fix: close issue #6586 make pg also show error as mysql (#6587)

* fix: close issue #6586 make pg also show error as mysql

Signed-off-by: yihong0618 <zouzou0208@gmail.com>

* fix: address comments

Signed-off-by: yihong0618 <zouzou0208@gmail.com>

* fix: drop useless debug print

Signed-off-by: yihong0618 <zouzou0208@gmail.com>

* fix: revert wrong change

Signed-off-by: yihong0618 <zouzou0208@gmail.com>

* refactor: convert error types

* refactor: inline

* chore: minimize changes

* fix: address comments

Signed-off-by: yihong0618 <zouzou0208@gmail.com>

* fix: make clippy happy

Signed-off-by: yihong0618 <zouzou0208@gmail.com>

* fix: address comments

Signed-off-by: yihong0618 <zouzou0208@gmail.com>

* refactor: convert datafusion error to ErrorExt

Signed-off-by: Ning Sun <sunning@greptime.com>

* fix: headers ?

Signed-off-by: yihong0618 <zouzou0208@gmail.com>

---------

Signed-off-by: yihong0618 <zouzou0208@gmail.com>
Signed-off-by: Ning Sun <sunning@greptime.com>
Co-authored-by: Ning Sun <sunning@greptime.com>

.cargo/config.toml

@@ -12,3 +12,6 @@ fetch = true
 checkout = true
 list_files = true
 internal_use_git2 = false
+
+[env]
+CARGO_WORKSPACE_DIR = { value = "", relative = true }

.gitignore

@@ -60,4 +60,7 @@ tests-fuzz/corpus/
 greptimedb_data
 
 # github
-!/.github
+!/.github
+
+# Claude code
+CLAUDE.md

AUTHOR.md

@@ -10,12 +10,10 @@
 * [NiwakaDev](https://github.com/NiwakaDev)
 * [tisonkun](https://github.com/tisonkun)
 
-
 ## Team Members (in alphabetical order)
 
 * [apdong2022](https://github.com/apdong2022)
 * [beryl678](https://github.com/beryl678)
-* [Breeze-P](https://github.com/Breeze-P)
 * [daviderli614](https://github.com/daviderli614)
 * [discord9](https://github.com/discord9)
 * [evenyag](https://github.com/evenyag)

Cargo.lock

@@ -1602,6 +1602,17 @@ version = "1.0.7"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "acbc26382d871df4b7442e3df10a9402bf3cf5e55cbd66f12be38861425f0564"
 
+[[package]]
+name = "cargo-manifest"
+version = "0.19.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "a1d8af896b707212cd0e99c112a78c9497dd32994192a463ed2f7419d29bd8c6"
+dependencies = [
+ "serde",
+ "thiserror 2.0.12",
+ "toml 0.8.19",
+]
+
 [[package]]
 name = "cast"
 version = "0.3.0"
@@ -2291,6 +2302,27 @@ dependencies = [
  "tonic 0.12.3",
 ]
 
+[[package]]
+name = "common-event-recorder"
+version = "0.16.0"
+dependencies = [
+ "api",
+ "async-trait",
+ "backon",
+ "client",
+ "common-error",
+ "common-macro",
+ "common-meta",
+ "common-telemetry",
+ "common-time",
+ "serde",
+ "serde_json",
+ "snafu 0.8.5",
+ "store-api",
+ "tokio",
+ "tokio-util",
+]
+
 [[package]]
 name = "common-frontend"
 version = "0.16.0"
@@ -2315,6 +2347,8 @@ dependencies = [
  "api",
  "approx 0.5.1",
  "arc-swap",
+ "arrow 54.2.1",
+ "arrow-schema 54.3.1",
  "async-trait",
  "bincode",
  "catalog",
@@ -2333,8 +2367,10 @@ dependencies = [
  "datafusion-common",
  "datafusion-expr",
  "datafusion-functions-aggregate-common",
+ "datafusion-physical-expr",
  "datatypes",
  "derive_more",
+ "futures",
  "geo",
  "geo-types",
  "geohash",
@@ -2347,6 +2383,7 @@ dependencies = [
  "num-traits",
  "once_cell",
  "paste",
+ "pretty_assertions",
  "s2",
  "serde",
  "serde_json",
@@ -2408,6 +2445,7 @@ dependencies = [
  "tokio-util",
  "tonic 0.12.3",
  "tower 0.5.2",
+ "vec1",
 ]
 
 [[package]]
@@ -2520,6 +2558,7 @@ dependencies = [
  "tokio",
  "tokio-postgres",
  "tonic 0.12.3",
+ "tracing",
  "typetag",
  "uuid",
 ]
@@ -2668,12 +2707,31 @@ dependencies = [
  "strum 0.27.1",
 ]
 
+[[package]]
+name = "common-sql"
+version = "0.16.0"
+dependencies = [
+ "common-base",
+ "common-datasource",
+ "common-decimal",
+ "common-error",
+ "common-macro",
+ "common-time",
+ "datafusion-sql",
+ "datatypes",
+ "hex",
+ "jsonb",
+ "snafu 0.8.5",
+ "sqlparser 0.54.0 (git+https://github.com/GreptimeTeam/sqlparser-rs.git?rev=0cf6c04490d59435ee965edd2078e8855bd8471e)",
+]
+
 [[package]]
 name = "common-telemetry"
 version = "0.16.0"
 dependencies = [
  "backtrace",
  "common-error",
+ "common-version",
  "console-subscriber",
  "greptime-proto",
  "humantime-serde",
@@ -2731,6 +2789,7 @@ name = "common-version"
 version = "0.16.0"
 dependencies = [
  "build-data",
+ "cargo-manifest",
  "const_format",
  "serde",
  "shadow-rs",
@@ -2964,9 +3023,9 @@ dependencies = [
 
 [[package]]
 name = "crc"
-version = "3.2.1"
+version = "3.3.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "69e6e4d7b33a94f0991c26729976b10ebde1d34c3ee82408fb536164fa10d636"
+checksum = "9710d3b3739c2e349eb44fe848ad0b7c8cb1e42bd87ee49371df2f7acaf3e675"
 dependencies = [
  "crc-catalog",
 ]
@@ -3007,6 +3066,7 @@ dependencies = [
  "ciborium",
  "clap 3.2.25",
  "criterion-plot",
+ "futures",
  "itertools 0.10.5",
  "lazy_static",
  "num-traits",
@@ -3018,6 +3078,7 @@ dependencies = [
  "serde_derive",
  "serde_json",
  "tinytemplate",
+ "tokio",
  "walkdir",
 ]
 
@@ -3775,6 +3836,7 @@ dependencies = [
  "tokio",
  "toml 0.8.19",
  "tonic 0.12.3",
+ "tracing",
 ]
 
 [[package]]
@@ -3797,7 +3859,7 @@ dependencies = [
  "jsonb",
  "num",
  "num-traits",
- "ordered-float 3.9.2",
+ "ordered-float 4.3.0",
  "paste",
  "serde",
  "serde_json",
@@ -4118,12 +4180,16 @@ dependencies = [
 
 [[package]]
 name = "domain"
-version = "0.10.4"
+version = "0.11.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "4c84070523f8ba0f9127ff156920f27eb27b302b425efe60bf5f41ec244d1c60"
+checksum = "a11dd7f04a6a6d2aea0153c6e31f5ea7af8b2efdf52cdaeea7a9a592c7fefef9"
 dependencies = [
+ "bumpalo",
  "bytes",
+ "domain-macros",
  "futures-util",
+ "hashbrown 0.14.5",
+ "log",
  "moka",
  "octseq",
  "rand 0.8.5",
@@ -4134,6 +4200,17 @@ dependencies = [
  "tracing",
 ]
 
+[[package]]
+name = "domain-macros"
+version = "0.11.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "0e197fdfd2cdb5fdeb7f8ddcf3aed5d5d04ecde2890d448b14ffb716f7376b70"
+dependencies = [
+ "proc-macro2",
+ "quote",
+ "syn 2.0.100",
+]
+
 [[package]]
 name = "dotenv"
 version = "0.15.0"
@@ -4619,6 +4696,7 @@ dependencies = [
  "get-size2",
  "greptime-proto",
  "http 1.1.0",
+ "humantime-serde",
  "itertools 0.14.0",
  "lazy_static",
  "meta-client",
@@ -4763,6 +4841,7 @@ dependencies = [
  "toml 0.8.19",
  "tonic 0.12.3",
  "tower 0.5.2",
+ "tracing",
  "uuid",
 ]
 
@@ -5146,7 +5225,7 @@ dependencies = [
 [[package]]
 name = "greptime-proto"
 version = "0.1.0"
-source = "git+https://github.com/GreptimeTeam/greptime-proto.git?rev=ceb1af4fa9309ce65bda0367db7b384df2bb4d4f#ceb1af4fa9309ce65bda0367db7b384df2bb4d4f"
+source = "git+https://github.com/GreptimeTeam/greptime-proto.git?rev=7fcaa3e413947a7a28d9af95812af26c1939ce78#7fcaa3e413947a7a28d9af95812af26c1939ce78"
 dependencies = [
  "prost 0.13.5",
  "serde",
@@ -6697,7 +6776,7 @@ source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "4979f22fdb869068da03c9f7528f8297c6fd2606bc3a4affe42e6a823fdb8da4"
 dependencies = [
  "cfg-if",
- "windows-targets 0.52.6",
+ "windows-targets 0.48.5",
 ]
 
 [[package]]
@@ -7144,6 +7223,9 @@ version = "0.16.0"
 dependencies = [
  "api",
  "async-trait",
+ "axum 0.8.1",
+ "axum-extra",
+ "axum-macros",
  "bytes",
  "chrono",
  "clap 4.5.19",
@@ -7177,6 +7259,7 @@ dependencies = [
  "http-body-util",
  "humantime",
  "humantime-serde",
+ "hyper 0.14.30",
  "hyper-util",
  "itertools 0.14.0",
  "lazy_static",
@@ -7204,6 +7287,7 @@ dependencies = [
  "toml 0.8.19",
  "tonic 0.12.3",
  "tower 0.5.2",
+ "tower-http 0.6.2",
  "tracing",
  "tracing-subscriber",
  "typetag",
@@ -7266,6 +7350,7 @@ dependencies = [
  "snafu 0.8.5",
  "store-api",
  "tokio",
+ "tracing",
 ]
 
 [[package]]
@@ -7379,6 +7464,7 @@ dependencies = [
  "datafusion-expr",
  "datatypes",
  "dotenv",
+ "either",
  "futures",
  "humantime-serde",
  "index",
@@ -7396,6 +7482,7 @@ dependencies = [
  "prost 0.13.5",
  "puffin",
  "rand 0.9.0",
+ "rayon",
  "regex",
  "rskafka",
  "rstest",
@@ -7414,6 +7501,7 @@ dependencies = [
  "tokio-stream",
  "tokio-util",
  "toml 0.8.19",
+ "tracing",
  "uuid",
 ]
 
@@ -8457,6 +8545,7 @@ dependencies = [
  "common-query",
  "common-recordbatch",
  "common-runtime",
+ "common-sql",
  "common-telemetry",
  "common-test-util",
  "common-time",
@@ -8492,6 +8581,7 @@ dependencies = [
  "tokio",
  "tokio-util",
  "tonic 0.12.3",
+ "tracing",
 ]
 
 [[package]]
@@ -8528,17 +8618,6 @@ dependencies = [
  "num-traits",
 ]
 
-[[package]]
-name = "ordered-float"
-version = "3.9.2"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "f1e1c390732d15f1d48471625cd92d154e66db2c56645e29a9cd26f4699f72dc"
-dependencies = [
- "num-traits",
- "rand 0.8.5",
- "serde",
-]
-
 [[package]]
 name = "ordered-float"
 version = "4.3.0"
@@ -8546,6 +8625,8 @@ source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "44d501f1a72f71d3c063a6bbc8f7271fa73aa09fe5d6283b6571e2ed176a2537"
 dependencies = [
  "num-traits",
+ "rand 0.8.5",
+ "serde",
 ]
 
 [[package]]
@@ -9082,6 +9163,7 @@ dependencies = [
  "moka",
  "once_cell",
  "operator",
+ "ordered-float 4.3.0",
  "paste",
  "prometheus",
  "query",
@@ -9522,8 +9604,9 @@ dependencies = [
 
 [[package]]
 name = "promql-parser"
-version = "0.5.1"
-source = "git+https://github.com/GreptimeTeam/promql-parser.git?rev=0410e8b459dda7cb222ce9596f8bf3971bd07bd2#0410e8b459dda7cb222ce9596f8bf3971bd07bd2"
+version = "0.6.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "328fe69c2443ec4f8e6c33ea925dde04a1026e6c95928e89ed02343944cac9bf"
 dependencies = [
  "cfgrammar",
  "chrono",
@@ -9594,7 +9677,7 @@ source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "be769465445e8c1474e9c5dac2018218498557af32d9ed057325ec9a41ae81bf"
 dependencies = [
  "heck 0.5.0",
- "itertools 0.14.0",
+ "itertools 0.11.0",
  "log",
  "multimap",
  "once_cell",
@@ -9640,7 +9723,7 @@ source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "8a56d757972c98b346a9b766e3f02746cde6dd1cd1d1d563472929fdd74bec4d"
 dependencies = [
  "anyhow",
- "itertools 0.14.0",
+ "itertools 0.11.0",
  "proc-macro2",
  "quote",
  "syn 2.0.100",
@@ -9892,6 +9975,7 @@ dependencies = [
  "table",
  "tokio",
  "tokio-stream",
+ "tracing",
  "unescaper",
  "uuid",
 ]
@@ -11243,6 +11327,7 @@ dependencies = [
  "common-recordbatch",
  "common-runtime",
  "common-session",
+ "common-sql",
  "common-telemetry",
  "common-test-util",
  "common-time",
@@ -11324,8 +11409,10 @@ dependencies = [
  "tonic-reflection",
  "tower 0.5.2",
  "tower-http 0.6.2",
+ "tracing",
  "urlencoding",
  "uuid",
+ "vrl",
  "zstd 0.13.2",
 ]
 
@@ -11681,6 +11768,7 @@ dependencies = [
  "common-error",
  "common-macro",
  "common-query",
+ "common-sql",
  "common-time",
  "datafusion",
  "datafusion-common",
@@ -12439,6 +12527,7 @@ dependencies = [
  "humantime",
  "humantime-serde",
  "lazy_static",
+ "once_cell",
  "parquet",
  "paste",
  "serde",
@@ -12987,9 +13076,9 @@ checksum = "1f3ccbac311fea05f86f61904b462b55fb3df8837a366dfc601a0161d0532f20"
 
 [[package]]
 name = "tokio"
-version = "1.44.2"
+version = "1.45.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "e6b88822cbe49de4185e3a4cbf8321dd487cf5fe0c5c65695fef6346371e9c48"
+checksum = "75ef51a33ef1da925cea3e4eb122833cb377c61439ca401b770f54902b806779"
 dependencies = [
  "backtrace",
  "bytes",
@@ -13155,6 +13244,7 @@ version = "0.8.19"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "a1ed1f98e3fdc28d6d910e6737ae6ab1a93bf1985935a1193e68f93eeb68d24e"
 dependencies = [
+ "indexmap 2.9.0",
  "serde",
  "serde_spanned",
  "toml_datetime",
@@ -13914,6 +14004,12 @@ version = "0.2.15"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "accd4ea62f7bb7a82fe23066fb0957d48ef677f6eeb8215f372f52e48bb32426"
 
+[[package]]
+name = "vec1"
+version = "1.12.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "eab68b56840f69efb0fefbe3ab6661499217ffdc58e2eef7c3f6f69835386322"
+
 [[package]]
 name = "vergen"
 version = "8.3.2"
@@ -13944,9 +14040,9 @@ dependencies = [
 
 [[package]]
 name = "vrl"
-version = "0.24.0"
+version = "0.25.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "f9ceadaa40aef567a26079ff014ca7a567ba85344f1b81090b5ec7d7bb16a219"
+checksum = "4f49394b948406ea1564aa00152e011d87a38ad35d277ebddda257a9ee39c419"
 dependencies = [
  "aes",
  "aes-siv",
@@ -14266,7 +14362,7 @@ version = "0.1.9"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "cf221c93e13a30d793f7645a0e7762c55d169dbb0a49671918a2319d289b10bb"
 dependencies = [
- "windows-sys 0.59.0",
+ "windows-sys 0.48.0",
 ]
 
 [[package]]

Cargo.toml

@@ -13,6 +13,7 @@ members = [
     "src/common/datasource",
     "src/common/decimal",
     "src/common/error",
+    "src/common/event-recorder",
     "src/common/frontend",
     "src/common/function",
     "src/common/greptimedb-telemetry",
@@ -30,6 +31,7 @@ members = [
     "src/common/recordbatch",
     "src/common/runtime",
     "src/common/session",
+    "src/common/sql",
     "src/common/stat",
     "src/common/substrait",
     "src/common/telemetry",
@@ -138,7 +140,7 @@ etcd-client = "0.14"
 fst = "0.4.7"
 futures = "0.3"
 futures-util = "0.3"
-greptime-proto = { git = "https://github.com/GreptimeTeam/greptime-proto.git", rev = "ceb1af4fa9309ce65bda0367db7b384df2bb4d4f" }
+greptime-proto = { git = "https://github.com/GreptimeTeam/greptime-proto.git", rev = "7fcaa3e413947a7a28d9af95812af26c1939ce78" }
 hex = "0.4"
 http = "1"
 humantime = "2.1"
@@ -166,14 +168,13 @@ opentelemetry-proto = { version = "0.27", features = [
     "with-serde",
     "logs",
 ] }
+ordered-float = { version = "4.3", features = ["serde"] }
 parking_lot = "0.12"
 parquet = { version = "54.2", default-features = false, features = ["arrow", "async", "object_store"] }
 paste = "1.0"
 pin-project = "1.0"
 prometheus = { version = "0.13.3", features = ["process"] }
-promql-parser = { git = "https://github.com/GreptimeTeam/promql-parser.git", rev = "0410e8b459dda7cb222ce9596f8bf3971bd07bd2", features = [
-    "ser",
-] }
+promql-parser = { version = "0.6", features = ["ser"] }
 prost = { version = "0.13", features = ["no-recursion-limit"] }
 raft-engine = { version = "0.4.1", default-features = false }
 rand = "0.9"
@@ -224,10 +225,13 @@ tokio-util = { version = "0.7", features = ["io-util", "compat"] }
 toml = "0.8.8"
 tonic = { version = "0.12", features = ["tls", "gzip", "zstd"] }
 tower = "0.5"
+tower-http = "0.6"
+tracing = "0.1"
 tracing-appender = "0.2"
 tracing-subscriber = { version = "0.3", features = ["env-filter", "json", "fmt"] }
 typetag = "0.2"
 uuid = { version = "1.7", features = ["serde", "v4", "fast-rng"] }
+vrl = "0.25"
 zstd = "0.13"
 # DO_NOT_REMOVE_THIS: END_OF_EXTERNAL_DEPENDENCIES
 
@@ -245,6 +249,7 @@ common-config = { path = "src/common/config" }
 common-datasource = { path = "src/common/datasource" }
 common-decimal = { path = "src/common/decimal" }
 common-error = { path = "src/common/error" }
+common-event-recorder = { path = "src/common/event-recorder" }
 common-frontend = { path = "src/common/frontend" }
 common-function = { path = "src/common/function" }
 common-greptimedb-telemetry = { path = "src/common/greptimedb-telemetry" }
@@ -262,6 +267,7 @@ common-query = { path = "src/common/query" }
 common-recordbatch = { path = "src/common/recordbatch" }
 common-runtime = { path = "src/common/runtime" }
 common-session = { path = "src/common/session" }
+common-sql = { path = "src/common/sql" }
 common-telemetry = { path = "src/common/telemetry" }
 common-test-util = { path = "src/common/test-util" }
 common-time = { path = "src/common/time" }

config/config.md

@@ -562,6 +562,16 @@
 | `node_id` | Integer | Unset | The flownode identifier and should be unique in the cluster. |
 | `flow` | -- | -- | flow engine options. |
 | `flow.num_workers` | Integer | `0` | The number of flow worker in flownode.<br/>Not setting(or set to 0) this value will use the number of CPU cores divided by 2. |
+| `flow.batching_mode` | -- | -- | -- |
+| `flow.batching_mode.query_timeout` | String | `600s` | The default batching engine query timeout is 10 minutes. |
+| `flow.batching_mode.slow_query_threshold` | String | `60s` | will output a warn log for any query that runs for more that this threshold |
+| `flow.batching_mode.experimental_min_refresh_duration` | String | `5s` | The minimum duration between two queries execution by batching mode task |
+| `flow.batching_mode.grpc_conn_timeout` | String | `5s` | The gRPC connection timeout |
+| `flow.batching_mode.experimental_grpc_max_retries` | Integer | `3` | The gRPC max retry number |
+| `flow.batching_mode.experimental_frontend_scan_timeout` | String | `30s` | Flow wait for available frontend timeout,<br/>if failed to find available frontend after frontend_scan_timeout elapsed, return error<br/>which prevent flownode from starting |
+| `flow.batching_mode.experimental_frontend_activity_timeout` | String | `60s` | Frontend activity timeout<br/>if frontend is down(not sending heartbeat) for more than frontend_activity_timeout,<br/>it will be removed from the list that flownode use to connect |
+| `flow.batching_mode.experimental_max_filter_num_per_query` | Integer | `20` | Maximum number of filters allowed in a single query |
+| `flow.batching_mode.experimental_time_window_merge_threshold` | Integer | `3` | Time window merge distance |
 | `grpc` | -- | -- | The gRPC server options. |
 | `grpc.bind_addr` | String | `127.0.0.1:6800` | The address to bind the gRPC server. |
 | `grpc.server_addr` | String | `127.0.0.1:6800` | The address advertised to the metasrv,<br/>and used for connections from outside the host |

config/flownode.example.toml

@@ -7,6 +7,29 @@ node_id = 14
 ## The number of flow worker in flownode.
 ## Not setting(or set to 0) this value will use the number of CPU cores divided by 2.
 #+num_workers=0
+[flow.batching_mode]
+## The default batching engine query timeout is 10 minutes.
+#+query_timeout="600s"
+## will output a warn log for any query that runs for more that this threshold
+#+slow_query_threshold="60s"
+## The minimum duration between two queries execution by batching mode task
+#+experimental_min_refresh_duration="5s"
+## The gRPC connection timeout
+#+grpc_conn_timeout="5s"
+## The gRPC max retry number
+#+experimental_grpc_max_retries=3
+## Flow wait for available frontend timeout,
+## if failed to find available frontend after frontend_scan_timeout elapsed, return error
+## which prevent flownode from starting
+#+experimental_frontend_scan_timeout="30s"
+## Frontend activity timeout
+## if frontend is down(not sending heartbeat) for more than frontend_activity_timeout,
+## it will be removed from the list that flownode use to connect
+#+experimental_frontend_activity_timeout="60s"
+## Maximum number of filters allowed in a single query
+#+experimental_max_filter_num_per_query=20
+## Time window merge distance
+#+experimental_time_window_merge_threshold=3
 
 ## The gRPC server options.
 [grpc]

docs/how-to/how-to-write-aggregate-function.md

@@ -1,4 +1,4 @@
-Currently, our query engine is based on DataFusion, so all aggregate function is executed by DataFusion, through its UDAF interface. You can find DataFusion's UDAF example [here](https://github.com/apache/arrow-datafusion/blob/arrow2/datafusion-examples/examples/simple_udaf.rs). Basically, we provide the same way as DataFusion to write aggregate functions: both are centered in a struct called "Accumulator" to accumulates states along the way in aggregation.
+Currently, our query engine is based on DataFusion, so all aggregate function is executed by DataFusion, through its UDAF interface. You can find DataFusion's UDAF example [here](https://github.com/apache/datafusion/tree/main/datafusion-examples/examples/simple_udaf.rs). Basically, we provide the same way as DataFusion to write aggregate functions: both are centered in a struct called "Accumulator" to accumulates states along the way in aggregation.
 
 However, DataFusion's UDAF implementation has a huge restriction, that it requires user to provide a concrete "Accumulator". Take `Median` aggregate function for example, to aggregate a `u32` datatype column, you have to write a `MedianU32`, and use `SELECT MEDIANU32(x)` in SQL. `MedianU32` cannot be used to aggregate a `i32` datatype column. Or, there's another way: you can use a special type that can hold all kinds of data (like our `Value` enum or Arrow's `ScalarValue`), and `match` all the way up to do aggregate calculations. It might work, though rather tedious. (But I think it's DataFusion's preferred way to write UDAF.)
 

