chore: bump version to 0.9.5 (#4853 )

fix: flush metric metadata region (#4852 )
* fix: flush metric metadata region * chore: apply suggestions from CR
2025-12-25 07:30:02 +00:00 · 2024-10-18 08:07:13 +00:00 · 2024-10-18 07:21:35 +00:00 · 2024-10-18 06:31:02 +00:00 · 2024-10-18 06:26:41 +00:00 · 2024-10-18 04:10:22 +00:00
130 changed files with 7393 additions and 3374 deletions
--- a/Cargo.lock
+++ b/Cargo.lock
@@ -214,7 +214,7 @@ checksum = "d301b3b94cb4b2f23d7917810addbbaff90738e0ca2be692bd027e70d7e0330c"

 [[package]]
 name = "api"
-version = "0.9.3"
+version = "0.9.5"
 dependencies = [
 "common-base",
 "common-decimal",
@@ -230,6 +230,15 @@ dependencies = [
 "tonic-build",
 ]

+[[package]]
+name = "approx"
+version = "0.4.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "3f2a05fd1bd10b2527e20a2cd32d8873d115b8b39fe219ee25f42a8aca6ba278"
+dependencies = [
+ "num-traits",
+]
+
 [[package]]
 name = "approx"
 version = "0.5.1"
@@ -766,7 +775,7 @@ dependencies = [

 [[package]]
 name = "auth"
-version = "0.9.3"
+version = "0.9.5"
 dependencies = [
 "api",
 "async-trait",
@@ -985,6 +994,7 @@ dependencies = [
 "num-bigint",
 "num-integer",
 "num-traits",
+ "serde",
 ]

 [[package]]
@@ -1375,7 +1385,7 @@ dependencies = [

 [[package]]
 name = "cache"
-version = "0.9.3"
+version = "0.9.5"
 dependencies = [
 "catalog",
 "common-error",
@@ -1383,7 +1393,7 @@ dependencies = [
 "common-meta",
 "moka",
 "snafu 0.8.5",
- "substrait 0.9.3",
+ "substrait 0.9.5",
 ]

 [[package]]
@@ -1410,7 +1420,7 @@ checksum = "37b2a672a2cb129a2e41c10b1224bb368f9f37a2b16b612598138befd7b37eb5"

 [[package]]
 name = "catalog"
-version = "0.9.3"
+version = "0.9.5"
 dependencies = [
 "api",
 "arrow",
@@ -1548,6 +1558,16 @@ dependencies = [
 "vob",
 ]

+[[package]]
+name = "cgmath"
+version = "0.18.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "1a98d30140e3296250832bbaaff83b27dcd6fa3cc70fb6f1f3e5c9c0023b5317"
+dependencies = [
+ "approx 0.4.0",
+ "num-traits",
+]
+
 [[package]]
 name = "chrono"
 version = "0.4.38"
@@ -1739,7 +1759,7 @@ checksum = "1462739cb27611015575c0c11df5df7601141071f07518d56fcc1be504cbec97"

 [[package]]
 name = "client"
-version = "0.9.3"
+version = "0.9.5"
 dependencies = [
 "api",
 "arc-swap",
@@ -1769,7 +1789,7 @@ dependencies = [
 "serde_json",
 "snafu 0.8.5",
 "substrait 0.37.3",
- "substrait 0.9.3",
+ "substrait 0.9.5",
 "tokio",
 "tokio-stream",
 "tonic 0.11.0",
@@ -1799,7 +1819,7 @@ dependencies = [

 [[package]]
 name = "cmd"
-version = "0.9.3"
+version = "0.9.5"
 dependencies = [
 "async-trait",
 "auth",
@@ -1856,7 +1876,7 @@ dependencies = [
 "similar-asserts",
 "snafu 0.8.5",
 "store-api",
- "substrait 0.9.3",
+ "substrait 0.9.5",
 "table",
 "temp-env",
 "tempfile",
@@ -1902,7 +1922,7 @@ checksum = "55b672471b4e9f9e95499ea597ff64941a309b2cdbffcc46f2cc5e2d971fd335"

 [[package]]
 name = "common-base"
-version = "0.9.3"
+version = "0.9.5"
 dependencies = [
 "anymap2",
 "async-trait",
@@ -1920,7 +1940,7 @@ dependencies = [

 [[package]]
 name = "common-catalog"
-version = "0.9.3"
+version = "0.9.5"
 dependencies = [
 "chrono",
 "common-error",
@@ -1931,7 +1951,7 @@ dependencies = [

 [[package]]
 name = "common-config"
-version = "0.9.3"
+version = "0.9.5"
 dependencies = [
 "common-base",
 "common-error",
@@ -1954,7 +1974,7 @@ dependencies = [

 [[package]]
 name = "common-datasource"
-version = "0.9.3"
+version = "0.9.5"
 dependencies = [
 "arrow",
 "arrow-schema",
@@ -1991,7 +2011,7 @@ dependencies = [

 [[package]]
 name = "common-decimal"
-version = "0.9.3"
+version = "0.9.5"
 dependencies = [
 "bigdecimal 0.4.5",
 "common-error",
@@ -2004,7 +2024,7 @@ dependencies = [

 [[package]]
 name = "common-error"
-version = "0.9.3"
+version = "0.9.5"
 dependencies = [
 "snafu 0.8.5",
 "strum 0.25.0",
@@ -2013,7 +2033,7 @@ dependencies = [

