConfig {
+                        event_column: source_row.event_time_field,
+                        watermark_strategy_column: watermark_field,
+                        liveness_timeout: None,
+                    },
+                    key_constraints: Vec::new(),
+                    payload_format: None,
+                    connection_format: None,
+                    description: source_row.description,
+                    catalog_with_options,
+                    registry_id: None,
+                    inferred_fields,
+                },
+            )))
         }
     }
 
-    fn decode_catalog_table(&self, proto_def: pb::TableDefinition) -> DFResult<CatalogTable> {
+    fn decode_catalog_table(&self, proto_def: pb::TableDefinition) -> DFResult<CatalogEntity> {
         let Some(table_type) = proto_def.table_type else {
             return internal_err!(
                 "Corrupted catalog row: missing table_type for {}",
@@ -577,7 +533,7 @@ impl CatalogManager {
 
     fn load_catalog_tables_map(
         &self,
-    ) -> DFResult<std::collections::HashMap<crate::sql::schema::ObjectName, Arc<CatalogTable>>>
+    ) -> DFResult<std::collections::HashMap<crate::sql::schema::ObjectName, Arc<CatalogEntity>>>
     {
         let mut out = std::collections::HashMap::new();
         let records = self.store.scan_prefix(CATALOG_KEY_PREFIX)?;
@@ -611,6 +567,52 @@ impl CatalogManager {
     }
 }
 
+fn build_connector_config_for_role<M>(
+    connector: &str,
+    role: TableRole,
+    with_options: &M,
+    physical_schema: &datafusion::arrow::datatypes::Schema,
+) -> DFResult<ConnectorConfig>
+where
+    for<'a> &'a M: IntoIterator<Item = (&'a String, &'a String)>,
+{
+    let flat: std::collections::HashMap<String, String> = with_options
+        .into_iter()
+        .map(|(k, v)| (k.clone(), v.clone()))
+        .collect();
+    crate::sql::connector::factory::build_connector_config_from_catalog(
+        connector,
+        role,
+        flat,
+        physical_schema,
+    )
+}
+
+fn external_to_stream_table(table: &CatalogEntity) -> Option<StreamTable> {
+    match table {
+        CatalogEntity::ExternalConnector(b) => match b.as_ref() {
+            ExternalTable::Source(s) => Some(StreamTable::Source {
+                name: s.name().to_string(),
+                connector: s.connector().to_string(),
+                schema: Arc::new(s.produce_physical_schema()),
+                event_time_field: s.event_time_field().map(str::to_string),
+                watermark_field: s.stream_catalog_watermark_field(),
+                with_options: s.catalog_with_options().clone(),
+            }),
+            ExternalTable::Lookup(l) => Some(StreamTable::Source {
+                name: l.name().to_string(),
+                connector: l.connector().to_string(),
+                schema: Arc::new(l.produce_physical_schema()),
+                event_time_field: None,
+                watermark_field: None,
+                with_options: l.catalog_with_options().clone(),
+            }),
+            ExternalTable::Sink(_) => None,
+        },
+        CatalogEntity::ComputedTable { .. } => None,
+    }
+}
+
 pub fn restore_global_catalog_from_store() {
     let Some(mgr) = CatalogManager::try_global() else {
         return;
@@ -677,22 +679,11 @@ pub fn restore_streaming_jobs_from_store() {
             program,
             checkpoint_interval_ms: interval_ms,
             latest_checkpoint_epoch: latest_epoch,
-            kafka_source_checkpoints,
+            source_checkpoints,
         } = job;
         let jm = job_manager.clone();
         let name = table_name.clone();
 
-        let job_dir = jm.job_state_directory(&table_name);
-        if let Err(e) =
-            materialize_kafka_source_checkpoints_from_catalog(&job_dir, &kafka_source_checkpoints)
-        {
-            warn!(
-                table = %table_name,
-                error = %e,
-                "Failed to materialize Kafka checkpoints from catalog before job restore"
-            );
-        }
-
         let custom_interval = if interval_ms > 0 {
             Some(interval_ms)
         } else {
@@ -704,7 +695,13 @@ pub fn restore_streaming_jobs_from_store() {
             None
         };
 
-        match rt.block_on(jm.submit_job(name.clone(), program, custom_interval, recovery_epoch)) {
+        match rt.block_on(jm.submit_job(
+            name.clone(),
+            program,
+            custom_interval,
+            recovery_epoch,
+            source_checkpoints,
+        )) {
             Ok(job_id) => {
                 info!(
                     table = %table_name, job_id = %job_id,
@@ -727,36 +724,6 @@ pub fn restore_streaming_jobs_from_store() {
     );
 }
 
-pub fn initialize_stream_catalog(config: &crate::config::GlobalConfig) -> anyhow::Result<()> {
-    if !config.stream_catalog.persist {
-        return CatalogManager::init_global_in_memory()
-            .context("Stream catalog (CatalogManager) in-memory init failed");
-    }
-
-    let path = config
-        .stream_catalog
-        .db_path
-        .as_ref()
-        .map(|p| crate::config::resolve_path(p))
-        .unwrap_or_else(|| crate::config::get_data_dir().join("stream_catalog"));
-
-    std::fs::create_dir_all(&path).with_context(|| {
-        format!(
-            "Failed to create stream catalog directory {}",
-            path.display()
-        )
-    })?;
-
-    let store = std::sync::Arc::new(super::RocksDbMetaStore::open(&path).with_context(|| {
-        format!(
-            "Failed to open stream catalog RocksDB at {}",
-            path.display()
-        )
-    })?);
-
-    CatalogManager::init_global(store).context("Stream catalog (CatalogManager) init failed")
-}
-
 #[allow(clippy::unwrap_or_default)]
 pub fn planning_schema_provider() -> StreamPlanningContext {
     CatalogManager::try_global()
@@ -770,10 +737,11 @@ mod tests {
 
     use datafusion::arrow::datatypes::{DataType, Field};
 
+    use crate::sql::connector::config::ConnectorConfig;
+    use crate::sql::schema::catalog::{ExternalTable, SourceTable};
     use crate::sql::schema::column_descriptor::ColumnDescriptor;
-    use crate::sql::schema::connection_type::ConnectionType;
-    use crate::sql::schema::source_table::SourceTable;
-    use crate::sql::schema::table::Table as CatalogTable;
+    use crate::sql::schema::table::CatalogEntity;
+    use crate::sql::schema::temporal_pipeline_config::TemporalPipelineConfig;
     use crate::storage::stream_catalog::InMemoryMetaStore;
 
     use super::CatalogManager;
@@ -782,34 +750,34 @@ mod tests {
         CatalogManager::new(Arc::new(InMemoryMetaStore::new()))
     }
 