docs/rfcs/2023-02-01-table-compaction.md

@@ -76,7 +76,7 @@ pub trait CompactionStrategy {
 ```
 
 The most suitable compaction strategy for time-series scenario would be 
-a hybrid strategy that combines time window compaction with size-tired compaction, just like [Cassandra](https://cassandra.apache.org/doc/latest/cassandra/operating/compaction/twcs.html) and [ScyllaDB](https://docs.scylladb.com/stable/architecture/compaction/compaction-strategies.html#time-window-compaction-strategy-twcs) does.
+a hybrid strategy that combines time window compaction with size-tired compaction, just like [Cassandra](https://cassandra.apache.org/doc/latest/cassandra/managing/operating/compaction/twcs.html) and [ScyllaDB](https://docs.scylladb.com/stable/architecture/compaction/compaction-strategies.html#time-window-compaction-strategy-twcs) does.
 
 We can first group SSTs in level n into buckets according to some predefined time window. Within that window, 
 SSTs are compacted in a size-tired manner (find SSTs with similar size and compact them to level n+1). 

docs/rfcs/2024-01-17-dataflow-framework.md

@@ -28,7 +28,7 @@ In order to do those things while maintaining a low memory footprint, you need t
 - Greptime Flow's is built on top of [Hydroflow](https://github.com/hydro-project/hydroflow).
 - We have three choices for the Dataflow/Streaming process framework for our simple continuous aggregation feature:
 1. Based on the timely/differential dataflow crate that [materialize](https://github.com/MaterializeInc/materialize) based on. Later, it's proved too obscure for a simple usage, and is hard to customize memory usage control.
-2. Based on a simple dataflow framework that we write from ground up, like what [arroyo](https://www.arroyo.dev/) or [risingwave](https://www.risingwave.dev/) did, for example the core streaming logic of [arroyo](https://github.com/ArroyoSystems/arroyo/blob/master/arroyo-datastream/src/lib.rs) only takes up to 2000 line of codes. However, it means maintaining another layer of dataflow framework, which might seem easy in the beginning, but I fear it might be too burdensome to maintain once we need more features.
+2. Based on a simple dataflow framework that we write from ground up, like what [arroyo](https://www.arroyo.dev/) or [risingwave](https://www.risingwave.dev/) did, for example the core streaming logic of [arroyo](https://github.com/ArroyoSystems/arroyo/blob/master/crates/arroyo-datastream/src/lib.rs) only takes up to 2000 line of codes. However, it means maintaining another layer of dataflow framework, which might seem easy in the beginning, but I fear it might be too burdensome to maintain once we need more features.
 3. Based on a simple and lower level dataflow framework that someone else write, like [hydroflow](https://github.com/hydro-project/hydroflow), this approach combines the best of both worlds. Firstly, it boasts ease of comprehension and customization. Secondly, the dataflow framework offers precisely the necessary features for crafting uncomplicated single-node dataflow programs while delivering decent performance.
 
 Hence, we choose the third option, and use a simple logical plan that's anagonistic to the underlying dataflow framework, as it only describe how the dataflow graph should be doing, not how it do that. And we built operator in hydroflow to execute the plan. And the result hydroflow graph is wrapped in a engine that only support data in/out and tick event to flush and compute the result. This provide a thin middle layer that's easy to maintain and allow switching to other dataflow framework if necessary.

grafana/README.md

@@ -83,7 +83,7 @@ If you use the [Helm Chart](https://github.com/GreptimeTeam/helm-charts) to depl
 - `monitoring.enabled=true`: Deploys a standalone GreptimeDB instance dedicated to monitoring the cluster;
 - `grafana.enabled=true`: Deploys Grafana and automatically imports the monitoring dashboard;
 
-The standalone GreptimeDB instance will collect metrics from your cluster, and the dashboard will be available in the Grafana UI. For detailed deployment instructions, please refer to our [Kubernetes deployment guide](https://docs.greptime.com/user-guide/deployments-administration-administration/deploy-on-kubernetes/getting-started).
+The standalone GreptimeDB instance will collect metrics from your cluster, and the dashboard will be available in the Grafana UI. For detailed deployment instructions, please refer to our [Kubernetes deployment guide](https://docs.greptime.com/user-guide/deployments-administration/deploy-on-kubernetes/overview).
 
 ### Self-host Prometheus and import dashboards manually
 

grafana/dashboards/metrics/cluster/dashboard.json

@@ -5954,9 +5954,9 @@
                 "uid": "${metrics}"
               },
               "editorMode": "code",
-              "expr": "sum by(instance, pod, scheme) (rate(opendal_operation_duration_seconds_count{instance=~\"$datanode\", operation=\"read\"}[$__rate_interval]))",
+              "expr": "sum by(instance, pod, scheme, operation) (rate(opendal_operation_duration_seconds_count{instance=~\"$datanode\", operation=~\"read|Reader::read\"}[$__rate_interval]))",
               "instant": false,
-              "legendFormat": "[{{instance}}]-[{{pod}}]-[{{scheme}}]",
+              "legendFormat": "[{{instance}}]-[{{pod}}]-[{{scheme}}]-[{{operation}}]",
               "range": true,
               "refId": "A"
             }
@@ -6055,9 +6055,9 @@
                 "uid": "${metrics}"
               },
               "editorMode": "code",
-              "expr": "histogram_quantile(0.99, sum by(instance, pod, le, scheme) (rate(opendal_operation_duration_seconds_bucket{instance=~\"$datanode\",operation=\"read\"}[$__rate_interval])))",
+              "expr": "histogram_quantile(0.99, sum by(instance, pod, le, scheme, operation) (rate(opendal_operation_duration_seconds_bucket{instance=~\"$datanode\",operation=~\"read|Reader::read\"}[$__rate_interval])))",
               "instant": false,
-              "legendFormat": "[{{instance}}]-[{{pod}}]-{{scheme}}",
+              "legendFormat": "[{{instance}}]-[{{pod}}]-[{{scheme}}]-[{{operation}}]",
               "range": true,
               "refId": "A"
             }
@@ -6156,9 +6156,9 @@
                 "uid": "${metrics}"
               },
               "editorMode": "code",
-              "expr": "sum by(instance, pod, scheme) (rate(opendal_operation_duration_seconds_count{instance=~\"$datanode\", operation=\"write\"}[$__rate_interval]))",
+              "expr": "sum by(instance, pod, scheme, operation) (rate(opendal_operation_duration_seconds_count{instance=~\"$datanode\", operation=~\"write|Writer::write|Writer::close\"}[$__rate_interval]))",
               "instant": false,
-              "legendFormat": "[{{instance}}]-[{{pod}}]-{{scheme}}",
+              "legendFormat": "[{{instance}}]-[{{pod}}]-[{{scheme}}]-[{{operation}}]",
               "range": true,
               "refId": "A"
             }
@@ -6257,9 +6257,9 @@
                 "uid": "${metrics}"
               },
               "editorMode": "code",
-              "expr": "histogram_quantile(0.99, sum by(instance, pod, le, scheme) (rate(opendal_operation_duration_seconds_bucket{instance=~\"$datanode\", operation=\"write\"}[$__rate_interval])))",
+              "expr": "histogram_quantile(0.99, sum by(instance, pod, le, scheme, operation) (rate(opendal_operation_duration_seconds_bucket{instance=~\"$datanode\", operation =~ \"Writer::write|Writer::close|write\"}[$__rate_interval])))",
               "instant": false,
-              "legendFormat": "[{{instance}}]-[{{pod}}]-[{{scheme}}]",
+              "legendFormat": "[{{instance}}]-[{{pod}}]-[{{scheme}}]-[{{operation}}]",
               "range": true,
               "refId": "A"
             }
@@ -6663,7 +6663,7 @@
                 "uid": "${metrics}"
               },
               "editorMode": "code",
-              "expr": "histogram_quantile(0.99, sum by(instance, pod, le, scheme, operation) (rate(opendal_operation_duration_seconds_bucket{instance=~\"$datanode\", operation!~\"read|write|list\"}[$__rate_interval])))",
+              "expr": "histogram_quantile(0.99, sum by(instance, pod, le, scheme, operation) (rate(opendal_operation_duration_seconds_bucket{instance=~\"$datanode\", operation!~\"read|write|list|Writer::write|Writer::close|Reader::read\"}[$__rate_interval])))",
               "instant": false,
               "legendFormat": "[{{instance}}]-[{{pod}}]-[{{scheme}}]-[{{operation}}]",
               "range": true,

grafana/dashboards/metrics/cluster/dashboard.md

@@ -76,14 +76,14 @@
 | Title | Query | Type | Description | Datasource | Unit | Legend Format |
 | --- | --- | --- | --- | --- | --- | --- |
 | QPS per Instance | `sum by(instance, pod, scheme, operation) (rate(opendal_operation_duration_seconds_count{instance=~"$datanode"}[$__rate_interval]))` | `timeseries` | QPS per Instance. | `prometheus` | `ops` | `[{{instance}}]-[{{pod}}]-[{{scheme}}]-[{{operation}}]` |
-| Read QPS per Instance | `sum by(instance, pod, scheme) (rate(opendal_operation_duration_seconds_count{instance=~"$datanode", operation="read"}[$__rate_interval]))` | `timeseries` | Read QPS per Instance. | `prometheus` | `ops` | `[{{instance}}]-[{{pod}}]-[{{scheme}}]` |
-| Read P99 per Instance | `histogram_quantile(0.99, sum by(instance, pod, le, scheme) (rate(opendal_operation_duration_seconds_bucket{instance=~"$datanode",operation="read"}[$__rate_interval])))` | `timeseries` | Read P99 per Instance. | `prometheus` | `s` | `[{{instance}}]-[{{pod}}]-{{scheme}}` |
-| Write QPS per Instance | `sum by(instance, pod, scheme) (rate(opendal_operation_duration_seconds_count{instance=~"$datanode", operation="write"}[$__rate_interval]))` | `timeseries` | Write QPS per Instance. | `prometheus` | `ops` | `[{{instance}}]-[{{pod}}]-{{scheme}}` |
-| Write P99 per Instance | `histogram_quantile(0.99, sum by(instance, pod, le, scheme) (rate(opendal_operation_duration_seconds_bucket{instance=~"$datanode", operation="write"}[$__rate_interval])))` | `timeseries` | Write P99 per Instance. | `prometheus` | `s` | `[{{instance}}]-[{{pod}}]-[{{scheme}}]` |
+| Read QPS per Instance | `sum by(instance, pod, scheme, operation) (rate(opendal_operation_duration_seconds_count{instance=~"$datanode", operation=~"read\|Reader::read"}[$__rate_interval]))` | `timeseries` | Read QPS per Instance. | `prometheus` | `ops` | `[{{instance}}]-[{{pod}}]-[{{scheme}}]-[{{operation}}]` |
+| Read P99 per Instance | `histogram_quantile(0.99, sum by(instance, pod, le, scheme, operation) (rate(opendal_operation_duration_seconds_bucket{instance=~"$datanode",operation=~"read\|Reader::read"}[$__rate_interval])))` | `timeseries` | Read P99 per Instance. | `prometheus` | `s` | `[{{instance}}]-[{{pod}}]-[{{scheme}}]-[{{operation}}]` |
+| Write QPS per Instance | `sum by(instance, pod, scheme, operation) (rate(opendal_operation_duration_seconds_count{instance=~"$datanode", operation=~"write\|Writer::write\|Writer::close"}[$__rate_interval]))` | `timeseries` | Write QPS per Instance. | `prometheus` | `ops` | `[{{instance}}]-[{{pod}}]-[{{scheme}}]-[{{operation}}]` |
+| Write P99 per Instance | `histogram_quantile(0.99, sum by(instance, pod, le, scheme, operation) (rate(opendal_operation_duration_seconds_bucket{instance=~"$datanode", operation =~ "Writer::write\|Writer::close\|write"}[$__rate_interval])))` | `timeseries` | Write P99 per Instance. | `prometheus` | `s` | `[{{instance}}]-[{{pod}}]-[{{scheme}}]-[{{operation}}]` |
 | List QPS per Instance | `sum by(instance, pod, scheme) (rate(opendal_operation_duration_seconds_count{instance=~"$datanode", operation="list"}[$__rate_interval]))` | `timeseries` | List QPS per Instance. | `prometheus` | `ops` | `[{{instance}}]-[{{pod}}]-[{{scheme}}]` |
 | List P99 per Instance | `histogram_quantile(0.99, sum by(instance, pod, le, scheme) (rate(opendal_operation_duration_seconds_bucket{instance=~"$datanode", operation="list"}[$__rate_interval])))` | `timeseries` | List P99 per Instance. | `prometheus` | `s` | `[{{instance}}]-[{{pod}}]-[{{scheme}}]` |
 | Other Requests per Instance | `sum by(instance, pod, scheme, operation) (rate(opendal_operation_duration_seconds_count{instance=~"$datanode",operation!~"read\|write\|list\|stat"}[$__rate_interval]))` | `timeseries` | Other Requests per Instance. | `prometheus` | `ops` | `[{{instance}}]-[{{pod}}]-[{{scheme}}]-[{{operation}}]` |
-| Other Request P99 per Instance | `histogram_quantile(0.99, sum by(instance, pod, le, scheme, operation) (rate(opendal_operation_duration_seconds_bucket{instance=~"$datanode", operation!~"read\|write\|list"}[$__rate_interval])))` | `timeseries` | Other Request P99 per Instance. | `prometheus` | `s` | `[{{instance}}]-[{{pod}}]-[{{scheme}}]-[{{operation}}]` |
+| Other Request P99 per Instance | `histogram_quantile(0.99, sum by(instance, pod, le, scheme, operation) (rate(opendal_operation_duration_seconds_bucket{instance=~"$datanode", operation!~"read\|write\|list\|Writer::write\|Writer::close\|Reader::read"}[$__rate_interval])))` | `timeseries` | Other Request P99 per Instance. | `prometheus` | `s` | `[{{instance}}]-[{{pod}}]-[{{scheme}}]-[{{operation}}]` |
 | Opendal traffic | `sum by(instance, pod, scheme, operation) (rate(opendal_operation_bytes_sum{instance=~"$datanode"}[$__rate_interval]))` | `timeseries` | Total traffic as in bytes by instance and operation | `prometheus` | `decbytes` | `[{{instance}}]-[{{pod}}]-[{{scheme}}]-[{{operation}}]` |
 | OpenDAL errors per Instance | `sum by(instance, pod, scheme, operation, error) (rate(opendal_operation_errors_total{instance=~"$datanode", error!="NotFound"}[$__rate_interval]))` | `timeseries` | OpenDAL error counts per Instance. | `prometheus` | -- | `[{{instance}}]-[{{pod}}]-[{{scheme}}]-[{{operation}}]-[{{error}}]` |
 # Metasrv

grafana/dashboards/metrics/cluster/dashboard.yaml

@@ -659,41 +659,41 @@ groups:
           description: Read QPS per Instance.
           unit: ops
           queries:
-            - expr: sum by(instance, pod, scheme) (rate(opendal_operation_duration_seconds_count{instance=~"$datanode", operation="read"}[$__rate_interval]))
+            - expr: sum by(instance, pod, scheme, operation) (rate(opendal_operation_duration_seconds_count{instance=~"$datanode", operation=~"read|Reader::read"}[$__rate_interval]))
               datasource:
                 type: prometheus
                 uid: ${metrics}
-              legendFormat: '[{{instance}}]-[{{pod}}]-[{{scheme}}]'
+              legendFormat: '[{{instance}}]-[{{pod}}]-[{{scheme}}]-[{{operation}}]'
         - title: Read P99 per Instance
           type: timeseries
           description: Read P99 per Instance.
           unit: s
           queries:
-            - expr: histogram_quantile(0.99, sum by(instance, pod, le, scheme) (rate(opendal_operation_duration_seconds_bucket{instance=~"$datanode",operation="read"}[$__rate_interval])))
+            - expr: histogram_quantile(0.99, sum by(instance, pod, le, scheme, operation) (rate(opendal_operation_duration_seconds_bucket{instance=~"$datanode",operation=~"read|Reader::read"}[$__rate_interval])))
               datasource:
                 type: prometheus
                 uid: ${metrics}
-              legendFormat: '[{{instance}}]-[{{pod}}]-{{scheme}}'
+              legendFormat: '[{{instance}}]-[{{pod}}]-[{{scheme}}]-[{{operation}}]'
         - title: Write QPS per Instance
           type: timeseries
           description: Write QPS per Instance.
           unit: ops
           queries:
-            - expr: sum by(instance, pod, scheme) (rate(opendal_operation_duration_seconds_count{instance=~"$datanode", operation="write"}[$__rate_interval]))
+            - expr: sum by(instance, pod, scheme, operation) (rate(opendal_operation_duration_seconds_count{instance=~"$datanode", operation=~"write|Writer::write|Writer::close"}[$__rate_interval]))
               datasource:
                 type: prometheus
                 uid: ${metrics}
-              legendFormat: '[{{instance}}]-[{{pod}}]-{{scheme}}'
+              legendFormat: '[{{instance}}]-[{{pod}}]-[{{scheme}}]-[{{operation}}]'
         - title: Write P99 per Instance
           type: timeseries
           description: Write P99 per Instance.
           unit: s
           queries:
-            - expr: histogram_quantile(0.99, sum by(instance, pod, le, scheme) (rate(opendal_operation_duration_seconds_bucket{instance=~"$datanode", operation="write"}[$__rate_interval])))
+            - expr: histogram_quantile(0.99, sum by(instance, pod, le, scheme, operation) (rate(opendal_operation_duration_seconds_bucket{instance=~"$datanode", operation =~ "Writer::write|Writer::close|write"}[$__rate_interval])))
               datasource:
                 type: prometheus
                 uid: ${metrics}
-              legendFormat: '[{{instance}}]-[{{pod}}]-[{{scheme}}]'
+              legendFormat: '[{{instance}}]-[{{pod}}]-[{{scheme}}]-[{{operation}}]'
         - title: List QPS per Instance
           type: timeseries
           description: List QPS per Instance.
@@ -729,7 +729,7 @@ groups:
           description: Other Request P99 per Instance.
           unit: s
           queries:
-            - expr: histogram_quantile(0.99, sum by(instance, pod, le, scheme, operation) (rate(opendal_operation_duration_seconds_bucket{instance=~"$datanode", operation!~"read|write|list"}[$__rate_interval])))
+            - expr: histogram_quantile(0.99, sum by(instance, pod, le, scheme, operation) (rate(opendal_operation_duration_seconds_bucket{instance=~"$datanode", operation!~"read|write|list|Writer::write|Writer::close|Reader::read"}[$__rate_interval])))
               datasource:
                 type: prometheus
                 uid: ${metrics}

grafana/dashboards/metrics/standalone/dashboard.json

@@ -5954,9 +5954,9 @@
                 "uid": "${metrics}"
               },
               "editorMode": "code",
-              "expr": "sum by(instance, pod, scheme) (rate(opendal_operation_duration_seconds_count{ operation=\"read\"}[$__rate_interval]))",
+              "expr": "sum by(instance, pod, scheme, operation) (rate(opendal_operation_duration_seconds_count{ operation=~\"read|Reader::read\"}[$__rate_interval]))",
               "instant": false,
-              "legendFormat": "[{{instance}}]-[{{pod}}]-[{{scheme}}]",
+              "legendFormat": "[{{instance}}]-[{{pod}}]-[{{scheme}}]-[{{operation}}]",
               "range": true,
               "refId": "A"
             }
@@ -6055,9 +6055,9 @@
                 "uid": "${metrics}"
               },
               "editorMode": "code",
-              "expr": "histogram_quantile(0.99, sum by(instance, pod, le, scheme) (rate(opendal_operation_duration_seconds_bucket{operation=\"read\"}[$__rate_interval])))",
+              "expr": "histogram_quantile(0.99, sum by(instance, pod, le, scheme, operation) (rate(opendal_operation_duration_seconds_bucket{operation=~\"read|Reader::read\"}[$__rate_interval])))",
               "instant": false,
-              "legendFormat": "[{{instance}}]-[{{pod}}]-{{scheme}}",
+              "legendFormat": "[{{instance}}]-[{{pod}}]-[{{scheme}}]-[{{operation}}]",
               "range": true,
               "refId": "A"
             }
@@ -6156,9 +6156,9 @@
                 "uid": "${metrics}"
               },
               "editorMode": "code",
-              "expr": "sum by(instance, pod, scheme) (rate(opendal_operation_duration_seconds_count{ operation=\"write\"}[$__rate_interval]))",
+              "expr": "sum by(instance, pod, scheme, operation) (rate(opendal_operation_duration_seconds_count{ operation=~\"write|Writer::write|Writer::close\"}[$__rate_interval]))",
               "instant": false,
-              "legendFormat": "[{{instance}}]-[{{pod}}]-{{scheme}}",
+              "legendFormat": "[{{instance}}]-[{{pod}}]-[{{scheme}}]-[{{operation}}]",
               "range": true,
               "refId": "A"
             }
@@ -6257,9 +6257,9 @@
                 "uid": "${metrics}"
               },
               "editorMode": "code",
-              "expr": "histogram_quantile(0.99, sum by(instance, pod, le, scheme) (rate(opendal_operation_duration_seconds_bucket{ operation=\"write\"}[$__rate_interval])))",
+              "expr": "histogram_quantile(0.99, sum by(instance, pod, le, scheme, operation) (rate(opendal_operation_duration_seconds_bucket{ operation =~ \"Writer::write|Writer::close|write\"}[$__rate_interval])))",
               "instant": false,
-              "legendFormat": "[{{instance}}]-[{{pod}}]-[{{scheme}}]",
+              "legendFormat": "[{{instance}}]-[{{pod}}]-[{{scheme}}]-[{{operation}}]",
               "range": true,
               "refId": "A"
             }
@@ -6663,7 +6663,7 @@
                 "uid": "${metrics}"
               },
               "editorMode": "code",
-              "expr": "histogram_quantile(0.99, sum by(instance, pod, le, scheme, operation) (rate(opendal_operation_duration_seconds_bucket{ operation!~\"read|write|list\"}[$__rate_interval])))",
+              "expr": "histogram_quantile(0.99, sum by(instance, pod, le, scheme, operation) (rate(opendal_operation_duration_seconds_bucket{ operation!~\"read|write|list|Writer::write|Writer::close|Reader::read\"}[$__rate_interval])))",
               "instant": false,
               "legendFormat": "[{{instance}}]-[{{pod}}]-[{{scheme}}]-[{{operation}}]",
               "range": true,

grafana/dashboards/metrics/standalone/dashboard.md

@@ -76,14 +76,14 @@
 | Title | Query | Type | Description | Datasource | Unit | Legend Format |
 | --- | --- | --- | --- | --- | --- | --- |
 | QPS per Instance | `sum by(instance, pod, scheme, operation) (rate(opendal_operation_duration_seconds_count{}[$__rate_interval]))` | `timeseries` | QPS per Instance. | `prometheus` | `ops` | `[{{instance}}]-[{{pod}}]-[{{scheme}}]-[{{operation}}]` |
-| Read QPS per Instance | `sum by(instance, pod, scheme) (rate(opendal_operation_duration_seconds_count{ operation="read"}[$__rate_interval]))` | `timeseries` | Read QPS per Instance. | `prometheus` | `ops` | `[{{instance}}]-[{{pod}}]-[{{scheme}}]` |
-| Read P99 per Instance | `histogram_quantile(0.99, sum by(instance, pod, le, scheme) (rate(opendal_operation_duration_seconds_bucket{operation="read"}[$__rate_interval])))` | `timeseries` | Read P99 per Instance. | `prometheus` | `s` | `[{{instance}}]-[{{pod}}]-{{scheme}}` |
-| Write QPS per Instance | `sum by(instance, pod, scheme) (rate(opendal_operation_duration_seconds_count{ operation="write"}[$__rate_interval]))` | `timeseries` | Write QPS per Instance. | `prometheus` | `ops` | `[{{instance}}]-[{{pod}}]-{{scheme}}` |
-| Write P99 per Instance | `histogram_quantile(0.99, sum by(instance, pod, le, scheme) (rate(opendal_operation_duration_seconds_bucket{ operation="write"}[$__rate_interval])))` | `timeseries` | Write P99 per Instance. | `prometheus` | `s` | `[{{instance}}]-[{{pod}}]-[{{scheme}}]` |
+| Read QPS per Instance | `sum by(instance, pod, scheme, operation) (rate(opendal_operation_duration_seconds_count{ operation=~"read\|Reader::read"}[$__rate_interval]))` | `timeseries` | Read QPS per Instance. | `prometheus` | `ops` | `[{{instance}}]-[{{pod}}]-[{{scheme}}]-[{{operation}}]` |
+| Read P99 per Instance | `histogram_quantile(0.99, sum by(instance, pod, le, scheme, operation) (rate(opendal_operation_duration_seconds_bucket{operation=~"read\|Reader::read"}[$__rate_interval])))` | `timeseries` | Read P99 per Instance. | `prometheus` | `s` | `[{{instance}}]-[{{pod}}]-[{{scheme}}]-[{{operation}}]` |
+| Write QPS per Instance | `sum by(instance, pod, scheme, operation) (rate(opendal_operation_duration_seconds_count{ operation=~"write\|Writer::write\|Writer::close"}[$__rate_interval]))` | `timeseries` | Write QPS per Instance. | `prometheus` | `ops` | `[{{instance}}]-[{{pod}}]-[{{scheme}}]-[{{operation}}]` |
+| Write P99 per Instance | `histogram_quantile(0.99, sum by(instance, pod, le, scheme, operation) (rate(opendal_operation_duration_seconds_bucket{ operation =~ "Writer::write\|Writer::close\|write"}[$__rate_interval])))` | `timeseries` | Write P99 per Instance. | `prometheus` | `s` | `[{{instance}}]-[{{pod}}]-[{{scheme}}]-[{{operation}}]` |
 | List QPS per Instance | `sum by(instance, pod, scheme) (rate(opendal_operation_duration_seconds_count{ operation="list"}[$__rate_interval]))` | `timeseries` | List QPS per Instance. | `prometheus` | `ops` | `[{{instance}}]-[{{pod}}]-[{{scheme}}]` |
 | List P99 per Instance | `histogram_quantile(0.99, sum by(instance, pod, le, scheme) (rate(opendal_operation_duration_seconds_bucket{ operation="list"}[$__rate_interval])))` | `timeseries` | List P99 per Instance. | `prometheus` | `s` | `[{{instance}}]-[{{pod}}]-[{{scheme}}]` |
 | Other Requests per Instance | `sum by(instance, pod, scheme, operation) (rate(opendal_operation_duration_seconds_count{operation!~"read\|write\|list\|stat"}[$__rate_interval]))` | `timeseries` | Other Requests per Instance. | `prometheus` | `ops` | `[{{instance}}]-[{{pod}}]-[{{scheme}}]-[{{operation}}]` |
-| Other Request P99 per Instance | `histogram_quantile(0.99, sum by(instance, pod, le, scheme, operation) (rate(opendal_operation_duration_seconds_bucket{ operation!~"read\|write\|list"}[$__rate_interval])))` | `timeseries` | Other Request P99 per Instance. | `prometheus` | `s` | `[{{instance}}]-[{{pod}}]-[{{scheme}}]-[{{operation}}]` |
+| Other Request P99 per Instance | `histogram_quantile(0.99, sum by(instance, pod, le, scheme, operation) (rate(opendal_operation_duration_seconds_bucket{ operation!~"read\|write\|list\|Writer::write\|Writer::close\|Reader::read"}[$__rate_interval])))` | `timeseries` | Other Request P99 per Instance. | `prometheus` | `s` | `[{{instance}}]-[{{pod}}]-[{{scheme}}]-[{{operation}}]` |
 | Opendal traffic | `sum by(instance, pod, scheme, operation) (rate(opendal_operation_bytes_sum{}[$__rate_interval]))` | `timeseries` | Total traffic as in bytes by instance and operation | `prometheus` | `decbytes` | `[{{instance}}]-[{{pod}}]-[{{scheme}}]-[{{operation}}]` |
 | OpenDAL errors per Instance | `sum by(instance, pod, scheme, operation, error) (rate(opendal_operation_errors_total{ error!="NotFound"}[$__rate_interval]))` | `timeseries` | OpenDAL error counts per Instance. | `prometheus` | -- | `[{{instance}}]-[{{pod}}]-[{{scheme}}]-[{{operation}}]-[{{error}}]` |
 # Metasrv

grafana/dashboards/metrics/standalone/dashboard.yaml

@@ -659,41 +659,41 @@ groups:
           description: Read QPS per Instance.
           unit: ops
           queries:
-            - expr: sum by(instance, pod, scheme) (rate(opendal_operation_duration_seconds_count{ operation="read"}[$__rate_interval]))
+            - expr: sum by(instance, pod, scheme, operation) (rate(opendal_operation_duration_seconds_count{ operation=~"read|Reader::read"}[$__rate_interval]))
               datasource:
                 type: prometheus
                 uid: ${metrics}
-              legendFormat: '[{{instance}}]-[{{pod}}]-[{{scheme}}]'
+              legendFormat: '[{{instance}}]-[{{pod}}]-[{{scheme}}]-[{{operation}}]'
         - title: Read P99 per Instance
           type: timeseries
           description: Read P99 per Instance.
           unit: s
           queries:
-            - expr: histogram_quantile(0.99, sum by(instance, pod, le, scheme) (rate(opendal_operation_duration_seconds_bucket{operation="read"}[$__rate_interval])))
+            - expr: histogram_quantile(0.99, sum by(instance, pod, le, scheme, operation) (rate(opendal_operation_duration_seconds_bucket{operation=~"read|Reader::read"}[$__rate_interval])))
               datasource:
                 type: prometheus
                 uid: ${metrics}
-              legendFormat: '[{{instance}}]-[{{pod}}]-{{scheme}}'
+              legendFormat: '[{{instance}}]-[{{pod}}]-[{{scheme}}]-[{{operation}}]'
         - title: Write QPS per Instance
           type: timeseries
           description: Write QPS per Instance.
           unit: ops
           queries:
-            - expr: sum by(instance, pod, scheme) (rate(opendal_operation_duration_seconds_count{ operation="write"}[$__rate_interval]))
+            - expr: sum by(instance, pod, scheme, operation) (rate(opendal_operation_duration_seconds_count{ operation=~"write|Writer::write|Writer::close"}[$__rate_interval]))
               datasource:
                 type: prometheus
                 uid: ${metrics}
-              legendFormat: '[{{instance}}]-[{{pod}}]-{{scheme}}'
+              legendFormat: '[{{instance}}]-[{{pod}}]-[{{scheme}}]-[{{operation}}]'
         - title: Write P99 per Instance
           type: timeseries
           description: Write P99 per Instance.
           unit: s
           queries:
-            - expr: histogram_quantile(0.99, sum by(instance, pod, le, scheme) (rate(opendal_operation_duration_seconds_bucket{ operation="write"}[$__rate_interval])))
+            - expr: histogram_quantile(0.99, sum by(instance, pod, le, scheme, operation) (rate(opendal_operation_duration_seconds_bucket{ operation =~ "Writer::write|Writer::close|write"}[$__rate_interval])))
               datasource:
                 type: prometheus
                 uid: ${metrics}
-              legendFormat: '[{{instance}}]-[{{pod}}]-[{{scheme}}]'
+              legendFormat: '[{{instance}}]-[{{pod}}]-[{{scheme}}]-[{{operation}}]'
         - title: List QPS per Instance
           type: timeseries
           description: List QPS per Instance.
@@ -729,7 +729,7 @@ groups:
           description: Other Request P99 per Instance.
           unit: s
           queries:
-            - expr: histogram_quantile(0.99, sum by(instance, pod, le, scheme, operation) (rate(opendal_operation_duration_seconds_bucket{ operation!~"read|write|list"}[$__rate_interval])))
+            - expr: histogram_quantile(0.99, sum by(instance, pod, le, scheme, operation) (rate(opendal_operation_duration_seconds_bucket{ operation!~"read|write|list|Writer::write|Writer::close|Reader::read"}[$__rate_interval])))
               datasource:
                 type: prometheus
                 uid: ${metrics}

scripts/check-super-imports.py

@@ -13,8 +13,8 @@
 # limitations under the License.
 
 import os
-import re
 from multiprocessing import Pool
+from pathlib import Path
 
 
 def find_rust_files(directory):
@@ -24,6 +24,10 @@ def find_rust_files(directory):
         if "test" in root.lower():
             continue
 
+        # Skip the target directory
+        if "target" in Path(root).parts:
+            continue
+
         for file in files:
             # Skip files with "test" in the filename
             if "test" in file.lower():

scripts/install.sh

@@ -53,6 +53,54 @@ get_arch_type() {
   esac
 }
 
+# Verify SHA256 checksum
+verify_sha256() {
+  file="$1"
+  expected_sha256="$2"
+
+  if command -v sha256sum >/dev/null 2>&1; then
+    actual_sha256=$(sha256sum "$file" | cut -d' ' -f1)
+  elif command -v shasum >/dev/null 2>&1; then
+    actual_sha256=$(shasum -a 256 "$file" | cut -d' ' -f1)
+  else
+    echo "Warning: No SHA256 verification tool found (sha256sum or shasum). Skipping checksum verification."
+    return 0
+  fi
+
+  if [ "$actual_sha256" = "$expected_sha256" ]; then
+    echo "SHA256 checksum verified successfully."
+    return 0
+  else
+    echo "Error: SHA256 checksum verification failed!"
+    echo "Expected: $expected_sha256"
+    echo "Actual: $actual_sha256"
+    return 1
+  fi
+}
+
+# Prompt for user confirmation (compatible with different shells)
+prompt_confirmation() {
+  message="$1"
+  printf "%s (y/N): " "$message"
+
+  # Try to read user input, fallback if read fails
+  answer=""
+  if read answer </dev/tty 2>/dev/null; then
+    case "$answer" in
+      [Yy]|[Yy][Ee][Ss])
+        return 0
+        ;;
+      *)
+        return 1
+        ;;
+    esac
+  else
+    echo ""
+    echo "Cannot read user input. Defaulting to No."
+    return 1
+  fi
+}
+
 download_artifact() {
   if [ -n "${OS_TYPE}" ] && [ -n "${ARCH_TYPE}" ]; then
     # Use the latest stable released version.
@@ -71,17 +119,104 @@ download_artifact() {
     fi
 
     echo "Downloading ${BIN}, OS: ${OS_TYPE}, Arch: ${ARCH_TYPE}, Version: ${VERSION}"
-    PACKAGE_NAME="${BIN}-${OS_TYPE}-${ARCH_TYPE}-${VERSION}.tar.gz"
+    PKG_NAME="${BIN}-${OS_TYPE}-${ARCH_TYPE}-${VERSION}"
+    PACKAGE_NAME="${PKG_NAME}.tar.gz"
+    SHA256_FILE="${PKG_NAME}.sha256sum"
 
     if [ -n "${PACKAGE_NAME}" ]; then
-      wget "https://github.com/${GITHUB_ORG}/${GITHUB_REPO}/releases/download/${VERSION}/${PACKAGE_NAME}"
+      # Check if files already exist and prompt for override
+      if [ -f "${PACKAGE_NAME}" ]; then
+        echo "File ${PACKAGE_NAME} already exists."
+        if prompt_confirmation "Do you want to override it?"; then
+          echo "Overriding existing file..."
+          rm -f "${PACKAGE_NAME}"
+        else
+          echo "Skipping download. Using existing file."
+        fi
+      fi
+
+      if [ -f "${BIN}" ]; then
+        echo "Binary ${BIN} already exists."
+        if prompt_confirmation "Do you want to override it?"; then
+          echo "Will override existing binary..."
+          rm -f "${BIN}"
+        else
+          echo "Installation cancelled."
+          exit 0
+        fi
+      fi
+
+      # Download package if not exists
+      if [ ! -f "${PACKAGE_NAME}" ]; then
+        echo "Downloading ${PACKAGE_NAME}..."
+        # Use curl instead of wget for better compatibility
+        if command -v curl >/dev/null 2>&1; then
+          if ! curl -L -o "${PACKAGE_NAME}" "https://github.com/${GITHUB_ORG}/${GITHUB_REPO}/releases/download/${VERSION}/${PACKAGE_NAME}"; then
+            echo "Error: Failed to download ${PACKAGE_NAME}"
+            exit 1
+          fi
+        elif command -v wget >/dev/null 2>&1; then
+          if ! wget -O "${PACKAGE_NAME}" "https://github.com/${GITHUB_ORG}/${GITHUB_REPO}/releases/download/${VERSION}/${PACKAGE_NAME}"; then
+            echo "Error: Failed to download ${PACKAGE_NAME}"
+            exit 1
+          fi
+        else
+          echo "Error: Neither curl nor wget is available for downloading."
+          exit 1
+        fi
+      fi
+
+      # Download and verify SHA256 checksum
+      echo "Downloading SHA256 checksum..."
+      sha256_download_success=0
+      if command -v curl >/dev/null 2>&1; then
+        if curl -L -s -o "${SHA256_FILE}" "https://github.com/${GITHUB_ORG}/${GITHUB_REPO}/releases/download/${VERSION}/${SHA256_FILE}" 2>/dev/null; then
+          sha256_download_success=1
+        fi
+      elif command -v wget >/dev/null 2>&1; then
+        if wget -q -O "${SHA256_FILE}" "https://github.com/${GITHUB_ORG}/${GITHUB_REPO}/releases/download/${VERSION}/${SHA256_FILE}" 2>/dev/null; then
+          sha256_download_success=1
+        fi
+      fi
+
+      if [ $sha256_download_success -eq 1 ] && [ -f "${SHA256_FILE}" ]; then
+        expected_sha256=$(cat "${SHA256_FILE}" | cut -d' ' -f1)
+        if [ -n "$expected_sha256" ]; then
+          if ! verify_sha256 "${PACKAGE_NAME}" "${expected_sha256}"; then
+            echo "SHA256 verification failed. Removing downloaded file."
+            rm -f "${PACKAGE_NAME}" "${SHA256_FILE}"
+            exit 1
+          fi
+        else
+          echo "Warning: Could not parse SHA256 checksum from file."
+        fi
+        rm -f "${SHA256_FILE}"
+      else
+        echo "Warning: Could not download SHA256 checksum file. Skipping verification."
+      fi
 
       # Extract the binary and clean the rest.
-      tar xvf "${PACKAGE_NAME}" && \
-      mv "${PACKAGE_NAME%.tar.gz}/${BIN}" "${PWD}" && \
-      rm -r "${PACKAGE_NAME}" && \
-      rm -r "${PACKAGE_NAME%.tar.gz}" && \
-      echo "Run './${BIN} --help' to get started"
+      echo "Extracting ${PACKAGE_NAME}..."
+      if ! tar xf "${PACKAGE_NAME}"; then
+        echo "Error: Failed to extract ${PACKAGE_NAME}"
+        exit 1
+      fi
+
+      # Find the binary in the extracted directory
+      extracted_dir="${PACKAGE_NAME%.tar.gz}"
+      if [ -f "${extracted_dir}/${BIN}" ]; then
+        mv "${extracted_dir}/${BIN}" "${PWD}/"
+        rm -f "${PACKAGE_NAME}"
+        rm -rf "${extracted_dir}"
+        chmod +x "${BIN}"
+        echo "Installation completed successfully!"
+        echo "Run './${BIN} --help' to get started"
+      else
+        echo "Error: Binary ${BIN} not found in extracted archive"
+        rm -f "${PACKAGE_NAME}"
+        rm -rf "${extracted_dir}"
+        exit 1
+      fi
     fi
   fi
 }

src/catalog/src/system_schema/information_schema.rs

@@ -162,6 +162,16 @@ impl SystemSchemaProviderInner for InformationSchemaProvider {
     }
 
     fn system_table(&self, name: &str) -> Option<SystemTableRef> {
+        #[cfg(feature = "enterprise")]
+        if let Some(factory) = self.extra_table_factories.get(name) {
+            let req = MakeInformationTableRequest {
+                catalog_name: self.catalog_name.clone(),
+                catalog_manager: self.catalog_manager.clone(),
+                kv_backend: self.kv_backend.clone(),
+            };
+            return Some(factory.make_information_table(req));
+        }
+
         match name.to_ascii_lowercase().as_str() {
             TABLES => Some(Arc::new(InformationSchemaTables::new(
                 self.catalog_name.clone(),
@@ -240,22 +250,7 @@ impl SystemSchemaProviderInner for InformationSchemaProvider {
                 .process_manager
                 .as_ref()
                 .map(|p| Arc::new(InformationSchemaProcessList::new(p.clone())) as _),
-            table_name => {
-                #[cfg(feature = "enterprise")]
-                return self.extra_table_factories.get(table_name).map(|factory| {
-                    let req = MakeInformationTableRequest {
-                        catalog_name: self.catalog_name.clone(),
-                        catalog_manager: self.catalog_manager.clone(),
-                        kv_backend: self.kv_backend.clone(),
-                    };
-                    factory.make_information_table(req)
-                });
-                #[cfg(not(feature = "enterprise"))]
-                {
-                    let _ = table_name;
-                    None
-                }
-            }
+            _ => None,
         }
     }
 }

src/catalog/src/system_schema/information_schema/information_memory_table.rs

@@ -15,7 +15,8 @@
 use std::sync::Arc;
 
 use common_catalog::consts::{METRIC_ENGINE, MITO_ENGINE};
-use datatypes::schema::{Schema, SchemaRef};
+use datatypes::data_type::ConcreteDataType;
+use datatypes::schema::{ColumnSchema, Schema, SchemaRef};
 use datatypes::vectors::{Int64Vector, StringVector, VectorRef};
 
 use crate::system_schema::information_schema::table_names::*;
@@ -367,28 +368,18 @@ pub(super) fn get_schema_columns(table_name: &str) -> (SchemaRef, Vec<VectorRef>
 
         TRIGGERS => (
             vec![
-                string_column("TRIGGER_CATALOG"),
-                string_column("TRIGGER_SCHEMA"),
                 string_column("TRIGGER_NAME"),
-                string_column("EVENT_MANIPULATION"),
-                string_column("EVENT_OBJECT_CATALOG"),
-                string_column("EVENT_OBJECT_SCHEMA"),
-                string_column("EVENT_OBJECT_TABLE"),
-                bigint_column("ACTION_ORDER"),
-                string_column("ACTION_CONDITION"),
-                string_column("ACTION_STATEMENT"),
-                string_column("ACTION_ORIENTATION"),
-                string_column("ACTION_TIMING"),
-                string_column("ACTION_REFERENCE_OLD_TABLE"),
-                string_column("ACTION_REFERENCE_NEW_TABLE"),
-                string_column("ACTION_REFERENCE_OLD_ROW"),
-                string_column("ACTION_REFERENCE_NEW_ROW"),
-                timestamp_micro_column("CREATED"),
-                string_column("SQL_MODE"),
-                string_column("DEFINER"),
-                string_column("CHARACTER_SET_CLIENT"),
-                string_column("COLLATION_CONNECTION"),
-                string_column("DATABASE_COLLATION"),
+                ColumnSchema::new(
+                    "trigger_id",
+                    ConcreteDataType::uint64_datatype(),
+                    false,
+                ),
+                string_column("TRIGGER_DEFINITION"),
+                ColumnSchema::new(
+                    "flownode_id",
+                    ConcreteDataType::uint64_datatype(),
+                    true,
+                ),
             ],
             vec![],
         ),

src/catalog/src/system_schema/information_schema/partitions.rs

@@ -329,13 +329,8 @@ impl InformationSchemaPartitionsBuilder {
             self.partition_names.push(Some(&partition_name));
             self.partition_ordinal_positions
                 .push(Some((index + 1) as i64));
-            let expressions = if partition.partition.partition_columns().is_empty() {
-                None
-            } else {
-                Some(partition.partition.to_string())
-            };
-
-            self.partition_expressions.push(expressions.as_deref());
+            let expression = partition.partition_expr.as_ref().map(|e| e.to_string());
+            self.partition_expressions.push(expression.as_deref());
             self.create_times.push(Some(TimestampMicrosecond::from(
                 table_info.meta.created_on.timestamp_millis(),
             )));

src/catalog/src/system_schema/information_schema/region_statistics.rs

@@ -44,6 +44,7 @@ const DISK_SIZE: &str = "disk_size";
 const MEMTABLE_SIZE: &str = "memtable_size";
 const MANIFEST_SIZE: &str = "manifest_size";
 const SST_SIZE: &str = "sst_size";
+const SST_NUM: &str = "sst_num";
 const INDEX_SIZE: &str = "index_size";
 const ENGINE: &str = "engine";
 const REGION_ROLE: &str = "region_role";
@@ -87,6 +88,7 @@ impl InformationSchemaRegionStatistics {
             ColumnSchema::new(MEMTABLE_SIZE, ConcreteDataType::uint64_datatype(), true),
             ColumnSchema::new(MANIFEST_SIZE, ConcreteDataType::uint64_datatype(), true),
             ColumnSchema::new(SST_SIZE, ConcreteDataType::uint64_datatype(), true),
+            ColumnSchema::new(SST_NUM, ConcreteDataType::uint64_datatype(), true),
             ColumnSchema::new(INDEX_SIZE, ConcreteDataType::uint64_datatype(), true),
             ColumnSchema::new(ENGINE, ConcreteDataType::string_datatype(), true),
             ColumnSchema::new(REGION_ROLE, ConcreteDataType::string_datatype(), true),
@@ -149,6 +151,7 @@ struct InformationSchemaRegionStatisticsBuilder {
     memtable_sizes: UInt64VectorBuilder,
     manifest_sizes: UInt64VectorBuilder,
     sst_sizes: UInt64VectorBuilder,
+    sst_nums: UInt64VectorBuilder,
     index_sizes: UInt64VectorBuilder,
     engines: StringVectorBuilder,
     region_roles: StringVectorBuilder,
@@ -167,6 +170,7 @@ impl InformationSchemaRegionStatisticsBuilder {
             memtable_sizes: UInt64VectorBuilder::with_capacity(INIT_CAPACITY),
             manifest_sizes: UInt64VectorBuilder::with_capacity(INIT_CAPACITY),
             sst_sizes: UInt64VectorBuilder::with_capacity(INIT_CAPACITY),
+            sst_nums: UInt64VectorBuilder::with_capacity(INIT_CAPACITY),
             index_sizes: UInt64VectorBuilder::with_capacity(INIT_CAPACITY),
             engines: StringVectorBuilder::with_capacity(INIT_CAPACITY),
             region_roles: StringVectorBuilder::with_capacity(INIT_CAPACITY),
@@ -197,6 +201,7 @@ impl InformationSchemaRegionStatisticsBuilder {
             (MEMTABLE_SIZE, &Value::from(region_stat.memtable_size)),
             (MANIFEST_SIZE, &Value::from(region_stat.manifest_size)),
             (SST_SIZE, &Value::from(region_stat.sst_size)),
+            (SST_NUM, &Value::from(region_stat.sst_num)),
             (INDEX_SIZE, &Value::from(region_stat.index_size)),
             (ENGINE, &Value::from(region_stat.engine.as_str())),
             (REGION_ROLE, &Value::from(region_stat.role.to_string())),
@@ -215,6 +220,7 @@ impl InformationSchemaRegionStatisticsBuilder {
         self.memtable_sizes.push(Some(region_stat.memtable_size));
         self.manifest_sizes.push(Some(region_stat.manifest_size));
         self.sst_sizes.push(Some(region_stat.sst_size));
+        self.sst_nums.push(Some(region_stat.sst_num));
         self.index_sizes.push(Some(region_stat.index_size));
         self.engines.push(Some(&region_stat.engine));
         self.region_roles.push(Some(&region_stat.role.to_string()));
@@ -230,6 +236,7 @@ impl InformationSchemaRegionStatisticsBuilder {
             Arc::new(self.memtable_sizes.finish()),
             Arc::new(self.manifest_sizes.finish()),
             Arc::new(self.sst_sizes.finish()),
+            Arc::new(self.sst_nums.finish()),
             Arc::new(self.index_sizes.finish()),
             Arc::new(self.engines.finish()),
             Arc::new(self.region_roles.finish()),

src/catalog/src/system_schema/information_schema/table_names.rs

@@ -48,4 +48,3 @@ pub const FLOWS: &str = "flows";
 pub const PROCEDURE_INFO: &str = "procedure_info";
 pub const REGION_STATISTICS: &str = "region_statistics";
 pub const PROCESS_LIST: &str = "process_list";
-pub const TRIGGER_LIST: &str = "trigger_list";

src/catalog/src/system_schema/pg_catalog/pg_class.rs

@@ -169,7 +169,7 @@ impl DfPartitionStream for PGClass {
 }
 
 /// Builds the `pg_catalog.pg_class` table row by row
-/// TODO(J0HN50N133): `relowner` is always the [`DUMMY_OWNER_ID`] cuz we don't have user.
+/// TODO(J0HN50N133): `relowner` is always the [`DUMMY_OWNER_ID`] because we don't have users.
 /// Once we have user system, make it the actual owner of the table.
 struct PGClassBuilder {
     schema: SchemaRef,

src/cli/src/bench.rs

@@ -188,6 +188,7 @@ fn create_region_routes(regions: Vec<RegionNumber>) -> Vec<RegionRoute> {
                 name: String::new(),
                 partition: None,
                 attrs: BTreeMap::new(),
+                partition_expr: Default::default(),
             },
             leader_peer: Some(Peer {
                 id: rng.random_range(0..10),

src/cli/src/metadata/repair.rs

@@ -241,7 +241,6 @@ impl RepairTool {
             let alter_table_request = alter_table::make_alter_region_request_for_peer(
                 logical_table_id,
                 &alter_table_expr,
-                full_table_metadata.table_info.ident.version,
                 peer,
                 physical_region_routes,
             )?;

src/cli/src/metadata/repair/alter_table.rs

@@ -66,7 +66,6 @@ pub fn generate_alter_table_expr_for_all_columns(
 pub fn make_alter_region_request_for_peer(
     logical_table_id: TableId,
     alter_table_expr: &AlterTableExpr,
-    schema_version: u64,
     peer: &Peer,
     region_routes: &[RegionRoute],
 ) -> Result<RegionRequest> {
@@ -74,7 +73,7 @@ pub fn make_alter_region_request_for_peer(
     let mut requests = Vec::with_capacity(regions_on_this_peer.len());
     for region_number in &regions_on_this_peer {
         let region_id = RegionId::new(logical_table_id, *region_number);
-        let request = make_alter_region_request(region_id, alter_table_expr, schema_version);
+        let request = make_alter_region_request(region_id, alter_table_expr);
         requests.push(request);
     }
 

src/client/src/database.rs

@@ -23,7 +23,7 @@ use api::v1::greptime_request::Request;
 use api::v1::query_request::Query;
 use api::v1::{
     AlterTableExpr, AuthHeader, Basic, CreateTableExpr, DdlRequest, GreptimeRequest,
-    InsertRequests, QueryRequest, RequestHeader,
+    InsertRequests, QueryRequest, RequestHeader, RowInsertRequests,
 };
 use arrow_flight::{FlightData, Ticket};
 use async_stream::stream;
@@ -42,7 +42,7 @@ use common_telemetry::{error, warn};
 use futures::future;
 use futures_util::{Stream, StreamExt, TryStreamExt};
 use prost::Message;
-use snafu::{ensure, ResultExt};
+use snafu::{ensure, OptionExt, ResultExt};
 use tonic::metadata::{AsciiMetadataKey, AsciiMetadataValue, MetadataMap, MetadataValue};
 use tonic::transport::Channel;
 
@@ -118,6 +118,7 @@ impl Database {
         }
     }
 
+    /// Set the catalog for the database client.
     pub fn set_catalog(&mut self, catalog: impl Into<String>) {
         self.catalog = catalog.into();
     }
@@ -130,6 +131,7 @@ impl Database {
         }
     }
 
+    /// Set the schema for the database client.
     pub fn set_schema(&mut self, schema: impl Into<String>) {
         self.schema = schema.into();
     }
@@ -142,20 +144,24 @@ impl Database {
         }
     }
 
+    /// Set the timezone for the database client.
     pub fn set_timezone(&mut self, timezone: impl Into<String>) {
         self.timezone = timezone.into();
     }
 
+    /// Set the auth scheme for the database client.
     pub fn set_auth(&mut self, auth: AuthScheme) {
         self.ctx.auth_header = Some(AuthHeader {
             auth_scheme: Some(auth),
         });
     }
 
+    /// Make an InsertRequests request to the database.
     pub async fn insert(&self, requests: InsertRequests) -> Result<u32> {
         self.handle(Request::Inserts(requests)).await
     }
 
+    /// Make an InsertRequests request to the database with hints.
     pub async fn insert_with_hints(
         &self,
         requests: InsertRequests,
@@ -172,6 +178,28 @@ impl Database {
         from_grpc_response(response)
     }
 
+    /// Make a RowInsertRequests request to the database.
+    pub async fn row_inserts(&self, requests: RowInsertRequests) -> Result<u32> {
+        self.handle(Request::RowInserts(requests)).await
+    }
+
+    /// Make a RowInsertRequests request to the database with hints.
+    pub async fn row_inserts_with_hints(
+        &self,
+        requests: RowInsertRequests,
+        hints: &[(&str, &str)],
+    ) -> Result<u32> {
+        let mut client = make_database_client(&self.client)?.inner;
+        let request = self.to_rpc_request(Request::RowInserts(requests));
+
+        let mut request = tonic::Request::new(request);
+        let metadata = request.metadata_mut();
+        Self::put_hints(metadata, hints)?;
+
+        let response = client.handle(request).await?.into_inner();
+        from_grpc_response(response)
+    }
+
     fn put_hints(metadata: &mut MetadataMap, hints: &[(&str, &str)]) -> Result<()> {
         let Some(value) = hints
             .iter()
@@ -187,6 +215,7 @@ impl Database {
         Ok(())
     }
 
+    /// Make a request to the database.
     pub async fn handle(&self, request: Request) -> Result<u32> {
         let mut client = make_database_client(&self.client)?.inner;
         let request = self.to_rpc_request(request);
@@ -221,12 +250,18 @@ impl Database {
                         retries += 1;
                         warn!("Retrying {} times with error = {:?}", retries, err);
                         continue;
+                    } else {
+                        error!(
+                            err; "Failed to send request to grpc handle, retries = {}, not retryable error, aborting",
+                            retries
+                        );
+                        return Err(err.into());
                     }
                 }
                 (Err(err), false) => {
                     error!(
-                        "Failed to send request to grpc handle after {} retries, error = {:?}",
-                        retries, err
+                        err; "Failed to send request to grpc handle after {} retries",
+                        retries,
                     );
                     return Err(err.into());
                 }
@@ -250,6 +285,7 @@ impl Database {
         }
     }
 
+    /// Executes a SQL query without any hints.
     pub async fn sql<S>(&self, sql: S) -> Result<Output>
     where
         S: AsRef<str>,
@@ -257,6 +293,7 @@ impl Database {
         self.sql_with_hint(sql, &[]).await
     }
 
+    /// Executes a SQL query with optional hints for query optimization.
     pub async fn sql_with_hint<S>(&self, sql: S, hints: &[(&str, &str)]) -> Result<Output>
     where
         S: AsRef<str>,
@@ -267,6 +304,7 @@ impl Database {
         self.do_get(request, hints).await
     }
 
+    /// Executes a logical plan directly without SQL parsing.
     pub async fn logical_plan(&self, logical_plan: Vec<u8>) -> Result<Output> {
         let request = Request::Query(QueryRequest {
             query: Some(Query::LogicalPlan(logical_plan)),
@@ -274,6 +312,7 @@ impl Database {
         self.do_get(request, &[]).await
     }
 
+    /// Creates a new table using the provided table expression.
     pub async fn create(&self, expr: CreateTableExpr) -> Result<Output> {
         let request = Request::Ddl(DdlRequest {
             expr: Some(DdlExpr::CreateTable(expr)),
@@ -281,6 +320,7 @@ impl Database {
         self.do_get(request, &[]).await
     }
 
+    /// Alters an existing table using the provided alter expression.
     pub async fn alter(&self, expr: AlterTableExpr) -> Result<Output> {
         let request = Request::Ddl(DdlRequest {
             expr: Some(DdlExpr::AlterTable(expr)),
@@ -321,7 +361,10 @@ impl Database {
         let mut flight_message_stream = flight_data_stream.map(move |flight_data| {
             flight_data
                 .map_err(Error::from)
-                .and_then(|data| decoder.try_decode(&data).context(ConvertFlightDataSnafu))
+                .and_then(|data| decoder.try_decode(&data).context(ConvertFlightDataSnafu))?
+                .context(IllegalFlightMessagesSnafu {
+                    reason: "none message",
+                })
         });
 
         let Some(first_flight_message) = flight_message_stream.next().await else {

src/client/src/region.rs

@@ -128,7 +128,10 @@ impl RegionRequester {
         let mut flight_message_stream = flight_data_stream.map(move |flight_data| {
             flight_data
                 .map_err(Error::from)
-                .and_then(|data| decoder.try_decode(&data).context(ConvertFlightDataSnafu))
+                .and_then(|data| decoder.try_decode(&data).context(ConvertFlightDataSnafu))?
+                .context(IllegalFlightMessagesSnafu {
+                    reason: "none message",
+                })
         });
 
         let Some(first_flight_message) = flight_message_stream.next().await else {
@@ -157,19 +160,70 @@ impl RegionRequester {
             let _span = tracing_context.attach(common_telemetry::tracing::info_span!(
                 "poll_flight_data_stream"
             ));
-            while let Some(flight_message) = flight_message_stream.next().await {
-                let flight_message = flight_message
-                    .map_err(BoxedError::new)
-                    .context(ExternalSnafu)?;
+
+            let mut buffered_message: Option<FlightMessage> = None;
+            let mut stream_ended = false;
+
+            while !stream_ended {
+                // get the next message from the buffered message or read from the flight message stream
+                let flight_message_item = if let Some(msg) = buffered_message.take() {
+                    Some(Ok(msg))
+                } else {
+                    flight_message_stream.next().await
+                };
+
+                let flight_message = match flight_message_item {
+                    Some(Ok(message)) => message,
+                    Some(Err(e)) => {
+                        yield Err(BoxedError::new(e)).context(ExternalSnafu);
+                        break;
+                    }
+                    None => break,
+                };
 
                 match flight_message {
                     FlightMessage::RecordBatch(record_batch) => {
-                        yield RecordBatch::try_from_df_record_batch(
+                        let result_to_yield = RecordBatch::try_from_df_record_batch(
                             schema_cloned.clone(),
                             record_batch,
-                        )
+                        );
+
+                        // get the next message from the stream. normally it should be a metrics message.
+                        if let Some(next_flight_message_result) = flight_message_stream.next().await
+                        {
+                            match next_flight_message_result {
+                                Ok(FlightMessage::Metrics(s)) => {
+                                    let m = serde_json::from_str(&s).ok().map(Arc::new);
+                                    metrics_ref.swap(m);
+                                }
+                                Ok(FlightMessage::RecordBatch(rb)) => {
+                                    // for some reason it's not a metrics message, so we need to buffer this record batch
+                                    // and yield it in the next iteration.
+                                    buffered_message = Some(FlightMessage::RecordBatch(rb));
+                                }
+                                Ok(_) => {
+                                    yield IllegalFlightMessagesSnafu {
+                                        reason: "A RecordBatch message can only be succeeded by a Metrics message or another RecordBatch message"
+                                    }
+                                    .fail()
+                                    .map_err(BoxedError::new)
+                                    .context(ExternalSnafu);
+                                    break;
+                                }
+                                Err(e) => {
+                                    yield Err(BoxedError::new(e)).context(ExternalSnafu);
+                                    break;
+                                }
+                            }
+                        } else {
+                            // the stream has ended
+                            stream_ended = true;
+                        }
+
+                        yield result_to_yield;
                     }
                     FlightMessage::Metrics(s) => {
+                        // just a branch in case of some metrics message comes after other things.
                         let m = serde_json::from_str(&s).ok().map(Arc::new);
                         metrics_ref.swap(m);
                         break;

src/cmd/src/bin/greptime.rs

@@ -20,11 +20,11 @@ use cmd::error::{InitTlsProviderSnafu, Result};
 use cmd::options::GlobalOptions;
 use cmd::{cli, datanode, flownode, frontend, metasrv, standalone, App};
 use common_base::Plugins;
-use common_version::version;
+use common_version::{verbose_version, version};
 use servers::install_ring_crypto_provider;
 
 #[derive(Parser)]
-#[command(name = "greptime", author, version, long_version = version(), about)]
+#[command(name = "greptime", author, version, long_version = verbose_version(), about)]
 #[command(propagate_version = true)]
 pub(crate) struct Command {
     #[clap(subcommand)]
@@ -143,10 +143,8 @@ async fn start(cli: Command) -> Result<()> {
 }
 
 fn setup_human_panic() {
-    human_panic::setup_panic!(
-        human_panic::Metadata::new("GreptimeDB", env!("CARGO_PKG_VERSION"))
-            .homepage("https://github.com/GreptimeTeam/greptimedb/discussions")
-    );
+    human_panic::setup_panic!(human_panic::Metadata::new("GreptimeDB", version())
+        .homepage("https://github.com/GreptimeTeam/greptimedb/discussions"));
 
     common_telemetry::set_panic_hook();
 }

src/cmd/src/datanode/builder.rs

@@ -19,7 +19,7 @@ use catalog::kvbackend::MetaKvBackend;
 use common_base::Plugins;
 use common_meta::cache::LayeredCacheRegistryBuilder;
 use common_telemetry::info;
-use common_version::{short_version, version};
+use common_version::{short_version, verbose_version};
 use datanode::datanode::DatanodeBuilder;
 use datanode::service::DatanodeServiceBuilder;
 use meta_client::MetaClientType;
@@ -67,7 +67,7 @@ impl InstanceBuilder {
             None,
         );
 
-        log_versions(version(), short_version(), APP_NAME);
+        log_versions(verbose_version(), short_version(), APP_NAME);
         create_resource_limit_metrics(APP_NAME);
 
         plugins::setup_datanode_plugins(plugins, &opts.plugins, dn_opts)

src/cmd/src/flownode.rs

@@ -32,7 +32,7 @@ use common_meta::key::flow::FlowMetadataManager;
 use common_meta::key::TableMetadataManager;
 use common_telemetry::info;
 use common_telemetry::logging::{TracingOptions, DEFAULT_LOGGING_DIR};
-use common_version::{short_version, version};
+use common_version::{short_version, verbose_version};
 use flow::{
     get_flow_auth_options, FlownodeBuilder, FlownodeInstance, FlownodeServiceBuilder,
     FrontendClient, FrontendInvoker,
@@ -279,7 +279,7 @@ impl StartCommand {
             None,
         );
 
-        log_versions(version(), short_version(), APP_NAME);
+        log_versions(verbose_version(), short_version(), APP_NAME);
         create_resource_limit_metrics(APP_NAME);
 
         info!("Flownode start command: {:#?}", self);
@@ -374,6 +374,7 @@ impl StartCommand {
             meta_client.clone(),
             flow_auth_header,
             opts.query.clone(),
+            opts.flow.batching_mode.clone(),
         );
         let frontend_client = Arc::new(frontend_client);
         let flownode_builder = FlownodeBuilder::new(

src/cmd/src/frontend.rs

@@ -33,7 +33,7 @@ use common_meta::heartbeat::handler::HandlerGroupExecutor;
 use common_telemetry::info;
 use common_telemetry::logging::{TracingOptions, DEFAULT_LOGGING_DIR};
 use common_time::timezone::set_default_timezone;
-use common_version::{short_version, version};
+use common_version::{short_version, verbose_version};
 use frontend::frontend::Frontend;
 use frontend::heartbeat::HeartbeatTask;
 use frontend::instance::builder::FrontendBuilder;
@@ -102,7 +102,7 @@ impl App for Instance {
 #[derive(Parser)]
 pub struct Command {
     #[clap(subcommand)]
-    subcmd: SubCommand,
+    pub subcmd: SubCommand,
 }
 
 impl Command {
@@ -116,7 +116,7 @@ impl Command {
 }
 
 #[derive(Parser)]
-enum SubCommand {
+pub enum SubCommand {
     Start(StartCommand),
 }
 
@@ -153,7 +153,7 @@ pub struct StartCommand {
     #[clap(long)]
     postgres_addr: Option<String>,
     #[clap(short, long)]
-    config_file: Option<String>,
+    pub config_file: Option<String>,
     #[clap(short, long)]
     influxdb_enable: Option<bool>,
     #[clap(long, value_delimiter = ',', num_args = 1..)]
@@ -169,7 +169,7 @@ pub struct StartCommand {
     #[clap(long)]
     disable_dashboard: Option<bool>,
     #[clap(long, default_value = "GREPTIMEDB_FRONTEND")]
-    env_prefix: String,
+    pub env_prefix: String,
 }
 
 impl StartCommand {
@@ -282,7 +282,7 @@ impl StartCommand {
             opts.component.slow_query.as_ref(),
         );
 
-        log_versions(version(), short_version(), APP_NAME);
+        log_versions(verbose_version(), short_version(), APP_NAME);
         create_resource_limit_metrics(APP_NAME);
 
         info!("Frontend start command: {:#?}", self);

src/cmd/src/lib.rs

@@ -112,7 +112,7 @@ pub trait App: Send {
 pub fn log_versions(version: &str, short_version: &str, app: &str) {
     // Report app version as gauge.
     APP_VERSION
-        .with_label_values(&[env!("CARGO_PKG_VERSION"), short_version, app])
+        .with_label_values(&[common_version::version(), short_version, app])
         .inc();
 
     // Log version and argument flags.

src/cmd/src/metasrv.rs

@@ -22,7 +22,7 @@ use common_base::Plugins;
 use common_config::Configurable;
 use common_telemetry::info;
 use common_telemetry::logging::{TracingOptions, DEFAULT_LOGGING_DIR};
-use common_version::{short_version, version};
+use common_version::{short_version, verbose_version};
 use meta_srv::bootstrap::MetasrvInstance;
 use meta_srv::metasrv::BackendImpl;
 use snafu::ResultExt;
@@ -324,7 +324,7 @@ impl StartCommand {
             None,
         );
 
-        log_versions(version(), short_version(), APP_NAME);
+        log_versions(verbose_version(), short_version(), APP_NAME);
         create_resource_limit_metrics(APP_NAME);
 
         info!("Metasrv start command: {:#?}", self);
@@ -340,12 +340,12 @@ impl StartCommand {
             .await
             .context(StartMetaServerSnafu)?;
 
-        let builder = meta_srv::bootstrap::metasrv_builder(&opts, plugins.clone(), None)
+        let builder = meta_srv::bootstrap::metasrv_builder(&opts, plugins, None)
             .await
             .context(error::BuildMetaServerSnafu)?;
         let metasrv = builder.build().await.context(error::BuildMetaServerSnafu)?;
 
-        let instance = MetasrvInstance::new(opts, plugins, metasrv)
+        let instance = MetasrvInstance::new(metasrv)
             .await
             .context(error::BuildMetaServerSnafu)?;
 

src/cmd/src/standalone.rs

@@ -51,7 +51,7 @@ use common_telemetry::logging::{
     LoggingOptions, SlowQueryOptions, TracingOptions, DEFAULT_LOGGING_DIR,
 };
 use common_time::timezone::set_default_timezone;
-use common_version::{short_version, version};
+use common_version::{short_version, verbose_version};
 use common_wal::config::DatanodeWalConfig;
 use datanode::config::{DatanodeOptions, ProcedureConfig, RegionEngineConfig, StorageConfig};
 use datanode::datanode::{Datanode, DatanodeBuilder};
@@ -485,7 +485,7 @@ impl StartCommand {
             opts.component.slow_query.as_ref(),
         );
 
-        log_versions(version(), short_version(), APP_NAME);
+        log_versions(verbose_version(), short_version(), APP_NAME);
         create_resource_limit_metrics(APP_NAME);
 
         info!("Standalone start command: {:#?}", self);
@@ -821,6 +821,7 @@ impl InformationExtension for StandaloneInformationExtension {
                     memtable_size: region_stat.memtable_size,
                     manifest_size: region_stat.manifest_size,
                     sst_size: region_stat.sst_size,
+                    sst_num: region_stat.sst_num,
                     index_size: region_stat.index_size,
                     region_manifest: region_stat.manifest.into(),
                     data_topic_latest_entry_id: region_stat.data_topic_latest_entry_id,

src/common/catalog/src/consts.rs

@@ -104,8 +104,6 @@ pub const INFORMATION_SCHEMA_PROCEDURE_INFO_TABLE_ID: u32 = 34;
 pub const INFORMATION_SCHEMA_REGION_STATISTICS_TABLE_ID: u32 = 35;
 /// id for information_schema.process_list
 pub const INFORMATION_SCHEMA_PROCESS_LIST_TABLE_ID: u32 = 36;
-/// id for information_schema.trigger_list (for greptimedb trigger)
-pub const INFORMATION_SCHEMA_TRIGGER_TABLE_ID: u32 = 37;
 
 // ----- End of information_schema tables -----
 

src/common/datasource/src/lib.rs

@@ -21,6 +21,7 @@ pub mod error;
 pub mod file_format;
 pub mod lister;
 pub mod object_store;
+pub mod parquet_writer;
 pub mod share_buffer;
 #[cfg(test)]
 pub mod test_util;

src/common/datasource/src/object_store/oss.rs

@@ -77,6 +77,11 @@ pub fn build_oss_backend(
 
     let op = ObjectStore::new(builder)
         .context(error::BuildBackendSnafu)?
+        .layer(
+            object_store::layers::RetryLayer::new()
+                .with_jitter()
+                .with_notify(object_store::util::PrintDetailedError),
+        )
         .layer(object_store::layers::LoggingLayer::default())
         .layer(object_store::layers::TracingLayer)
         .layer(object_store::layers::build_prometheus_metrics_layer(true))

src/common/datasource/src/object_store/s3.rs

@@ -85,6 +85,11 @@ pub fn build_s3_backend(
     // TODO(weny): Consider finding a better way to eliminate duplicate code.
     Ok(ObjectStore::new(builder)
         .context(error::BuildBackendSnafu)?
+        .layer(
+            object_store::layers::RetryLayer::new()
+                .with_jitter()
+                .with_notify(object_store::util::PrintDetailedError),
+        )
         .layer(object_store::layers::LoggingLayer::new(
             DefaultLoggingInterceptor,
         ))

src/common/datasource/src/parquet_writer.rs

@@ -0,0 +1,52 @@
+// Copyright 2023 Greptime Team
+//
+// 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 bytes::Bytes;
+use futures::future::BoxFuture;
+use object_store::Writer;
+use parquet::arrow::async_writer::AsyncFileWriter;
+use parquet::errors::ParquetError;
+
+/// Bridges opendal [Writer] with parquet [AsyncFileWriter].
+pub struct AsyncWriter {
+    inner: Writer,
+}
+
+impl AsyncWriter {
+    /// Create a [`AsyncWriter`] by given [`Writer`].
+    pub fn new(writer: Writer) -> Self {
+        Self { inner: writer }
+    }
+}
+
+impl AsyncFileWriter for AsyncWriter {
+    fn write(&mut self, bs: Bytes) -> BoxFuture<'_, parquet::errors::Result<()>> {
+        Box::pin(async move {
+            self.inner
+                .write(bs)
+                .await
+                .map_err(|err| ParquetError::External(Box::new(err)))
+        })
+    }
+
+    fn complete(&mut self) -> BoxFuture<'_, parquet::errors::Result<()>> {
+        Box::pin(async move {
+            self.inner
+                .close()
+                .await
+                .map(|_| ())
+                .map_err(|err| ParquetError::External(Box::new(err)))
+        })
+    }
+}

src/common/error/tests/ext.rs

@@ -84,24 +84,33 @@ fn test_to_string() {
     assert_eq!(result.unwrap_err().to_string(), "<root cause>");
 }
 
+fn normalize_path(s: &str) -> String {
+    s.replace('\\', "/")
+}
+
 #[test]
 fn test_debug_format() {
     let result = normal_error();
     let debug_output = format!("{:?}", result.unwrap_err());
-    let normalized_output = debug_output.replace('\\', "/");
+
     assert_eq!(
-        normalized_output,
-        r#"0: A normal error with "display" attribute, message "blabla", at src/common/error/tests/ext.rs:55:22
-1: PlainError { msg: "<root cause>", status_code: Unexpected }"#
+        normalize_path(&debug_output),
+        format!(
+            r#"0: A normal error with "display" attribute, message "blabla", at {}:55:22
+1: PlainError {{ msg: "<root cause>", status_code: Unexpected }}"#,
+            normalize_path(file!())
+        )
     );
 
     let result = transparent_error();
     let debug_output = format!("{:?}", result.unwrap_err());
-    let normalized_output = debug_output.replace('\\', "/");
     assert_eq!(
-        normalized_output,
-        r#"0: <transparent>, at src/common/error/tests/ext.rs:60:5
-1: PlainError { msg: "<root cause>", status_code: Unexpected }"#
+        normalize_path(&debug_output),
+        format!(
+            r#"0: <transparent>, at {}:60:5
+1: PlainError {{ msg: "<root cause>", status_code: Unexpected }}"#,
+            normalize_path(file!())
+        )
     );
 }
 

src/common/event-recorder/Cargo.toml

@@ -0,0 +1,25 @@
+[package]
+name = "common-event-recorder"
+version.workspace = true
+edition.workspace = true
+license.workspace = true
+
+[dependencies]
+api.workspace = true
+async-trait.workspace = true
+backon.workspace = true
+client.workspace = true
+common-error.workspace = true
+common-macro.workspace = true
+common-meta.workspace = true
+common-telemetry.workspace = true
+common-time.workspace = true
+serde.workspace = true
+serde_json.workspace = true
+snafu.workspace = true
+store-api.workspace = true
+tokio.workspace = true
+tokio-util.workspace = true
+
+[lints]
+workspace = true

src/common/event-recorder/src/error.rs

@@ -0,0 +1,53 @@
+// Copyright 2023 Greptime Team
+//
+// 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 api::v1::ColumnSchema;
+use common_error::ext::ErrorExt;
+use common_error::status_code::StatusCode;
+use common_macro::stack_trace_debug;
+use snafu::{Location, Snafu};
+
+#[derive(Snafu)]
+#[snafu(visibility(pub))]
+#[stack_trace_debug]
+pub enum Error {
+    #[snafu(display("No available frontend"))]
+    NoAvailableFrontend {
+        #[snafu(implicit)]
+        location: Location,
+    },
+
+    #[snafu(display("Mismatched schema, expected: {:?}, actual: {:?}", expected, actual))]
+    MismatchedSchema {
+        #[snafu(implicit)]
+        location: Location,
+        expected: Vec<ColumnSchema>,
+        actual: Vec<ColumnSchema>,
+    },
+}
+
+pub type Result<T> = std::result::Result<T, Error>;
+
+impl ErrorExt for Error {
+    fn status_code(&self) -> StatusCode {
+        match self {
+            Error::MismatchedSchema { .. } => StatusCode::InvalidArguments,
+            Error::NoAvailableFrontend { .. } => StatusCode::Internal,
+        }
+    }
+
+    fn as_any(&self) -> &dyn std::any::Any {
+        self
+    }
+}

src/common/event-recorder/src/lib.rs

@@ -0,0 +1,18 @@
+// Copyright 2023 Greptime Team
+//
+// 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 error;
+pub mod recorder;
+
+pub use recorder::*;

src/common/event-recorder/src/recorder.rs

@@ -0,0 +1,527 @@
+// Copyright 2023 Greptime Team
+//
+// 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::any::Any;
+use std::collections::HashMap;
+use std::fmt::Debug;
+use std::sync::{Arc, OnceLock};
+use std::time::Duration;
+
+use api::v1::column_data_type_extension::TypeExt;
+use api::v1::value::ValueData;
+use api::v1::{
+    ColumnDataType, ColumnDataTypeExtension, ColumnSchema, JsonTypeExtension, Row,
+    RowInsertRequest, RowInsertRequests, Rows, SemanticType,
+};
+use async_trait::async_trait;
+use backon::{BackoffBuilder, ExponentialBuilder};
+use common_telemetry::{debug, error, info, warn};
+use common_time::timestamp::{TimeUnit, Timestamp};
+use serde::{Deserialize, Serialize};
+use store_api::mito_engine_options::{APPEND_MODE_KEY, TTL_KEY};
+use tokio::sync::mpsc::{channel, Receiver, Sender};
+use tokio::task::JoinHandle;
+use tokio::time::sleep;
+use tokio_util::sync::CancellationToken;
+
+use crate::error::{MismatchedSchemaSnafu, Result};
+
+/// The default table name for storing the events.
+pub const DEFAULT_EVENTS_TABLE_NAME: &str = "events";
+
+/// The column name for the event type.
+pub const EVENTS_TABLE_TYPE_COLUMN_NAME: &str = "type";
+/// The column name for the event payload.
+pub const EVENTS_TABLE_PAYLOAD_COLUMN_NAME: &str = "payload";
+/// The column name for the event timestamp.
+pub const EVENTS_TABLE_TIMESTAMP_COLUMN_NAME: &str = "timestamp";
+
+/// EventRecorderRef is the reference to the event recorder.
+pub type EventRecorderRef = Arc<dyn EventRecorder>;
+
+static EVENTS_TABLE_TTL: OnceLock<String> = OnceLock::new();
+
+/// The time interval for flushing batched events to the event handler.
+pub const DEFAULT_FLUSH_INTERVAL_SECONDS: Duration = Duration::from_secs(5);
+// The default TTL for the events table.
+const DEFAULT_EVENTS_TABLE_TTL: &str = "30d";
+// The capacity of the tokio channel for transmitting events to background processor.
+const DEFAULT_CHANNEL_SIZE: usize = 2048;
+// The size of the buffer for batching events before flushing to event handler.
+const DEFAULT_BUFFER_SIZE: usize = 100;
+// The maximum number of retry attempts when event handler processing fails.
+const DEFAULT_MAX_RETRY_TIMES: u64 = 3;
+
+/// Event trait defines the interface for events that can be recorded and persisted as the system table.
+/// By default, the event will be persisted as the system table with the following schema:
+///
+/// - `type`: the type of the event.
+/// - `payload`: the JSON bytes of the event.
+/// - `timestamp`: the timestamp of the event.
+///
+/// The event can also add the extra schema and row to the event by overriding the `extra_schema` and `extra_row` methods.
+pub trait Event: Send + Sync + Debug {
+    /// Returns the type of the event.
+    fn event_type(&self) -> &str;
+
+    /// Returns the timestamp of the event. Default to the current time.
+    fn timestamp(&self) -> Timestamp {
+        Timestamp::current_time(TimeUnit::Nanosecond)
+    }
+
+    /// Returns the JSON bytes of the event as the payload. It will use JSON type to store the payload.
+    fn json_payload(&self) -> Result<String>;
+
+    /// Add the extra schema to the event with the default schema.
+    fn extra_schema(&self) -> Vec<ColumnSchema> {
+        vec![]
+    }
+
+    /// Add the extra row to the event with the default row.
+    fn extra_row(&self) -> Result<Row> {
+        Ok(Row { values: vec![] })
+    }
+
+    /// Returns the event as any type.
+    fn as_any(&self) -> &dyn Any;
+}
+
+/// Returns the hints for the insert operation.
+pub fn insert_hints() -> Vec<(&'static str, &'static str)> {
+    vec![
+        (
+            TTL_KEY,
+            EVENTS_TABLE_TTL
+                .get()
+                .map(|s| s.as_str())
+                .unwrap_or(DEFAULT_EVENTS_TABLE_TTL),
+        ),
+        (APPEND_MODE_KEY, "true"),
+    ]
+}
+
+/// Builds the row inserts request for the events that will be persisted to the events table.
+pub fn build_row_inserts_request(events: &[Box<dyn Event>]) -> Result<RowInsertRequests> {
+    // Aggregate the events by the event type.
+    let mut event_groups: HashMap<&str, Vec<&Box<dyn Event>>> = HashMap::new();
+
+    for event in events {
+        event_groups
+            .entry(event.event_type())
+            .or_default()
+            .push(event);
+    }
+
+    let mut row_insert_requests = RowInsertRequests {
+        inserts: Vec::with_capacity(event_groups.len()),
+    };
+
+    for (_, events) in event_groups {
+        validate_events(&events)?;
+
+        // We already validated the events, so it's safe to get the first event to build the schema for the RowInsertRequest.
+        let event = &events[0];
+        let mut schema = vec![
+            ColumnSchema {
+                column_name: EVENTS_TABLE_TYPE_COLUMN_NAME.to_string(),
+                datatype: ColumnDataType::String.into(),
+                semantic_type: SemanticType::Tag.into(),
+                ..Default::default()
+            },
+            ColumnSchema {
+                column_name: EVENTS_TABLE_PAYLOAD_COLUMN_NAME.to_string(),
+                datatype: ColumnDataType::Binary as i32,
+                semantic_type: SemanticType::Field as i32,
+                datatype_extension: Some(ColumnDataTypeExtension {
+                    type_ext: Some(TypeExt::JsonType(JsonTypeExtension::JsonBinary.into())),
+                }),
+                ..Default::default()
+            },
+            ColumnSchema {
+                column_name: EVENTS_TABLE_TIMESTAMP_COLUMN_NAME.to_string(),
+                datatype: ColumnDataType::TimestampNanosecond.into(),
+                semantic_type: SemanticType::Timestamp.into(),
+                ..Default::default()
+            },
+        ];
+        schema.extend(event.extra_schema());
+
+        let rows = events
+            .iter()
+            .map(|event| {
+                let mut row = Row {
+                    values: vec![
+                        ValueData::StringValue(event.event_type().to_string()).into(),
+                        ValueData::BinaryValue(event.json_payload()?.as_bytes().to_vec()).into(),
+                        ValueData::TimestampNanosecondValue(event.timestamp().value()).into(),
+                    ],
+                };
+                row.values.extend(event.extra_row()?.values);
+                Ok(row)
+            })
+            .collect::<Result<Vec<_>>>()?;
+
+        row_insert_requests.inserts.push(RowInsertRequest {
+            table_name: DEFAULT_EVENTS_TABLE_NAME.to_string(),
+            rows: Some(Rows { schema, rows }),
+        });
+    }
+
+    Ok(row_insert_requests)
+}
+
+// Ensure the events with the same event type have the same extra schema.
+#[allow(clippy::borrowed_box)]
+fn validate_events(events: &[&Box<dyn Event>]) -> Result<()> {
+    // It's safe to get the first event because the events are already grouped by the event type.
+    let extra_schema = events[0].extra_schema();
+    for event in events {
+        if event.extra_schema() != extra_schema {
+            MismatchedSchemaSnafu {
+                expected: extra_schema.clone(),
+                actual: event.extra_schema(),
+            }
+            .fail()?;
+        }
+    }
+    Ok(())
+}
+
+/// EventRecorder trait defines the interface for recording events.
+pub trait EventRecorder: Send + Sync + 'static {
+    /// Records an event for persistence and processing by [EventHandler].
+    fn record(&self, event: Box<dyn Event>);
+
+    /// Cancels the event recorder.
+    fn close(&self);
+}
+
+/// EventHandler trait defines the interface for how to handle the event.
+#[async_trait]
+pub trait EventHandler: Send + Sync + 'static {
+    /// Processes and handles incoming events. The [DefaultEventHandlerImpl] implementation forwards events to frontend instances for persistence.
+    /// We use `&[Box<dyn Event>]` to avoid consuming the events, so the caller can buffer the events and retry if the handler fails.
+    async fn handle(&self, events: &[Box<dyn Event>]) -> Result<()>;
+}
+
+/// Configuration options for the event recorder.
+#[derive(Debug, Clone, Serialize, Deserialize, PartialEq, Eq)]
+pub struct EventRecorderOptions {
+    /// TTL for the events table that will be used to store the events.
+    pub ttl: String,
+}
+
+impl Default for EventRecorderOptions {
+    fn default() -> Self {
+        Self {
+            ttl: DEFAULT_EVENTS_TABLE_TTL.to_string(),
+        }
+    }
+}
+
+/// Implementation of [EventRecorder] that records the events and processes them in the background by the [EventHandler].
+pub struct EventRecorderImpl {
+    // The channel to send the events to the background processor.
+    tx: Sender<Box<dyn Event>>,
+    // The cancel token to cancel the background processor.
+    cancel_token: CancellationToken,
+    // The background processor to process the events.
+    handle: Option<JoinHandle<()>>,
+}
+
+impl EventRecorderImpl {
+    pub fn new(event_handler: Box<dyn EventHandler>, opts: EventRecorderOptions) -> Self {
+        info!("Creating event recorder with options: {:?}", opts);
+
+        let (tx, rx) = channel(DEFAULT_CHANNEL_SIZE);
+        let cancel_token = CancellationToken::new();
+
+        let mut recorder = Self {
+            tx,
+            handle: None,
+            cancel_token: cancel_token.clone(),
+        };
+
+        let processor = EventProcessor::new(
+            rx,
+            event_handler,
+            DEFAULT_FLUSH_INTERVAL_SECONDS,
+            DEFAULT_MAX_RETRY_TIMES,
+        )
+        .with_cancel_token(cancel_token);
+
+        // Spawn a background task to process the events.