-    #[test]
-    fn add_table_roundtrip_snapshot() {
-        let mgr = create_test_manager();
-        let mut source = SourceTable::new("t1", "kafka", ConnectionType::Source);
-        source.schema_specs = vec![ColumnDescriptor::new_physical(Field::new(
-            "a",
-            DataType::Int32,
-            false,
-        ))];
-        source.temporal_config.event_column = Some("ts".into());
-        let table = CatalogTable::ConnectorTable(source);
-
-        mgr.add_catalog_table(table).unwrap();
-
-        let got = mgr.get_catalog_table("t1").unwrap().expect("table present");
-        assert_eq!(got.name(), "t1");
+    fn make_test_source(name: &str) -> SourceTable {
+        SourceTable {
+            table_identifier: name.to_string(),
+            adapter_type: "kafka".to_string(),
+            schema_specs: vec![ColumnDescriptor::new_physical(Field::new(
+                "a",
+                DataType::Int32,
+                false,
+            ))],
+            connector_config: ConnectorConfig::KafkaSource(
+                protocol::function_stream_graph::KafkaSourceConfig::default(),
+            ),
+            temporal_config: TemporalPipelineConfig::default(),
+            key_constraints: Vec::new(),
+            payload_format: None,
+            connection_format: None,
+            description: String::new(),
+            catalog_with_options: std::collections::BTreeMap::new(),
+            registry_id: None,
+            inferred_fields: None,
+        }
     }
 
     #[test]
     fn drop_table_if_exists() {
         let mgr = create_test_manager();
-        let mut source = SourceTable::new("t_drop", "kafka", ConnectionType::Source);
-        source.schema_specs = vec![ColumnDescriptor::new_physical(Field::new(
-            "a",
-            DataType::Int32,
-            false,
-        ))];
-        mgr.add_catalog_table(CatalogTable::ConnectorTable(source))
+        let source = make_test_source("t_drop");
+        mgr.add_catalog_table(CatalogEntity::external(ExternalTable::Source(source)))
             .unwrap();
 
         mgr.drop_catalog_table("t_drop", false).unwrap();
diff --git a/src/storage/stream_catalog/meta_store.rs b/src/storage/stream_catalog/meta_store.rs
index 6f61b3f7..741273cf 100644
--- a/src/storage/stream_catalog/meta_store.rs
+++ b/src/storage/stream_catalog/meta_store.rs
@@ -23,6 +23,19 @@ pub trait MetaStore: Send + Sync {
     fn get(&self, key: &str) -> Result<Option<Vec<u8>>>;
     fn delete(&self, key: &str) -> Result<()>;
     fn scan_prefix(&self, prefix: &str) -> Result<Vec<(String, Vec<u8>)>>;
+
+    /// Atomic apply of many puts (`Some(value)`) and/or deletes (`None`).
+    /// Default implementation is sequential; backends should override with a
+    /// single transaction / `WriteBatch` when available.
+    fn write_batch(&self, batch: Vec<(String, Option<Vec<u8>>)>) -> Result<()> {
+        for (k, v) in batch {
+            match v {
+                Some(val) => self.put(&k, val)?,
+                None => self.delete(&k)?,
+            }
+        }
+        Ok(())
+    }
 }
 
 /// In-process KV store for single-node deployments and tests.
@@ -67,4 +80,19 @@ impl MetaStore for InMemoryMetaStore {
             .map(|(k, v)| (k.clone(), v.clone()))
             .collect())
     }
+
+    fn write_batch(&self, batch: Vec<(String, Option<Vec<u8>>)>) -> Result<()> {
+        let mut db = self.db.write();
+        for (k, v) in batch {
+            match v {
+                Some(val) => {
+                    db.insert(k, val);
+                }
+                None => {
+                    db.remove(&k);
+                }
+            }
+        }
+        Ok(())
+    }
 }
diff --git a/src/storage/stream_catalog/mod.rs b/src/storage/stream_catalog/mod.rs
index ef176c40..08a987dd 100644
--- a/src/storage/stream_catalog/mod.rs
+++ b/src/storage/stream_catalog/mod.rs
@@ -17,10 +17,11 @@ mod manager;
 mod meta_store;
 mod rocksdb_meta_store;
 
+pub use super::initialize_stream_catalog;
+
 #[allow(unused_imports)]
 pub use manager::{
-    CatalogManager, StoredStreamingJob, initialize_stream_catalog,
-    materialize_kafka_source_checkpoints_from_catalog, restore_global_catalog_from_store,
+    CatalogManager, StoredStreamingJob, restore_global_catalog_from_store,
     restore_streaming_jobs_from_store,
 };
 pub use meta_store::{InMemoryMetaStore, MetaStore};
diff --git a/src/storage/stream_catalog/rocksdb_meta_store.rs b/src/storage/stream_catalog/rocksdb_meta_store.rs
index 5315454f..1537f278 100644
--- a/src/storage/stream_catalog/rocksdb_meta_store.rs
+++ b/src/storage/stream_catalog/rocksdb_meta_store.rs
@@ -16,7 +16,7 @@ use std::sync::Arc;
 
 use anyhow::Context;
 use datafusion::common::Result;
-use rocksdb::{DB, Direction, IteratorMode, Options};
+use rocksdb::{DB, Direction, IteratorMode, Options, WriteBatch};
 
 use super::MetaStore;
 
@@ -84,6 +84,24 @@ impl MetaStore for RocksDbMetaStore {
         }
         Ok(out)
     }
+
+    fn write_batch(&self, batch: Vec<(String, Option<Vec<u8>>)>) -> Result<()> {
+        if batch.is_empty() {
+            return Ok(());
+        }
+        let mut wb = WriteBatch::default();
+        for (k, v) in batch {
+            match v {
+                Some(val) => wb.put(k.as_bytes(), val.as_slice()),
+                None => wb.delete(k.as_bytes()),
+            }
+        }
+        self.db.write(wb).map_err(|e| {
+            datafusion::common::DataFusionError::Execution(format!(
+                "stream catalog store write_batch: {e}"
+            ))
+        })
+    }
 }
 