+        let handle = tokio::spawn(async move {
+            processor.process(DEFAULT_BUFFER_SIZE).await;
+        });
+
+        recorder.handle = Some(handle);
+
+        // It only sets the ttl once, so it's safe to skip the error.
+        if EVENTS_TABLE_TTL.set(opts.ttl.clone()).is_err() {
+            info!(
+                "Events table ttl already set to {}, skip setting it",
+                opts.ttl
+            );
+        }
+
+        recorder
+    }
+}
+
+impl EventRecorder for EventRecorderImpl {
+    // Accepts an event and send it to the background handler.
+    fn record(&self, event: Box<dyn Event>) {
+        if let Err(e) = self.tx.try_send(event) {
+            error!("Failed to send event to the background processor: {}", e);
+        }
+    }
+
+    // Closes the event recorder. It will stop the background processor and flush the buffer.
+    fn close(&self) {
+        self.cancel_token.cancel();
+    }
+}
+
+impl Drop for EventRecorderImpl {
+    fn drop(&mut self) {
+        if let Some(handle) = self.handle.take() {
+            handle.abort();
+            info!("Aborted the background processor in event recorder");
+        }
+    }
+}
+
+struct EventProcessor {
+    rx: Receiver<Box<dyn Event>>,
+    event_handler: Box<dyn EventHandler>,
+    max_retry_times: u64,
+    process_interval: Duration,
+    cancel_token: CancellationToken,
+}
+
+impl EventProcessor {
+    fn new(
+        rx: Receiver<Box<dyn Event>>,
+        event_handler: Box<dyn EventHandler>,
+        process_interval: Duration,
+        max_retry_times: u64,
+    ) -> Self {
+        Self {
+            rx,
+            event_handler,
+            max_retry_times,
+            process_interval,
+            cancel_token: CancellationToken::new(),
+        }
+    }
+
+    fn with_cancel_token(mut self, cancel_token: CancellationToken) -> Self {
+        self.cancel_token = cancel_token;
+        self
+    }
+
+    async fn process(mut self, buffer_size: usize) {
+        info!("Start the background processor in event recorder to handle the received events.");
+
+        let mut buffer = Vec::with_capacity(buffer_size);
+        let mut interval = tokio::time::interval(self.process_interval);
+
+        loop {
+            tokio::select! {
+                maybe_event = self.rx.recv() => {
+                    if let Some(maybe_event) = maybe_event {
+                        debug!("Received event: {:?}", maybe_event);
+
+                        if buffer.len() >= buffer_size {
+                            debug!(
+                                "Flushing events to the event handler because the buffer is full with {} events",
+                                buffer.len()
+                            );
+                            self.flush_events_to_handler(&mut buffer).await;
+                        }
+
+                        // Push the event to the buffer, the buffer will be flushed when the interval is triggered or received a closed signal.
+                        buffer.push(maybe_event);
+                    } else {
+                        // When received a closed signal, flush the buffer and exit the loop.
+                        self.flush_events_to_handler(&mut buffer).await;
+                        break;
+                    }
+                }
+                // Cancel the processor through the cancel token.
+                _ = self.cancel_token.cancelled() => {
+                    warn!("Received a cancel signal, flushing the buffer and exiting the loop");
+                    self.flush_events_to_handler(&mut buffer).await;
+                    break;
+                }
+                // When the interval is triggered, flush the buffer and send the events to the event handler.
+                _ = interval.tick() => {
+                    self.flush_events_to_handler(&mut buffer).await;
+                }
+            }
+        }
+    }
+
+    // NOTE: While we implement a retry mechanism for failed event handling, there is no guarantee that all events will be processed successfully.
+    async fn flush_events_to_handler(&self, buffer: &mut Vec<Box<dyn Event>>) {
+        if !buffer.is_empty() {
+            debug!("Flushing {} events to the event handler", buffer.len());
+
+            let mut backoff = ExponentialBuilder::default()
+                .with_min_delay(Duration::from_millis(
+                    DEFAULT_FLUSH_INTERVAL_SECONDS.as_millis() as u64 / self.max_retry_times.max(1),
+                ))
+                .with_max_delay(Duration::from_millis(
+                    DEFAULT_FLUSH_INTERVAL_SECONDS.as_millis() as u64,
+                ))
+                .with_max_times(self.max_retry_times as usize)
+                .build();
+
+            loop {
+                match self.event_handler.handle(buffer).await {
+                    Ok(()) => {
+                        debug!("Successfully handled {} events", buffer.len());
+                        break;
+                    }
+                    Err(e) => {
+                        if let Some(d) = backoff.next() {
+                            warn!(e; "Failed to handle events, retrying...");
+                            sleep(d).await;
+                            continue;
+                        } else {
+                            warn!(
+                                e; "Failed to handle events after {} retries",
+                                self.max_retry_times
+                            );
+                            break;
+                        }
+                    }
+                }
+            }
+        }
+
+        // Clear the buffer to prevent unbounded memory growth, regardless of whether event processing succeeded or failed.
+        buffer.clear();
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[derive(Debug)]
+    struct TestEvent {}
+
+    impl Event for TestEvent {
+        fn event_type(&self) -> &str {
+            "test_event"
+        }
+
+        fn json_payload(&self) -> Result<String> {
+            Ok("{\"procedure_id\": \"1234567890\"}".to_string())
+        }
+
+        fn as_any(&self) -> &dyn Any {
+            self
+        }
+    }
+
+    struct TestEventHandlerImpl {}
+
+    #[async_trait]
+    impl EventHandler for TestEventHandlerImpl {
+        async fn handle(&self, events: &[Box<dyn Event>]) -> Result<()> {
+            let event = events
+                .first()
+                .unwrap()
+                .as_any()
+                .downcast_ref::<TestEvent>()
+                .unwrap();
+            assert_eq!(
+                event.json_payload().unwrap(),
+                "{\"procedure_id\": \"1234567890\"}"
+            );
+            assert_eq!(event.event_type(), "test_event");
+            Ok(())
+        }
+    }
+
+    #[tokio::test]
+    async fn test_event_recorder() {
+        let mut event_recorder = EventRecorderImpl::new(
+            Box::new(TestEventHandlerImpl {}),
+            EventRecorderOptions::default(),
+        );
+        event_recorder.record(Box::new(TestEvent {}));
+
+        // Sleep for a while to let the event be sent to the event handler.
+        sleep(Duration::from_millis(500)).await;
+
+        // Close the event recorder to flush the buffer.
+        event_recorder.close();
+
+        // Sleep for a while to let the background task process the event.
+        sleep(Duration::from_millis(500)).await;
+
+        if let Some(handle) = event_recorder.handle.take() {
+            assert!(handle.await.is_ok());
+        }
+    }
+
+    struct TestEventHandlerImplShouldPanic {}
+
+    #[async_trait]
+    impl EventHandler for TestEventHandlerImplShouldPanic {
+        async fn handle(&self, events: &[Box<dyn Event>]) -> Result<()> {
+            let event = events
+                .first()
+                .unwrap()
+                .as_any()
+                .downcast_ref::<TestEvent>()
+                .unwrap();
+
+            // Set the incorrect payload and event type to trigger the panic.
+            assert_eq!(
+                event.json_payload().unwrap(),
+                "{\"procedure_id\": \"should_panic\"}"
+            );
+            assert_eq!(event.event_type(), "should_panic");
+            Ok(())
+        }
+    }
+
+    #[tokio::test]
+    async fn test_event_recorder_should_panic() {
+        let mut event_recorder = EventRecorderImpl::new(
+            Box::new(TestEventHandlerImplShouldPanic {}),
+            EventRecorderOptions::default(),
+        );
+
+        event_recorder.record(Box::new(TestEvent {}));
+
+        // Sleep for a while to let the event be sent to the event handler.
+        sleep(Duration::from_millis(500)).await;
+
+        // Close the event recorder to flush the buffer.
+        event_recorder.close();
+
+        // Sleep for a while to let the background task process the event.
+        sleep(Duration::from_millis(500)).await;
+
+        if let Some(handle) = event_recorder.handle.take() {
+            assert!(handle.await.unwrap_err().is_panic());
+        }
+    }
+}

src/common/function/Cargo.toml

@@ -16,6 +16,8 @@ geo = ["geohash", "h3o", "s2", "wkt", "geo-types", "dep:geo"]
 ahash.workspace = true
 api.workspace = true
 arc-swap = "1.0"
+arrow.workspace = true
+arrow-schema.workspace = true
 async-trait.workspace = true
 bincode = "1.3"
 catalog.workspace = true
@@ -34,6 +36,7 @@ datafusion.workspace = true
 datafusion-common.workspace = true
 datafusion-expr.workspace = true
 datafusion-functions-aggregate-common.workspace = true
+datafusion-physical-expr.workspace = true
 datatypes.workspace = true
 derive_more = { version = "1", default-features = false, features = ["display"] }
 geo = { version = "0.29", optional = true }
@@ -62,5 +65,7 @@ wkt = { version = "0.11", optional = true }
 
 [dev-dependencies]
 approx = "0.5"
+futures.workspace = true
+pretty_assertions = "1.4.0"
 serde = { version = "1.0", features = ["derive"] }
 tokio.workspace = true

src/common/function/src/aggrs.rs

@@ -17,3 +17,5 @@ pub mod count_hash;
 #[cfg(feature = "geo")]
 pub mod geo;
 pub mod vector;
+
+pub mod aggr_wrapper;

src/common/function/src/aggrs/aggr_wrapper.rs

@@ -0,0 +1,538 @@
+// Copyright 2023 Greptime Team
+//
+// 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.
+
+//! Wrapper for making aggregate functions out of state/merge functions of original aggregate functions.
+//!
+//! i.e. for a aggregate function `foo`, we will have a state function `foo_state` and a merge function `foo_merge`.
+//!
+//! `foo_state`'s input args is the same as `foo`'s, and its output is a state object.
+//! Note that `foo_state` might have multiple output columns, so it's a struct array
+//! that each output column is a struct field.
+//! `foo_merge`'s input arg is the same as `foo_state`'s output, and its output is the same as `foo`'s input.
+//!
+
+use std::sync::Arc;
+
+use arrow::array::StructArray;
+use arrow_schema::Fields;
+use datafusion::optimizer::analyzer::type_coercion::TypeCoercion;
+use datafusion::optimizer::AnalyzerRule;
+use datafusion::physical_planner::create_aggregate_expr_and_maybe_filter;
+use datafusion_common::{Column, ScalarValue};
+use datafusion_expr::expr::AggregateFunction;
+use datafusion_expr::function::StateFieldsArgs;
+use datafusion_expr::{
+    Accumulator, Aggregate, AggregateUDF, AggregateUDFImpl, Expr, ExprSchemable, LogicalPlan,
+    Signature,
+};
+use datafusion_physical_expr::aggregate::AggregateFunctionExpr;
+use datatypes::arrow::datatypes::{DataType, Field};
+
+/// Returns the name of the state function for the given aggregate function name.
+/// The state function is used to compute the state of the aggregate function.
+/// The state function's name is in the format `__<aggr_name>_state
+pub fn aggr_state_func_name(aggr_name: &str) -> String {
+    format!("__{}_state", aggr_name)
+}
+
+/// Returns the name of the merge function for the given aggregate function name.
+/// The merge function is used to merge the states of the state functions.
+/// The merge function's name is in the format `__<aggr_name>_merge
+pub fn aggr_merge_func_name(aggr_name: &str) -> String {
+    format!("__{}_merge", aggr_name)
+}
+
+/// A wrapper to make an aggregate function out of the state and merge functions of the original aggregate function.
+/// It contains the original aggregate function, the state functions, and the merge function.
+///
+/// Notice state functions may have multiple output columns, so it's return type is always a struct array, and the merge function is used to merge the states of the state functions.
+#[derive(Debug, Clone)]
+pub struct StateMergeHelper;
+
+/// A struct to hold the two aggregate plans, one for the state function(lower) and one for the merge function(upper).
+#[allow(unused)]
+#[derive(Debug, Clone)]
+pub struct StepAggrPlan {
+    /// Upper merge plan, which is the aggregate plan that merges the states of the state function.
+    pub upper_merge: Arc<LogicalPlan>,
+    /// Lower state plan, which is the aggregate plan that computes the state of the aggregate function.
+    pub lower_state: Arc<LogicalPlan>,
+}
+
+pub fn get_aggr_func(expr: &Expr) -> Option<&datafusion_expr::expr::AggregateFunction> {
+    let mut expr_ref = expr;
+    while let Expr::Alias(alias) = expr_ref {
+        expr_ref = &alias.expr;
+    }
+    if let Expr::AggregateFunction(aggr_func) = expr_ref {
+        Some(aggr_func)
+    } else {
+        None
+    }
+}
+
+impl StateMergeHelper {
+    /// Split an aggregate plan into two aggregate plans, one for the state function and one for the merge function.
+    pub fn split_aggr_node(aggr_plan: Aggregate) -> datafusion_common::Result<StepAggrPlan> {
+        let aggr = {
+            // certain aggr func need type coercion to work correctly, so we need to analyze the plan first.
+            let aggr_plan = TypeCoercion::new().analyze(
+                LogicalPlan::Aggregate(aggr_plan).clone(),
+                &Default::default(),
+            )?;
+            if let LogicalPlan::Aggregate(aggr) = aggr_plan {
+                aggr
+            } else {
+                return Err(datafusion_common::DataFusionError::Internal(format!(
+                    "Failed to coerce expressions in aggregate plan, expected Aggregate, got: {:?}",
+                    aggr_plan
+                )));
+            }
+        };
+        let mut lower_aggr_exprs = vec![];
+        let mut upper_aggr_exprs = vec![];
+
+        for aggr_expr in aggr.aggr_expr.iter() {
+            let Some(aggr_func) = get_aggr_func(aggr_expr) else {
+                return Err(datafusion_common::DataFusionError::NotImplemented(format!(
+                    "Unsupported aggregate expression for step aggr optimize: {:?}",
+                    aggr_expr
+                )));
+            };
+
+            let original_input_types = aggr_func
+                .args
+                .iter()
+                .map(|e| e.get_type(&aggr.input.schema()))
+                .collect::<Result<Vec<_>, _>>()?;
+
+            // first create the state function from the original aggregate function.
+            let state_func = StateWrapper::new((*aggr_func.func).clone())?;
+
+            let expr = AggregateFunction {
+                func: Arc::new(state_func.into()),
+                args: aggr_func.args.clone(),
+                distinct: aggr_func.distinct,
+                filter: aggr_func.filter.clone(),
+                order_by: aggr_func.order_by.clone(),
+                null_treatment: aggr_func.null_treatment,
+            };
+            let expr = Expr::AggregateFunction(expr);
+            let lower_state_output_col_name = expr.schema_name().to_string();
+
+            lower_aggr_exprs.push(expr);
+
+            let (original_phy_expr, _filter, _ordering) = create_aggregate_expr_and_maybe_filter(
+                aggr_expr,
+                aggr.input.schema(),
+                aggr.input.schema().as_arrow(),
+                &Default::default(),
+            )?;
+
+            let merge_func = MergeWrapper::new(
+                (*aggr_func.func).clone(),
+                original_phy_expr,
+                original_input_types,
+            )?;
+            let arg = Expr::Column(Column::new_unqualified(lower_state_output_col_name));
+            let expr = AggregateFunction {
+                func: Arc::new(merge_func.into()),
+                args: vec![arg],
+                distinct: aggr_func.distinct,
+                filter: aggr_func.filter.clone(),
+                order_by: aggr_func.order_by.clone(),
+                null_treatment: aggr_func.null_treatment,
+            };
+
+            // alias to the original aggregate expr's schema name, so parent plan can refer to it
+            // correctly.
+            let expr = Expr::AggregateFunction(expr).alias(aggr_expr.schema_name().to_string());
+            upper_aggr_exprs.push(expr);
+        }
+
+        let mut lower = aggr.clone();
+        lower.aggr_expr = lower_aggr_exprs;
+        let lower_plan = LogicalPlan::Aggregate(lower);
+
+        // update aggregate's output schema
+        let lower_plan = Arc::new(lower_plan.recompute_schema()?);
+
+        let mut upper = aggr.clone();
+        let aggr_plan = LogicalPlan::Aggregate(aggr);
+        upper.aggr_expr = upper_aggr_exprs;
+        upper.input = lower_plan.clone();
+        // upper schema's output schema should be the same as the original aggregate plan's output schema
+        let upper_check = upper.clone();
+        let upper_plan = Arc::new(LogicalPlan::Aggregate(upper_check).recompute_schema()?);
+        if *upper_plan.schema() != *aggr_plan.schema() {
+            return Err(datafusion_common::DataFusionError::Internal(format!(
+                 "Upper aggregate plan's schema is not the same as the original aggregate plan's schema: \n[transformed]:{}\n[   original]{}",
+                upper_plan.schema(), aggr_plan.schema()
+            )));
+        }
+
+        Ok(StepAggrPlan {
+            lower_state: lower_plan,
+            upper_merge: upper_plan,
+        })
+    }
+}
+
+/// Wrapper to make an aggregate function out of a state function.
+#[derive(Debug, Clone, PartialEq, Eq)]
+pub struct StateWrapper {
+    inner: AggregateUDF,
+    name: String,
+}
+
+impl StateWrapper {
+    /// `state_index`: The index of the state in the output of the state function.
+    pub fn new(inner: AggregateUDF) -> datafusion_common::Result<Self> {
+        let name = aggr_state_func_name(inner.name());
+        Ok(Self { inner, name })
+    }
+
+    pub fn inner(&self) -> &AggregateUDF {
+        &self.inner
+    }
+
+    /// Deduce the return type of the original aggregate function
+    /// based on the accumulator arguments.
+    ///
+    pub fn deduce_aggr_return_type(
+        &self,
+        acc_args: &datafusion_expr::function::AccumulatorArgs,
+    ) -> datafusion_common::Result<DataType> {
+        let input_exprs = acc_args.exprs;
+        let input_schema = acc_args.schema;
+        let input_types = input_exprs
+            .iter()
+            .map(|e| e.data_type(input_schema))
+            .collect::<Result<Vec<_>, _>>()?;
+        let return_type = self.inner.return_type(&input_types)?;
+        Ok(return_type)
+    }
+}
+
+impl AggregateUDFImpl for StateWrapper {
+    fn accumulator<'a, 'b>(
+        &'a self,
+        acc_args: datafusion_expr::function::AccumulatorArgs<'b>,
+    ) -> datafusion_common::Result<Box<dyn Accumulator>> {
+        // fix and recover proper acc args for the original aggregate function.
+        let state_type = acc_args.return_type.clone();
+        let inner = {
+            let old_return_type = self.deduce_aggr_return_type(&acc_args)?;
+            let acc_args = datafusion_expr::function::AccumulatorArgs {
+                return_type: &old_return_type,
+                schema: acc_args.schema,
+                ignore_nulls: acc_args.ignore_nulls,
+                ordering_req: acc_args.ordering_req,
+                is_reversed: acc_args.is_reversed,
+                name: acc_args.name,
+                is_distinct: acc_args.is_distinct,
+                exprs: acc_args.exprs,
+            };
+            self.inner.accumulator(acc_args)?
+        };
+        Ok(Box::new(StateAccum::new(inner, state_type)?))
+    }
+
+    fn as_any(&self) -> &dyn std::any::Any {
+        self
+    }
+    fn name(&self) -> &str {
+        self.name.as_str()
+    }
+
+    fn is_nullable(&self) -> bool {
+        self.inner.is_nullable()
+    }
+
+    /// Return state_fields as the output struct type.
+    ///
+    fn return_type(&self, arg_types: &[DataType]) -> datafusion_common::Result<DataType> {
+        let old_return_type = self.inner.return_type(arg_types)?;
+        let state_fields_args = StateFieldsArgs {
+            name: self.inner().name(),
+            input_types: arg_types,
+            return_type: &old_return_type,
+            // TODO(discord9): how to get this?, probably ok?
+            ordering_fields: &[],
+            is_distinct: false,
+        };
+        let state_fields = self.inner.state_fields(state_fields_args)?;
+        let struct_field = DataType::Struct(state_fields.into());
+        Ok(struct_field)
+    }
+
+    /// The state function's output fields are the same as the original aggregate function's state fields.
+    fn state_fields(
+        &self,
+        args: datafusion_expr::function::StateFieldsArgs,
+    ) -> datafusion_common::Result<Vec<Field>> {
+        let old_return_type = self.inner.return_type(args.input_types)?;
+        let state_fields_args = StateFieldsArgs {
+            name: args.name,
+            input_types: args.input_types,
+            return_type: &old_return_type,
+            ordering_fields: args.ordering_fields,
+            is_distinct: args.is_distinct,
+        };
+        self.inner.state_fields(state_fields_args)
+    }
+
+    /// The state function's signature is the same as the original aggregate function's signature,
+    fn signature(&self) -> &Signature {
+        self.inner.signature()
+    }
+
+    /// Coerce types also do nothing, as optimzer should be able to already make struct types
+    fn coerce_types(&self, arg_types: &[DataType]) -> datafusion_common::Result<Vec<DataType>> {
+        self.inner.coerce_types(arg_types)
+    }
+}
+
+/// The wrapper's input is the same as the original aggregate function's input,
+/// and the output is the state function's output.
+#[derive(Debug)]
+pub struct StateAccum {
+    inner: Box<dyn Accumulator>,
+    state_fields: Fields,
+}
+
+impl StateAccum {
+    pub fn new(
+        inner: Box<dyn Accumulator>,
+        state_type: DataType,
+    ) -> datafusion_common::Result<Self> {
+        let DataType::Struct(fields) = state_type else {
+            return Err(datafusion_common::DataFusionError::Internal(format!(
+                "Expected a struct type for state, got: {:?}",
+                state_type
+            )));
+        };
+        Ok(Self {
+            inner,
+            state_fields: fields,
+        })
+    }
+}
+
+impl Accumulator for StateAccum {
+    fn evaluate(&mut self) -> datafusion_common::Result<ScalarValue> {
+        let state = self.inner.state()?;
+
+        let array = state
+            .iter()
+            .map(|s| s.to_array())
+            .collect::<Result<Vec<_>, _>>()?;
+        let struct_array = StructArray::try_new(self.state_fields.clone(), array, None)?;
+        Ok(ScalarValue::Struct(Arc::new(struct_array)))
+    }
+
+    fn merge_batch(
+        &mut self,
+        states: &[datatypes::arrow::array::ArrayRef],
+    ) -> datafusion_common::Result<()> {
+        self.inner.merge_batch(states)
+    }
+
+    fn update_batch(
+        &mut self,
+        values: &[datatypes::arrow::array::ArrayRef],
+    ) -> datafusion_common::Result<()> {
+        self.inner.update_batch(values)
+    }
+
+    fn size(&self) -> usize {
+        self.inner.size()
+    }
+
+    fn state(&mut self) -> datafusion_common::Result<Vec<ScalarValue>> {
+        self.inner.state()
+    }
+}
+
+/// TODO(discord9): mark this function as non-ser/de able
+///
+/// This wrapper shouldn't be register as a udaf, as it contain extra data that is not serializable.
+/// and changes for different logical plans.
+#[derive(Debug, Clone)]
+pub struct MergeWrapper {
+    inner: AggregateUDF,
+    name: String,
+    merge_signature: Signature,
+    /// The original physical expression of the aggregate function, can't store the original aggregate function directly, as PhysicalExpr didn't implement Any
+    original_phy_expr: Arc<AggregateFunctionExpr>,
+    original_input_types: Vec<DataType>,
+}
+impl MergeWrapper {
+    pub fn new(
+        inner: AggregateUDF,
+        original_phy_expr: Arc<AggregateFunctionExpr>,
+        original_input_types: Vec<DataType>,
+    ) -> datafusion_common::Result<Self> {
+        let name = aggr_merge_func_name(inner.name());
+        // the input type is actually struct type, which is the state fields of the original aggregate function.
+        let merge_signature = Signature::user_defined(datafusion_expr::Volatility::Immutable);
+
+        Ok(Self {
+            inner,
+            name,
+            merge_signature,
+            original_phy_expr,
+            original_input_types,
+        })
+    }
+
+    pub fn inner(&self) -> &AggregateUDF {
+        &self.inner
+    }
+}
+
+impl AggregateUDFImpl for MergeWrapper {
+    fn accumulator<'a, 'b>(
+        &'a self,
+        acc_args: datafusion_expr::function::AccumulatorArgs<'b>,
+    ) -> datafusion_common::Result<Box<dyn Accumulator>> {
+        if acc_args.schema.fields().len() != 1
+            || !matches!(acc_args.schema.field(0).data_type(), DataType::Struct(_))
+        {
+            return Err(datafusion_common::DataFusionError::Internal(format!(
+                "Expected one struct type as input, got: {:?}",
+                acc_args.schema
+            )));
+        }
+        let input_type = acc_args.schema.field(0).data_type();
+        let DataType::Struct(fields) = input_type else {
+            return Err(datafusion_common::DataFusionError::Internal(format!(
+                "Expected a struct type for input, got: {:?}",
+                input_type
+            )));
+        };
+
+        let inner_accum = self.original_phy_expr.create_accumulator()?;
+        Ok(Box::new(MergeAccum::new(inner_accum, fields)))
+    }
+
+    fn as_any(&self) -> &dyn std::any::Any {
+        self
+    }
+    fn name(&self) -> &str {
+        self.name.as_str()
+    }
+
+    fn is_nullable(&self) -> bool {
+        self.inner.is_nullable()
+    }
+
+    /// Notice here the `arg_types` is actually the `state_fields`'s data types,
+    /// so return fixed return type instead of using `arg_types` to determine the return type.
+    fn return_type(&self, _arg_types: &[DataType]) -> datafusion_common::Result<DataType> {
+        // The return type is the same as the original aggregate function's return type.
+        let ret_type = self.inner.return_type(&self.original_input_types)?;
+        Ok(ret_type)
+    }
+    fn signature(&self) -> &Signature {
+        &self.merge_signature
+    }
+
+    /// Coerce types also do nothing, as optimzer should be able to already make struct types
+    fn coerce_types(&self, arg_types: &[DataType]) -> datafusion_common::Result<Vec<DataType>> {
+        // just check if the arg_types are only one and is struct array
+        if arg_types.len() != 1 || !matches!(arg_types.first(), Some(DataType::Struct(_))) {
+            return Err(datafusion_common::DataFusionError::Internal(format!(
+                "Expected one struct type as input, got: {:?}",
+                arg_types
+            )));
+        }
+        Ok(arg_types.to_vec())
+    }
+
+    /// Just return the original aggregate function's state fields.
+    fn state_fields(
+        &self,
+        _args: datafusion_expr::function::StateFieldsArgs,
+    ) -> datafusion_common::Result<Vec<Field>> {
+        self.original_phy_expr.state_fields()
+    }
+}
+
+/// The merge accumulator, which modify `update_batch`'s behavior to accept one struct array which
+/// include the state fields of original aggregate function, and merge said states into original accumulator
+/// the output is the same as original aggregate function
+#[derive(Debug)]
+pub struct MergeAccum {
+    inner: Box<dyn Accumulator>,
+    state_fields: Fields,
+}
+
+impl MergeAccum {
+    pub fn new(inner: Box<dyn Accumulator>, state_fields: &Fields) -> Self {
+        Self {
+            inner,
+            state_fields: state_fields.clone(),
+        }
+    }
+}
+
+impl Accumulator for MergeAccum {
+    fn evaluate(&mut self) -> datafusion_common::Result<ScalarValue> {
+        self.inner.evaluate()
+    }
+
+    fn merge_batch(&mut self, states: &[arrow::array::ArrayRef]) -> datafusion_common::Result<()> {
+        self.inner.merge_batch(states)
+    }
+
+    fn update_batch(&mut self, values: &[arrow::array::ArrayRef]) -> datafusion_common::Result<()> {
+        let value = values.first().ok_or_else(|| {
+            datafusion_common::DataFusionError::Internal("No values provided for merge".to_string())
+        })?;
+        // The input values are states from other accumulators, so we merge them.
+        let struct_arr = value
+            .as_any()
+            .downcast_ref::<StructArray>()
+            .ok_or_else(|| {
+                datafusion_common::DataFusionError::Internal(format!(
+                    "Expected StructArray, got: {:?}",
+                    value.data_type()
+                ))
+            })?;
+        let fields = struct_arr.fields();
+        if fields != &self.state_fields {
+            return Err(datafusion_common::DataFusionError::Internal(format!(
+                "Expected state fields: {:?}, got: {:?}",
+                self.state_fields, fields
+            )));
+        }
+
+        // now fields should be the same, so we can merge the batch
+        // by pass the columns as order should be the same
+        let state_columns = struct_arr.columns();
+        self.inner.merge_batch(state_columns)
+    }
+
+    fn size(&self) -> usize {
+        self.inner.size()
+    }
+
+    fn state(&mut self) -> datafusion_common::Result<Vec<ScalarValue>> {
+        self.inner.state()
+    }
+}
+
+#[cfg(test)]
+mod tests;

src/common/function/src/aggrs/aggr_wrapper/tests.rs

@@ -0,0 +1,804 @@
+// Copyright 2023 Greptime Team
+//
+// 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::any::Any;
+use std::pin::Pin;
+use std::sync::{Arc, Mutex};
+use std::task::{Context, Poll};
+
+use arrow::array::{ArrayRef, Float64Array, Int64Array, UInt64Array};
+use arrow::record_batch::RecordBatch;
+use arrow_schema::SchemaRef;
+use datafusion::catalog::{Session, TableProvider};
+use datafusion::datasource::DefaultTableSource;
+use datafusion::execution::{RecordBatchStream, SendableRecordBatchStream, TaskContext};
+use datafusion::functions_aggregate::average::avg_udaf;
+use datafusion::functions_aggregate::sum::sum_udaf;
+use datafusion::optimizer::analyzer::type_coercion::TypeCoercion;
+use datafusion::optimizer::AnalyzerRule;
+use datafusion::physical_plan::aggregates::AggregateExec;
+use datafusion::physical_plan::execution_plan::{Boundedness, EmissionType};
+use datafusion::physical_plan::{DisplayAs, DisplayFormatType, ExecutionPlan, PlanProperties};
+use datafusion::physical_planner::{DefaultPhysicalPlanner, PhysicalPlanner};
+use datafusion::prelude::SessionContext;
+use datafusion_common::{Column, TableReference};
+use datafusion_expr::expr::AggregateFunction;
+use datafusion_expr::sqlparser::ast::NullTreatment;
+use datafusion_expr::{Aggregate, Expr, LogicalPlan, SortExpr, TableScan};
+use datafusion_physical_expr::aggregate::AggregateExprBuilder;
+use datafusion_physical_expr::{EquivalenceProperties, Partitioning};
+use datatypes::arrow_array::StringArray;
+use futures::{Stream, StreamExt as _};
+use pretty_assertions::assert_eq;
+
+use super::*;
+use crate::aggrs::approximate::hll::HllState;
+use crate::aggrs::approximate::uddsketch::UddSketchState;
+use crate::aggrs::count_hash::CountHash;
+use crate::function::Function as _;
+use crate::scalars::hll_count::HllCalcFunction;
+use crate::scalars::uddsketch_calc::UddSketchCalcFunction;
+
+#[derive(Debug)]
+pub struct MockInputExec {
+    input: Vec<RecordBatch>,
+    schema: SchemaRef,
+    properties: PlanProperties,
+}
+
+impl MockInputExec {
+    pub fn new(input: Vec<RecordBatch>, schema: SchemaRef) -> Self {
+        Self {
+            properties: PlanProperties::new(
+                EquivalenceProperties::new(schema.clone()),
+                Partitioning::UnknownPartitioning(1),
+                EmissionType::Incremental,
+                Boundedness::Bounded,
+            ),
+            input,
+            schema,
+        }
+    }
+}
+
+impl DisplayAs for MockInputExec {
+    fn fmt_as(&self, _t: DisplayFormatType, _f: &mut std::fmt::Formatter) -> std::fmt::Result {
+        unimplemented!()
+    }
+}
+
+impl ExecutionPlan for MockInputExec {
+    fn name(&self) -> &str {
+        "MockInputExec"
+    }
+
+    fn as_any(&self) -> &dyn Any {
+        self
+    }
+
+    fn properties(&self) -> &PlanProperties {
+        &self.properties
+    }
+
+    fn children(&self) -> Vec<&Arc<dyn ExecutionPlan>> {
+        vec![]
+    }
+
+    fn with_new_children(
+        self: Arc<Self>,
+        _children: Vec<Arc<dyn ExecutionPlan>>,
+    ) -> datafusion_common::Result<Arc<dyn ExecutionPlan>> {
+        Ok(self)
+    }
+
+    fn execute(
+        &self,
+        _partition: usize,
+        _context: Arc<TaskContext>,
+    ) -> datafusion_common::Result<SendableRecordBatchStream> {
+        let stream = MockStream {
+            stream: self.input.clone(),
+            schema: self.schema.clone(),
+            idx: 0,
+        };
+        Ok(Box::pin(stream))
+    }
+}
+
+struct MockStream {
+    stream: Vec<RecordBatch>,
+    schema: SchemaRef,
+    idx: usize,
+}
+
+impl Stream for MockStream {
+    type Item = datafusion_common::Result<RecordBatch>;
+    fn poll_next(
+        mut self: Pin<&mut Self>,
+        _cx: &mut Context<'_>,
+    ) -> Poll<Option<datafusion_common::Result<RecordBatch>>> {
+        if self.idx < self.stream.len() {
+            let ret = self.stream[self.idx].clone();
+            self.idx += 1;
+            Poll::Ready(Some(Ok(ret)))
+        } else {
+            Poll::Ready(None)
+        }
+    }
+}
+
+impl RecordBatchStream for MockStream {
+    fn schema(&self) -> SchemaRef {
+        self.schema.clone()
+    }
+}
+
+#[derive(Debug)]
+struct DummyTableProvider {
+    schema: Arc<arrow_schema::Schema>,
+    record_batch: Mutex<Option<RecordBatch>>,
+}
+
+impl DummyTableProvider {
+    #[allow(unused)]
+    pub fn new(schema: Arc<arrow_schema::Schema>, record_batch: Option<RecordBatch>) -> Self {
+        Self {
+            schema,
+            record_batch: Mutex::new(record_batch),
+        }
+    }
+}
+
+impl Default for DummyTableProvider {
+    fn default() -> Self {
+        Self {
+            schema: Arc::new(arrow_schema::Schema::new(vec![Field::new(
+                "number",
+                DataType::Int64,
+                true,
+            )])),
+            record_batch: Mutex::new(None),
+        }
+    }
+}
+
+#[async_trait::async_trait]
+impl TableProvider for DummyTableProvider {
+    fn as_any(&self) -> &dyn std::any::Any {
+        self
+    }
+
+    fn schema(&self) -> Arc<arrow_schema::Schema> {
+        self.schema.clone()
+    }
+
+    fn table_type(&self) -> datafusion_expr::TableType {
+        datafusion_expr::TableType::Base
+    }
+
+    async fn scan(
+        &self,
+        _state: &dyn Session,
+        _projection: Option<&Vec<usize>>,
+        _filters: &[Expr],
+        _limit: Option<usize>,
+    ) -> datafusion::error::Result<Arc<dyn ExecutionPlan>> {
+        let input: Vec<RecordBatch> = self
+            .record_batch
+            .lock()
+            .unwrap()
+            .clone()
+            .map(|r| vec![r])
+            .unwrap_or_default();
+        Ok(Arc::new(MockInputExec::new(input, self.schema.clone())))
+    }
+}
+
+fn dummy_table_scan() -> LogicalPlan {
+    let table_provider = Arc::new(DummyTableProvider::default());
+    let table_source = DefaultTableSource::new(table_provider);
+    LogicalPlan::TableScan(
+        TableScan::try_new(
+            TableReference::bare("Number"),
+            Arc::new(table_source),
+            None,
+            vec![],
+            None,
+        )
+        .unwrap(),
+    )
+}
+
+#[tokio::test]
+async fn test_sum_udaf() {
+    let ctx = SessionContext::new();
+
+    let sum = datafusion::functions_aggregate::sum::sum_udaf();
+    let sum = (*sum).clone();
+    let original_aggr = Aggregate::try_new(
+        Arc::new(dummy_table_scan()),
+        vec![],
+        vec![Expr::AggregateFunction(AggregateFunction::new_udf(
+            Arc::new(sum.clone()),
+            vec![Expr::Column(Column::new_unqualified("number"))],
+            false,
+            None,
+            None,
+            None,
+        ))],
+    )
+    .unwrap();
+    let res = StateMergeHelper::split_aggr_node(original_aggr).unwrap();
+
+    let expected_lower_plan = LogicalPlan::Aggregate(
+        Aggregate::try_new(
+            Arc::new(dummy_table_scan()),
+            vec![],
+            vec![Expr::AggregateFunction(AggregateFunction::new_udf(
+                Arc::new(StateWrapper::new(sum.clone()).unwrap().into()),
+                vec![Expr::Column(Column::new_unqualified("number"))],
+                false,
+                None,
+                None,
+                None,
+            ))],
+        )
+        .unwrap(),
+    )
+    .recompute_schema()
+    .unwrap();
+    assert_eq!(res.lower_state.as_ref(), &expected_lower_plan);
+
+    let expected_merge_plan = LogicalPlan::Aggregate(
+        Aggregate::try_new(
+            Arc::new(expected_lower_plan),
+            vec![],
+            vec![Expr::AggregateFunction(AggregateFunction::new_udf(
+                Arc::new(
+                    MergeWrapper::new(
+                        sum.clone(),
+                        Arc::new(
+                            AggregateExprBuilder::new(
+                                Arc::new(sum.clone()),
+                                vec![Arc::new(
+                                    datafusion::physical_expr::expressions::Column::new(
+                                        "number", 0,
+                                    ),
+                                )],
+                            )
+                            .schema(Arc::new(dummy_table_scan().schema().as_arrow().clone()))
+                            .alias("sum(number)")
+                            .build()
+                            .unwrap(),
+                        ),
+                        vec![DataType::Int64],
+                    )
+                    .unwrap()
+                    .into(),
+                ),
+                vec![Expr::Column(Column::new_unqualified("__sum_state(number)"))],
+                false,
+                None,
+                None,
+                None,
+            ))
+            .alias("sum(number)")],
+        )
+        .unwrap(),
+    );
+    assert_eq!(res.upper_merge.as_ref(), &expected_merge_plan);
+
+    let phy_aggr_state_plan = DefaultPhysicalPlanner::default()
+        .create_physical_plan(&res.lower_state, &ctx.state())
+        .await
+        .unwrap();
+    let aggr_exec = phy_aggr_state_plan
+        .as_any()
+        .downcast_ref::<AggregateExec>()
+        .unwrap();
+    let aggr_func_expr = &aggr_exec.aggr_expr()[0];
+    let mut state_accum = aggr_func_expr.create_accumulator().unwrap();
+
+    // evaluate the state function
+    let input = Int64Array::from(vec![Some(1), Some(2), None, Some(3)]);
+    let values = vec![Arc::new(input) as arrow::array::ArrayRef];
+
+    state_accum.update_batch(&values).unwrap();
+    let state = state_accum.state().unwrap();
+    assert_eq!(state.len(), 1);
+    assert_eq!(state[0], ScalarValue::Int64(Some(6)));
+
+    let eval_res = state_accum.evaluate().unwrap();
+    assert_eq!(
+        eval_res,
+        ScalarValue::Struct(Arc::new(
+            StructArray::try_new(
+                vec![Field::new("sum[sum]", DataType::Int64, true)].into(),
+                vec![Arc::new(Int64Array::from(vec![Some(6)]))],
+                None,
+            )
+            .unwrap(),
+        ))
+    );
+
+    let phy_aggr_merge_plan = DefaultPhysicalPlanner::default()
+        .create_physical_plan(&res.upper_merge, &ctx.state())
+        .await
+        .unwrap();
+    let aggr_exec = phy_aggr_merge_plan
+        .as_any()
+        .downcast_ref::<AggregateExec>()
+        .unwrap();
+    let aggr_func_expr = &aggr_exec.aggr_expr()[0];
+    let mut merge_accum = aggr_func_expr.create_accumulator().unwrap();
+
+    let merge_input =
+        vec![Arc::new(Int64Array::from(vec![Some(6), Some(42), None])) as arrow::array::ArrayRef];
+    let merge_input_struct_arr = StructArray::try_new(
+        vec![Field::new("sum[sum]", DataType::Int64, true)].into(),
+        merge_input,
+        None,
+    )
+    .unwrap();
+
+    merge_accum
+        .update_batch(&[Arc::new(merge_input_struct_arr)])
+        .unwrap();
+    let merge_state = merge_accum.state().unwrap();
+    assert_eq!(merge_state.len(), 1);
+    assert_eq!(merge_state[0], ScalarValue::Int64(Some(48)));
+
+    let merge_eval_res = merge_accum.evaluate().unwrap();
+    assert_eq!(merge_eval_res, ScalarValue::Int64(Some(48)));
+}
+
+#[tokio::test]
+async fn test_avg_udaf() {
+    let ctx = SessionContext::new();
+
+    let avg = datafusion::functions_aggregate::average::avg_udaf();
+    let avg = (*avg).clone();
+
+    let original_aggr = Aggregate::try_new(
+        Arc::new(dummy_table_scan()),
+        vec![],
+        vec![Expr::AggregateFunction(AggregateFunction::new_udf(
+            Arc::new(avg.clone()),
+            vec![Expr::Column(Column::new_unqualified("number"))],
+            false,
+            None,
+            None,
+            None,
+        ))],
+    )
+    .unwrap();
+    let res = StateMergeHelper::split_aggr_node(original_aggr).unwrap();
+
+    let state_func: Arc<AggregateUDF> = Arc::new(StateWrapper::new(avg.clone()).unwrap().into());
+    let expected_aggr_state_plan = LogicalPlan::Aggregate(
+        Aggregate::try_new(
+            Arc::new(dummy_table_scan()),
+            vec![],
+            vec![Expr::AggregateFunction(AggregateFunction::new_udf(
+                state_func,
+                vec![Expr::Column(Column::new_unqualified("number"))],
+                false,
+                None,
+                None,
+                None,
+            ))],
+        )
+        .unwrap(),
+    );
+    // type coerced so avg aggr function can function correctly
+    let coerced_aggr_state_plan = TypeCoercion::new()
+        .analyze(expected_aggr_state_plan.clone(), &Default::default())
+        .unwrap();
+    assert_eq!(res.lower_state.as_ref(), &coerced_aggr_state_plan);
+    assert_eq!(
+        res.lower_state.schema().as_arrow(),
+        &arrow_schema::Schema::new(vec![Field::new(
+            "__avg_state(number)",
+            DataType::Struct(
+                vec![
+                    Field::new("avg[count]", DataType::UInt64, true),
+                    Field::new("avg[sum]", DataType::Float64, true)
+                ]
+                .into()
+            ),
+            true,
+        )])
+    );
+
+    let expected_merge_fn = MergeWrapper::new(
+        avg.clone(),
+        Arc::new(
+            AggregateExprBuilder::new(
+                Arc::new(avg.clone()),
+                vec![Arc::new(
+                    datafusion::physical_expr::expressions::Column::new("number", 0),
+                )],
+            )
+            .schema(Arc::new(dummy_table_scan().schema().as_arrow().clone()))
+            .alias("avg(number)")
+            .build()
+            .unwrap(),
+        ),
+        // coerced to float64
+        vec![DataType::Float64],
+    )
+    .unwrap();
+
+    let expected_merge_plan = LogicalPlan::Aggregate(
+        Aggregate::try_new(
+            Arc::new(coerced_aggr_state_plan.clone()),
+            vec![],
+            vec![Expr::AggregateFunction(AggregateFunction::new_udf(
+                Arc::new(expected_merge_fn.into()),
+                vec![Expr::Column(Column::new_unqualified("__avg_state(number)"))],
+                false,
+                None,
+                None,
+                None,
+            ))
+            .alias("avg(number)")],
+        )
+        .unwrap(),
+    );
+    assert_eq!(res.upper_merge.as_ref(), &expected_merge_plan);
+
+    let phy_aggr_state_plan = DefaultPhysicalPlanner::default()
+        .create_physical_plan(&coerced_aggr_state_plan, &ctx.state())
+        .await
+        .unwrap();
+    let aggr_exec = phy_aggr_state_plan
+        .as_any()
+        .downcast_ref::<AggregateExec>()
+        .unwrap();
+    let aggr_func_expr = &aggr_exec.aggr_expr()[0];
+    let mut state_accum = aggr_func_expr.create_accumulator().unwrap();
+
+    // evaluate the state function
+    let input = Float64Array::from(vec![Some(1.), Some(2.), None, Some(3.)]);
+    let values = vec![Arc::new(input) as arrow::array::ArrayRef];
+
+    state_accum.update_batch(&values).unwrap();
+    let state = state_accum.state().unwrap();
+    assert_eq!(state.len(), 2);
+    assert_eq!(state[0], ScalarValue::UInt64(Some(3)));
+    assert_eq!(state[1], ScalarValue::Float64(Some(6.)));
+
+    let eval_res = state_accum.evaluate().unwrap();
+    let expected = Arc::new(
+        StructArray::try_new(
+            vec![
+                Field::new("avg[count]", DataType::UInt64, true),
+                Field::new("avg[sum]", DataType::Float64, true),
+            ]
+            .into(),
+            vec![
+                Arc::new(UInt64Array::from(vec![Some(3)])),
+                Arc::new(Float64Array::from(vec![Some(6.)])),
+            ],
+            None,
+        )
+        .unwrap(),
+    );
+    assert_eq!(eval_res, ScalarValue::Struct(expected));
+
+    let phy_aggr_merge_plan = DefaultPhysicalPlanner::default()
+        .create_physical_plan(&res.upper_merge, &ctx.state())
+        .await
+        .unwrap();
+    let aggr_exec = phy_aggr_merge_plan
+        .as_any()
+        .downcast_ref::<AggregateExec>()
+        .unwrap();
+    let aggr_func_expr = &aggr_exec.aggr_expr()[0];
+
+    let mut merge_accum = aggr_func_expr.create_accumulator().unwrap();
+
+    let merge_input = vec![
+        Arc::new(UInt64Array::from(vec![Some(3), Some(42), None])) as arrow::array::ArrayRef,
+        Arc::new(Float64Array::from(vec![Some(48.), Some(84.), None])),
+    ];
+    let merge_input_struct_arr = StructArray::try_new(
+        vec![
+            Field::new("avg[count]", DataType::UInt64, true),
+            Field::new("avg[sum]", DataType::Float64, true),
+        ]
+        .into(),
+        merge_input,
+        None,
+    )
+    .unwrap();
+
+    merge_accum
+        .update_batch(&[Arc::new(merge_input_struct_arr)])
+        .unwrap();
+    let merge_state = merge_accum.state().unwrap();
+    assert_eq!(merge_state.len(), 2);
+    assert_eq!(merge_state[0], ScalarValue::UInt64(Some(45)));
+    assert_eq!(merge_state[1], ScalarValue::Float64(Some(132.)));
+
+    let merge_eval_res = merge_accum.evaluate().unwrap();
+    // the merge function returns the average, which is 132 / 45
+    assert_eq!(merge_eval_res, ScalarValue::Float64(Some(132. / 45_f64)));
+}
+
+/// For testing whether the UDAF state fields are correctly implemented.
+/// esp. for our own custom UDAF's state fields.
+/// By compare eval results before and after split to state/merge functions.
+#[tokio::test]
+async fn test_udaf_correct_eval_result() {
+    struct TestCase {
+        func: Arc<AggregateUDF>,
+        args: Vec<Expr>,
+        input_schema: SchemaRef,
+        input: Vec<ArrayRef>,
+        expected_output: Option<ScalarValue>,
+        expected_fn: Option<ExpectedFn>,
+        distinct: bool,
+        filter: Option<Box<Expr>>,
+        order_by: Option<Vec<SortExpr>>,
+        null_treatment: Option<NullTreatment>,
+    }
+    type ExpectedFn = fn(ArrayRef) -> bool;
+
+    let test_cases = vec![
+        TestCase {
+            func: sum_udaf(),
+            input_schema: Arc::new(arrow_schema::Schema::new(vec![Field::new(
+                "number",
+                DataType::Int64,
+                true,
+            )])),
+            args: vec![Expr::Column(Column::new_unqualified("number"))],
+            input: vec![Arc::new(Int64Array::from(vec![
+                Some(1),
+                Some(2),
+                None,
+                Some(3),
+            ]))],
+            expected_output: Some(ScalarValue::Int64(Some(6))),
+            expected_fn: None,
+            distinct: false,
+            filter: None,
+            order_by: None,
+            null_treatment: None,
+        },
+        TestCase {
+            func: avg_udaf(),
+            input_schema: Arc::new(arrow_schema::Schema::new(vec![Field::new(
+                "number",
+                DataType::Int64,
+                true,
+            )])),
+            args: vec![Expr::Column(Column::new_unqualified("number"))],
+            input: vec![Arc::new(Int64Array::from(vec![
+                Some(1),
+                Some(2),
+                None,
+                Some(3),
+            ]))],
+            expected_output: Some(ScalarValue::Float64(Some(2.0))),
+            expected_fn: None,
+            distinct: false,
+            filter: None,
+            order_by: None,
+            null_treatment: None,
+        },
+        TestCase {
+            func: Arc::new(CountHash::udf_impl()),
+            input_schema: Arc::new(arrow_schema::Schema::new(vec![Field::new(
+                "number",
+                DataType::Int64,
+                true,
+            )])),
+            args: vec![Expr::Column(Column::new_unqualified("number"))],
+            input: vec![Arc::new(Int64Array::from(vec![
+                Some(1),
+                Some(2),
+                None,
+                Some(3),
+                Some(3),
+                Some(3),
+            ]))],
+            expected_output: Some(ScalarValue::Int64(Some(4))),
+            expected_fn: None,
+            distinct: false,
+            filter: None,
+            order_by: None,
+            null_treatment: None,
+        },
+        TestCase {
+            func: Arc::new(UddSketchState::state_udf_impl()),
+            input_schema: Arc::new(arrow_schema::Schema::new(vec![Field::new(
+                "number",
+                DataType::Float64,
+                true,
+            )])),
+            args: vec![
+                Expr::Literal(ScalarValue::Int64(Some(128))),
+                Expr::Literal(ScalarValue::Float64(Some(0.05))),
+                Expr::Column(Column::new_unqualified("number")),
+            ],
+            input: vec![Arc::new(Float64Array::from(vec![
+                Some(1.),
+                Some(2.),
+                None,
+                Some(3.),
+                Some(3.),
+                Some(3.),
+            ]))],
+            expected_output: None,
+            expected_fn: Some(|arr| {
+                let percent = ScalarValue::Float64(Some(0.5)).to_array().unwrap();
+                let percent = datatypes::vectors::Helper::try_into_vector(percent).unwrap();
+                let state = datatypes::vectors::Helper::try_into_vector(arr).unwrap();
+                let udd_calc = UddSketchCalcFunction;
+                let res = udd_calc
+                    .eval(&Default::default(), &[percent, state])
+                    .unwrap();
+                let binding = res.to_arrow_array();
+                let res_arr = binding.as_any().downcast_ref::<Float64Array>().unwrap();
+                assert!(res_arr.len() == 1);
+                assert!((res_arr.value(0) - 2.856578984907706f64).abs() <= f64::EPSILON);
+                true
+            }),
+            distinct: false,
+            filter: None,
+            order_by: None,
+            null_treatment: None,
+        },
+        TestCase {
+            func: Arc::new(HllState::state_udf_impl()),
+            input_schema: Arc::new(arrow_schema::Schema::new(vec![Field::new(
+                "word",
+                DataType::Utf8,
+                true,
+            )])),
+            args: vec![Expr::Column(Column::new_unqualified("word"))],
+            input: vec![Arc::new(StringArray::from(vec![
+                Some("foo"),
+                Some("bar"),
+                None,
+                Some("baz"),
+                Some("baz"),
+            ]))],
+            expected_output: None,
+            expected_fn: Some(|arr| {
+                let state = datatypes::vectors::Helper::try_into_vector(arr).unwrap();
+                let hll_calc = HllCalcFunction;
+                let res = hll_calc.eval(&Default::default(), &[state]).unwrap();
+                let binding = res.to_arrow_array();
+                let res_arr = binding.as_any().downcast_ref::<UInt64Array>().unwrap();
+                assert!(res_arr.len() == 1);
+                assert_eq!(res_arr.value(0), 3);
+                true
+            }),
+            distinct: false,
+            filter: None,
+            order_by: None,
+            null_treatment: None,
+        },
+        // TODO(discord9): udd_merge/hll_merge/geo_path/quantile_aggr tests
+    ];
+    let test_table_ref = TableReference::bare("TestTable");
+
+    for case in test_cases {
+        let ctx = SessionContext::new();
+        let table_provider = DummyTableProvider::new(
+            case.input_schema.clone(),
+            Some(RecordBatch::try_new(case.input_schema.clone(), case.input.clone()).unwrap()),
+        );
+        let table_source = DefaultTableSource::new(Arc::new(table_provider));
+        let logical_plan = LogicalPlan::TableScan(
+            TableScan::try_new(
+                test_table_ref.clone(),
+                Arc::new(table_source),
+                None,
+                vec![],
+                None,
+            )
+            .unwrap(),
+        );
+
+        let args = case.args;
+
+        let aggr_expr = Expr::AggregateFunction(AggregateFunction::new_udf(
+            case.func.clone(),
+            args,
+            case.distinct,
+            case.filter,
+            case.order_by,
+            case.null_treatment,
+        ));
+
+        let aggr_plan = LogicalPlan::Aggregate(
+            Aggregate::try_new(Arc::new(logical_plan), vec![], vec![aggr_expr]).unwrap(),
+        );
+
+        // make sure the aggr_plan is type coerced
+        let aggr_plan = TypeCoercion::new()
+            .analyze(aggr_plan, &Default::default())
+            .unwrap();
+
+        // first eval the original aggregate function
+        let phy_full_aggr_plan = DefaultPhysicalPlanner::default()
+            .create_physical_plan(&aggr_plan, &ctx.state())
+            .await
+            .unwrap();
+
+        {
+            let unsplit_result = execute_phy_plan(&phy_full_aggr_plan).await.unwrap();
+            assert_eq!(unsplit_result.len(), 1);
+            let unsplit_batch = &unsplit_result[0];
+            assert_eq!(unsplit_batch.num_columns(), 1);
+            assert_eq!(unsplit_batch.num_rows(), 1);
+            let unsplit_col = unsplit_batch.column(0);
+            if let Some(expected_output) = &case.expected_output {
+                assert_eq!(unsplit_col.data_type(), &expected_output.data_type());
+                assert_eq!(unsplit_col.len(), 1);
+                assert_eq!(unsplit_col, &expected_output.to_array().unwrap());
+            }
+
+            if let Some(expected_fn) = &case.expected_fn {
+                assert!(expected_fn(unsplit_col.clone()));
+            }
+        }
+        let LogicalPlan::Aggregate(aggr_plan) = aggr_plan else {
+            panic!("Expected Aggregate plan");
+        };
+        let split_plan = StateMergeHelper::split_aggr_node(aggr_plan).unwrap();
+
+        let phy_upper_plan = DefaultPhysicalPlanner::default()
+            .create_physical_plan(&split_plan.upper_merge, &ctx.state())
+            .await
+            .unwrap();
+
+        // since upper plan use lower plan as input, execute upper plan should also execute lower plan
+        // which should give the same result as the original aggregate function
+        {
+            let split_res = execute_phy_plan(&phy_upper_plan).await.unwrap();
+
+            assert_eq!(split_res.len(), 1);
+            let split_batch = &split_res[0];
+            assert_eq!(split_batch.num_columns(), 1);
+            assert_eq!(split_batch.num_rows(), 1);
+            let split_col = split_batch.column(0);
+            if let Some(expected_output) = &case.expected_output {
+                assert_eq!(split_col.data_type(), &expected_output.data_type());
+                assert_eq!(split_col.len(), 1);
+                assert_eq!(split_col, &expected_output.to_array().unwrap());
+            }
+
+            if let Some(expected_fn) = &case.expected_fn {
+                assert!(expected_fn(split_col.clone()));
+            }
+        }
+    }
+}
+
+async fn execute_phy_plan(
+    phy_plan: &Arc<dyn ExecutionPlan>,
+) -> datafusion_common::Result<Vec<RecordBatch>> {
+    let task_ctx = Arc::new(TaskContext::default());
+    let mut stream = phy_plan.execute(0, task_ctx)?;
+    let mut batches = Vec::new();
+    while let Some(batch) = stream.next().await {
+        batches.push(batch?);
+    }
+    Ok(batches)
+}

src/common/function/src/system/pg_catalog/version.rs

@@ -12,8 +12,8 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
+use std::fmt;
 use std::sync::Arc;
-use std::{env, fmt};
 
 use common_query::error::Result;
 use common_query::prelude::{Signature, Volatility};
@@ -47,7 +47,7 @@ impl Function for PGVersionFunction {
     fn eval(&self, _func_ctx: &FunctionContext, _columns: &[VectorRef]) -> Result<VectorRef> {
         let result = StringVector::from(vec![format!(
             "PostgreSQL 16.3 GreptimeDB {}",
-            env!("CARGO_PKG_VERSION")
+            common_version::version()
         )]);
         Ok(Arc::new(result))
     }

src/common/function/src/system/version.rs

@@ -12,8 +12,8 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
+use std::fmt;
 use std::sync::Arc;
-use std::{env, fmt};
 
 use common_query::error::Result;
 use common_query::prelude::{Signature, Volatility};
@@ -52,13 +52,13 @@ impl Function for VersionFunction {
                     "{}-greptimedb-{}",
                     std::env::var("GREPTIMEDB_MYSQL_SERVER_VERSION")
                         .unwrap_or_else(|_| "8.4.2".to_string()),
-                    env!("CARGO_PKG_VERSION")
+                    common_version::version()
                 )
             }
             Channel::Postgres => {
-                format!("16.3-greptimedb-{}", env!("CARGO_PKG_VERSION"))
+                format!("16.3-greptimedb-{}", common_version::version())
             }
-            _ => env!("CARGO_PKG_VERSION").to_string(),
+            _ => common_version::version().to_string(),
         };
         let result = StringVector::from(vec![version]);
         Ok(Arc::new(result))

src/common/greptimedb-telemetry/src/lib.rs

@@ -16,8 +16,8 @@ use std::env;
 use std::io::ErrorKind;
 use std::path::{Path, PathBuf};
 use std::sync::atomic::{AtomicBool, Ordering};
-use std::sync::Arc;
-use std::time::Duration;
+use std::sync::{Arc, LazyLock};
+use std::time::{Duration, SystemTime};
 
 use common_runtime::error::{Error, Result};
 use common_runtime::{BoxedTaskFunction, RepeatedTask, TaskFunction};
@@ -31,6 +31,9 @@ pub const TELEMETRY_URL: &str = "https://telemetry.greptimestats.com/db/otel/sta
 /// The local installation uuid cache file
 const UUID_FILE_NAME: &str = ".greptimedb-telemetry-uuid";
 
+/// System start time for uptime calculation
+static START_TIME: LazyLock<SystemTime> = LazyLock::new(SystemTime::now);
+
 /// The default interval of reporting telemetry data to greptime cloud
 pub static TELEMETRY_INTERVAL: Duration = Duration::from_secs(60 * 30);
 /// The default connect timeout to greptime cloud.
@@ -103,6 +106,8 @@ struct StatisticData {
     pub nodes: Option<i32>,
     /// The local installation uuid
     pub uuid: String,
+    /// System uptime range (e.g., "hours", "days", "weeks")
+    pub uptime: String,
 }
 
 #[derive(Serialize, Deserialize, Debug, Eq, PartialEq)]
@@ -171,6 +176,25 @@ fn print_anonymous_usage_data_disclaimer() {
     info!("https://docs.greptime.com/reference/telemetry");
 }
 
+/// Format uptime duration into a general time range string
+/// Returns privacy-friendly descriptions like "hours", "days", etc.
+fn format_uptime() -> String {
+    let uptime_duration = START_TIME.elapsed().unwrap_or(Duration::ZERO);
+    let total_seconds = uptime_duration.as_secs();
+
+    if total_seconds < 86400 {
+        "hours".to_string()
+    } else if total_seconds < 604800 {
+        "days".to_string()
+    } else if total_seconds < 2629746 {
+        "weeks".to_string()
+    } else if total_seconds < 31556952 {
+        "months".to_string()
+    } else {
+        "years".to_string()
+    }
+}
+
 pub fn default_get_uuid(working_home: &Option<String>) -> Option<String> {
     let temp_dir = env::temp_dir();
 
@@ -260,6 +284,7 @@ impl GreptimeDBTelemetry {
                     mode: self.statistics.get_mode(),
                     nodes: self.statistics.get_nodes().await,
                     uuid,
+                    uptime: format_uptime(),
                 };
 
                 if let Some(client) = self.client.as_ref() {
@@ -294,7 +319,9 @@ mod tests {
     use reqwest::{Client, Response};
     use tokio::spawn;
 
-    use crate::{default_get_uuid, Collector, GreptimeDBTelemetry, Mode, StatisticData};
+    use crate::{
+        default_get_uuid, format_uptime, Collector, GreptimeDBTelemetry, Mode, StatisticData,
+    };
 
     static COUNT: AtomicUsize = std::sync::atomic::AtomicUsize::new(0);
 
@@ -438,6 +465,7 @@ mod tests {
         assert_eq!(build_info().commit, body.git_commit);
         assert_eq!(Mode::Standalone, body.mode);
         assert_eq!(1, body.nodes.unwrap());
+        assert!(!body.uptime.is_empty());
 
         let failed_statistic = Box::new(FailedStatistic);
         let failed_report = GreptimeDBTelemetry::new(
@@ -477,4 +505,18 @@ mod tests {
         assert_eq!(uuid, default_get_uuid(&Some(working_home.clone())));
         assert_eq!(uuid, default_get_uuid(&Some(working_home)));
     }
+
+    #[test]
+    fn test_format_uptime() {
+        let uptime = format_uptime();
+        assert!(!uptime.is_empty());
+        // Should be a valid general time range (no specific numbers)
+        assert!(
+            uptime == "hours"
+                || uptime == "days"
+                || uptime == "weeks"
+                || uptime == "months"
+                || uptime == "years"
+        );
+    }
 }

src/common/grpc-expr/src/alter.rs

@@ -29,8 +29,8 @@ use snafu::{ensure, OptionExt, ResultExt};
 use store_api::region_request::{SetRegionOption, UnsetRegionOption};
 use table::metadata::TableId;
 use table::requests::{
-    AddColumnRequest, AlterKind, AlterTableRequest, ModifyColumnTypeRequest, SetIndexOptions,
-    UnsetIndexOptions,
+    AddColumnRequest, AlterKind, AlterTableRequest, ModifyColumnTypeRequest, SetIndexOption,
+    UnsetIndexOption,
 };
 
 use crate::error::{
@@ -43,6 +43,59 @@ use crate::error::{
 const LOCATION_TYPE_FIRST: i32 = LocationType::First as i32;
 const LOCATION_TYPE_AFTER: i32 = LocationType::After as i32;
 
+fn set_index_option_from_proto(set_index: api::v1::SetIndex) -> Result<SetIndexOption> {
+    let options = set_index.options.context(MissingAlterIndexOptionSnafu)?;
+    Ok(match options {
+        api::v1::set_index::Options::Fulltext(f) => SetIndexOption::Fulltext {
+            column_name: f.column_name.clone(),
+            options: FulltextOptions::new(
+                f.enable,
+                as_fulltext_option_analyzer(
+                    Analyzer::try_from(f.analyzer).context(InvalidSetFulltextOptionRequestSnafu)?,
+                ),
+                f.case_sensitive,
+                as_fulltext_option_backend(
+                    PbFulltextBackend::try_from(f.backend)
+                        .context(InvalidSetFulltextOptionRequestSnafu)?,
+                ),
+                f.granularity as u32,
+                f.false_positive_rate,
+            )
+            .context(InvalidIndexOptionSnafu)?,
+        },
+        api::v1::set_index::Options::Inverted(i) => SetIndexOption::Inverted {
+            column_name: i.column_name,
+        },
+        api::v1::set_index::Options::Skipping(s) => SetIndexOption::Skipping {
+            column_name: s.column_name,
+            options: SkippingIndexOptions::new(
+                s.granularity as u32,
+                s.false_positive_rate,
+                as_skipping_index_type(
+                    PbSkippingIndexType::try_from(s.skipping_index_type)
+                        .context(InvalidSetSkippingIndexOptionRequestSnafu)?,
+                ),
+            )
+            .context(InvalidIndexOptionSnafu)?,
+        },
+    })
+}
+
+fn unset_index_option_from_proto(unset_index: api::v1::UnsetIndex) -> Result<UnsetIndexOption> {
+    let options = unset_index.options.context(MissingAlterIndexOptionSnafu)?;
+    Ok(match options {
+        api::v1::unset_index::Options::Fulltext(f) => UnsetIndexOption::Fulltext {
+            column_name: f.column_name,
+        },
+        api::v1::unset_index::Options::Inverted(i) => UnsetIndexOption::Inverted {
+            column_name: i.column_name,
+        },
+        api::v1::unset_index::Options::Skipping(s) => UnsetIndexOption::Skipping {
+            column_name: s.column_name,
+        },
+    })
+}
+
 /// Convert an [`AlterTableExpr`] to an [`AlterTableRequest`]
 pub fn alter_expr_to_request(table_id: TableId, expr: AlterTableExpr) -> Result<AlterTableRequest> {
     let catalog_name = expr.catalog_name;
@@ -121,70 +174,34 @@ pub fn alter_expr_to_request(table_id: TableId, expr: AlterTableExpr) -> Result<
                     .context(InvalidUnsetTableOptionRequestSnafu)?,
             }
         }
-        Kind::SetIndex(o) => match o.options {
-            Some(opt) => match opt {
-                api::v1::set_index::Options::Fulltext(f) => AlterKind::SetIndex {
-                    options: SetIndexOptions::Fulltext {
-                        column_name: f.column_name.clone(),
-                        options: FulltextOptions::new(
-                            f.enable,
-                            as_fulltext_option_analyzer(
-                                Analyzer::try_from(f.analyzer)
-                                    .context(InvalidSetFulltextOptionRequestSnafu)?,
-                            ),
-                            f.case_sensitive,
-                            as_fulltext_option_backend(
-                                PbFulltextBackend::try_from(f.backend)
-                                    .context(InvalidSetFulltextOptionRequestSnafu)?,
-                            ),
-                            f.granularity as u32,
-                            f.false_positive_rate,
-                        )
-                        .context(InvalidIndexOptionSnafu)?,
-                    },
-                },
-                api::v1::set_index::Options::Inverted(i) => AlterKind::SetIndex {
-                    options: SetIndexOptions::Inverted {
-                        column_name: i.column_name,
-                    },
-                },
-                api::v1::set_index::Options::Skipping(s) => AlterKind::SetIndex {
-                    options: SetIndexOptions::Skipping {
-                        column_name: s.column_name,
-                        options: SkippingIndexOptions::new(
-                            s.granularity as u32,
-                            s.false_positive_rate,
-                            as_skipping_index_type(
-                                PbSkippingIndexType::try_from(s.skipping_index_type)
-                                    .context(InvalidSetSkippingIndexOptionRequestSnafu)?,
-                            ),
-                        )
-                        .context(InvalidIndexOptionSnafu)?,
-                    },
-                },
-            },
-            None => return MissingAlterIndexOptionSnafu.fail(),
-        },
-        Kind::UnsetIndex(o) => match o.options {
-            Some(opt) => match opt {
-                api::v1::unset_index::Options::Fulltext(f) => AlterKind::UnsetIndex {
-                    options: UnsetIndexOptions::Fulltext {
-                        column_name: f.column_name,
-                    },
-                },
-                api::v1::unset_index::Options::Inverted(i) => AlterKind::UnsetIndex {
-                    options: UnsetIndexOptions::Inverted {
-                        column_name: i.column_name,
-                    },
-                },
-                api::v1::unset_index::Options::Skipping(s) => AlterKind::UnsetIndex {
-                    options: UnsetIndexOptions::Skipping {
-                        column_name: s.column_name,
-                    },
-                },
-            },
-            None => return MissingAlterIndexOptionSnafu.fail(),
-        },
+        Kind::SetIndex(o) => {
+            let option = set_index_option_from_proto(o)?;
+            AlterKind::SetIndexes {
+                options: vec![option],
+            }
+        }
+        Kind::UnsetIndex(o) => {
+            let option = unset_index_option_from_proto(o)?;
+            AlterKind::UnsetIndexes {
+                options: vec![option],
+            }
+        }
+        Kind::SetIndexes(o) => {
+            let options = o
+                .set_indexes
+                .into_iter()
+                .map(set_index_option_from_proto)
+                .collect::<Result<Vec<_>>>()?;
+            AlterKind::SetIndexes { options }
+        }
+        Kind::UnsetIndexes(o) => {
+            let options = o
+                .unset_indexes
+                .into_iter()
+                .map(unset_index_option_from_proto)
+                .collect::<Result<Vec<_>>>()?;
+            AlterKind::UnsetIndexes { options }
+        }
         Kind::DropDefaults(o) => {
             let names = o
                 .drop_defaults

src/common/grpc/Cargo.toml

@@ -31,6 +31,7 @@ tokio.workspace = true
 tokio-util.workspace = true
 tonic.workspace = true
 tower.workspace = true
+vec1 = "1.12"
 
 [dev-dependencies]
 criterion = "0.4"

src/common/grpc/benches/bench_flight_decoder.rs

@@ -84,9 +84,12 @@ fn prepare_random_record_batch(
 fn prepare_flight_data(num_rows: usize) -> (FlightData, FlightData) {
     let schema = schema();
     let mut encoder = FlightEncoder::default();
-    let schema_data = encoder.encode(FlightMessage::Schema(schema.clone()));
+    let schema_data = encoder.encode_schema(schema.as_ref());
     let rb = prepare_random_record_batch(schema, num_rows);
-    let rb_data = encoder.encode(FlightMessage::RecordBatch(rb));
+    let [rb_data] = encoder
+        .encode(FlightMessage::RecordBatch(rb))
+        .try_into()
+        .unwrap();
     (schema_data, rb_data)
 }
 
@@ -96,7 +99,7 @@ fn decode_flight_data_from_protobuf(schema: &Bytes, payload: &Bytes) -> DfRecord
     let mut decoder = FlightDecoder::default();
     let _schema = decoder.try_decode(&schema).unwrap();
     let message = decoder.try_decode(&payload).unwrap();
-    let FlightMessage::RecordBatch(batch) = message else {
+    let Some(FlightMessage::RecordBatch(batch)) = message else {
         unreachable!("unexpected message");
     };
     batch

src/common/grpc/src/flight.rs

@@ -23,6 +23,7 @@ use arrow_flight::{FlightData, SchemaAsIpc};
 use common_base::bytes::Bytes;
 use common_recordbatch::DfRecordBatch;
 use datatypes::arrow;
+use datatypes::arrow::array::ArrayRef;
 use datatypes::arrow::buffer::Buffer;
 use datatypes::arrow::datatypes::{Schema as ArrowSchema, SchemaRef};
 use datatypes::arrow::error::ArrowError;
@@ -31,6 +32,7 @@ use flatbuffers::FlatBufferBuilder;
 use prost::bytes::Bytes as ProstBytes;
 use prost::Message;
 use snafu::{OptionExt, ResultExt};
+use vec1::{vec1, Vec1};
 
 use crate::error;
 use crate::error::{DecodeFlightDataSnafu, InvalidFlightDataSnafu, Result};
@@ -77,9 +79,19 @@ impl FlightEncoder {
         }
     }
 
-    pub fn encode(&mut self, flight_message: FlightMessage) -> FlightData {
+    /// Encode the Arrow schema to [FlightData].
+    pub fn encode_schema(&self, schema: &ArrowSchema) -> FlightData {
+        SchemaAsIpc::new(schema, &self.write_options).into()
+    }
+
+    /// Encode the [FlightMessage] to a list (at least one element) of [FlightData]s.
+    ///
+    /// Normally only when the [FlightMessage] is an Arrow [RecordBatch] with dictionary arrays
+    /// will the encoder produce more than one [FlightData]s. Other types of [FlightMessage] should
+    /// be encoded to exactly one [FlightData].
+    pub fn encode(&mut self, flight_message: FlightMessage) -> Vec1<FlightData> {
         match flight_message {
-            FlightMessage::Schema(schema) => SchemaAsIpc::new(&schema, &self.write_options).into(),
+            FlightMessage::Schema(schema) => vec1![self.encode_schema(schema.as_ref())],
             FlightMessage::RecordBatch(record_batch) => {
                 let (encoded_dictionaries, encoded_batch) = self
                     .data_gen
@@ -90,14 +102,10 @@ impl FlightEncoder {
                     )
                     .expect("DictionaryTracker configured above to not fail on replacement");
 
-                // TODO(LFC): Handle dictionary as FlightData here, when we supported Arrow's Dictionary DataType.
-                // Currently we don't have a datatype corresponding to Arrow's Dictionary DataType,
-                // so there won't be any "dictionaries" here. Assert to be sure about it, and
-                // perform a "testing guard" in case we forgot to handle the possible "dictionaries"
-                // here in the future.
-                debug_assert_eq!(encoded_dictionaries.len(), 0);
-
-                encoded_batch.into()
+                Vec1::from_vec_push(
+                    encoded_dictionaries.into_iter().map(Into::into).collect(),
+                    encoded_batch.into(),
+                )
             }
             FlightMessage::AffectedRows(rows) => {
                 let metadata = FlightMetadata {
@@ -105,12 +113,12 @@ impl FlightEncoder {
                     metrics: None,
                 }
                 .encode_to_vec();
-                FlightData {
+                vec1![FlightData {
                     flight_descriptor: None,
                     data_header: build_none_flight_msg().into(),
                     app_metadata: metadata.into(),
                     data_body: ProstBytes::default(),
-                }
+                }]
             }
             FlightMessage::Metrics(s) => {
                 let metadata = FlightMetadata {
@@ -120,12 +128,12 @@ impl FlightEncoder {
                     }),
                 }
                 .encode_to_vec();
-                FlightData {
+                vec1![FlightData {
                     flight_descriptor: None,
                     data_header: build_none_flight_msg().into(),
                     app_metadata: metadata.into(),
                     data_body: ProstBytes::default(),
-                }
+                }]
             }
         }
     }
@@ -135,6 +143,7 @@ impl FlightEncoder {
 pub struct FlightDecoder {
     schema: Option<SchemaRef>,
     schema_bytes: Option<bytes::Bytes>,
+    dictionaries_by_id: HashMap<i64, ArrayRef>,
 }
 
 impl FlightDecoder {
@@ -145,6 +154,7 @@ impl FlightDecoder {
         Ok(Self {
             schema: Some(Arc::new(arrow_schema)),
             schema_bytes: Some(schema_bytes.clone()),
+            dictionaries_by_id: HashMap::new(),
         })
     }
 
@@ -186,7 +196,13 @@ impl FlightDecoder {
         Ok(result)
     }
 
-    pub fn try_decode(&mut self, flight_data: &FlightData) -> Result<FlightMessage> {
+    /// Try to decode the [FlightData] to a [FlightMessage].
+    ///
+    /// If the [FlightData] is of type `DictionaryBatch` (produced while encoding an Arrow
+    /// [RecordBatch] with dictionary arrays), the decoder will not return any [FlightMessage]s.
+    /// Instead, it will update its internal dictionary cache. Other types of [FlightData] will
+    /// be decoded to exactly one [FlightMessage].
+    pub fn try_decode(&mut self, flight_data: &FlightData) -> Result<Option<FlightMessage>> {
         let message = root_as_message(&flight_data.data_header).map_err(|e| {
             InvalidFlightDataSnafu {
                 reason: e.to_string(),
@@ -198,12 +214,12 @@ impl FlightDecoder {
                 let metadata = FlightMetadata::decode(flight_data.app_metadata.clone())
                     .context(DecodeFlightDataSnafu)?;
                 if let Some(AffectedRows { value }) = metadata.affected_rows {
-                    return Ok(FlightMessage::AffectedRows(value as _));
+                    return Ok(Some(FlightMessage::AffectedRows(value as _)));
                 }
                 if let Some(Metrics { metrics }) = metadata.metrics {
-                    return Ok(FlightMessage::Metrics(
+                    return Ok(Some(FlightMessage::Metrics(
                         String::from_utf8_lossy(&metrics).to_string(),
-                    ));
+                    )));
                 }
                 InvalidFlightDataSnafu {
                     reason: "Expecting FlightMetadata have some meaningful content.",
@@ -219,21 +235,46 @@ impl FlightDecoder {
                 })?);
                 self.schema = Some(arrow_schema.clone());
                 self.schema_bytes = Some(flight_data.data_header.clone());
-                Ok(FlightMessage::Schema(arrow_schema))
+                Ok(Some(FlightMessage::Schema(arrow_schema)))
             }
             MessageHeader::RecordBatch => {
                 let schema = self.schema.clone().context(InvalidFlightDataSnafu {
                     reason: "Should have decoded schema first!",
                 })?;
-                let arrow_batch =
-                    flight_data_to_arrow_batch(flight_data, schema.clone(), &HashMap::new())
-                        .map_err(|e| {
-                            InvalidFlightDataSnafu {
-                                reason: e.to_string(),
-                            }
-                            .build()
+                let arrow_batch = flight_data_to_arrow_batch(
+                    flight_data,
+                    schema.clone(),
+                    &self.dictionaries_by_id,
+                )
+                .map_err(|e| {
+                    InvalidFlightDataSnafu {
+                        reason: e.to_string(),
+                    }
+                    .build()
+                })?;
+                Ok(Some(FlightMessage::RecordBatch(arrow_batch)))
+            }
+            MessageHeader::DictionaryBatch => {
+                let dictionary_batch =
+                    message
+                        .header_as_dictionary_batch()
+                        .context(InvalidFlightDataSnafu {
+                            reason: "could not get dictionary batch from DictionaryBatch message",
                         })?;
-                Ok(FlightMessage::RecordBatch(arrow_batch))
+
+                let schema = self.schema.as_ref().context(InvalidFlightDataSnafu {
+                    reason: "schema message is not present previously",
+                })?;
+
+                reader::read_dictionary(
+                    &flight_data.data_body.clone().into(),
+                    dictionary_batch,
+                    schema,
+                    &mut self.dictionaries_by_id,
+                    &message.version(),
+                )
+                .context(error::ArrowSnafu)?;
+                Ok(None)
             }
             other => {
                 let name = other.variant_name().unwrap_or("UNKNOWN");
@@ -305,14 +346,16 @@ fn build_none_flight_msg() -> Bytes {
 #[cfg(test)]
 mod test {
     use arrow_flight::utils::batches_to_flight_data;
-    use datatypes::arrow::array::Int32Array;
+    use datatypes::arrow::array::{
+        DictionaryArray, Int32Array, StringArray, UInt32Array, UInt8Array,
+    };
     use datatypes::arrow::datatypes::{DataType, Field, Schema};
 
     use super::*;
     use crate::Error;
 
     #[test]
-    fn test_try_decode() {
+    fn test_try_decode() -> Result<()> {
         let schema = Arc::new(ArrowSchema::new(vec![Field::new(
             "n",
             DataType::Int32,
@@ -347,7 +390,7 @@ mod test {
             .to_string()
             .contains("Should have decoded schema first!"));
 
-        let message = decoder.try_decode(d1).unwrap();
+        let message = decoder.try_decode(d1)?.unwrap();
         assert!(matches!(message, FlightMessage::Schema(_)));
         let FlightMessage::Schema(decoded_schema) = message else {
             unreachable!()
@@ -356,19 +399,20 @@ mod test {
 
         let _ = decoder.schema.as_ref().unwrap();
 
-        let message = decoder.try_decode(d2).unwrap();
+        let message = decoder.try_decode(d2)?.unwrap();
         assert!(matches!(message, FlightMessage::RecordBatch(_)));
         let FlightMessage::RecordBatch(actual_batch) = message else {
             unreachable!()
         };
         assert_eq!(actual_batch, batch1);
 
-        let message = decoder.try_decode(d3).unwrap();
+        let message = decoder.try_decode(d3)?.unwrap();
         assert!(matches!(message, FlightMessage::RecordBatch(_)));
         let FlightMessage::RecordBatch(actual_batch) = message else {