 #[cfg(test)]
@@ -120,4 +138,27 @@ mod tests {
 
         let _ = std::fs::remove_dir_all(&dir);
     }
+
+    #[test]
+    fn write_batch_put_and_delete() {
+        let dir: PathBuf =
+            std::env::temp_dir().join(format!("fs_stream_catalog_wb_{}", Uuid::new_v4()));
+        let _ = std::fs::remove_dir_all(&dir);
+
+        let store = RocksDbMetaStore::open(&dir).expect("open");
+        store
+            .write_batch(vec![
+                ("k1".to_string(), Some(vec![1])),
+                ("k2".to_string(), Some(vec![2])),
+            ])
+            .unwrap();
+        assert_eq!(store.get("k1").unwrap(), Some(vec![1]));
+        assert_eq!(store.get("k2").unwrap(), Some(vec![2]));
+
+        store.write_batch(vec![("k1".to_string(), None)]).unwrap();
+        assert!(store.get("k1").unwrap().is_none());
+        assert_eq!(store.get("k2").unwrap(), Some(vec![2]));
+
+        let _ = std::fs::remove_dir_all(&dir);
+    }
 }
diff --git a/tests/integration/Makefile b/tests/integration/Makefile
index f16d640b..990436ac 100644
--- a/tests/integration/Makefile
+++ b/tests/integration/Makefile
@@ -15,7 +15,7 @@
 # -----------------------------------------------------------------------
 # Usage:
 #   make test        — Setup env + run pytest  (PYTEST_ARGS="-k xxx")
-#   make clean       — Remove .venv and test output
+#   make clean       — Remove test output
 #
 # Prerequisites:
 #   The FunctionStream binary must already be built (make build / make build-lite
@@ -24,7 +24,7 @@
 
 PROJECT_ROOT := $(shell git -C $(CURDIR) rev-parse --show-toplevel)
 PYTHON_ROOT  := $(PROJECT_ROOT)/python
-VENV         := $(CURDIR)/.venv
+VENV         := $(PROJECT_ROOT)/.venv
 PIP          := $(VENV)/bin/pip
 PY           := $(VENV)/bin/python
 
@@ -41,7 +41,7 @@ help:
 	@echo "Integration Test Targets:"
 	@echo ""
 	@echo "  test    Setup Python env + run pytest (PYTEST_ARGS=...)"
-	@echo "  clean   Remove .venv and target/tests output"
+	@echo "  clean   Remove target/tests output"
 
 install: requirements.txt $(PYTHON_ROOT)/functionstream-api/pyproject.toml $(PYTHON_ROOT)/functionstream-client/pyproject.toml
 	$(call log,ENV,Setting up Python virtual environment)
@@ -58,7 +58,6 @@ test: install
 
 clean:
 	$(call log,CLEAN,Removing test artifacts)
-	@rm -rf $(VENV)
 	@rm -rf $(CURDIR)/target
 	@rm -rf $(CURDIR)/install
 	$(call success,Clean complete)
diff --git a/tests/integration/README.md b/tests/integration/README.md
index 23096720..55995055 100644
--- a/tests/integration/README.md
+++ b/tests/integration/README.md
@@ -1,88 +1,25 @@
 # Integration Tests
 
-## Prerequisites
+This directory contains the end-to-end (E2E) integration test suite for FunctionStream.
 
-| Dependency | Version  | Purpose                                         |
-|------------|----------|-------------------------------------------------|
-| Python     | >= 3.9   | Test framework runtime                          |
-| Rust       | stable   | Build the FunctionStream binary                 |
-| Docker     | >= 20.10 | Run a Kafka broker for streaming integration tests |
+## 📋 Prerequisites
 
-> **Docker is required.** The test framework automatically pulls and manages
-> an `apache/kafka:3.7.0` container in KRaft mode to provide a real Kafka
-> broker for tests that involve Kafka input/output. Tests will fail if the
-> Docker daemon is not running.
+Ensure the following dependencies are available in your `PATH` before running the suite:
 
-## Quick Start
+| Dependency | Version  | Note                                         |
+|------------|----------|----------------------------------------------|
+| Python     | `>= 3.9` | Test framework runtime                       |
+| Rust       | `stable` | For compiling the `function-stream` binary   |
+| Docker     | `>= 20.10`| Required for containerized infrastructure (Kafka, MinIO) |
 
-```bash
-# From the project root
-make build          # Build the release binary (with --features python)
-make integration-test
+*Note: Infrastructure containers (e.g., `apache/kafka:3.7.0` in KRaft mode, `minio/minio`) are automatically provisioned and torn down by the test framework via the Docker daemon.*
 
-# Or run directly from this directory
-cd tests/integration
-make test
-```
+## 🚀 Quick Start
 
-## Directory Layout
-
-```
-tests/integration/
-├── Makefile              # test / clean targets
-├── requirements.txt      # Python dependencies (pytest, grpcio, docker, ...)
-├── pytest.ini            # Pytest configuration
-├── framework/            # Reusable test infrastructure
-│   ├── instance.py       # FunctionStreamInstance facade
-│   ├── workspace.py      # Per-test directory management
-│   ├── config.py         # Server config generation
-│   ├── process.py        # OS process lifecycle (start/stop/kill)
-│   ├── utils.py          # Port allocation, readiness probes
-│   └── kafka_manager.py  # Docker-managed Kafka broker (KRaft mode)
-├── test/                 # Test suites
-│   ├── wasm/             # WASM function tests
-│   │   └── python_sdk/   # Python SDK integration tests
-│   └── streaming/        # Streaming engine tests (future)
-└── target/               # Test output (git-ignored)
-    ├── .shared_cache/    # Shared WASM compilation cache across tests
-    └── <suite>/<class>/<method>/<timestamp>/logs/
-```
-
-## Test Output
-
-Each test gets an isolated server instance with its own log directory:
-
-```
-target/wasm/python_sdk/TestFunctionLifecycle/test_full_lifecycle_transitions/20260416_221655/
-  logs/
-    app.log       # FunctionStream application log
-    stdout.log    # Server stdout
-    stderr.log    # Server stderr
-```
-
-Only `logs/` is retained after tests complete; `conf/` and `data/` are
-automatically cleaned up.
-
-## Python Dependencies
-
-All Python packages are listed in `requirements.txt` and installed
-automatically by `make test`. Key dependencies:
-
-- `pytest` — test runner
-- `grpcio` / `protobuf` — gRPC client communication
-- `docker` — Docker SDK for managing the Kafka container
-- `confluent-kafka` — Kafka admin client for topic management
-- `functionstream-api` / `functionstream-client` — local editable installs
-
-## Running Specific Tests
+The test suite requires a fresh release build of the engine with the `python` feature enabled.
 
 ```bash
-# Single test
-make test PYTEST_ARGS="-k test_full_lifecycle_transitions"
-
-# Single file
-make test PYTEST_ARGS="test/wasm/python_sdk/test_lifecycle.py"
-
-# Verbose with live log
-make test PYTEST_ARGS="-v --log-cli-level=DEBUG"
+# From the project root
+make build
+make integration-test
 ```
diff --git a/tests/integration/framework/__init__.py b/tests/integration/framework/__init__.py
index b735753a..b8c08d67 100644
--- a/tests/integration/framework/__init__.py
+++ b/tests/integration/framework/__init__.py
@@ -12,11 +12,14 @@
 