             unreachable!()
         };
         assert_eq!(actual_batch, batch2);
+        Ok(())
     }
 
     #[test]
@@ -407,4 +451,86 @@ mod test {
         let actual = flight_messages_to_recordbatches(vec![m1, m2, m3]).unwrap();
         assert_eq!(actual, recordbatches);
     }
+
+    #[test]
+    fn test_flight_encode_decode_with_dictionary_array() -> Result<()> {
+        let schema = Arc::new(Schema::new(vec![
+            Field::new("i", DataType::UInt8, true),
+            Field::new_dictionary("s", DataType::UInt32, DataType::Utf8, true),
+        ]));
+        let batch1 = DfRecordBatch::try_new(
+            schema.clone(),
+            vec![
+                Arc::new(UInt8Array::from_iter_values(vec![1, 2, 3])) as _,
+                Arc::new(DictionaryArray::new(
+                    UInt32Array::from_value(0, 3),
+                    Arc::new(StringArray::from_iter_values(["x"])),
+                )) as _,
+            ],
+        )
+        .unwrap();
+        let batch2 = DfRecordBatch::try_new(
+            schema.clone(),
+            vec![
+                Arc::new(UInt8Array::from_iter_values(vec![4, 5, 6, 7, 8])) as _,
+                Arc::new(DictionaryArray::new(
+                    UInt32Array::from_iter_values([0, 1, 2, 2, 3]),
+                    Arc::new(StringArray::from_iter_values(["h", "e", "l", "o"])),
+                )) as _,
+            ],
+        )
+        .unwrap();
+
+        let message_1 = FlightMessage::Schema(schema.clone());
+        let message_2 = FlightMessage::RecordBatch(batch1);
+        let message_3 = FlightMessage::RecordBatch(batch2);
+
+        let mut encoder = FlightEncoder::default();
+        let encoded_1 = encoder.encode(message_1);
+        let encoded_2 = encoder.encode(message_2);
+        let encoded_3 = encoder.encode(message_3);
+        // message 1 is Arrow Schema, should be encoded to one FlightData:
+        assert_eq!(encoded_1.len(), 1);
+        // message 2 and 3 are Arrow RecordBatch with dictionary arrays, should be encoded to
+        // multiple FlightData:
+        assert_eq!(encoded_2.len(), 2);
+        assert_eq!(encoded_3.len(), 2);
+
+        let mut decoder = FlightDecoder::default();
+        let decoded_1 = decoder.try_decode(encoded_1.first())?;
+        let Some(FlightMessage::Schema(actual_schema)) = decoded_1 else {
+            unreachable!()
+        };
+        assert_eq!(actual_schema, schema);
+        let decoded_2 = decoder.try_decode(&encoded_2[0])?;
+        // expected to be a dictionary batch message, decoder should return none:
+        assert!(decoded_2.is_none());
+        let Some(FlightMessage::RecordBatch(decoded_2)) = decoder.try_decode(&encoded_2[1])? else {
+            unreachable!()
+        };
+        let decoded_3 = decoder.try_decode(&encoded_3[0])?;
+        // expected to be a dictionary batch message, decoder should return none:
+        assert!(decoded_3.is_none());
+        let Some(FlightMessage::RecordBatch(decoded_3)) = decoder.try_decode(&encoded_3[1])? else {
+            unreachable!()
+        };
+        let actual = arrow::util::pretty::pretty_format_batches(&[decoded_2, decoded_3])
+            .unwrap()
+            .to_string();
+        let expected = r"
++---+---+
+| i | s |
++---+---+
+| 1 | x |
+| 2 | x |
+| 3 | x |
+| 4 | h |
+| 5 | e |
+| 6 | l |
+| 7 | l |
+| 8 | o |
++---+---+";
+        assert_eq!(actual, expected.trim());
+        Ok(())
+    }
 }

src/common/meta/Cargo.toml

@@ -69,6 +69,7 @@ table = { workspace = true, features = ["testing"] }
 tokio.workspace = true
 tokio-postgres = { workspace = true, optional = true }
 tonic.workspace = true
+tracing.workspace = true
 typetag.workspace = true
 
 [dev-dependencies]

src/common/meta/src/cache/flow/table_flownode.rs

@@ -15,6 +15,7 @@
 use std::collections::HashMap;
 use std::sync::Arc;
 
+use common_telemetry::info;
 use futures::future::BoxFuture;
 use moka::future::Cache;
 use moka::ops::compute::Op;
@@ -89,6 +90,12 @@ fn init_factory(table_flow_manager: TableFlowManagerRef) -> Initializer<TableId,
                 // we have a corresponding cache invalidation mechanism to invalidate `(Key, EmptyHashSet)`.
                 .map(Arc::new)
                 .map(Some)
+                .inspect(|set| {
+                    info!(
+                        "Initialized table_flownode cache for table_id: {}, set: {:?}",
+                        table_id, set
+                    );
+                })
         })
     })
 }
@@ -167,6 +174,13 @@ fn invalidator<'a>(
         match ident {
             CacheIdent::CreateFlow(create_flow) => handle_create_flow(cache, create_flow).await,
             CacheIdent::DropFlow(drop_flow) => handle_drop_flow(cache, drop_flow).await,
+            CacheIdent::FlowNodeAddressChange(node_id) => {
+                info!(
+                    "Invalidate flow node cache for node_id in table_flownode: {}",
+                    node_id
+                );
+                cache.invalidate_all();
+            }
             _ => {}
         }
         Ok(())
@@ -174,7 +188,10 @@ fn invalidator<'a>(
 }
 
 fn filter(ident: &CacheIdent) -> bool {
-    matches!(ident, CacheIdent::CreateFlow(_) | CacheIdent::DropFlow(_))
+    matches!(
+        ident,
+        CacheIdent::CreateFlow(_) | CacheIdent::DropFlow(_) | CacheIdent::FlowNodeAddressChange(_)
+    )
 }
 
 #[cfg(test)]

src/common/meta/src/cache_invalidator.rs

@@ -22,6 +22,7 @@ use crate::key::flow::flow_name::FlowNameKey;
 use crate::key::flow::flow_route::FlowRouteKey;
 use crate::key::flow::flownode_flow::FlownodeFlowKey;
 use crate::key::flow::table_flow::TableFlowKey;
+use crate::key::node_address::NodeAddressKey;
 use crate::key::schema_name::SchemaNameKey;
 use crate::key::table_info::TableInfoKey;
 use crate::key::table_name::TableNameKey;
@@ -53,6 +54,10 @@ pub struct Context {
 #[async_trait::async_trait]
 pub trait CacheInvalidator: Send + Sync {
     async fn invalidate(&self, ctx: &Context, caches: &[CacheIdent]) -> Result<()>;
+
+    fn name(&self) -> &'static str {
+        std::any::type_name::<Self>()
+    }
 }
 
 pub type CacheInvalidatorRef = Arc<dyn CacheInvalidator>;
@@ -137,6 +142,13 @@ where
                     let key = FlowInfoKey::new(*flow_id);
                     self.invalidate_key(&key.to_bytes()).await;
                 }
+                CacheIdent::FlowNodeAddressChange(node_id) => {
+                    // other caches doesn't need to be invalidated
+                    // since this is only for flownode address change not id change
+                    common_telemetry::info!("Invalidate flow node cache for node_id: {}", node_id);
+                    let key = NodeAddressKey::with_flownode(*node_id);
+                    self.invalidate_key(&key.to_bytes()).await;
+                }
             }
         }
         Ok(())

src/common/meta/src/datanode.rs

@@ -93,6 +93,8 @@ pub struct RegionStat {
     pub manifest_size: u64,
     /// The size of the SST data files in bytes.
     pub sst_size: u64,
+    /// The num of the SST data files.
+    pub sst_num: u64,
     /// The size of the SST index files in bytes.
     pub index_size: u64,
     /// The manifest infoof the region.
@@ -173,8 +175,8 @@ impl RegionStat {
         std::mem::size_of::<RegionId>() +
         // rcus, wcus, approximate_bytes, num_rows
         std::mem::size_of::<i64>() * 4 +
-        // memtable_size, manifest_size, sst_size, index_size
-        std::mem::size_of::<u64>() * 4 +
+        // memtable_size, manifest_size, sst_size, sst_num, index_size
+        std::mem::size_of::<u64>() * 5 +
         // engine
         std::mem::size_of::<String>() + self.engine.capacity() +
         // region_manifest
@@ -275,6 +277,7 @@ impl From<&api::v1::meta::RegionStat> for RegionStat {
             memtable_size: region_stat.memtable_size,
             manifest_size: region_stat.manifest_size,
             sst_size: region_stat.sst_size,
+            sst_num: region_stat.sst_num,
             index_size: region_stat.index_size,
             region_manifest: region_stat.manifest.into(),
             data_topic_latest_entry_id: region_stat.data_topic_latest_entry_id,

src/common/meta/src/ddl.rs

@@ -50,7 +50,6 @@ pub mod drop_flow;
 pub mod drop_table;
 pub mod drop_view;
 pub mod flow_meta;
-mod physical_table_metadata;
 pub mod table_meta;
 #[cfg(any(test, feature = "testing"))]
 pub mod test_util;

src/common/meta/src/ddl/alter_logical_tables.rs

@@ -12,20 +12,17 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
-mod check;
-mod metadata;
-mod region_request;
-mod table_cache_keys;
+mod executor;
 mod update_metadata;
+mod validator;
 
 use api::region::RegionResponse;
 use async_trait::async_trait;
 use common_catalog::format_full_table_name;
 use common_procedure::error::{FromJsonSnafu, Result as ProcedureResult, ToJsonSnafu};
 use common_procedure::{Context, LockKey, Procedure, Status};
-use common_telemetry::{error, info, warn};
-use futures_util::future;
-pub use region_request::make_alter_region_request;
+use common_telemetry::{debug, error, info, warn};
+pub use executor::make_alter_region_request;
 use serde::{Deserialize, Serialize};
 use snafu::ResultExt;
 use store_api::metadata::ColumnMetadata;
@@ -33,10 +30,12 @@ use store_api::metric_engine_consts::ALTER_PHYSICAL_EXTENSION_KEY;
 use strum::AsRefStr;
 use table::metadata::TableId;
 
-use crate::ddl::utils::{
-    add_peer_context_if_needed, extract_column_metadatas, map_to_procedure_error,
-    sync_follower_regions,
+use crate::cache_invalidator::Context as CacheContext;
+use crate::ddl::alter_logical_tables::executor::AlterLogicalTablesExecutor;
+use crate::ddl::alter_logical_tables::validator::{
+    retain_unskipped, AlterLogicalTableValidator, ValidatorResult,
 };
+use crate::ddl::utils::{extract_column_metadatas, map_to_procedure_error, sync_follower_regions};
 use crate::ddl::DdlContext;
 use crate::error::Result;
 use crate::instruction::CacheIdent;
@@ -46,13 +45,38 @@ use crate::key::DeserializedValueWithBytes;
 use crate::lock_key::{CatalogLock, SchemaLock, TableLock};
 use crate::metrics;
 use crate::rpc::ddl::AlterTableTask;
-use crate::rpc::router::{find_leaders, RegionRoute};
+use crate::rpc::router::RegionRoute;
 
 pub struct AlterLogicalTablesProcedure {
     pub context: DdlContext,
     pub data: AlterTablesData,
 }
 
+/// Builds the validator from the [`AlterTablesData`].
+fn build_validator_from_alter_table_data<'a>(
+    data: &'a AlterTablesData,
+) -> AlterLogicalTableValidator<'a> {
+    let phsycial_table_id = data.physical_table_id;
+    let alters = data
+        .tasks
+        .iter()
+        .map(|task| &task.alter_table)
+        .collect::<Vec<_>>();
+    AlterLogicalTableValidator::new(phsycial_table_id, alters)
+}
+
+/// Builds the executor from the [`AlterTablesData`].
+fn build_executor_from_alter_expr<'a>(data: &'a AlterTablesData) -> AlterLogicalTablesExecutor<'a> {
+    debug_assert_eq!(data.tasks.len(), data.table_info_values.len());
+    let alters = data
+        .tasks
+        .iter()
+        .zip(data.table_info_values.iter())
+        .map(|(task, table_info)| (table_info.table_info.ident.table_id, &task.alter_table))
+        .collect::<Vec<_>>();
+    AlterLogicalTablesExecutor::new(alters)
+}
+
 impl AlterLogicalTablesProcedure {
     pub const TYPE_NAME: &'static str = "metasrv-procedure::AlterLogicalTables";
 
@@ -82,35 +106,44 @@ impl AlterLogicalTablesProcedure {
     }
 
     pub(crate) async fn on_prepare(&mut self) -> Result<Status> {
-        // Checks all the tasks
-        self.check_input_tasks()?;
-        // Fills the table info values
-        self.fill_table_info_values().await?;
-        // Checks the physical table, must after [fill_table_info_values]
-        self.check_physical_table().await?;
-        // Fills the physical table info
-        self.fill_physical_table_info().await?;
-        // Filter the finished tasks
-        let finished_tasks = self.check_finished_tasks()?;
-        let already_finished_count = finished_tasks
-            .iter()
-            .map(|x| if *x { 1 } else { 0 })
-            .sum::<usize>();
-        let apply_tasks_count = self.data.tasks.len();
-        if already_finished_count == apply_tasks_count {
+        let validator = build_validator_from_alter_table_data(&self.data);
+        let ValidatorResult {
+            num_skipped,
+            skip_alter,
+            table_info_values,
+            physical_table_info,
+            physical_table_route,
+        } = validator
+            .validate(&self.context.table_metadata_manager)
+            .await?;
+
+        let num_tasks = self.data.tasks.len();
+        if num_skipped == num_tasks {
             info!("All the alter tasks are finished, will skip the procedure.");
+            let cache_ident_keys = AlterLogicalTablesExecutor::build_cache_ident_keys(
+                &physical_table_info,
+                &table_info_values
+                    .iter()
+                    .map(|v| v.get_inner_ref())
+                    .collect::<Vec<_>>(),
+            );
+            self.data.table_cache_keys_to_invalidate = cache_ident_keys;
             // Re-invalidate the table cache
             self.data.state = AlterTablesState::InvalidateTableCache;
             return Ok(Status::executing(true));
-        } else if already_finished_count > 0 {
+        } else if num_skipped > 0 {
             info!(
                 "There are {} alter tasks, {} of them were already finished.",
-                apply_tasks_count, already_finished_count
+                num_tasks, num_skipped
             );
         }
-        self.filter_task(&finished_tasks)?;
 
-        // Next state
+        // Updates the procedure state.
+        retain_unskipped(&mut self.data.tasks, &skip_alter);
+        self.data.physical_table_info = Some(physical_table_info);
+        self.data.physical_table_route = Some(physical_table_route);
+        self.data.table_info_values = table_info_values;
+        debug_assert_eq!(self.data.tasks.len(), self.data.table_info_values.len());
         self.data.state = AlterTablesState::SubmitAlterRegionRequests;
         Ok(Status::executing(true))
     }
@@ -118,25 +151,13 @@ impl AlterLogicalTablesProcedure {
     pub(crate) async fn on_submit_alter_region_requests(&mut self) -> Result<Status> {
         // Safety: we have checked the state in on_prepare
         let physical_table_route = &self.data.physical_table_route.as_ref().unwrap();
-        let leaders = find_leaders(&physical_table_route.region_routes);
-        let mut alter_region_tasks = Vec::with_capacity(leaders.len());
-
-        for peer in leaders {
-            let requester = self.context.node_manager.datanode(&peer).await;
-            let request = self.make_request(&peer, &physical_table_route.region_routes)?;
-
-            alter_region_tasks.push(async move {
-                requester
-                    .handle(request)
-                    .await
-                    .map_err(add_peer_context_if_needed(peer))
-            });
-        }
-
-        let mut results = future::join_all(alter_region_tasks)
-            .await
-            .into_iter()
-            .collect::<Result<Vec<_>>>()?;
+        let executor = build_executor_from_alter_expr(&self.data);
+        let mut results = executor
+            .on_alter_regions(
+                &self.context.node_manager,
+                &physical_table_route.region_routes,
+            )
+            .await?;
 
         if let Some(column_metadatas) =
             extract_column_metadatas(&mut results, ALTER_PHYSICAL_EXTENSION_KEY)?
@@ -177,7 +198,18 @@ impl AlterLogicalTablesProcedure {
         self.update_physical_table_metadata().await?;
         self.update_logical_tables_metadata().await?;
 
-        self.data.build_cache_keys_to_invalidate();
+        let logical_table_info_values = self
+            .data
+            .table_info_values
+            .iter()
+            .map(|v| v.get_inner_ref())
+            .collect::<Vec<_>>();
+
+        let cache_ident_keys = AlterLogicalTablesExecutor::build_cache_ident_keys(
+            self.data.physical_table_info.as_ref().unwrap(),
+            &logical_table_info_values,
+        );
+        self.data.table_cache_keys_to_invalidate = cache_ident_keys;
         self.data.clear_metadata_fields();
 
         self.data.state = AlterTablesState::InvalidateTableCache;
@@ -187,9 +219,16 @@ impl AlterLogicalTablesProcedure {
     pub(crate) async fn on_invalidate_table_cache(&mut self) -> Result<Status> {
         let to_invalidate = &self.data.table_cache_keys_to_invalidate;
 
+        let ctx = CacheContext {
+            subject: Some(format!(
+                "Invalidate table cache by altering logical tables, physical_table_id: {}",
+                self.data.physical_table_id,
+            )),
+        };
+
         self.context
             .cache_invalidator
-            .invalidate(&Default::default(), to_invalidate)
+            .invalidate(&ctx, to_invalidate)
             .await?;
         Ok(Status::done())
     }
@@ -209,6 +248,10 @@ impl Procedure for AlterLogicalTablesProcedure {
         let _timer = metrics::METRIC_META_PROCEDURE_ALTER_TABLE
             .with_label_values(&[step])
             .start_timer();
+        debug!(
+            "Executing alter logical tables procedure, state: {:?}",
+            state
+        );
 
         match state {
             AlterTablesState::Prepare => self.on_prepare().await,

src/common/meta/src/ddl/alter_logical_tables/check.rs

@@ -1,136 +0,0 @@
-// Copyright 2023 Greptime Team
-//
-// 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::HashSet;
-
-use api::v1::alter_table_expr::Kind;
-use snafu::{ensure, OptionExt};
-
-use crate::ddl::alter_logical_tables::AlterLogicalTablesProcedure;
-use crate::error::{AlterLogicalTablesInvalidArgumentsSnafu, Result};
-use crate::key::table_info::TableInfoValue;
-use crate::key::table_route::TableRouteValue;
-use crate::rpc::ddl::AlterTableTask;
-
-impl AlterLogicalTablesProcedure {
-    pub(crate) fn check_input_tasks(&self) -> Result<()> {
-        self.check_schema()?;
-        self.check_alter_kind()?;
-        Ok(())
-    }
-
-    pub(crate) async fn check_physical_table(&self) -> Result<()> {
-        let table_route_manager = self.context.table_metadata_manager.table_route_manager();
-        let table_ids = self
-            .data
-            .table_info_values
-            .iter()
-            .map(|v| v.table_info.ident.table_id)
-            .collect::<Vec<_>>();
-        let table_routes = table_route_manager
-            .table_route_storage()
-            .batch_get(&table_ids)
-            .await?;
-        let physical_table_id = self.data.physical_table_id;
-        let is_same_physical_table = table_routes.iter().all(|r| {
-            if let Some(TableRouteValue::Logical(r)) = r {
-                r.physical_table_id() == physical_table_id
-            } else {
-                false
-            }
-        });
-
-        ensure!(
-            is_same_physical_table,
-            AlterLogicalTablesInvalidArgumentsSnafu {
-                err_msg: "All the tasks should have the same physical table id"
-            }
-        );
-
-        Ok(())
-    }
-
-    pub(crate) fn check_finished_tasks(&self) -> Result<Vec<bool>> {
-        let task = &self.data.tasks;
-        let table_info_values = &self.data.table_info_values;
-
-        Ok(task
-            .iter()
-            .zip(table_info_values.iter())
-            .map(|(task, table)| Self::check_finished_task(task, table))
-            .collect())
-    }
-
-    // Checks if the schemas of the tasks are the same
-    fn check_schema(&self) -> Result<()> {
-        let is_same_schema = self.data.tasks.windows(2).all(|pair| {
-            pair[0].alter_table.catalog_name == pair[1].alter_table.catalog_name
-                && pair[0].alter_table.schema_name == pair[1].alter_table.schema_name
-        });
-
-        ensure!(
-            is_same_schema,
-            AlterLogicalTablesInvalidArgumentsSnafu {
-                err_msg: "Schemas of the tasks are not the same"
-            }
-        );
-
-        Ok(())
-    }
-
-    fn check_alter_kind(&self) -> Result<()> {
-        for task in &self.data.tasks {
-            let kind = task.alter_table.kind.as_ref().context(
-                AlterLogicalTablesInvalidArgumentsSnafu {
-                    err_msg: "Alter kind is missing",
-                },
-            )?;
-            let Kind::AddColumns(_) = kind else {
-                return AlterLogicalTablesInvalidArgumentsSnafu {
-                    err_msg: "Only support add columns operation",
-                }
-                .fail();
-            };
-        }
-
-        Ok(())
-    }
-
-    fn check_finished_task(task: &AlterTableTask, table: &TableInfoValue) -> bool {
-        let columns = table
-            .table_info
-            .meta
-            .schema
-            .column_schemas
-            .iter()
-            .map(|c| &c.name)
-            .collect::<HashSet<_>>();
-
-        let Some(kind) = task.alter_table.kind.as_ref() else {
-            return true; // Never get here since we have checked it in `check_alter_kind`
-        };
-        let Kind::AddColumns(add_columns) = kind else {
-            return true; // Never get here since we have checked it in `check_alter_kind`
-        };
-
-        // We only check that all columns have been finished. That is to say,
-        // if one part is finished but another part is not, it will be considered
-        // unfinished.
-        add_columns
-            .add_columns
-            .iter()
-            .map(|add_column| add_column.column_def.as_ref().map(|c| &c.name))
-            .all(|column| column.map(|c| columns.contains(c)).unwrap_or(false))
-    }
-}

src/common/meta/src/ddl/alter_logical_tables/executor.rs

@@ -0,0 +1,216 @@
+// Copyright 2023 Greptime Team
+//
+// 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 api::region::RegionResponse;
+use api::v1::alter_table_expr::Kind;
+use api::v1::region::{
+    alter_request, region_request, AddColumn, AddColumns, AlterRequest, AlterRequests,
+    RegionColumnDef, RegionRequest, RegionRequestHeader,
+};
+use api::v1::{self, AlterTableExpr};
+use common_telemetry::tracing_context::TracingContext;
+use common_telemetry::{debug, warn};
+use futures::future;
+use store_api::metadata::ColumnMetadata;
+use store_api::storage::{RegionId, RegionNumber, TableId};
+
+use crate::ddl::utils::{add_peer_context_if_needed, raw_table_info};
+use crate::error::Result;
+use crate::instruction::CacheIdent;
+use crate::key::table_info::TableInfoValue;
+use crate::key::{DeserializedValueWithBytes, RegionDistribution, TableMetadataManagerRef};
+use crate::node_manager::NodeManagerRef;
+use crate::rpc::router::{find_leaders, region_distribution, RegionRoute};
+
+/// [AlterLogicalTablesExecutor] performs:
+/// - Alters logical regions on the datanodes.
+/// - Updates table metadata for alter table operation.
+pub struct AlterLogicalTablesExecutor<'a> {
+    /// The alter table expressions.
+    ///
+    /// The first element is the logical table id, the second element is the alter table expression.
+    alters: Vec<(TableId, &'a AlterTableExpr)>,
+}
+
+impl<'a> AlterLogicalTablesExecutor<'a> {
+    pub fn new(alters: Vec<(TableId, &'a AlterTableExpr)>) -> Self {
+        Self { alters }
+    }
+
+    /// Alters logical regions on the datanodes.
+    pub(crate) async fn on_alter_regions(
+        &self,
+        node_manager: &NodeManagerRef,
+        region_routes: &[RegionRoute],
+    ) -> Result<Vec<RegionResponse>> {
+        let region_distribution = region_distribution(region_routes);
+        let leaders = find_leaders(region_routes)
+            .into_iter()
+            .map(|p| (p.id, p))
+            .collect::<HashMap<_, _>>();
+        let mut alter_region_tasks = Vec::with_capacity(leaders.len());
+        for (datanode_id, region_role_set) in region_distribution {
+            if region_role_set.leader_regions.is_empty() {
+                continue;
+            }
+            // Safety: must exists.
+            let peer = leaders.get(&datanode_id).unwrap();
+            let requester = node_manager.datanode(peer).await;
+            let requests = self.make_alter_region_request(&region_role_set.leader_regions);
+            let requester = requester.clone();
+            let peer = peer.clone();
+
+            debug!("Sending alter region requests to datanode {}", peer);
+            alter_region_tasks.push(async move {
+                requester
+                    .handle(make_request(requests))
+                    .await
+                    .map_err(add_peer_context_if_needed(peer))
+            });
+        }
+
+        future::join_all(alter_region_tasks)
+            .await
+            .into_iter()
+            .collect::<Result<Vec<_>>>()
+    }
+
+    fn make_alter_region_request(&self, region_numbers: &[RegionNumber]) -> AlterRequests {
+        let mut requests = Vec::with_capacity(region_numbers.len() * self.alters.len());
+        for (table_id, alter) in self.alters.iter() {
+            for region_number in region_numbers {
+                let region_id = RegionId::new(*table_id, *region_number);
+                let request = make_alter_region_request(region_id, alter);
+                requests.push(request);
+            }
+        }
+
+        AlterRequests { requests }
+    }
+
+    /// Updates table metadata for alter table operation.
+    ///
+    /// ## Panic:
+    /// - If the region distribution is not set when updating table metadata.
+    pub(crate) async fn on_alter_metadata(
+        physical_table_id: TableId,
+        table_metadata_manager: &TableMetadataManagerRef,
+        current_table_info_value: &DeserializedValueWithBytes<TableInfoValue>,
+        region_distribution: RegionDistribution,
+        physical_columns: &[ColumnMetadata],
+    ) -> Result<()> {
+        if physical_columns.is_empty() {
+            warn!("No physical columns found, leaving the physical table's schema unchanged when altering logical tables");
+            return Ok(());
+        }
+
+        let table_ref = current_table_info_value.table_ref();
+        let table_id = physical_table_id;
+
+        // Generates new table info
+        let old_raw_table_info = current_table_info_value.table_info.clone();
+        let new_raw_table_info =
+            raw_table_info::build_new_physical_table_info(old_raw_table_info, physical_columns);
+
+        debug!(
+            "Starting update table: {} metadata, table_id: {}, new table info: {:?}",
+            table_ref, table_id, new_raw_table_info
+        );
+
+        table_metadata_manager
+            .update_table_info(
+                current_table_info_value,
+                Some(region_distribution),
+                new_raw_table_info,
+            )
+            .await?;
+
+        Ok(())
+    }
+
+    /// Builds the cache ident keys for the alter logical tables.
+    ///
+    /// The cache ident keys are:
+    /// - The table id of the logical tables.
+    /// - The table name of the logical tables.
+    /// - The table id of the physical table.
+    pub(crate) fn build_cache_ident_keys(
+        physical_table_info: &TableInfoValue,
+        logical_table_info_values: &[&TableInfoValue],
+    ) -> Vec<CacheIdent> {
+        let mut cache_keys = Vec::with_capacity(logical_table_info_values.len() * 2 + 2);
+        cache_keys.extend(logical_table_info_values.iter().flat_map(|table| {
+            vec![
+                CacheIdent::TableId(table.table_info.ident.table_id),
+                CacheIdent::TableName(table.table_name()),
+            ]
+        }));
+        cache_keys.push(CacheIdent::TableId(
+            physical_table_info.table_info.ident.table_id,
+        ));
+        cache_keys.push(CacheIdent::TableName(physical_table_info.table_name()));
+
+        cache_keys
+    }
+}
+
+fn make_request(alter_requests: AlterRequests) -> RegionRequest {
+    RegionRequest {
+        header: Some(RegionRequestHeader {
+            tracing_context: TracingContext::from_current_span().to_w3c(),
+            ..Default::default()
+        }),
+        body: Some(region_request::Body::Alters(alter_requests)),
+    }
+}
+
+/// Makes an alter region request.
+pub fn make_alter_region_request(
+    region_id: RegionId,
+    alter_table_expr: &AlterTableExpr,
+) -> AlterRequest {
+    let region_id = region_id.as_u64();
+    let kind = match &alter_table_expr.kind {
+        Some(Kind::AddColumns(add_columns)) => Some(alter_request::Kind::AddColumns(
+            to_region_add_columns(add_columns),
+        )),
+        _ => unreachable!(), // Safety: we have checked the kind in check_input_tasks
+    };
+
+    AlterRequest {
+        region_id,
+        schema_version: 0,
+        kind,
+    }
+}
+
+fn to_region_add_columns(add_columns: &v1::AddColumns) -> AddColumns {
+    let add_columns = add_columns
+        .add_columns
+        .iter()
+        .map(|add_column| {
+            let region_column_def = RegionColumnDef {
+                column_def: add_column.column_def.clone(),
+                ..Default::default() // other fields are not used in alter logical table
+            };
+            AddColumn {
+                column_def: Some(region_column_def),
+                ..Default::default() // other fields are not used in alter logical table
+            }
+        })
+        .collect();
+    AddColumns { add_columns }
+}

src/common/meta/src/ddl/alter_logical_tables/metadata.rs

@@ -1,158 +0,0 @@
-// Copyright 2023 Greptime Team
-//
-// 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 common_catalog::format_full_table_name;
-use snafu::OptionExt;
-use table::metadata::TableId;
-
-use crate::ddl::alter_logical_tables::AlterLogicalTablesProcedure;
-use crate::error::{
-    AlterLogicalTablesInvalidArgumentsSnafu, Result, TableInfoNotFoundSnafu, TableNotFoundSnafu,
-    TableRouteNotFoundSnafu,
-};
-use crate::key::table_info::TableInfoValue;
-use crate::key::table_name::TableNameKey;
-use crate::key::table_route::TableRouteValue;
-use crate::key::DeserializedValueWithBytes;
-use crate::rpc::ddl::AlterTableTask;
-
-impl AlterLogicalTablesProcedure {
-    pub(crate) fn filter_task(&mut self, finished_tasks: &[bool]) -> Result<()> {
-        debug_assert_eq!(finished_tasks.len(), self.data.tasks.len());
-        debug_assert_eq!(finished_tasks.len(), self.data.table_info_values.len());
-        self.data.tasks = self
-            .data
-            .tasks
-            .drain(..)
-            .zip(finished_tasks.iter())
-            .filter_map(|(task, finished)| if *finished { None } else { Some(task) })
-            .collect();
-        self.data.table_info_values = self
-            .data
-            .table_info_values
-            .drain(..)
-            .zip(finished_tasks.iter())
-            .filter_map(|(table_info_value, finished)| {
-                if *finished {
-                    None
-                } else {
-                    Some(table_info_value)
-                }
-            })
-            .collect();
-
-        Ok(())
-    }
-
-    pub(crate) async fn fill_physical_table_info(&mut self) -> Result<()> {
-        let (physical_table_info, physical_table_route) = self
-            .context
-            .table_metadata_manager
-            .get_full_table_info(self.data.physical_table_id)
-            .await?;
-
-        let physical_table_info = physical_table_info.with_context(|| TableInfoNotFoundSnafu {
-            table: format!("table id - {}", self.data.physical_table_id),
-        })?;
-        let physical_table_route = physical_table_route
-            .context(TableRouteNotFoundSnafu {
-                table_id: self.data.physical_table_id,
-            })?
-            .into_inner();
-
-        self.data.physical_table_info = Some(physical_table_info);
-        let TableRouteValue::Physical(physical_table_route) = physical_table_route else {
-            return AlterLogicalTablesInvalidArgumentsSnafu {
-                err_msg: format!(
-                    "expected a physical table but got a logical table: {:?}",
-                    self.data.physical_table_id
-                ),
-            }
-            .fail();
-        };
-        self.data.physical_table_route = Some(physical_table_route);
-
-        Ok(())
-    }
-
-    pub(crate) async fn fill_table_info_values(&mut self) -> Result<()> {
-        let table_ids = self.get_all_table_ids().await?;
-        let table_info_values = self.get_all_table_info_values(&table_ids).await?;
-        debug_assert_eq!(table_info_values.len(), self.data.tasks.len());
-        self.data.table_info_values = table_info_values;
-
-        Ok(())
-    }
-
-    async fn get_all_table_info_values(
-        &self,
-        table_ids: &[TableId],
-    ) -> Result<Vec<DeserializedValueWithBytes<TableInfoValue>>> {
-        let table_info_manager = self.context.table_metadata_manager.table_info_manager();
-        let mut table_info_map = table_info_manager.batch_get_raw(table_ids).await?;
-        let mut table_info_values = Vec::with_capacity(table_ids.len());
-        for (table_id, task) in table_ids.iter().zip(self.data.tasks.iter()) {
-            let table_info_value =
-                table_info_map
-                    .remove(table_id)
-                    .with_context(|| TableInfoNotFoundSnafu {
-                        table: extract_table_name(task),
-                    })?;
-            table_info_values.push(table_info_value);
-        }
-
-        Ok(table_info_values)
-    }
-
-    async fn get_all_table_ids(&self) -> Result<Vec<TableId>> {
-        let table_name_manager = self.context.table_metadata_manager.table_name_manager();
-        let table_name_keys = self
-            .data
-            .tasks
-            .iter()
-            .map(|task| extract_table_name_key(task))
-            .collect();
-
-        let table_name_values = table_name_manager.batch_get(table_name_keys).await?;
-        let mut table_ids = Vec::with_capacity(table_name_values.len());
-        for (value, task) in table_name_values.into_iter().zip(self.data.tasks.iter()) {
-            let table_id = value
-                .with_context(|| TableNotFoundSnafu {
-                    table_name: extract_table_name(task),
-                })?
-                .table_id();
-            table_ids.push(table_id);
-        }
-
-        Ok(table_ids)
-    }
-}
-
-#[inline]
-fn extract_table_name(task: &AlterTableTask) -> String {
-    format_full_table_name(
-        &task.alter_table.catalog_name,
-        &task.alter_table.schema_name,
-        &task.alter_table.table_name,
-    )
-}
-
-#[inline]
-fn extract_table_name_key(task: &AlterTableTask) -> TableNameKey {
-    TableNameKey::new(
-        &task.alter_table.catalog_name,
-        &task.alter_table.schema_name,
-        &task.alter_table.table_name,
-    )
-}

src/common/meta/src/ddl/alter_logical_tables/region_request.rs

@@ -1,113 +0,0 @@
-// Copyright 2023 Greptime Team
-//
-// 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 api::v1::alter_table_expr::Kind;
-use api::v1::region::{
-    alter_request, region_request, AddColumn, AddColumns, AlterRequest, AlterRequests,
-    RegionColumnDef, RegionRequest, RegionRequestHeader,
-};
-use api::v1::{self, AlterTableExpr};
-use common_telemetry::tracing_context::TracingContext;
-use store_api::storage::RegionId;
-
-use crate::ddl::alter_logical_tables::AlterLogicalTablesProcedure;
-use crate::error::Result;
-use crate::peer::Peer;
-use crate::rpc::router::{find_leader_regions, RegionRoute};
-
-impl AlterLogicalTablesProcedure {
-    pub(crate) fn make_request(
-        &self,
-        peer: &Peer,
-        region_routes: &[RegionRoute],
-    ) -> Result<RegionRequest> {
-        let alter_requests = self.make_alter_region_requests(peer, region_routes)?;
-        let request = RegionRequest {
-            header: Some(RegionRequestHeader {
-                tracing_context: TracingContext::from_current_span().to_w3c(),
-                ..Default::default()
-            }),
-            body: Some(region_request::Body::Alters(alter_requests)),
-        };
-
-        Ok(request)
-    }
-
-    fn make_alter_region_requests(
-        &self,
-        peer: &Peer,
-        region_routes: &[RegionRoute],
-    ) -> Result<AlterRequests> {
-        let tasks = &self.data.tasks;
-        let regions_on_this_peer = find_leader_regions(region_routes, peer);
-        let mut requests = Vec::with_capacity(tasks.len() * regions_on_this_peer.len());
-        for (task, table) in self
-            .data
-            .tasks
-            .iter()
-            .zip(self.data.table_info_values.iter())
-        {
-            for region_number in &regions_on_this_peer {
-                let region_id = RegionId::new(table.table_info.ident.table_id, *region_number);
-                let request = make_alter_region_request(
-                    region_id,
-                    &task.alter_table,
-                    table.table_info.ident.version,
-                );
-                requests.push(request);
-            }
-        }
-
-        Ok(AlterRequests { requests })
-    }
-}
-
-/// Makes an alter region request.
-pub fn make_alter_region_request(
-    region_id: RegionId,
-    alter_table_expr: &AlterTableExpr,
-    schema_version: u64,
-) -> AlterRequest {
-    let region_id = region_id.as_u64();
-    let kind = match &alter_table_expr.kind {
-        Some(Kind::AddColumns(add_columns)) => Some(alter_request::Kind::AddColumns(
-            to_region_add_columns(add_columns),
-        )),
-        _ => unreachable!(), // Safety: we have checked the kind in check_input_tasks
-    };
-
-    AlterRequest {
-        region_id,
-        schema_version,
-        kind,
-    }
-}
-
-fn to_region_add_columns(add_columns: &v1::AddColumns) -> AddColumns {
-    let add_columns = add_columns
-        .add_columns
-        .iter()
-        .map(|add_column| {
-            let region_column_def = RegionColumnDef {
-                column_def: add_column.column_def.clone(),
-                ..Default::default() // other fields are not used in alter logical table
-            };
-            AddColumn {
-                column_def: Some(region_column_def),
-                ..Default::default() // other fields are not used in alter logical table
-            }
-        })
-        .collect();
-    AddColumns { add_columns }
-}

src/common/meta/src/ddl/alter_logical_tables/table_cache_keys.rs

@@ -1,50 +0,0 @@
-// Copyright 2023 Greptime Team
-//
-// 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 table::metadata::RawTableInfo;
-use table::table_name::TableName;
-
-use crate::ddl::alter_logical_tables::AlterTablesData;
-use crate::instruction::CacheIdent;
-
-impl AlterTablesData {
-    pub(crate) fn build_cache_keys_to_invalidate(&mut self) {
-        let mut cache_keys = self
-            .table_info_values
-            .iter()
-            .flat_map(|table| {
-                vec![
-                    CacheIdent::TableId(table.table_info.ident.table_id),
-                    CacheIdent::TableName(extract_table_name(&table.table_info)),
-                ]
-            })
-            .collect::<Vec<_>>();
-        cache_keys.push(CacheIdent::TableId(self.physical_table_id));
-        // Safety: physical_table_info already filled in previous steps
-        let physical_table_info = &self.physical_table_info.as_ref().unwrap().table_info;
-        cache_keys.push(CacheIdent::TableName(extract_table_name(
-            physical_table_info,
-        )));
-
-        self.table_cache_keys_to_invalidate = cache_keys;
-    }
-}
-
-fn extract_table_name(table_info: &RawTableInfo) -> TableName {
-    TableName::new(
-        &table_info.catalog_name,
-        &table_info.schema_name,
-        &table_info.name,
-    )
-}

src/common/meta/src/ddl/alter_logical_tables/update_metadata.rs

@@ -13,41 +13,35 @@
 // limitations under the License.
 
 use common_grpc_expr::alter_expr_to_request;
-use common_telemetry::warn;
 use itertools::Itertools;
 use snafu::ResultExt;
 use table::metadata::{RawTableInfo, TableInfo};
 
+use crate::ddl::alter_logical_tables::executor::AlterLogicalTablesExecutor;
 use crate::ddl::alter_logical_tables::AlterLogicalTablesProcedure;
-use crate::ddl::physical_table_metadata;
 use crate::error;
 use crate::error::{ConvertAlterTableRequestSnafu, Result};
 use crate::key::table_info::TableInfoValue;
 use crate::key::DeserializedValueWithBytes;
 use crate::rpc::ddl::AlterTableTask;
+use crate::rpc::router::region_distribution;
 
 impl AlterLogicalTablesProcedure {
     pub(crate) async fn update_physical_table_metadata(&mut self) -> Result<()> {
-        if self.data.physical_columns.is_empty() {
-            warn!("No physical columns found, leaving the physical table's schema unchanged when altering logical tables");
-            return Ok(());
-        }
-
         // Safety: must exist.
         let physical_table_info = self.data.physical_table_info.as_ref().unwrap();
+        let physical_table_route = self.data.physical_table_route.as_ref().unwrap();
+        let region_distribution = region_distribution(&physical_table_route.region_routes);
 
-        // Generates new table info
-        let old_raw_table_info = physical_table_info.table_info.clone();
-        let new_raw_table_info = physical_table_metadata::build_new_physical_table_info(
-            old_raw_table_info,
+        // Updates physical table's metadata.
+        AlterLogicalTablesExecutor::on_alter_metadata(
+            self.data.physical_table_id,
+            &self.context.table_metadata_manager,
+            physical_table_info,
+            region_distribution,
             &self.data.physical_columns,
-        );
-
-        // Updates physical table's metadata, and we don't need to touch per-region settings.
-        self.context
-            .table_metadata_manager
-            .update_table_info(physical_table_info, None, new_raw_table_info)
-            .await?;
+        )
+        .await?;
 
         Ok(())
     }

src/common/meta/src/ddl/alter_logical_tables/validator.rs

@@ -0,0 +1,284 @@
+// Copyright 2023 Greptime Team
+//
+// 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::HashSet;
+
+use api::v1::alter_table_expr::Kind;
+use api::v1::AlterTableExpr;
+use snafu::{ensure, OptionExt};
+use store_api::storage::TableId;
+use table::table_reference::TableReference;
+
+use crate::ddl::utils::table_id::get_all_table_ids_by_names;
+use crate::ddl::utils::table_info::get_all_table_info_values_by_table_ids;
+use crate::error::{
+    AlterLogicalTablesInvalidArgumentsSnafu, Result, TableInfoNotFoundSnafu,
+    TableRouteNotFoundSnafu,
+};
+use crate::key::table_info::TableInfoValue;
+use crate::key::table_route::{PhysicalTableRouteValue, TableRouteManager, TableRouteValue};
+use crate::key::{DeserializedValueWithBytes, TableMetadataManagerRef};
+
+/// [AlterLogicalTableValidator] validates the alter logical expressions.
+pub struct AlterLogicalTableValidator<'a> {
+    physical_table_id: TableId,
+    alters: Vec<&'a AlterTableExpr>,
+}
+
+impl<'a> AlterLogicalTableValidator<'a> {
+    pub fn new(physical_table_id: TableId, alters: Vec<&'a AlterTableExpr>) -> Self {
+        Self {
+            physical_table_id,
+            alters,
+        }
+    }
+
+    /// Validates all alter table expressions have the same schema and catalog.
+    fn validate_schema(&self) -> Result<()> {
+        let is_same_schema = self.alters.windows(2).all(|pair| {
+            pair[0].catalog_name == pair[1].catalog_name
+                && pair[0].schema_name == pair[1].schema_name
+        });
+
+        ensure!(
+            is_same_schema,
+            AlterLogicalTablesInvalidArgumentsSnafu {
+                err_msg: "Schemas of the alter table expressions are not the same"
+            }
+        );
+
+        Ok(())
+    }
+
+    /// Validates that all alter table expressions are of the supported kind.
+    /// Currently only supports `AddColumns` operations.
+    fn validate_alter_kind(&self) -> Result<()> {
+        for alter in &self.alters {
+            let kind = alter
+                .kind
+                .as_ref()
+                .context(AlterLogicalTablesInvalidArgumentsSnafu {
+                    err_msg: "Alter kind is missing",
+                })?;
+
+            let Kind::AddColumns(_) = kind else {
+                return AlterLogicalTablesInvalidArgumentsSnafu {
+                    err_msg: "Only support add columns operation",
+                }
+                .fail();
+            };
+        }
+
+        Ok(())
+    }
+
+    fn table_names(&self) -> Vec<TableReference> {
+        self.alters
+            .iter()
+            .map(|alter| {
+                TableReference::full(&alter.catalog_name, &alter.schema_name, &alter.table_name)
+            })
+            .collect()
+    }
+
+    /// Validates that the physical table info and route exist.
+    ///
+    /// This method performs the following validations:
+    /// 1. Retrieves the full table info and route for the given physical table id
+    /// 2. Ensures the table info and table route exists
+    /// 3. Verifies that the table route is actually a physical table route, not a logical one
+    ///
+    /// Returns a tuple containing the validated table info and physical table route.
+    async fn validate_physical_table(
+        &self,
+        table_metadata_manager: &TableMetadataManagerRef,
+    ) -> Result<(
+        DeserializedValueWithBytes<TableInfoValue>,
+        PhysicalTableRouteValue,
+    )> {
+        let (table_info, table_route) = table_metadata_manager
+            .get_full_table_info(self.physical_table_id)
+            .await?;
+
+        let table_info = table_info.with_context(|| TableInfoNotFoundSnafu {
+            table: format!("table id - {}", self.physical_table_id),
+        })?;
+
+        let physical_table_route = table_route
+            .context(TableRouteNotFoundSnafu {
+                table_id: self.physical_table_id,
+            })?
+            .into_inner();
+
+        let TableRouteValue::Physical(table_route) = physical_table_route else {
+            return AlterLogicalTablesInvalidArgumentsSnafu {
+                err_msg: format!(
+                    "expected a physical table but got a logical table: {:?}",
+                    self.physical_table_id
+                ),
+            }
+            .fail();
+        };
+
+        Ok((table_info, table_route))
+    }
+
+    /// Validates that all logical table routes have the same physical table id.
+    ///
+    /// This method performs the following validations:
+    /// 1. Retrieves table routes for all the given table ids.
+    /// 2. Ensures that all retrieved routes are logical table routes (not physical)
+    /// 3. Verifies that all logical table routes reference the same physical table id.
+    /// 4. Returns an error if any route is not logical or references a different physical table.
+    async fn validate_logical_table_routes(
+        &self,
+        table_route_manager: &TableRouteManager,
+        table_ids: &[TableId],
+    ) -> Result<()> {
+        let table_routes = table_route_manager
+            .table_route_storage()
+            .batch_get(table_ids)
+            .await?;
+
+        let physical_table_id = self.physical_table_id;
+
+        let is_same_physical_table = table_routes.iter().all(|r| {
+            if let Some(TableRouteValue::Logical(r)) = r {
+                r.physical_table_id() == physical_table_id
+            } else {
+                false
+            }
+        });
+
+        ensure!(
+            is_same_physical_table,
+            AlterLogicalTablesInvalidArgumentsSnafu {
+                err_msg: "All the tasks should have the same physical table id"
+            }
+        );
+
+        Ok(())
+    }
+
+    /// Validates the alter logical expressions.
+    ///
+    /// This method performs the following validations:
+    /// 1. Validates that all alter table expressions have the same schema and catalog.
+    /// 2. Validates that all alter table expressions are of the supported kind.
+    /// 3. Validates that the physical table info and route exist.
+    /// 4. Validates that all logical table routes have the same physical table id.
+    ///
+    /// Returns a [ValidatorResult] containing the validation results.
+    pub async fn validate(
+        &self,
+        table_metadata_manager: &TableMetadataManagerRef,
+    ) -> Result<ValidatorResult> {
+        self.validate_schema()?;
+        self.validate_alter_kind()?;
+        let (physical_table_info, physical_table_route) =
+            self.validate_physical_table(table_metadata_manager).await?;
+        let table_names = self.table_names();
+        let table_ids =
+            get_all_table_ids_by_names(table_metadata_manager.table_name_manager(), &table_names)
+                .await?;
+        let mut table_info_values = get_all_table_info_values_by_table_ids(
+            table_metadata_manager.table_info_manager(),
+            &table_ids,
+            &table_names,
+        )
+        .await?;
+        self.validate_logical_table_routes(
+            table_metadata_manager.table_route_manager(),
+            &table_ids,
+        )
+        .await?;
+        let skip_alter = self
+            .alters
+            .iter()
+            .zip(table_info_values.iter())
+            .map(|(task, table)| skip_alter_logical_region(task, table))
+            .collect::<Vec<_>>();
+        retain_unskipped(&mut table_info_values, &skip_alter);
+        let num_skipped = skip_alter.iter().filter(|&&x| x).count();
+
+        Ok(ValidatorResult {
+            num_skipped,
+            skip_alter,
+            table_info_values,
+            physical_table_info,
+            physical_table_route,
+        })
+    }
+}
+
+/// The result of the validator.
+pub(crate) struct ValidatorResult {
+    pub(crate) num_skipped: usize,
+    pub(crate) skip_alter: Vec<bool>,
+    pub(crate) table_info_values: Vec<DeserializedValueWithBytes<TableInfoValue>>,
+    pub(crate) physical_table_info: DeserializedValueWithBytes<TableInfoValue>,
+    pub(crate) physical_table_route: PhysicalTableRouteValue,
+}
+
+/// Retains the elements that are not skipped.
+pub(crate) fn retain_unskipped<T>(target: &mut Vec<T>, skipped: &[bool]) {
+    debug_assert_eq!(target.len(), skipped.len());
+    let mut iter = skipped.iter();
+    target.retain(|_| !iter.next().unwrap());
+}
+
+/// Returns true if does not required to alter the logical region.
+fn skip_alter_logical_region(alter: &AlterTableExpr, table: &TableInfoValue) -> bool {
+    let existing_columns = table
+        .table_info
+        .meta
+        .schema
+        .column_schemas
+        .iter()
+        .map(|c| &c.name)
+        .collect::<HashSet<_>>();
+
+    let Some(kind) = alter.kind.as_ref() else {
+        return true; // Never get here since we have checked it in `validate_alter_kind`
+    };
+    let Kind::AddColumns(add_columns) = kind else {
+        return true; // Never get here since we have checked it in `validate_alter_kind`
+    };
+
+    // We only check that all columns have been finished. That is to say,
+    // if one part is finished but another part is not, it will be considered
+    // unfinished.
+    add_columns
+        .add_columns
+        .iter()
+        .map(|add_column| add_column.column_def.as_ref().map(|c| &c.name))
+        .all(|column| {
+            column
+                .map(|c| existing_columns.contains(c))
+                .unwrap_or(false)
+        })
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn test_retain_unskipped() {
+        let mut target = vec![1, 2, 3, 4, 5];
+        let skipped = vec![false, true, false, true, false];
+        retain_unskipped(&mut target, &skipped);
+        assert_eq!(target, vec![1, 3, 5]);
+    }
+}

src/common/meta/src/ddl/alter_table.rs

@@ -12,10 +12,9 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
-mod check;
+mod executor;
 mod metadata;
 mod region_request;
-mod update_metadata;
 
 use std::vec;
 
@@ -29,33 +28,29 @@ use common_procedure::{
     Context as ProcedureContext, ContextProvider, Error as ProcedureError, LockKey, PoisonKey,
     PoisonKeys, Procedure, ProcedureId, Status, StringKey,
 };
-use common_telemetry::{debug, error, info, warn};
-use futures::future::{self};
+use common_telemetry::{error, info, warn};
 use serde::{Deserialize, Serialize};
 use snafu::{ensure, ResultExt};
 use store_api::metadata::ColumnMetadata;
 use store_api::metric_engine_consts::TABLE_COLUMN_METADATA_EXTENSION_KEY;
-use store_api::storage::RegionId;
 use strum::AsRefStr;
 use table::metadata::{RawTableInfo, TableId, TableInfo};
 use table::table_reference::TableReference;
 
-use crate::cache_invalidator::Context;
-use crate::ddl::physical_table_metadata::update_table_info_column_ids;
+use crate::ddl::alter_table::executor::AlterTableExecutor;
 use crate::ddl::utils::{
-    add_peer_context_if_needed, extract_column_metadatas, handle_multiple_results,
-    map_to_procedure_error, sync_follower_regions, MultipleResults,
+    extract_column_metadatas, handle_multiple_results, map_to_procedure_error,
+    sync_follower_regions, MultipleResults,
 };
 use crate::ddl::DdlContext;
 use crate::error::{AbortProcedureSnafu, NoLeaderSnafu, PutPoisonSnafu, Result, RetryLaterSnafu};
-use crate::instruction::CacheIdent;
 use crate::key::table_info::TableInfoValue;
 use crate::key::{DeserializedValueWithBytes, RegionDistribution};
 use crate::lock_key::{CatalogLock, SchemaLock, TableLock, TableNameLock};
 use crate::metrics;
 use crate::poison_key::table_poison_key;
 use crate::rpc::ddl::AlterTableTask;
-use crate::rpc::router::{find_leader_regions, find_leaders, region_distribution, RegionRoute};
+use crate::rpc::router::{find_leaders, region_distribution, RegionRoute};
 
 /// The alter table procedure
 pub struct AlterTableProcedure {
@@ -67,6 +62,24 @@ pub struct AlterTableProcedure {
     /// If we recover the procedure from json, then the table info value is not cached.
     /// But we already validated it in the prepare step.
     new_table_info: Option<TableInfo>,
+    /// The alter table executor.
+    executor: AlterTableExecutor,
+}
+
+/// Builds the executor from the [`AlterTableData`].
+///
+/// # Panics
+/// - If the alter kind is not set.
+fn build_executor_from_alter_expr(alter_data: &AlterTableData) -> AlterTableExecutor {
+    let table_name = alter_data.table_ref().into();
+    let table_id = alter_data.table_id;
+    let alter_kind = alter_data.task.alter_table.kind.as_ref().unwrap();
+    let new_table_name = if let Kind::RenameTable(RenameTable { new_table_name }) = alter_kind {
+        Some(new_table_name.to_string())
+    } else {
+        None
+    };
+    AlterTableExecutor::new(table_name, table_id, new_table_name)
 }
 
 impl AlterTableProcedure {
@@ -74,33 +87,42 @@ impl AlterTableProcedure {
 
     pub fn new(table_id: TableId, task: AlterTableTask, context: DdlContext) -> Result<Self> {
         task.validate()?;
+        let data = AlterTableData::new(task, table_id);
+        let executor = build_executor_from_alter_expr(&data);
         Ok(Self {
             context,
-            data: AlterTableData::new(task, table_id),
+            data,
             new_table_info: None,
+            executor,
         })
     }
 
     pub fn from_json(json: &str, context: DdlContext) -> ProcedureResult<Self> {
         let data: AlterTableData = serde_json::from_str(json).context(FromJsonSnafu)?;
+        let executor = build_executor_from_alter_expr(&data);
+
         Ok(AlterTableProcedure {
             context,
             data,
             new_table_info: None,
+            executor,
         })
     }
 
     // Checks whether the table exists.
     pub(crate) async fn on_prepare(&mut self) -> Result<Status> {
-        self.check_alter().await?;
+        self.executor
+            .on_prepare(&self.context.table_metadata_manager)
+            .await?;
         self.fill_table_info().await?;
 
-        // Validates the request and builds the new table info.
-        // We need to build the new table info here because we should ensure the alteration
-        // is valid in `UpdateMeta` state as we already altered the region.
-        // Safety: `fill_table_info()` already set it.
+        // Safety: filled in `fill_table_info`.
         let table_info_value = self.data.table_info_value.as_ref().unwrap();
-        self.new_table_info = Some(self.build_new_table_info(&table_info_value.table_info)?);
+        let new_table_info = AlterTableExecutor::validate_alter_table_expr(
+            &table_info_value.table_info,
+            self.data.task.alter_table.clone(),
+        )?;
+        self.new_table_info = Some(new_table_info);
 
         // Safety: Checked in `AlterTableProcedure::new`.
         let alter_kind = self.data.task.alter_table.kind.as_ref().unwrap();
@@ -143,9 +165,7 @@ impl AlterTableProcedure {
 
         self.data.region_distribution =
             Some(region_distribution(&physical_table_route.region_routes));
-
         let leaders = find_leaders(&physical_table_route.region_routes);
-        let mut alter_region_tasks = Vec::with_capacity(leaders.len());
         let alter_kind = self.make_region_alter_kind()?;
 
         info!(
@@ -158,31 +178,14 @@ impl AlterTableProcedure {
         ensure!(!leaders.is_empty(), NoLeaderSnafu { table_id });
         // Puts the poison before submitting alter region requests to datanodes.
         self.put_poison(ctx_provider, procedure_id).await?;
-        for datanode in leaders {
-            let requester = self.context.node_manager.datanode(&datanode).await;
-            let regions = find_leader_regions(&physical_table_route.region_routes, &datanode);
-
-            for region in regions {
-                let region_id = RegionId::new(table_id, region);
-                let request = self.make_alter_region_request(region_id, alter_kind.clone())?;
-                debug!("Submitting {request:?} to {datanode}");
-
-                let datanode = datanode.clone();
-                let requester = requester.clone();
-
-                alter_region_tasks.push(async move {
-                    requester
-                        .handle(request)
-                        .await
-                        .map_err(add_peer_context_if_needed(datanode))
-                });
-            }
-        }
-
-        let results = future::join_all(alter_region_tasks)
-            .await
-            .into_iter()
-            .collect::<Vec<_>>();
+        let results = self
+            .executor
+            .on_alter_regions(
+                &self.context.node_manager,
+                &physical_table_route.region_routes,
+                alter_kind,
+            )
+            .await;
 
         match handle_multiple_results(results) {
             MultipleResults::PartialRetryable(error) => {
@@ -224,7 +227,6 @@ impl AlterTableProcedure {
     }
 
     fn handle_alter_region_response(&mut self, mut results: Vec<RegionResponse>) -> Result<()> {
-        self.data.state = AlterTableState::UpdateMetadata;
         if let Some(column_metadatas) =
             extract_column_metadatas(&mut results, TABLE_COLUMN_METADATA_EXTENSION_KEY)?
         {
@@ -232,7 +234,7 @@ impl AlterTableProcedure {
         } else {
             warn!("altering table result doesn't contains extension key `{TABLE_COLUMN_METADATA_EXTENSION_KEY}`,leaving the table's column metadata unchanged");
         }
-
+        self.data.state = AlterTableState::UpdateMetadata;
         Ok(())
     }
 
@@ -260,43 +262,34 @@ impl AlterTableProcedure {
     pub(crate) async fn on_update_metadata(&mut self) -> Result<Status> {
         let table_id = self.data.table_id();
         let table_ref = self.data.table_ref();
-        // Safety: checked before.
+        // Safety: filled in `fill_table_info`.
         let table_info_value = self.data.table_info_value.as_ref().unwrap();
+        // Safety: Checked in `AlterTableProcedure::new`.
+        let alter_kind = self.data.task.alter_table.kind.as_ref().unwrap();
+
         // Gets the table info from the cache or builds it.
-        let new_info = match &self.new_table_info {
+        let  new_info = match &self.new_table_info {
             Some(cached) => cached.clone(),
-            None => self.build_new_table_info(&table_info_value.table_info)
+            None => AlterTableExecutor::validate_alter_table_expr(
+                &table_info_value.table_info,
+                self.data.task.alter_table.clone(),
+               )
                 .inspect_err(|e| {
                     // We already check the table info in the prepare step so this should not happen.
                     error!(e; "Unable to build info for table {} in update metadata step, table_id: {}", table_ref, table_id);
                 })?,
         };
 
-        debug!(
-            "Starting update table: {} metadata, new table info {:?}",
-            table_ref.to_string(),
-            new_info
-        );
-
-        // Safety: Checked in `AlterTableProcedure::new`.
-        let alter_kind = self.data.task.alter_table.kind.as_ref().unwrap();
-        if let Kind::RenameTable(RenameTable { new_table_name }) = alter_kind {
-            self.on_update_metadata_for_rename(new_table_name.to_string(), table_info_value)
-                .await?;
-        } else {
-            let mut raw_table_info = new_info.into();
-            if !self.data.column_metadatas.is_empty() {
-                update_table_info_column_ids(&mut raw_table_info, &self.data.column_metadatas);
-            }
-            // region distribution is set in submit_alter_region_requests
-            let region_distribution = self.data.region_distribution.as_ref().unwrap().clone();
-            self.on_update_metadata_for_alter(
-                raw_table_info,
-                region_distribution,
+        // Safety: region distribution is set in `submit_alter_region_requests`.
+        self.executor
+            .on_alter_metadata(
+                &self.context.table_metadata_manager,
                 table_info_value,
+                self.data.region_distribution.as_ref(),
+                new_info.into(),
+                &self.data.column_metadatas,
             )
             .await?;
-        }
 
         info!("Updated table metadata for table {table_ref}, table_id: {table_id}, kind: {alter_kind:?}");
         self.data.state = AlterTableState::InvalidateTableCache;
@@ -305,18 +298,9 @@ impl AlterTableProcedure {
 
     /// Broadcasts the invalidating table cache instructions.
     async fn on_broadcast(&mut self) -> Result<Status> {
-        let cache_invalidator = &self.context.cache_invalidator;
-
-        cache_invalidator
-            .invalidate(
-                &Context::default(),
-                &[
-                    CacheIdent::TableId(self.data.table_id()),
-                    CacheIdent::TableName(self.data.table_ref().into()),
-                ],
-            )
+        self.executor
+            .invalidate_table_cache(&self.context.cache_invalidator)
             .await?;
-
         Ok(Status::done())
     }
 

src/common/meta/src/ddl/alter_table/check.rs

@@ -1,62 +0,0 @@
-// Copyright 2023 Greptime Team
-//
-// 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 api::v1::alter_table_expr::Kind;
-use api::v1::RenameTable;
-use common_catalog::format_full_table_name;
-use snafu::ensure;
-
-use crate::ddl::alter_table::AlterTableProcedure;
-use crate::error::{self, Result};
-use crate::key::table_name::TableNameKey;
-
-impl AlterTableProcedure {
-    /// Checks:
-    /// - The new table name doesn't exist (rename).
-    /// - Table exists.
-    pub(crate) async fn check_alter(&self) -> Result<()> {
-        let alter_expr = &self.data.task.alter_table;
-        let catalog = &alter_expr.catalog_name;
-        let schema = &alter_expr.schema_name;
-        let table_name = &alter_expr.table_name;
-        // Safety: Checked in `AlterTableProcedure::new`.
-        let alter_kind = self.data.task.alter_table.kind.as_ref().unwrap();
-
-        let manager = &self.context.table_metadata_manager;
-        if let Kind::RenameTable(RenameTable { new_table_name }) = alter_kind {
-            let new_table_name_key = TableNameKey::new(catalog, schema, new_table_name);
-            let exists = manager
-                .table_name_manager()
-                .exists(new_table_name_key)
-                .await?;
-            ensure!(
-                !exists,
-                error::TableAlreadyExistsSnafu {
-                    table_name: format_full_table_name(catalog, schema, new_table_name),
-                }
-            )
-        }
-
-        let table_name_key = TableNameKey::new(catalog, schema, table_name);
-        let exists = manager.table_name_manager().exists(table_name_key).await?;
-        ensure!(
-            exists,
-            error::TableNotFoundSnafu {
-                table_name: format_full_table_name(catalog, schema, &alter_expr.table_name),
-            }
-        );
-
-        Ok(())
-    }
-}

src/common/meta/src/ddl/alter_table/executor.rs

@@ -0,0 +1,313 @@
+// Copyright 2023 Greptime Team
+//
+// 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 api::region::RegionResponse;
+use api::v1::region::region_request::Body;
+use api::v1::region::{alter_request, AlterRequest, RegionRequest, RegionRequestHeader};
+use api::v1::AlterTableExpr;
+use common_catalog::format_full_table_name;
+use common_grpc_expr::alter_expr_to_request;
+use common_telemetry::tracing_context::TracingContext;
+use common_telemetry::{debug, info};
+use futures::future;
+use snafu::{ensure, ResultExt};
+use store_api::metadata::ColumnMetadata;
+use store_api::storage::{RegionId, TableId};
+use table::metadata::{RawTableInfo, TableInfo};
+use table::requests::AlterKind;
+use table::table_name::TableName;
+
+use crate::cache_invalidator::{CacheInvalidatorRef, Context};
+use crate::ddl::utils::{add_peer_context_if_needed, raw_table_info};
+use crate::error::{self, Result, UnexpectedSnafu};
+use crate::instruction::CacheIdent;
+use crate::key::table_info::TableInfoValue;
+use crate::key::table_name::TableNameKey;
+use crate::key::{DeserializedValueWithBytes, RegionDistribution, TableMetadataManagerRef};
+use crate::node_manager::NodeManagerRef;
+use crate::rpc::router::{find_leaders, region_distribution, RegionRoute};
+
+/// [AlterTableExecutor] performs:
+/// - Alters the metadata of the table.
+/// - Alters regions on the datanode nodes.
+pub struct AlterTableExecutor {
+    table: TableName,
+    table_id: TableId,
+    /// The new table name if the alter kind is rename table.
+    new_table_name: Option<String>,
+}
+
+impl AlterTableExecutor {
+    /// Creates a new [`AlterTableExecutor`].
+    pub fn new(table: TableName, table_id: TableId, new_table_name: Option<String>) -> Self {
+        Self {
+            table,
+            table_id,
+            new_table_name,
+        }
+    }
+
+    /// Prepares to alter the table.
+    ///
+    /// ## Checks:
+    /// - The new table name doesn't exist (rename).
+    /// - Table exists.
+    pub(crate) async fn on_prepare(
+        &self,
+        table_metadata_manager: &TableMetadataManagerRef,
+    ) -> Result<()> {
+        let catalog = &self.table.catalog_name;
+        let schema = &self.table.schema_name;
+        let table_name = &self.table.table_name;
+
+        let manager = table_metadata_manager;
+        if let Some(new_table_name) = &self.new_table_name {
+            let new_table_name_key = TableNameKey::new(catalog, schema, new_table_name);
+            let exists = manager
+                .table_name_manager()
+                .exists(new_table_name_key)
+                .await?;
+            ensure!(
+                !exists,
+                error::TableAlreadyExistsSnafu {
+                    table_name: format_full_table_name(catalog, schema, new_table_name),
+                }
+            )
+        }
+
+        let table_name_key = TableNameKey::new(catalog, schema, table_name);
+        let exists = manager.table_name_manager().exists(table_name_key).await?;
+        ensure!(
+            exists,
+            error::TableNotFoundSnafu {
+                table_name: format_full_table_name(catalog, schema, table_name),
+            }
+        );
+
+        Ok(())
+    }
+
+    /// Validates the alter table expression and builds the new table info.
+    ///
+    /// This validation is performed early to ensure the alteration is valid before
+    /// proceeding to the `on_alter_metadata` state, where regions have already been altered.
+    /// Building the new table info here allows us to catch any issues with the
+    /// alteration before committing metadata changes.
+    pub(crate) fn validate_alter_table_expr(
+        table_info: &RawTableInfo,
+        alter_table_expr: AlterTableExpr,
+    ) -> Result<TableInfo> {
+        build_new_table_info(table_info, alter_table_expr)
+    }
+
+    /// Updates table metadata for alter table operation.
+    pub(crate) async fn on_alter_metadata(
+        &self,
+        table_metadata_manager: &TableMetadataManagerRef,
+        current_table_info_value: &DeserializedValueWithBytes<TableInfoValue>,
+        region_distribution: Option<&RegionDistribution>,
+        mut raw_table_info: RawTableInfo,
+        column_metadatas: &[ColumnMetadata],
+    ) -> Result<()> {
+        let table_ref = self.table.table_ref();
+        let table_id = self.table_id;
+
+        if let Some(new_table_name) = &self.new_table_name {
+            debug!(
+                "Starting update table: {} metadata, table_id: {}, new table info: {:?}, new table name: {}",
+                table_ref, table_id, raw_table_info, new_table_name
+            );
+
+            table_metadata_manager
+                .rename_table(current_table_info_value, new_table_name.to_string())
+                .await?;
+        } else {
+            debug!(
+                "Starting update table: {} metadata, table_id: {}, new table info: {:?}",
+                table_ref, table_id, raw_table_info
+            );
+
+            ensure!(
+                region_distribution.is_some(),
+                UnexpectedSnafu {
+                    err_msg: "region distribution is not set when updating table metadata",
+                }
+            );
+
+            if !column_metadatas.is_empty() {
+                raw_table_info::update_table_info_column_ids(&mut raw_table_info, column_metadatas);
+            }
+            table_metadata_manager
+                .update_table_info(
+                    current_table_info_value,
+                    region_distribution.cloned(),
+                    raw_table_info,
+                )
+                .await?;
+        }
+
+        Ok(())
+    }
+
+    /// Alters regions on the datanode nodes.
+    pub(crate) async fn on_alter_regions(
+        &self,
+        node_manager: &NodeManagerRef,
+        region_routes: &[RegionRoute],
+        kind: Option<alter_request::Kind>,
+    ) -> Vec<Result<RegionResponse>> {
+        let region_distribution = region_distribution(region_routes);
+        let leaders = find_leaders(region_routes)
+            .into_iter()
+            .map(|p| (p.id, p))
+            .collect::<HashMap<_, _>>();
+        let total_num_region = region_distribution
+            .values()
+            .map(|r| r.leader_regions.len())
+            .sum::<usize>();
+        let mut alter_region_tasks = Vec::with_capacity(total_num_region);
+        for (datanode_id, region_role_set) in region_distribution {
+            if region_role_set.leader_regions.is_empty() {
+                continue;
+            }
+            // Safety: must exists.
+            let peer = leaders.get(&datanode_id).unwrap();
+            let requester = node_manager.datanode(peer).await;
+
+            for region_id in region_role_set.leader_regions {
+                let region_id = RegionId::new(self.table_id, region_id);
+                let request = make_alter_region_request(region_id, kind.clone());
+
+                let requester = requester.clone();
+                let peer = peer.clone();
+
+                alter_region_tasks.push(async move {
+                    requester
+                        .handle(request)
+                        .await
+                        .map_err(add_peer_context_if_needed(peer))
+                });
+            }
+        }
+
+        future::join_all(alter_region_tasks)
+            .await
+            .into_iter()
+            .collect::<Vec<_>>()
+    }
+
+    /// Invalidates cache for the table.
+    pub(crate) async fn invalidate_table_cache(
+        &self,
+        cache_invalidator: &CacheInvalidatorRef,
+    ) -> Result<()> {
+        let ctx = Context {
+            subject: Some(format!(
+                "Invalidate table cache by altering table {}, table_id: {}",
+                self.table.table_ref(),
+                self.table_id,
+            )),
+        };
+
+        cache_invalidator
+            .invalidate(
+                &ctx,
+                &[
+                    CacheIdent::TableName(self.table.clone()),
+                    CacheIdent::TableId(self.table_id),
+                ],
+            )
+            .await?;
+
+        Ok(())
+    }
+}
+
+/// Makes alter region request.
+pub(crate) fn make_alter_region_request(
+    region_id: RegionId,
+    kind: Option<alter_request::Kind>,
+) -> RegionRequest {
+    RegionRequest {
+        header: Some(RegionRequestHeader {
+            tracing_context: TracingContext::from_current_span().to_w3c(),
+            ..Default::default()
+        }),
+        body: Some(Body::Alter(AlterRequest {
+            region_id: region_id.as_u64(),
+            kind,
+            ..Default::default()
+        })),
+    }
+}
+
+/// Builds new table info after alteration.
+///
+/// This function creates a new table info by applying the alter table expression
+/// to the existing table info. For add column operations, it increments the
+/// `next_column_id` by the number of columns being added, which may result in gaps
+/// in the column id sequence.
+fn build_new_table_info(
+    table_info: &RawTableInfo,
+    alter_table_expr: AlterTableExpr,
+) -> Result<TableInfo> {
+    let table_info =
+        TableInfo::try_from(table_info.clone()).context(error::ConvertRawTableInfoSnafu)?;
+    let schema_name = &table_info.schema_name;
+    let catalog_name = &table_info.catalog_name;
+    let table_name = &table_info.name;
+    let table_id = table_info.ident.table_id;
+    let request = alter_expr_to_request(table_id, alter_table_expr)
+        .context(error::ConvertAlterTableRequestSnafu)?;
+
+    let new_meta = table_info
+        .meta
+        .builder_with_alter_kind(table_name, &request.alter_kind)
+        .context(error::TableSnafu)?
+        .build()
+        .with_context(|_| error::BuildTableMetaSnafu {
+            table_name: format_full_table_name(catalog_name, schema_name, table_name),
+        })?;
+
+    let mut new_info = table_info.clone();
+    new_info.meta = new_meta;
+    new_info.ident.version = table_info.ident.version + 1;
+    match request.alter_kind {
+        AlterKind::AddColumns { columns } => {
+            // Bumps the column id for the new columns.
+            // It may bump more than the actual number of columns added if there are
+            // existing columns, but it's fine.
+            new_info.meta.next_column_id += columns.len() as u32;
+        }
+        AlterKind::RenameTable { new_table_name } => {
+            new_info.name = new_table_name.to_string();
+        }
+        AlterKind::DropColumns { .. }
+        | AlterKind::ModifyColumnTypes { .. }
+        | AlterKind::SetTableOptions { .. }
+        | AlterKind::UnsetTableOptions { .. }
+        | AlterKind::SetIndexes { .. }
+        | AlterKind::UnsetIndexes { .. }
+        | AlterKind::DropDefaults { .. } => {}
+    }
+
+    info!(
+        "Built new table info: {:?} for table {}, table_id: {}",
+        new_info.meta, table_name, table_id
+    );
+
+    Ok(new_info)
+}

src/common/meta/src/ddl/alter_table/region_request.rs

@@ -15,43 +15,16 @@
 use std::collections::HashSet;
 
 use api::v1::alter_table_expr::Kind;
-use api::v1::region::region_request::Body;
 use api::v1::region::{
-    alter_request, AddColumn, AddColumns, AlterRequest, DropColumn, DropColumns, RegionColumnDef,
-    RegionRequest, RegionRequestHeader,
+    alter_request, AddColumn, AddColumns, DropColumn, DropColumns, RegionColumnDef,
 };
-use common_telemetry::tracing_context::TracingContext;
 use snafu::OptionExt;
-use store_api::storage::RegionId;
 use table::metadata::RawTableInfo;
 
 use crate::ddl::alter_table::AlterTableProcedure;
 use crate::error::{InvalidProtoMsgSnafu, Result};
 
 impl AlterTableProcedure {
-    /// Makes alter region request from existing an alter kind.
-    /// Region alter request always add columns if not exist.
-    pub(crate) fn make_alter_region_request(
-        &self,
-        region_id: RegionId,
-        kind: Option<alter_request::Kind>,
-    ) -> Result<RegionRequest> {
-        // Safety: checked
-        let table_info = self.data.table_info().unwrap();
-
-        Ok(RegionRequest {
-            header: Some(RegionRequestHeader {
-                tracing_context: TracingContext::from_current_span().to_w3c(),
-                ..Default::default()
-            }),
-            body: Some(Body::Alter(AlterRequest {
-                region_id: region_id.as_u64(),
-                schema_version: table_info.ident.version,
-                kind,
-            })),
-        })
-    }
-
     /// Makes alter kind proto that all regions can reuse.
     /// Region alter request always add columns if not exist.
     pub(crate) fn make_region_alter_kind(&self) -> Result<Option<alter_request::Kind>> {
@@ -135,6 +108,8 @@ fn create_proto_alter_kind(
         Kind::UnsetTableOptions(v) => Ok(Some(alter_request::Kind::UnsetTableOptions(v.clone()))),
         Kind::SetIndex(v) => Ok(Some(alter_request::Kind::SetIndex(v.clone()))),
         Kind::UnsetIndex(v) => Ok(Some(alter_request::Kind::UnsetIndex(v.clone()))),
+        Kind::SetIndexes(v) => Ok(Some(alter_request::Kind::SetIndexes(v.clone()))),
+        Kind::UnsetIndexes(v) => Ok(Some(alter_request::Kind::UnsetIndexes(v.clone()))),
         Kind::DropDefaults(v) => Ok(Some(alter_request::Kind::DropDefaults(v.clone()))),
     }
 }
@@ -155,6 +130,7 @@ mod tests {
     use common_catalog::consts::{DEFAULT_CATALOG_NAME, DEFAULT_SCHEMA_NAME};
     use store_api::storage::{RegionId, TableId};
 
+    use crate::ddl::alter_table::executor::make_alter_region_request;
     use crate::ddl::alter_table::AlterTableProcedure;
     use crate::ddl::test_util::columns::TestColumnDefBuilder;
     use crate::ddl::test_util::create_table::{
@@ -261,15 +237,13 @@ mod tests {
         let mut procedure = AlterTableProcedure::new(table_id, task, ddl_context).unwrap();
         procedure.on_prepare().await.unwrap();
         let alter_kind = procedure.make_region_alter_kind().unwrap();
-        let Some(Body::Alter(alter_region_request)) = procedure
-            .make_alter_region_request(region_id, alter_kind)
-            .unwrap()
-            .body
+        let Some(Body::Alter(alter_region_request)) =
+            make_alter_region_request(region_id, alter_kind).body
         else {
             unreachable!()
         };
         assert_eq!(alter_region_request.region_id, region_id.as_u64());
-        assert_eq!(alter_region_request.schema_version, 1);
+        assert_eq!(alter_region_request.schema_version, 0);
         assert_eq!(
             alter_region_request.kind,
             Some(region::alter_request::Kind::AddColumns(
@@ -319,15 +293,13 @@ mod tests {
         let mut procedure = AlterTableProcedure::new(table_id, task, ddl_context).unwrap();
         procedure.on_prepare().await.unwrap();
         let alter_kind = procedure.make_region_alter_kind().unwrap();
-        let Some(Body::Alter(alter_region_request)) = procedure
-            .make_alter_region_request(region_id, alter_kind)
-            .unwrap()
-            .body
+        let Some(Body::Alter(alter_region_request)) =
+            make_alter_region_request(region_id, alter_kind).body
         else {
             unreachable!()
         };
         assert_eq!(alter_region_request.region_id, region_id.as_u64());
-        assert_eq!(alter_region_request.schema_version, 1);
+        assert_eq!(alter_region_request.schema_version, 0);
         assert_eq!(
             alter_region_request.kind,
             Some(region::alter_request::Kind::ModifyColumnTypes(

src/common/meta/src/ddl/alter_table/update_metadata.rs

@@ -1,103 +0,0 @@
-// Copyright 2023 Greptime Team
-//
-// 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 common_grpc_expr::alter_expr_to_request;
-use snafu::ResultExt;
-use table::metadata::{RawTableInfo, TableInfo};
-use table::requests::AlterKind;
-
-use crate::ddl::alter_table::AlterTableProcedure;
-use crate::error::{self, Result};
-use crate::key::table_info::TableInfoValue;
-use crate::key::{DeserializedValueWithBytes, RegionDistribution};
-
-impl AlterTableProcedure {
-    /// Builds new table info after alteration.
-    /// It bumps the column id of the table by the number of the add column requests.
-    /// So there may be holes in the column id sequence.
-    pub(crate) fn build_new_table_info(&self, table_info: &RawTableInfo) -> Result<TableInfo> {
-        let table_info =
-            TableInfo::try_from(table_info.clone()).context(error::ConvertRawTableInfoSnafu)?;
-        let table_ref = self.data.table_ref();
-        let alter_expr = self.data.task.alter_table.clone();
-        let request = alter_expr_to_request(self.data.table_id(), alter_expr)
-            .context(error::ConvertAlterTableRequestSnafu)?;