 from .instance import FunctionStreamInstance
 
-__all__ = ["FunctionStreamInstance", "KafkaDockerManager"]
+__all__ = ["FunctionStreamInstance", "KafkaDockerManager", "MinioDockerManager"]
 
 
 def __getattr__(name: str):
     if name == "KafkaDockerManager":
         from .kafka_manager import KafkaDockerManager
         return KafkaDockerManager
+    if name == "MinioDockerManager":
+        from .minio_manager import MinioDockerManager
+        return MinioDockerManager
     raise AttributeError(f"module {__name__!r} has no attribute {name!r}")
diff --git a/tests/integration/framework/minio_manager.py b/tests/integration/framework/minio_manager.py
new file mode 100644
index 00000000..d10abf3e
--- /dev/null
+++ b/tests/integration/framework/minio_manager.py
@@ -0,0 +1,170 @@
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""
+Docker-managed MinIO service for integration tests.
+"""
+
+from __future__ import annotations
+
+import logging
+import time
+from dataclasses import dataclass
+from typing import Any, Dict, Optional
+
+import boto3
+import docker
+import requests
+from docker.errors import APIError, DockerException, NotFound
+
+from .utils import find_free_port
+
+logger = logging.getLogger(__name__)
+
+
+class MinioDockerManagerError(Exception):
+    """Base exception for MinioDockerManager errors."""
+
+
+@dataclass(frozen=True)
+class MinioConfig:
+    image: str = "minio/minio:RELEASE.2024-01-16T16-07-38Z"
+    container_name: str = "fs-integration-minio"
+    root_user: str = "minioadmin"
+    root_password: str = "minioadmin"
+    api_host: str = "127.0.0.1"
+    api_port: int = 9000
+    console_port: int = 9001
+    readiness_timeout_sec: int = 60
+
+    @property
+    def endpoint_url(self) -> str:
+        return f"http://{self.api_host}:{self.api_port}"
+
+
+class MinioDockerManager:
+    def __init__(
+        self,
+        config: Optional[MinioConfig] = None,
+        docker_client: Optional[docker.DockerClient] = None,
+    ) -> None:
+        if config is None:
+            config = MinioConfig(api_port=find_free_port(), console_port=find_free_port())
+        self.config = config
+        self._docker_client = docker_client
+
+    @property
+    def docker_client(self) -> docker.DockerClient:
+        if self._docker_client is None:
+            try:
+                self._docker_client = docker.from_env()
+            except DockerException as e:
+                raise MinioDockerManagerError(f"Failed to connect to Docker daemon: {e}") from e
+        return self._docker_client
+
+    @property
+    def s3_client(self):
+        return boto3.client(
+            "s3",
+            endpoint_url=self.config.endpoint_url,
+            aws_access_key_id=self.config.root_user,
+            aws_secret_access_key=self.config.root_password,
+            region_name="us-east-1",
+        )
+
+    def __enter__(self) -> "MinioDockerManager":
+        self.setup_minio()
+        return self
+
+    def __exit__(self, exc_type: Any, exc_val: Any, exc_tb: Any) -> None:
+        self.teardown_minio()
+
+    def setup_minio(self) -> None:
+        self._ensure_image()
+        self._ensure_container()
+        self._wait_for_readiness()
+        logger.info("MinIO ready at %s", self.config.endpoint_url)
+
+    def teardown_minio(self) -> None:
+        try:
+            container = self.docker_client.containers.get(self.config.container_name)
+            container.stop(timeout=5)
+        except NotFound:
+            logger.debug("MinIO container not found during teardown.")
+        except APIError as exc:
+            logger.warning("Docker API error while stopping MinIO: %s", exc)
+
+    def create_bucket_if_not_exists(self, bucket: str) -> None:
+        client = self.s3_client
+        buckets = [b["Name"] for b in client.list_buckets().get("Buckets", [])]
+        if bucket not in buckets:
+            client.create_bucket(Bucket=bucket)
+
+    def clear_bucket(self, bucket: str) -> None:
+        client = self.s3_client
+        token = None
+        while True:
+            kwargs: Dict[str, Any] = {"Bucket": bucket}
+            if token:
+                kwargs["ContinuationToken"] = token
+            resp = client.list_objects_v2(**kwargs)
+            contents = resp.get("Contents", [])
+            if contents:
+                for item in contents:
+                    client.delete_object(Bucket=bucket, Key=item["Key"])
+            if not resp.get("IsTruncated"):
+                break
+            token = resp.get("NextContinuationToken")
+
+    def _ensure_image(self) -> None:
+        try:
+            self.docker_client.images.get(self.config.image)
+        except NotFound:
+            logger.info("Pulling MinIO image '%s'...", self.config.image)
+            self.docker_client.images.pull(self.config.image)
+
+    def _ensure_container(self) -> None:
+        try:
+            container = self.docker_client.containers.get(self.config.container_name)
+            if container.status != "running":
+                container.start()
+            return
+        except NotFound:
+            pass
+
+        self.docker_client.containers.run(
+            image=self.config.image,
+            name=self.config.container_name,
+            command='server /data --console-address ":9001"',
+            ports={"9000/tcp": self.config.api_port, "9001/tcp": self.config.console_port},
+            environment={
+                "MINIO_ROOT_USER": self.config.root_user,
+                "MINIO_ROOT_PASSWORD": self.config.root_password,
+            },
+            detach=True,
+            remove=True,
+        )
+
+    def _wait_for_readiness(self) -> None:
+        deadline = time.time() + self.config.readiness_timeout_sec
+        url = f"{self.config.endpoint_url}/minio/health/ready"
+        while time.time() < deadline:
+            try:
+                r = requests.get(url, timeout=1.5)
+                if r.status_code == 200:
+                    return
+            except requests.RequestException:
+                pass
+            time.sleep(1.0)
+        raise MinioDockerManagerError(
+            f"MinIO not ready within {self.config.readiness_timeout_sec}s ({url})"
+        )
diff --git a/tests/integration/requirements.txt b/tests/integration/requirements.txt
index d1e597c3..ac613dd9 100644
--- a/tests/integration/requirements.txt
+++ b/tests/integration/requirements.txt
@@ -19,6 +19,11 @@ protobuf>=4.25.0
 # Docker + Kafka management for integration tests
 docker>=7.0
 confluent-kafka>=2.3.0
+boto3>=1.34.0
+requests>=2.31.0
+pyarrow>=17.0.0
+lance>=1.1.0
+python-dateutil>=2.9.0
 