-
-        let new_meta = table_info
-            .meta
-            .builder_with_alter_kind(table_ref.table, &request.alter_kind)
-            .context(error::TableSnafu)?
-            .build()
-            .with_context(|_| error::BuildTableMetaSnafu {
-                table_name: table_ref.table,
-            })?;
-
-        let mut new_info = table_info.clone();
-        new_info.meta = new_meta;
-        new_info.ident.version = table_info.ident.version + 1;
-        match request.alter_kind {
-            AlterKind::AddColumns { columns } => {
-                // Bumps the column id for the new columns.
-                // It may bump more than the actual number of columns added if there are
-                // existing columns, but it's fine.
-                new_info.meta.next_column_id += columns.len() as u32;
-            }
-            AlterKind::RenameTable { new_table_name } => {
-                new_info.name = new_table_name.to_string();
-            }
-            AlterKind::DropColumns { .. }
-            | AlterKind::ModifyColumnTypes { .. }
-            | AlterKind::SetTableOptions { .. }
-            | AlterKind::UnsetTableOptions { .. }
-            | AlterKind::SetIndex { .. }
-            | AlterKind::UnsetIndex { .. }
-            | AlterKind::DropDefaults { .. } => {}
-        }
-
-        Ok(new_info)
-    }
-
-    /// Updates table metadata for rename table operation.
-    pub(crate) async fn on_update_metadata_for_rename(
-        &self,
-        new_table_name: String,
-        current_table_info_value: &DeserializedValueWithBytes<TableInfoValue>,
-    ) -> Result<()> {
-        let table_metadata_manager = &self.context.table_metadata_manager;
-        table_metadata_manager
-            .rename_table(current_table_info_value, new_table_name)
-            .await?;
-
-        Ok(())
-    }
-
-    /// Updates table metadata for alter table operation.
-    pub(crate) async fn on_update_metadata_for_alter(
-        &self,
-        new_table_info: RawTableInfo,
-        region_distribution: RegionDistribution,
-        current_table_info_value: &DeserializedValueWithBytes<TableInfoValue>,
-    ) -> Result<()> {
-        let table_metadata_manager = &self.context.table_metadata_manager;
-        table_metadata_manager
-            .update_table_info(
-                current_table_info_value,
-                Some(region_distribution),
-                new_table_info,
-            )
-            .await?;
-
-        Ok(())
-    }
-}

src/common/meta/src/ddl/create_flow/metadata.rs

@@ -21,7 +21,7 @@ use crate::key::table_name::TableNameKey;
 impl CreateFlowProcedure {
     /// Allocates the [FlowId].
     pub(crate) async fn allocate_flow_id(&mut self) -> Result<()> {
-        //TODO(weny, ruihang): We doesn't support the partitions. It's always be 1, now.
+        // TODO(weny, ruihang): We don't support the partitions. It's always be 1, now.
         let partitions = 1;
         let (flow_id, peers) = self
             .context

src/common/meta/src/ddl/create_logical_tables/update_metadata.rs

@@ -22,7 +22,7 @@ use table::table_name::TableName;
 
 use crate::cache_invalidator::Context;
 use crate::ddl::create_logical_tables::CreateLogicalTablesProcedure;
-use crate::ddl::physical_table_metadata;
+use crate::ddl::utils::raw_table_info;
 use crate::error::{Result, TableInfoNotFoundSnafu};
 use crate::instruction::CacheIdent;
 
@@ -47,7 +47,7 @@ impl CreateLogicalTablesProcedure {
         // Generates new table info
         let raw_table_info = physical_table_info.deref().table_info.clone();
 
-        let new_table_info = physical_table_metadata::build_new_physical_table_info(
+        let new_table_info = raw_table_info::build_new_physical_table_info(
             raw_table_info,
             &self.data.physical_columns,
         );

src/common/meta/src/ddl/create_table.rs

@@ -21,7 +21,7 @@ use common_error::ext::BoxedError;
 use common_procedure::error::{
     ExternalSnafu, FromJsonSnafu, Result as ProcedureResult, ToJsonSnafu,
 };
-use common_procedure::{Context as ProcedureContext, LockKey, Procedure, Status};
+use common_procedure::{Context as ProcedureContext, LockKey, Procedure, ProcedureId, Status};
 use common_telemetry::tracing_context::TracingContext;
 use common_telemetry::{info, warn};
 use futures::future::join_all;
@@ -35,7 +35,7 @@ use table::metadata::{RawTableInfo, TableId};
 use table::table_reference::TableReference;
 
 use crate::ddl::create_table_template::{build_template, CreateRequestBuilder};
-use crate::ddl::physical_table_metadata::update_table_info_column_ids;
+use crate::ddl::utils::raw_table_info::update_table_info_column_ids;
 use crate::ddl::utils::{
     add_peer_context_if_needed, convert_region_routes_to_detecting_regions,
     extract_column_metadatas, map_to_procedure_error, region_storage_path,
@@ -246,8 +246,6 @@ impl CreateTableProcedure {
             }
         }
 
-        self.creator.data.state = CreateTableState::CreateMetadata;
-
         let mut results = join_all(create_region_tasks)
             .await
             .into_iter()
@@ -261,6 +259,7 @@ impl CreateTableProcedure {
             warn!("creating table result doesn't contains extension key `{TABLE_COLUMN_METADATA_EXTENSION_KEY}`,leaving the table's column metadata unchanged");
         }
 
+        self.creator.data.state = CreateTableState::CreateMetadata;
         Ok(Status::executing(true))
     }
 
@@ -268,8 +267,9 @@ impl CreateTableProcedure {
     ///
     /// Abort(not-retry):
     /// - Failed to create table metadata.
-    async fn on_create_metadata(&mut self) -> Result<Status> {
+    async fn on_create_metadata(&mut self, pid: ProcedureId) -> Result<Status> {
         let table_id = self.table_id();
+        let table_ref = self.creator.data.table_ref();
         let manager = &self.context.table_metadata_manager;
 
         let mut raw_table_info = self.table_info().clone();
@@ -289,7 +289,10 @@ impl CreateTableProcedure {
         self.context
             .register_failure_detectors(detecting_regions)
             .await;
-        info!("Created table metadata for table {table_id}");
+        info!(
+            "Successfully created table: {}, table_id: {}, procedure_id: {}",
+            table_ref, table_id, pid
+        );
 
         self.creator.opening_regions.clear();
         Ok(Status::done_with_output(table_id))
@@ -317,7 +320,7 @@ impl Procedure for CreateTableProcedure {
         Ok(())
     }
 
-    async fn execute(&mut self, _ctx: &ProcedureContext) -> ProcedureResult<Status> {
+    async fn execute(&mut self, ctx: &ProcedureContext) -> ProcedureResult<Status> {
         let state = &self.creator.data.state;
 
         let _timer = metrics::METRIC_META_PROCEDURE_CREATE_TABLE
@@ -327,7 +330,7 @@ impl Procedure for CreateTableProcedure {
         match state {
             CreateTableState::Prepare => self.on_prepare().await,
             CreateTableState::DatanodeCreateRegions => self.on_datanode_create_regions().await,
-            CreateTableState::CreateMetadata => self.on_create_metadata().await,
+            CreateTableState::CreateMetadata => self.on_create_metadata(ctx.procedure_id).await,
         }
         .map_err(map_to_procedure_error)
     }

src/common/meta/src/ddl/drop_table/executor.rs

@@ -185,11 +185,15 @@ impl DropTableExecutor {
             .await
     }
 
-    /// Invalidates frontend caches
+    /// Invalidates caches for the table.
     pub async fn invalidate_table_cache(&self, ctx: &DdlContext) -> Result<()> {
         let cache_invalidator = &ctx.cache_invalidator;
         let ctx = Context {
-            subject: Some("Invalidate table cache by dropping table".to_string()),
+            subject: Some(format!(
+                "Invalidate table cache by dropping table {}, table_id: {}",
+                self.table.table_ref(),
+                self.table_id,
+            )),
         };
 
         cache_invalidator

src/common/meta/src/ddl/table_meta.rs

@@ -21,7 +21,7 @@ use snafu::ensure;
 use store_api::storage::{RegionId, RegionNumber, TableId};
 
 use crate::ddl::TableMetadata;
-use crate::error::{self, Result, UnsupportedSnafu};
+use crate::error::{Result, UnsupportedSnafu};
 use crate::key::table_route::PhysicalTableRouteValue;
 use crate::peer::Peer;
 use crate::rpc::ddl::CreateTableTask;
@@ -113,24 +113,18 @@ impl TableMetadataAllocator {
         table_id: TableId,
         task: &CreateTableTask,
     ) -> Result<PhysicalTableRouteValue> {
-        let regions = task.partitions.len();
-        ensure!(
-            regions > 0,
-            error::UnexpectedSnafu {
-                err_msg: "The number of partitions must be greater than 0"
-            }
-        );
-
+        let regions = task.partitions.len().max(1);
         let peers = self.peer_allocator.alloc(regions).await?;
         debug!("Allocated peers {:?} for table {}", peers, table_id);
-        let region_routes = task
+
+        let mut region_routes = task
             .partitions
             .iter()
             .enumerate()
             .map(|(i, partition)| {
                 let region = Region {
                     id: RegionId::new(table_id, i as u32),
-                    partition: Some(partition.clone().into()),
+                    partition_expr: partition.expression.clone(),
                     ..Default::default()
                 };
 
@@ -144,6 +138,18 @@ impl TableMetadataAllocator {
             })
             .collect::<Vec<_>>();
 
+        // If the table has no partitions, we need to create a default region.
+        if region_routes.is_empty() {
+            region_routes.push(RegionRoute {
+                region: Region {
+                    id: RegionId::new(table_id, 0),
+                    ..Default::default()
+                },
+                leader_peer: Some(peers[0].clone()),
+                ..Default::default()
+            });
+        }
+
         Ok(PhysicalTableRouteValue::new(region_routes))
     }
 

src/common/meta/src/ddl/test_util.rs

@@ -17,6 +17,7 @@ pub mod columns;
 pub mod create_table;
 pub mod datanode_handler;
 pub mod flownode_handler;
+pub mod region_metadata;
 
 use std::assert_matches::assert_matches;
 use std::collections::HashMap;
@@ -145,10 +146,7 @@ pub fn test_create_logical_table_task(name: &str) -> CreateTableTask {
     CreateTableTask {
         create_table,
         // Single region
-        partitions: vec![Partition {
-            column_list: vec![],
-            value_list: vec![],
-        }],
+        partitions: vec![Partition::default()],
         table_info,
     }
 }
@@ -183,10 +181,7 @@ pub fn test_create_physical_table_task(name: &str) -> CreateTableTask {
     CreateTableTask {
         create_table,
         // Single region
-        partitions: vec![Partition {
-            column_list: vec![],
-            value_list: vec![],
-        }],
+        partitions: vec![Partition::default()],
         table_info,
     }
 }

src/common/meta/src/ddl/test_util/create_table.rs

@@ -175,10 +175,7 @@ pub fn test_create_table_task(name: &str, table_id: TableId) -> CreateTableTask
     CreateTableTask {
         create_table,
         // Single region
-        partitions: vec![Partition {
-            column_list: vec![],
-            value_list: vec![],
-        }],
+        partitions: vec![Partition::default()],
         table_info,
     }
 }

src/common/meta/src/ddl/test_util/datanode_handler.rs

@@ -12,9 +12,11 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
+use std::collections::HashMap;
 use std::sync::Arc;
 
 use api::region::RegionResponse;
+use api::v1::region::region_request::Body;
 use api::v1::region::RegionRequest;
 use common_error::ext::{BoxedError, ErrorExt, StackError};
 use common_error::status_code::StatusCode;
@@ -22,6 +24,8 @@ use common_query::request::QueryRequest;
 use common_recordbatch::SendableRecordBatchStream;
 use common_telemetry::debug;
 use snafu::{ResultExt, Snafu};
+use store_api::metadata::RegionMetadata;
+use store_api::storage::RegionId;
 use tokio::sync::mpsc;
 
 use crate::error::{self, Error, Result};
@@ -278,3 +282,47 @@ impl MockDatanodeHandler for AllFailureDatanodeHandler {
         unreachable!()
     }
 }
+
+#[derive(Clone)]
+pub struct ListMetadataDatanodeHandler {
+    pub region_metadatas: HashMap<RegionId, Option<RegionMetadata>>,
+}
+
+impl ListMetadataDatanodeHandler {
+    pub fn new(region_metadatas: HashMap<RegionId, Option<RegionMetadata>>) -> Self {
+        Self { region_metadatas }
+    }
+}
+
+#[async_trait::async_trait]
+impl MockDatanodeHandler for ListMetadataDatanodeHandler {
+    async fn handle(&self, _peer: &Peer, request: RegionRequest) -> Result<RegionResponse> {
+        let Some(Body::ListMetadata(req)) = request.body else {
+            unreachable!()
+        };
+        let mut response = RegionResponse::new(0);
+
+        let mut output = Vec::with_capacity(req.region_ids.len());
+        for region_id in req.region_ids {
+            match self.region_metadatas.get(&RegionId::from_u64(region_id)) {
+                Some(metadata) => {
+                    output.push(metadata.clone());
+                }
+                None => {
+                    output.push(None);
+                }
+            }
+        }
+
+        response.metadata = serde_json::to_vec(&output).unwrap();
+        Ok(response)
+    }
+
+    async fn handle_query(
+        &self,
+        _peer: &Peer,
+        _request: QueryRequest,
+    ) -> Result<SendableRecordBatchStream> {
+        unreachable!()
+    }
+}

src/common/meta/src/ddl/test_util/region_metadata.rs

@@ -0,0 +1,34 @@
+// Copyright 2023 Greptime Team
+//
+// 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 api::v1::SemanticType;
+use store_api::metadata::{ColumnMetadata, RegionMetadata, RegionMetadataBuilder};
+use store_api::storage::RegionId;
+
+/// Builds a region metadata with the given column metadatas.
+pub fn build_region_metadata(
+    region_id: RegionId,
+    column_metadatas: &[ColumnMetadata],
+) -> RegionMetadata {
+    let mut builder = RegionMetadataBuilder::new(region_id);
+    let mut primary_key = vec![];
+    for column_metadata in column_metadatas {
+        builder.push_column_metadata(column_metadata.clone());
+        if column_metadata.semantic_type == SemanticType::Tag {
+            primary_key.push(column_metadata.column_id);
+        }
+    }
+    builder.primary_key(primary_key);
+    builder.build().unwrap()
+}

src/common/meta/src/ddl/tests/create_table.rs

@@ -141,10 +141,7 @@ pub(crate) fn test_create_table_task(name: &str) -> CreateTableTask {
     CreateTableTask {
         create_table,
         // Single region
-        partitions: vec![Partition {
-            column_list: vec![],
-            value_list: vec![],
-        }],
+        partitions: vec![Partition::default()],
         table_info,
     }
 }
@@ -213,21 +210,6 @@ async fn test_on_prepare_without_create_if_table_exists() {
     assert_eq!(procedure.table_id(), 1024);
 }
 
-#[tokio::test]
-async fn test_on_prepare_with_no_partition_err() {
-    let node_manager = Arc::new(MockDatanodeManager::new(()));
-    let ddl_context = new_ddl_context(node_manager);
-    let mut task = test_create_table_task("foo");
-    task.partitions = vec![];
-    task.create_table.create_if_not_exists = true;
-    let mut procedure = CreateTableProcedure::new(task, ddl_context);
-    let err = procedure.on_prepare().await.unwrap_err();
-    assert_matches!(err, Error::Unexpected { .. });
-    assert!(err
-        .to_string()
-        .contains("The number of partitions must be greater than 0"),);
-}
-
 #[tokio::test]
 async fn test_on_datanode_create_regions_should_retry() {
     common_telemetry::init_default_ut_logging();

src/common/meta/src/ddl/utils.rs

@@ -12,6 +12,12 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
+pub(crate) mod raw_table_info;
+#[allow(dead_code)]
+pub(crate) mod region_metadata_lister;
+pub(crate) mod table_id;
+pub(crate) mod table_info;
+
 use std::collections::HashMap;
 use std::fmt::Debug;
 
@@ -442,6 +448,7 @@ pub fn extract_column_metadatas(
         .collect::<Vec<_>>();
 
     if schemas.is_empty() {
+        warn!("extract_column_metadatas: no extension key `{key}` found in results");
         return Ok(None);
     }
 

src/common/meta/src/ddl/utils/raw_table_info.rs

@@ -54,7 +54,10 @@ pub(crate) fn build_new_physical_table_info(
                 }
             }
             SemanticType::Field => value_indices.push(idx),
-            SemanticType::Timestamp => *time_index = Some(idx),
+            SemanticType::Timestamp => {
+                value_indices.push(idx);
+                *time_index = Some(idx);
+            }
         }
 
         columns.push(col.column_schema.clone());

src/common/meta/src/ddl/utils/region_metadata_lister.rs

@@ -0,0 +1,240 @@
+// Copyright 2023 Greptime Team
+//
+// 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 api::v1::region::region_request::Body as PbRegionRequest;
+use api::v1::region::{ListMetadataRequest, RegionRequest, RegionRequestHeader};
+use common_telemetry::tracing_context::TracingContext;
+use futures::future::join_all;
+use snafu::ResultExt;
+use store_api::metadata::RegionMetadata;
+use store_api::storage::{RegionId, TableId};
+
+use crate::ddl::utils::add_peer_context_if_needed;
+use crate::error::{DecodeJsonSnafu, Result};
+use crate::node_manager::NodeManagerRef;
+use crate::rpc::router::{find_leaders, region_distribution, RegionRoute};
+
+/// Collects the region metadata from the datanodes.
+pub struct RegionMetadataLister {
+    node_manager: NodeManagerRef,
+}
+
+impl RegionMetadataLister {
+    /// Creates a new [`RegionMetadataLister`] with the given [`NodeManagerRef`].
+    pub fn new(node_manager: NodeManagerRef) -> Self {
+        Self { node_manager }
+    }
+
+    /// Collects the region metadata from the datanodes.
+    pub async fn list(
+        &self,
+        table_id: TableId,
+        region_routes: &[RegionRoute],
+    ) -> Result<Vec<Option<RegionMetadata>>> {
+        let region_distribution = region_distribution(region_routes);
+        let leaders = find_leaders(region_routes)
+            .into_iter()
+            .map(|p| (p.id, p))
+            .collect::<HashMap<_, _>>();
+
+        let total_num_region = region_distribution
+            .values()
+            .map(|r| r.leader_regions.len())
+            .sum::<usize>();
+
+        let mut list_metadata_tasks = Vec::with_capacity(leaders.len());
+
+        // Build requests.
+        for (datanode_id, region_role_set) in region_distribution {
+            if region_role_set.leader_regions.is_empty() {
+                continue;
+            }
+            // Safety: must exists.
+            let peer = leaders.get(&datanode_id).unwrap();
+            let requester = self.node_manager.datanode(peer).await;
+            let region_ids = region_role_set
+                .leader_regions
+                .iter()
+                .map(|r| RegionId::new(table_id, *r).as_u64())
+                .collect();
+            let request = Self::build_list_metadata_request(region_ids);
+
+            let peer = peer.clone();
+            list_metadata_tasks.push(async move {
+                requester
+                    .handle(request)
+                    .await
+                    .map_err(add_peer_context_if_needed(peer))
+            });
+        }
+
+        let results = join_all(list_metadata_tasks)
+            .await
+            .into_iter()
+            .collect::<Result<Vec<_>>>()?
+            .into_iter()
+            .map(|r| r.metadata);
+
+        let mut output = Vec::with_capacity(total_num_region);
+        for result in results {
+            let region_metadatas: Vec<Option<RegionMetadata>> =
+                serde_json::from_slice(&result).context(DecodeJsonSnafu)?;
+            output.extend(region_metadatas);
+        }
+
+        Ok(output)
+    }
+
+    fn build_list_metadata_request(region_ids: Vec<u64>) -> RegionRequest {
+        RegionRequest {
+            header: Some(RegionRequestHeader {
+                tracing_context: TracingContext::from_current_span().to_w3c(),
+                ..Default::default()
+            }),
+            body: Some(PbRegionRequest::ListMetadata(ListMetadataRequest {
+                region_ids,
+            })),
+        }
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use std::collections::HashMap;
+    use std::sync::Arc;
+
+    use api::region::RegionResponse;
+    use api::v1::meta::Peer;
+    use api::v1::region::region_request::Body;
+    use api::v1::region::RegionRequest;
+    use store_api::metadata::RegionMetadata;
+    use store_api::storage::RegionId;
+    use tokio::sync::mpsc;
+
+    use crate::ddl::test_util::datanode_handler::{DatanodeWatcher, ListMetadataDatanodeHandler};
+    use crate::ddl::test_util::region_metadata::build_region_metadata;
+    use crate::ddl::test_util::test_column_metadatas;
+    use crate::ddl::utils::region_metadata_lister::RegionMetadataLister;
+    use crate::error::Result;
+    use crate::rpc::router::{Region, RegionRoute};
+    use crate::test_util::MockDatanodeManager;
+
+    fn assert_list_metadata_request(req: RegionRequest, expected_region_ids: &[RegionId]) {
+        let Some(Body::ListMetadata(req)) = req.body else {
+            unreachable!()
+        };
+
+        assert_eq!(req.region_ids.len(), expected_region_ids.len());
+        for region_id in expected_region_ids {
+            assert!(req.region_ids.contains(&region_id.as_u64()));
+        }
+    }
+
+    fn empty_list_metadata_handler(_peer: Peer, request: RegionRequest) -> Result<RegionResponse> {
+        let Some(Body::ListMetadata(req)) = request.body else {
+            unreachable!()
+        };
+
+        let mut output: Vec<Option<RegionMetadata>> = Vec::with_capacity(req.region_ids.len());
+        for _region_id in req.region_ids {
+            output.push(None);
+        }
+
+        Ok(RegionResponse::from_metadata(
+            serde_json::to_vec(&output).unwrap(),
+        ))
+    }
+
+    #[tokio::test]
+    async fn test_list_request() {
+        let (tx, mut rx) = mpsc::channel(8);
+        let handler = DatanodeWatcher::new(tx).with_handler(empty_list_metadata_handler);
+        let node_manager = Arc::new(MockDatanodeManager::new(handler));
+        let lister = RegionMetadataLister::new(node_manager);
+        let region_routes = vec![
+            RegionRoute {
+                region: Region::new_test(RegionId::new(1024, 1)),
+                leader_peer: Some(Peer::empty(1)),
+                follower_peers: vec![Peer::empty(5)],
+                leader_state: None,
+                leader_down_since: None,
+            },
+            RegionRoute {
+                region: Region::new_test(RegionId::new(1024, 2)),
+                leader_peer: Some(Peer::empty(3)),
+                follower_peers: vec![Peer::empty(4)],
+                leader_state: None,
+                leader_down_since: None,
+            },
+            RegionRoute {
+                region: Region::new_test(RegionId::new(1024, 3)),
+                leader_peer: Some(Peer::empty(3)),
+                follower_peers: vec![Peer::empty(4)],
+                leader_state: None,
+                leader_down_since: None,
+            },
+        ];
+        let region_metadatas = lister.list(1024, &region_routes).await.unwrap();
+        assert_eq!(region_metadatas.len(), 3);
+
+        let mut requests = vec![];
+        for _ in 0..2 {
+            let (peer, request) = rx.try_recv().unwrap();
+            requests.push((peer, request));
+        }
+        rx.try_recv().unwrap_err();
+
+        let (peer, request) = requests.remove(0);
+        assert_eq!(peer.id, 1);
+        assert_list_metadata_request(request, &[RegionId::new(1024, 1)]);
+        let (peer, request) = requests.remove(0);
+        assert_eq!(peer.id, 3);
+        assert_list_metadata_request(request, &[RegionId::new(1024, 2), RegionId::new(1024, 3)]);
+    }
+
+    #[tokio::test]
+    async fn test_list_region_metadata() {
+        let region_metadata =
+            build_region_metadata(RegionId::new(1024, 1), &test_column_metadatas(&["tag_0"]));
+        let region_metadatas = HashMap::from([
+            (RegionId::new(1024, 0), None),
+            (RegionId::new(1024, 1), Some(region_metadata.clone())),
+        ]);
+        let handler = ListMetadataDatanodeHandler::new(region_metadatas);
+        let node_manager = Arc::new(MockDatanodeManager::new(handler));
+        let lister = RegionMetadataLister::new(node_manager);
+        let region_routes = vec![
+            RegionRoute {
+                region: Region::new_test(RegionId::new(1024, 0)),
+                leader_peer: Some(Peer::empty(1)),
+                follower_peers: vec![],
+                leader_state: None,
+                leader_down_since: None,
+            },
+            RegionRoute {
+                region: Region::new_test(RegionId::new(1024, 1)),
+                leader_peer: Some(Peer::empty(3)),
+                follower_peers: vec![],
+                leader_state: None,
+                leader_down_since: None,
+            },
+        ];
+        let region_metadatas = lister.list(1024, &region_routes).await.unwrap();
+        assert_eq!(region_metadatas.len(), 2);
+        assert_eq!(region_metadatas[0], None);
+        assert_eq!(region_metadatas[1], Some(region_metadata));
+    }
+}

src/common/meta/src/ddl/utils/table_id.rs

@@ -0,0 +1,46 @@
+// Copyright 2023 Greptime Team
+//
+// 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 snafu::OptionExt;
+use store_api::storage::TableId;
+use table::table_reference::TableReference;
+
+use crate::error::{Result, TableNotFoundSnafu};
+use crate::key::table_name::{TableNameKey, TableNameManager};
+
+/// Get all the table ids from the table names.
+///
+/// Returns an error if any table does not exist.
+pub(crate) async fn get_all_table_ids_by_names<'a>(
+    table_name_manager: &TableNameManager,
+    table_names: &[TableReference<'a>],
+) -> Result<Vec<TableId>> {
+    let table_name_keys = table_names
+        .iter()
+        .map(TableNameKey::from)
+        .collect::<Vec<_>>();
+    let table_name_values = table_name_manager.batch_get(table_name_keys).await?;
+    let mut table_ids = Vec::with_capacity(table_name_values.len());
+    for (value, table_name) in table_name_values.into_iter().zip(table_names) {
+        let value = value
+            .with_context(|| TableNotFoundSnafu {
+                table_name: table_name.to_string(),
+            })?
+            .table_id();
+
+        table_ids.push(value);
+    }
+
+    Ok(table_ids)
+}

src/common/meta/src/ddl/utils/table_info.rs

@@ -0,0 +1,44 @@
+// Copyright 2023 Greptime Team
+//
+// 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 snafu::OptionExt;
+use store_api::storage::TableId;
+use table::table_reference::TableReference;
+
+use crate::error::{Result, TableInfoNotFoundSnafu};
+use crate::key::table_info::{TableInfoManager, TableInfoValue};
+use crate::key::DeserializedValueWithBytes;
+
+/// Get all table info values by table ids.
+///
+/// Returns an error if any table does not exist.
+pub(crate) async fn get_all_table_info_values_by_table_ids<'a>(
+    table_info_manager: &TableInfoManager,
+    table_ids: &[TableId],
+    table_names: &[TableReference<'a>],
+) -> Result<Vec<DeserializedValueWithBytes<TableInfoValue>>> {
+    let mut table_info_map = table_info_manager.batch_get_raw(table_ids).await?;
+    let mut table_info_values = Vec::with_capacity(table_ids.len());
+    for (table_id, table_name) in table_ids.iter().zip(table_names) {
+        let table_info_value =
+            table_info_map
+                .remove(table_id)
+                .with_context(|| TableInfoNotFoundSnafu {
+                    table: table_name.to_string(),
+                })?;
+        table_info_values.push(table_info_value);
+    }
+
+    Ok(table_info_values)
+}

src/common/meta/src/error.rs

@@ -877,6 +877,36 @@ pub enum Error {
         #[snafu(source)]
         error: object_store::Error,
     },
+
+    #[snafu(display(
+        "Missing column in column metadata: {}, table: {}, table_id: {}",
+        column_name,
+        table_name,
+        table_id,
+    ))]
+    MissingColumnInColumnMetadata {
+        column_name: String,
+        #[snafu(implicit)]
+        location: Location,
+        table_name: String,
+        table_id: TableId,
+    },
+
+    #[snafu(display(
+        "Mismatch column id: column_name: {}, column_id: {}, table: {}, table_id: {}",
+        column_name,
+        column_id,
+        table_name,
+        table_id,
+    ))]
+    MismatchColumnId {
+        column_name: String,
+        column_id: u32,
+        #[snafu(implicit)]
+        location: Location,
+        table_name: String,
+        table_id: TableId,
+    },
 }
 
 pub type Result<T> = std::result::Result<T, Error>;
@@ -896,7 +926,10 @@ impl ErrorExt for Error {
             | DeserializeFromJson { .. } => StatusCode::Internal,
 
             NoLeader { .. } => StatusCode::TableUnavailable,
-            ValueNotExist { .. } | ProcedurePoisonConflict { .. } => StatusCode::Unexpected,
+            ValueNotExist { .. }
+            | ProcedurePoisonConflict { .. }
+            | MissingColumnInColumnMetadata { .. }
+            | MismatchColumnId { .. } => StatusCode::Unexpected,
 
             Unsupported { .. } => StatusCode::Unsupported,
             WriteObject { .. } | ReadObject { .. } => StatusCode::StorageUnavailable,

src/common/meta/src/instruction.rs

@@ -174,6 +174,8 @@ pub struct UpgradeRegion {
 /// The identifier of cache.
 pub enum CacheIdent {
     FlowId(FlowId),
+    /// Indicate change of address of flownode.
+    FlowNodeAddressChange(u64),
     FlowName(FlowName),
     TableId(TableId),
     TableName(TableName),

src/common/meta/src/key.rs

@@ -1509,6 +1509,7 @@ mod tests {
                 name: "r1".to_string(),
                 partition: None,
                 attrs: BTreeMap::new(),
+                partition_expr: Default::default(),
             },
             leader_peer: Some(Peer::new(datanode, "a2")),
             follower_peers: vec![],
@@ -2001,6 +2002,7 @@ mod tests {
                     name: "r1".to_string(),
                     partition: None,
                     attrs: BTreeMap::new(),
+                    partition_expr: Default::default(),
                 },
                 leader_peer: Some(Peer::new(datanode, "a2")),
                 leader_state: Some(LeaderState::Downgrading),
@@ -2013,6 +2015,7 @@ mod tests {
                     name: "r2".to_string(),
                     partition: None,
                     attrs: BTreeMap::new(),
+                    partition_expr: Default::default(),
                 },
                 leader_peer: Some(Peer::new(datanode, "a1")),
                 leader_state: None,

src/common/meta/src/key/schema_name.rs

@@ -12,7 +12,7 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
-use std::collections::HashMap;
+use std::collections::{BTreeMap, HashMap};
 use std::fmt::Display;
 
 use common_catalog::consts::{DEFAULT_CATALOG_NAME, DEFAULT_SCHEMA_NAME};
@@ -58,12 +58,17 @@ impl Default for SchemaNameKey<'_> {
 pub struct SchemaNameValue {
     #[serde(default)]
     pub ttl: Option<DatabaseTimeToLive>,
+    #[serde(default)]
+    pub extra_options: BTreeMap<String, String>,
 }
 
 impl Display for SchemaNameValue {
     fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
         if let Some(ttl) = self.ttl.map(|i| i.to_string()) {
-            write!(f, "ttl='{}'", ttl)?;
+            writeln!(f, "'ttl'='{}'", ttl)?;
+        }
+        for (k, v) in self.extra_options.iter() {
+            writeln!(f, "'{k}'='{v}'")?;
         }
 
         Ok(())
@@ -87,7 +92,18 @@ impl TryFrom<&HashMap<String, String>> for SchemaNameValue {
             })
             .transpose()?
             .map(|ttl| ttl.into());
-        Ok(Self { ttl })
+        let extra_options = value
+            .iter()
+            .filter_map(|(k, v)| {
+                if k == OPT_KEY_TTL {
+                    None
+                } else {
+                    Some((k.clone(), v.clone()))
+                }
+            })
+            .collect();
+
+        Ok(Self { ttl, extra_options })
     }
 }
 
@@ -97,6 +113,12 @@ impl From<SchemaNameValue> for HashMap<String, String> {
         if let Some(ttl) = value.ttl.map(|ttl| ttl.to_string()) {
             opts.insert(OPT_KEY_TTL.to_string(), ttl);
         }
+        opts.extend(
+            value
+                .extra_options
+                .iter()
+                .map(|(k, v)| (k.clone(), v.clone())),
+        );
         opts
     }
 }
@@ -316,18 +338,23 @@ mod tests {
 
     #[test]
     fn test_display_schema_value() {
-        let schema_value = SchemaNameValue { ttl: None };
+        let schema_value = SchemaNameValue {
+            ttl: None,
+            ..Default::default()
+        };
         assert_eq!("", schema_value.to_string());
 
         let schema_value = SchemaNameValue {
             ttl: Some(Duration::from_secs(9).into()),
+            ..Default::default()
         };
-        assert_eq!("ttl='9s'", schema_value.to_string());
+        assert_eq!("'ttl'='9s'\n", schema_value.to_string());
 
         let schema_value = SchemaNameValue {
             ttl: Some(Duration::from_secs(0).into()),
+            ..Default::default()
         };
-        assert_eq!("ttl='forever'", schema_value.to_string());
+        assert_eq!("'ttl'='forever'\n", schema_value.to_string());
     }
 
     #[test]
@@ -341,6 +368,7 @@ mod tests {
 
         let value = SchemaNameValue {
             ttl: Some(Duration::from_secs(10).into()),
+            ..Default::default()
         };
         let mut opts: HashMap<String, String> = HashMap::new();
         opts.insert("ttl".to_string(), "10s".to_string());
@@ -355,6 +383,7 @@ mod tests {
 
         let forever = SchemaNameValue {
             ttl: Some(Default::default()),
+            ..Default::default()
         };
         let parsed = SchemaNameValue::try_from_raw_value(
             serde_json::json!({"ttl": "forever"}).to_string().as_bytes(),
@@ -374,6 +403,81 @@ mod tests {
         assert!(err_empty.is_err());
     }
 
+    #[test]
+    fn test_extra_options_compatibility() {
+        // Test with extra_options only
+        let mut opts: HashMap<String, String> = HashMap::new();
+        opts.insert("foo".to_string(), "bar".to_string());
+        opts.insert("baz".to_string(), "qux".to_string());
+        let value = SchemaNameValue::try_from(&opts).unwrap();
+        assert_eq!(value.ttl, None);
+        assert_eq!(value.extra_options.get("foo"), Some(&"bar".to_string()));
+        assert_eq!(value.extra_options.get("baz"), Some(&"qux".to_string()));
+
+        // Test round-trip conversion
+        let opts_back: HashMap<String, String> = value.clone().into();
+        assert_eq!(opts_back.get("foo"), Some(&"bar".to_string()));
+        assert_eq!(opts_back.get("baz"), Some(&"qux".to_string()));
+        assert!(!opts_back.contains_key("ttl"));
+
+        // Test with both ttl and extra_options
+        let mut opts: HashMap<String, String> = HashMap::new();
+        opts.insert("ttl".to_string(), "5m".to_string());
+        opts.insert("opt1".to_string(), "val1".to_string());
+        let value = SchemaNameValue::try_from(&opts).unwrap();
+        assert_eq!(value.ttl, Some(Duration::from_secs(300).into()));
+        assert_eq!(value.extra_options.get("opt1"), Some(&"val1".to_string()));
+
+        // Test serialization/deserialization compatibility
+        let json = serde_json::to_string(&value).unwrap();
+        let deserialized: SchemaNameValue = serde_json::from_str(&json).unwrap();
+        assert_eq!(value, deserialized);
+
+        // Test display includes extra_options
+        let mut value = SchemaNameValue::default();
+        value
+            .extra_options
+            .insert("foo".to_string(), "bar".to_string());
+        let display = value.to_string();
+        assert!(display.contains("'foo'='bar'"));
+    }
+
+    #[test]
+    fn test_backward_compatibility_with_old_format() {
+        // Simulate old format: only ttl, no extra_options
+        let json = r#"{"ttl":"10s"}"#;
+        let parsed = SchemaNameValue::try_from_raw_value(json.as_bytes()).unwrap();
+        assert_eq!(
+            parsed,
+            Some(SchemaNameValue {
+                ttl: Some(Duration::from_secs(10).into()),
+                extra_options: BTreeMap::new(),
+            })
+        );
+
+        // Simulate old format: null value
+        let json = r#"null"#;
+        let parsed = SchemaNameValue::try_from_raw_value(json.as_bytes()).unwrap();
+        assert!(parsed.is_none());
+    }
+
+    #[test]
+    fn test_forward_compatibility_with_new_options() {
+        // Simulate new format: ttl + extra_options
+        let json = r#"{"ttl":"15s","extra_options":{"foo":"bar","baz":"qux"}}"#;
+        let parsed = SchemaNameValue::try_from_raw_value(json.as_bytes()).unwrap();
+        let mut expected_options = BTreeMap::new();
+        expected_options.insert("foo".to_string(), "bar".to_string());
+        expected_options.insert("baz".to_string(), "qux".to_string());
+        assert_eq!(
+            parsed,
+            Some(SchemaNameValue {
+                ttl: Some(Duration::from_secs(15).into()),
+                extra_options: expected_options,
+            })
+        );
+    }
+
     #[tokio::test]
     async fn test_key_exist() {
         let manager = SchemaManager::new(Arc::new(MemoryKvBackend::default()));
@@ -396,6 +500,7 @@ mod tests {
         let current_schema_value = manager.get(schema_key).await.unwrap().unwrap();
         let new_schema_value = SchemaNameValue {
             ttl: Some(Duration::from_secs(10).into()),
+            ..Default::default()
         };
         manager
             .update(schema_key, &current_schema_value, &new_schema_value)
@@ -410,9 +515,11 @@ mod tests {
 
         let new_schema_value = SchemaNameValue {
             ttl: Some(Duration::from_secs(40).into()),
+            ..Default::default()
         };
         let incorrect_schema_value = SchemaNameValue {
             ttl: Some(Duration::from_secs(20).into()),
+            ..Default::default()
         }
         .try_as_raw_value()
         .unwrap();
@@ -425,7 +532,10 @@ mod tests {
             .unwrap_err();
 
         let current_schema_value = manager.get(schema_key).await.unwrap().unwrap();
-        let new_schema_value = SchemaNameValue { ttl: None };
+        let new_schema_value = SchemaNameValue {
+            ttl: None,
+            ..Default::default()
+        };
         manager
             .update(schema_key, &current_schema_value, &new_schema_value)
             .await

src/common/meta/src/key/table_name.rs

@@ -103,6 +103,26 @@ pub fn table_decoder(kv: KeyValue) -> Result<(String, TableNameValue)> {
     Ok((table_name_key.table.to_string(), table_name_value))
 }
 
+impl<'a> From<&TableReference<'a>> for TableNameKey<'a> {
+    fn from(value: &TableReference<'a>) -> Self {
+        Self {
+            catalog: value.catalog,
+            schema: value.schema,
+            table: value.table,
+        }
+    }
+}
+
+impl<'a> From<TableReference<'a>> for TableNameKey<'a> {
+    fn from(value: TableReference<'a>) -> Self {
+        Self {
+            catalog: value.catalog,
+            schema: value.schema,
+            table: value.table,
+        }
+    }
+}
+
 impl<'a> From<&'a TableName> for TableNameKey<'a> {
     fn from(value: &'a TableName) -> Self {
         Self {

src/common/meta/src/key/table_route.rs

@@ -711,6 +711,7 @@ mod tests {
                         name: "r1".to_string(),
                         partition: None,
                         attrs: Default::default(),
+                        partition_expr: Default::default(),
                     },
                     leader_peer: Some(Peer {
                         id: 2,
@@ -726,6 +727,7 @@ mod tests {
                         name: "r1".to_string(),
                         partition: None,
                         attrs: Default::default(),
+                        partition_expr: Default::default(),
                     },
                     leader_peer: Some(Peer {
                         id: 2,

src/common/meta/src/lib.rs

@@ -32,6 +32,7 @@ pub mod key;
 pub mod kv_backend;
 pub mod leadership_notifier;
 pub mod lock_key;
+pub mod maintenance;
 pub mod metrics;
 pub mod node_expiry_listener;
 pub mod node_manager;

src/common/meta/src/maintenance.rs

@@ -0,0 +1,15 @@
+// Copyright 2023 Greptime Team
+//
+// 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(crate) mod reconcile_table;

src/common/meta/src/maintenance/reconcile_table.rs

@@ -0,0 +1,17 @@
+// Copyright 2023 Greptime Team
+//
+// 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.
+
+// TODO(weny): Remove it
+#[allow(dead_code)]
+pub(crate) mod utils;