 # FunctionStream Python packages (local editable installs)
 -e ../../python/functionstream-api
diff --git a/tests/integration/test/wasm/python_sdk/conftest.py b/tests/integration/test/wasm/python_sdk/conftest.py
index e0acd26e..c83a4763 100644
--- a/tests/integration/test/wasm/python_sdk/conftest.py
+++ b/tests/integration/test/wasm/python_sdk/conftest.py
@@ -10,19 +10,27 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
+import concurrent.futures
 import logging
+import os
 import re
 import sys
 from pathlib import Path
-from typing import Generator, List
+from typing import Any, Generator, Set
 
 _CURRENT_DIR = Path(__file__).resolve().parent
-_INTEGRATION_ROOT = Path(__file__).resolve().parents[3]
+_INTEGRATION_ROOT = _CURRENT_DIR.parents[2]
 
-if str(_INTEGRATION_ROOT) not in sys.path:
-    sys.path.insert(0, str(_INTEGRATION_ROOT))
-if str(_CURRENT_DIR) not in sys.path:
-    sys.path.insert(0, str(_CURRENT_DIR))
+
+def _inject_path_safely(target_path: Path) -> None:
+    path_str = str(target_path)
+    if path_str not in sys.path:
+        sys.path.insert(0, path_str)
+        os.environ["PYTHONPATH"] = f"{path_str}{os.pathsep}{os.environ.get('PYTHONPATH', '')}"
+
+
+_inject_path_safely(_INTEGRATION_ROOT)
+_inject_path_safely(_CURRENT_DIR)
 
 import pytest
 from framework import FunctionStreamInstance, KafkaDockerManager
@@ -33,41 +41,32 @@
 
 @pytest.fixture(scope="session")
 def kafka() -> Generator[KafkaDockerManager, None, None]:
-    """
-    Session-scoped Kafka broker manager.
-    Leverages Context Manager for guaranteed teardown.
-    """
     with KafkaDockerManager() as mgr:
         yield mgr
-        try:
-            mgr.clear_all_topics()
-        except Exception as e:
-            logger.warning("Failed to clear topics during Kafka teardown: %s", e)
 
 
 @pytest.fixture(scope="session")
 def kafka_topics(kafka: KafkaDockerManager) -> str:
-    """
-    Pre-creates standard topics and returns the bootstrap server address.
-    """
     kafka.create_topics_if_not_exist(["in", "out", "events", "counts"])
     return kafka.config.bootstrap_servers
 
 
+@pytest.fixture(scope="session")
+def minio() -> Generator[Any, None, None]:
+    try:
+        from framework import MinioDockerManager
+    except ModuleNotFoundError as exc:
+        pytest.skip(f"MinIO tests require optional dependency: {exc}")
+    with MinioDockerManager() as mgr:
+        yield mgr
+
+
 def _sanitize_segment(segment: str) -> str:
     clean = re.sub(r"[^\w\-]+", "_", segment).strip("_")
     return clean or "unknown"
 
 
 def _nodeid_to_workspace_path(nodeid: str) -> str:
-    """
-    Convert pytest nodeid into a readable nested path under target/.
-
-    Example:
-        test/wasm/python_sdk/test_data_flow.py::TestDataFlow::test_single_word_counting
-    ->
-        test/wasm/python_sdk/test_data_flow/TestDataFlow/test_single_word_counting
-    """
     parts = nodeid.split("::")
     file_part = Path(parts[0]).with_suffix("")
     file_segments = [_sanitize_segment(seg) for seg in file_part.parts]
@@ -77,10 +76,6 @@ def _nodeid_to_workspace_path(nodeid: str) -> str:
 
 @pytest.fixture
 def fs_server(request: pytest.FixtureRequest) -> Generator[FunctionStreamInstance, None, None]:
-    """
-    Function-scoped FunctionStream instance.
-    Uses Context Manager to ensure SIGKILL and workspace cleanup.
-    """
     test_name = _nodeid_to_workspace_path(request.node.nodeid)
     with FunctionStreamInstance(test_name=test_name) as instance:
         yield instance
@@ -88,30 +83,54 @@ def fs_server(request: pytest.FixtureRequest) -> Generator[FunctionStreamInstanc
 
 @pytest.fixture
 def fs_client(fs_server: FunctionStreamInstance) -> Generator[FsClient, None, None]:
-    """
-    Function-scoped FsClient connected to the isolated fs_server.
-    """
     with fs_server.get_client() as client:
         yield client
 
 
-@pytest.fixture
-def function_registry(fs_client: FsClient) -> Generator[List[str], None, None]:
-    """
-    RAII-style registry for FunctionStream tasks.
-    Ensures absolute teardown of functions to prevent state leakage.
-    """
-    registered_names: List[str] = []
+class FunctionTracker:
+    def __init__(self, client: FsClient):
+        self._client = client
+        self._registered: Set[str] = set()
 
-    yield registered_names
+    def __contains__(self, name: str) -> bool:
+        return name in self._registered
 
-    for name in registered_names:
+    def append(self, name: str) -> None:
+        self._registered.add(name)
+
+    def extend(self, names) -> None:
+        for name in names:
+            self._registered.add(name)
+
+    def remove(self, name: str) -> None:
+        self._registered.discard(name)
+
+    def register(self, name: str) -> None:
+        self._registered.add(name)
+
+    def _teardown_single_function(self, name: str) -> None:
         try:
-            fs_client.stop_function(name)
+            self._client.stop_function(name)
         except Exception as e:
-            logger.debug("Failed to stop function '%s' during cleanup: %s", name, e)
+            logger.debug("Ignored stop error for '%s': %s", name, e)
 
         try:
-            fs_client.drop_function(name)
+            self._client.drop_function(name)
         except Exception as e:
-            logger.error("Failed to drop function '%s' during cleanup: %s", name, e)
\ No newline at end of file
+            logger.debug("Ignored drop error for '%s': %s", name, e)
+
+    def teardown_all(self) -> None:
+        if not self._registered:
+            return
+
+        with concurrent.futures.ThreadPoolExecutor(max_workers=10) as executor:
+            list(executor.map(self._teardown_single_function, self._registered))
+
+        self._registered.clear()
+
+
+@pytest.fixture
+def function_registry(fs_client: FsClient) -> Generator[FunctionTracker, None, None]:
+    tracker = FunctionTracker(fs_client)
+    yield tracker
+    tracker.teardown_all()
\ No newline at end of file
diff --git a/tests/integration/test/wasm/python_sdk/test_streaming_sql_sinks.py b/tests/integration/test/wasm/python_sdk/test_streaming_sql_sinks.py
new file mode 100644
index 00000000..a31d8956
--- /dev/null
+++ b/tests/integration/test/wasm/python_sdk/test_streaming_sql_sinks.py
@@ -0,0 +1,420 @@
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import csv
+import datetime as dt
+import io
+import json
+import shutil
+import tempfile
+import time
+import uuid
+from pathlib import Path
+from typing import Any, Dict, List
+
+import lance
+import pyarrow as pa
+import pyarrow.parquet as pq
+import pytest
+from dateutil import parser as dt_parser
+
+from .test_data_flow import produce_messages
+
+
+def _uid(prefix: str) -> str:
+    return f"{prefix}_{uuid.uuid4().hex[:8]}"
+
+
+def _bucket_name(prefix: str) -> str:
+    return f"{prefix}-{uuid.uuid4().hex[:12]}".lower()
+
+
+def _sql_ok(fs_server: Any, sql: str) -> Any:
+    resp = fs_server.execute_sql(sql)
+    assert resp.status_code == 200, f"SQL failed: {sql}\nstatus={resp.status_code}\nmsg={resp.message}"
+    return resp
+
+
+def _sql_drop_streaming_table(fs_server: Any, table_name: str) -> None:
+    resp = fs_server.execute_sql(f"DROP STREAMING TABLE {table_name};")
+    if resp.status_code == 200:
+        return
+    msg = (resp.message or "").lower()
+    if "not found" in msg or "does not exist" in msg:
+        return
+    raise AssertionError(
+        f"SQL failed: DROP STREAMING TABLE {table_name};\n"
+        f"status={resp.status_code}\nmsg={resp.message}"
+    )
+
+
+def _create_source(fs_server: Any, source_name: str, in_topic: str, bootstrap: str) -> None:
+    _sql_ok(
+        fs_server,
+        f"""
+        CREATE TABLE {source_name} (
+            id BIGINT,
+            value VARCHAR,
+            ts TIMESTAMP NOT NULL,
+            WATERMARK FOR ts AS ts - INTERVAL '1' SECOND
+        ) WITH (
+            'connector' = 'kafka',
+            'topic' = '{in_topic}',
+            'format' = 'json',
+            'scan.startup.mode' = 'earliest',
+            'bootstrap.servers' = '{bootstrap}'
+        );
+        """,
+    )
+
+
+def _create_sink_streaming_table(
+    fs_server: Any,
+    stream_name: str,
+    connector: str,
+    format_name: str,
+    with_extra: Dict[str, str],
+    source_name: str,
+    select_expr: str = "id, value, ts",
+) -> None:
+    with_pairs = {
+        "connector": connector,
+        "type": "sink",
+        "format": format_name,
+        "checkpoint.interval.ms": "1000",
+        **with_extra,
+    }
+    with_sql = ",\n".join([f"'{k}' = '{v}'" for k, v in with_pairs.items()])
+    _sql_ok(
+        fs_server,
+        f"""
+        CREATE STREAMING TABLE {stream_name} WITH (
+            {with_sql}
+        ) AS
+        SELECT {select_expr} FROM {source_name};
+        """,
+    )
+
+
+def _publish_rows(kafka_bootstrap: str, topic: str, rows: List[Dict[str, Any]]) -> None:
+    produce_messages(kafka_bootstrap, topic, [json.dumps(r) for r in rows])
+
+
+def _sample_rows() -> List[Dict[str, Any]]:
+    now = dt.datetime.now(dt.timezone.utc)
+    return [
+        {"id": 1, "value": "alpha", "ts": (now - dt.timedelta(seconds=2)).isoformat()},
+        {"id": 2, "value": "beta", "ts": (now - dt.timedelta(seconds=1)).isoformat()},
+        {"id": 3, "value": "gamma", "ts": now.isoformat()},
+    ]
+
+
+def _parse_timestamp(val: Any) -> float:
+    if isinstance(val, dt.datetime):
+        # Parquet/Iceberg may return naive datetimes that are in UTC without a tz marker;
+        # treat them as UTC to avoid local-timezone skew (e.g. UTC+8 adds 28800 s).
+        if val.tzinfo is None:
+            val = val.replace(tzinfo=dt.timezone.utc)
+        return val.timestamp()
+    if isinstance(val, str):
+        parsed = dt_parser.isoparse(val)
+        if parsed.tzinfo is None:
+            parsed = parsed.replace(tzinfo=dt.timezone.utc)
+        return parsed.timestamp()
+    raise TypeError(f"Unknown timestamp type: {type(val)}")
+
+
+def _assert_data_integrity(actual_rows: List[Dict[str, Any]], expected_rows: List[Dict[str, Any]]) -> None:
+    assert len(actual_rows) >= len(expected_rows), f"Expected at least {len(expected_rows)} rows, got {len(actual_rows)}"
+
+    actual_mapped = {str(r["id"]): r for r in actual_rows}
+    for expected in expected_rows:
+        exp_id = str(expected["id"])
+        assert exp_id in actual_mapped, f"Data Loss: Missing row with id {exp_id}"
+
+        actual = actual_mapped[exp_id]
+        if "value" in expected:
+            assert str(actual.get("value")) == str(expected["value"]), f"Data Corruption: Value mismatch for id {exp_id}"
+
+        assert "ts" in actual, f"Data Corruption: Missing timestamp column in output for id {exp_id}"
+
+        expected_ts = _parse_timestamp(expected["ts"])
+        actual_ts = _parse_timestamp(actual["ts"])
+        assert abs(actual_ts - expected_ts) < 1.0, f"Timestamp drift too large for id {exp_id}: expected {expected_ts}, got {actual_ts}"
+
+
+def _wait_and_verify_local_csv(dir_path: Path, expected_count: int, timeout_s: float = 30.0) -> List[Dict[str, Any]]:
+    deadline = time.time() + timeout_s
+    while time.time() < deadline:
+        files = list(dir_path.glob("*.csv"))
+        if files:
+            merged: List[Dict[str, Any]] = []
+            for f in files:
+                try:
+                    with f.open("r", encoding="utf-8") as fp:
+                        merged.extend(list(csv.DictReader(fp)))
+                except Exception:
+                    pass
+            if len(merged) >= expected_count:
+                return merged
+        time.sleep(1.0)
+    raise TimeoutError(f"Failed to verify {expected_count} rows in local CSV at {dir_path}")
+
+
+def _wait_and_verify_local_parquet(dir_path: Path, expected_count: int, timeout_s: float = 30.0) -> List[Dict[str, Any]]:
+    deadline = time.time() + timeout_s
+    while time.time() < deadline:
+        files = list(dir_path.glob("**/*.parquet"))
+        if files:
+            try:
+                tables = [pq.read_table(f) for f in files]
+                if tables:
+                    combined = pa.concat_tables(tables)
+                    if combined.num_rows >= expected_count:
+                        return combined.to_pylist()
+            except Exception:
+                pass
+        time.sleep(1.0)
+    raise TimeoutError(f"Failed to verify {expected_count} rows in local Parquet at {dir_path}")
+
+
+def _wait_and_verify_local_lance(dir_path: Path, expected_count: int, timeout_s: float = 30.0) -> List[Dict[str, Any]]:
+    deadline = time.time() + timeout_s
+    while time.time() < deadline:
+        try:
+            ds = lance.dataset(dir_path.as_posix())
+            if ds.count_rows() >= expected_count:
+                return ds.to_table().to_pylist()
+        except Exception:
+            pass
+        time.sleep(1.0)
+    raise TimeoutError(f"Failed to verify {expected_count} rows in local Lance dataset at {dir_path}")
+
+
+def _wait_and_verify_s3_parquet(minio: Any, bucket: str, prefix: str, expected_count: int, timeout_s: float = 30.0) -> List[Dict[str, Any]]:
+    deadline = time.time() + timeout_s
+    client = minio.s3_client
+    while time.time() < deadline:
+        resp = client.list_objects_v2(Bucket=bucket, Prefix=prefix)
+        keys = [obj["Key"] for obj in resp.get("Contents", []) if obj["Key"].endswith(".parquet")]
+        if keys:
+            try:
+                tables = []
+                for k in keys:
+                    body = client.get_object(Bucket=bucket, Key=k)["Body"].read()
+                    tables.append(pq.read_table(io.BytesIO(body)))
+                if tables:
+                    combined = pa.concat_tables(tables)
+                    if combined.num_rows >= expected_count:
+                        return combined.to_pylist()
+            except Exception:
+                pass
+        time.sleep(1.0)
+    raise TimeoutError(f"Failed to verify {expected_count} rows in S3 Parquet at s3://{bucket}/{prefix}")
+
+
+def _wait_and_verify_s3_lance(minio: Any, bucket: str, prefix: str, expected_count: int, timeout_s: float = 35.0) -> List[Dict[str, Any]]:
+    uri = f"s3://{bucket}/{prefix}"
+    deadline = time.time() + timeout_s
+    storage_options = {
+        "endpoint": minio.config.endpoint_url,
+        "access_key_id": minio.config.root_user,
+        "secret_access_key": minio.config.root_password,
+        "region": "us-east-1",
+        "allow_http": "true",
+    }
+    while time.time() < deadline:
+        try:
+            ds = lance.dataset(uri, storage_options=storage_options)
+            if ds.count_rows() >= expected_count:
+                return ds.to_table().to_pylist()
+        except Exception:
+            pass
+        time.sleep(1.0)
+    raise TimeoutError(f"Failed to verify {expected_count} rows in S3 Lance dataset at {uri}")
+
+
+class TestStreamingSqlSinks:
+    def test_filesystem_csv_sink(self, fs_server: Any, kafka: Any, kafka_topics: str) -> None:
+        source_name = _uid("src")
+        stream_name = _uid("st_fs_csv")
+        in_topic = _uid("topic_in")
+        kafka.create_topics_if_not_exist([in_topic])
+        _create_source(fs_server, source_name, in_topic, kafka_topics)
+
+        temp_dir = Path(tempfile.mkdtemp(prefix="fs_sink_csv_"))
+        try:
+            _create_sink_streaming_table(
+                fs_server,
+                stream_name,
+                connector="filesystem",
+                format_name="csv",
+                with_extra={"path": temp_dir.as_posix()},
+                source_name=source_name,
+            )
+            expected_data = _sample_rows()
+            _publish_rows(kafka_topics, in_topic, expected_data)
+            actual_data = _wait_and_verify_local_csv(temp_dir, len(expected_data))
+            _assert_data_integrity(actual_data, expected_data)
+        finally:
+            _sql_drop_streaming_table(fs_server, stream_name)
+            shutil.rmtree(temp_dir, ignore_errors=True)
+
+    def test_s3_parquet_sink(self, fs_server: Any, kafka: Any, kafka_topics: str, minio: Any) -> None:
+        source_name = _uid("src")
+        stream_name = _uid("st_s3_pq")
+        in_topic = _uid("topic_in")
+        kafka.create_topics_if_not_exist([in_topic])
+        _create_source(fs_server, source_name, in_topic, kafka_topics)
+
+        bucket = _bucket_name("sink-bucket")
+        prefix = f"sinks/{stream_name}"
+        minio.create_bucket_if_not_exists(bucket)
+        try:
+            _create_sink_streaming_table(
+                fs_server,
+                stream_name,
+                connector="s3",
+                format_name="parquet",
+                with_extra={
+                    "path": prefix,
+                    "s3.bucket": bucket,
+                    "s3.region": "us-east-1",
+                    "s3.endpoint": minio.config.endpoint_url,
+                    "s3.access_key_id": minio.config.root_user,
+                    "s3.secret_access_key": minio.config.root_password,
+                },
+                source_name=source_name,
+            )
+            expected_data = _sample_rows()
+            _publish_rows(kafka_topics, in_topic, expected_data)
+            actual_data = _wait_and_verify_s3_parquet(minio, bucket, prefix, len(expected_data))
+            _assert_data_integrity(actual_data, expected_data)
+        finally:
+            _sql_drop_streaming_table(fs_server, stream_name)
+            minio.clear_bucket(bucket)
+
+    def test_delta_parquet_sink(self, fs_server: Any, kafka: Any, kafka_topics: str) -> None:
+        source_name = _uid("src")
+        stream_name = _uid("st_delta_pq")
+        in_topic = _uid("topic_in")
+        kafka.create_topics_if_not_exist([in_topic])
+        _create_source(fs_server, source_name, in_topic, kafka_topics)
+
+        temp_dir = Path(tempfile.mkdtemp(prefix="delta_sink_pq_"))
+        try:
+            _create_sink_streaming_table(
+                fs_server,
+                stream_name,
+                connector="delta",
+                format_name="parquet",
+                with_extra={"path": temp_dir.as_posix()},
+                source_name=source_name,
+            )
+            expected_data = _sample_rows()
+            _publish_rows(kafka_topics, in_topic, expected_data)
+            actual_data = _wait_and_verify_local_parquet(temp_dir, len(expected_data))
+            _assert_data_integrity(actual_data, expected_data)
+        finally:
+            _sql_drop_streaming_table(fs_server, stream_name)
+            shutil.rmtree(temp_dir, ignore_errors=True)
+
+    def test_iceberg_parquet_sink(self, fs_server: Any, kafka: Any, kafka_topics: str) -> None:
+        source_name = _uid("src")
+        stream_name = _uid("st_iceberg_pq")
+        in_topic = _uid("topic_in")
+        kafka.create_topics_if_not_exist([in_topic])
+        _create_source(fs_server, source_name, in_topic, kafka_topics)
+
+        temp_dir = Path(tempfile.mkdtemp(prefix="iceberg_sink_pq_"))
+        try:
+            _create_sink_streaming_table(
+                fs_server,
+                stream_name,
+                connector="iceberg",
+                format_name="parquet",
+                with_extra={"path": temp_dir.as_posix()},
+                source_name=source_name,
+            )
+            expected_data = _sample_rows()
+            _publish_rows(kafka_topics, in_topic, expected_data)
+            actual_data = _wait_and_verify_local_parquet(temp_dir, len(expected_data))
+            _assert_data_integrity(actual_data, expected_data)
+        finally:
+            _sql_drop_streaming_table(fs_server, stream_name)
+            shutil.rmtree(temp_dir, ignore_errors=True)
+
+    def test_lancedb_lance_sink(self, fs_server: Any, kafka: Any, kafka_topics: str) -> None:
+        source_name = _uid("src")
+        stream_name = _uid("st_lancedb")
+        in_topic = _uid("topic_in")
+        kafka.create_topics_if_not_exist([in_topic])
+        _create_source(fs_server, source_name, in_topic, kafka_topics)
+
+        temp_dir = Path(tempfile.mkdtemp(prefix="lancedb_sink_"))
+        try:
+            _create_sink_streaming_table(
+                fs_server,
+                stream_name,
+                connector="lanceDB",
+                format_name="lance",
+                with_extra={"path": temp_dir.as_posix()},
+                source_name=source_name,
+                select_expr="id, ts",
+            )
+            expected_data = _sample_rows()
+            _publish_rows(kafka_topics, in_topic, expected_data)
+            actual_data = _wait_and_verify_local_lance(temp_dir, len(expected_data))
+            expected_truncated = [{"id": r["id"], "ts": r["ts"]} for r in expected_data]
+            _assert_data_integrity(actual_data, expected_truncated)
+        finally:
+            _sql_drop_streaming_table(fs_server, stream_name)
+            shutil.rmtree(temp_dir, ignore_errors=True)
+
+    def test_lancedb_lance_sink_to_s3(
+        self, fs_server: Any, kafka: Any, kafka_topics: str, minio: Any
+    ) -> None:
+        source_name = _uid("src")
+        stream_name = _uid("st_lancedb_s3")
+        in_topic = _uid("topic_in")
+        kafka.create_topics_if_not_exist([in_topic])
+        _create_source(fs_server, source_name, in_topic, kafka_topics)
+
+        bucket = _bucket_name("lance-bucket")
+        prefix = f"sinks/{stream_name}"
+        minio.create_bucket_if_not_exists(bucket)
+        s3_path = f"s3://{bucket}/{prefix}"
+        try:
+            _create_sink_streaming_table(
+                fs_server,
+                stream_name,
+                connector="lanceDB",
+                format_name="lance",
+                with_extra={
+                    "path": s3_path,
+                    "s3.bucket": bucket,
+                    "s3.region": "us-east-1",
+                    "s3.endpoint": minio.config.endpoint_url,
+                    "s3.access_key_id": minio.config.root_user,
+                    "s3.secret_access_key": minio.config.root_password,
+                },
+                source_name=source_name,
+                select_expr="id, ts",
+            )
+            expected_data = _sample_rows()
+            _publish_rows(kafka_topics, in_topic, expected_data)
+            actual_data = _wait_and_verify_s3_lance(minio, bucket, prefix, len(expected_data))
+            expected_truncated = [{"id": r["id"], "ts": r["ts"]} for r in expected_data]
+            _assert_data_integrity(actual_data, expected_truncated)
+        finally:
+            _sql_drop_streaming_table(fs_server, stream_name)
+            minio.clear_bucket(bucket)