 [[package]]
 name = "common-frontend"
-version = "0.9.3"
+version = "0.9.5"
 dependencies = [
 "api",
 "async-trait",
@@ -2028,7 +2048,7 @@ dependencies = [

 [[package]]
 name = "common-function"
-version = "0.9.3"
+version = "0.9.5"
 dependencies = [
 "api",
 "arc-swap",
@@ -2054,6 +2074,7 @@ dependencies = [
 "once_cell",
 "paste",
 "ron",
+ "s2",
 "serde",
 "serde_json",
 "session",
@@ -2067,7 +2088,7 @@ dependencies = [

 [[package]]
 name = "common-greptimedb-telemetry"
-version = "0.9.3"
+version = "0.9.5"
 dependencies = [
 "async-trait",
 "common-runtime",
@@ -2084,7 +2105,7 @@ dependencies = [

 [[package]]
 name = "common-grpc"
-version = "0.9.3"
+version = "0.9.5"
 dependencies = [
 "api",
 "arrow-flight",
@@ -2110,7 +2131,7 @@ dependencies = [

 [[package]]
 name = "common-grpc-expr"
-version = "0.9.3"
+version = "0.9.5"
 dependencies = [
 "api",
 "common-base",
@@ -2128,7 +2149,7 @@ dependencies = [

 [[package]]
 name = "common-macro"
-version = "0.9.3"
+version = "0.9.5"
 dependencies = [
 "arc-swap",
 "common-query",
@@ -2142,7 +2163,7 @@ dependencies = [

 [[package]]
 name = "common-mem-prof"
-version = "0.9.3"
+version = "0.9.5"
 dependencies = [
 "common-error",
 "common-macro",
@@ -2155,7 +2176,7 @@ dependencies = [

 [[package]]
 name = "common-meta"
-version = "0.9.3"
+version = "0.9.5"
 dependencies = [
 "anymap2",
 "api",
@@ -2212,11 +2233,23 @@ dependencies = [

 [[package]]
 name = "common-plugins"
-version = "0.9.3"
+version = "0.9.5"
+
+[[package]]
+name = "common-pprof"
+version = "0.9.5"
+dependencies = [
+ "common-error",
+ "common-macro",
+ "pprof",
+ "prost 0.12.6",
+ "snafu 0.8.5",
+ "tokio",
+]

 [[package]]
 name = "common-procedure"
-version = "0.9.3"
+version = "0.9.5"
 dependencies = [
 "async-stream",
 "async-trait",
@@ -2243,7 +2276,7 @@ dependencies = [

 [[package]]
 name = "common-procedure-test"
-version = "0.9.3"
+version = "0.9.5"
 dependencies = [
 "async-trait",
 "common-procedure",
@@ -2251,7 +2284,7 @@ dependencies = [

 [[package]]
 name = "common-query"
-version = "0.9.3"
+version = "0.9.5"
 dependencies = [
 "api",
 "async-trait",
@@ -2277,7 +2310,7 @@ dependencies = [

 [[package]]
 name = "common-recordbatch"
-version = "0.9.3"
+version = "0.9.5"
 dependencies = [
 "arc-swap",
 "common-error",
@@ -2296,7 +2329,7 @@ dependencies = [

 [[package]]
 name = "common-runtime"
-version = "0.9.3"
+version = "0.9.5"
 dependencies = [
 "async-trait",
 "common-error",
@@ -2318,7 +2351,7 @@ dependencies = [

 [[package]]
 name = "common-telemetry"
-version = "0.9.3"
+version = "0.9.5"
 dependencies = [
 "atty",
 "backtrace",
@@ -2346,7 +2379,7 @@ dependencies = [

 [[package]]
 name = "common-test-util"
-version = "0.9.3"
+version = "0.9.5"
 dependencies = [
 "client",
 "common-query",
@@ -2358,7 +2391,7 @@ dependencies = [

 [[package]]
 name = "common-time"
-version = "0.9.3"
+version = "0.9.5"
 dependencies = [
 "arrow",
 "chrono",
@@ -2374,7 +2407,7 @@ dependencies = [

 [[package]]
 name = "common-version"
-version = "0.9.3"
+version = "0.9.5"
 dependencies = [
 "build-data",
 "const_format",
@@ -2385,7 +2418,7 @@ dependencies = [

 [[package]]
 name = "common-wal"
-version = "0.9.3"
+version = "0.9.5"
 dependencies = [
 "common-base",
 "common-error",
@@ -3194,7 +3227,7 @@ dependencies = [

 [[package]]
 name = "datanode"
-version = "0.9.3"
+version = "0.9.5"
 dependencies = [
 "api",
 "arrow-flight",
@@ -3244,7 +3277,7 @@ dependencies = [
 "session",
 "snafu 0.8.5",
 "store-api",
- "substrait 0.9.3",
+ "substrait 0.9.5",
 "table",
 "tokio",
 "toml 0.8.19",
@@ -3253,7 +3286,7 @@ dependencies = [

 [[package]]
 name = "datatypes"
-version = "0.9.3"
+version = "0.9.5"
 dependencies = [
 "arrow",
 "arrow-array",
@@ -3859,7 +3892,7 @@ dependencies = [

 [[package]]
 name = "file-engine"
-version = "0.9.3"
+version = "0.9.5"
 dependencies = [
 "api",
 "async-trait",
@@ -3959,9 +3992,18 @@ version = "1.0.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "28a80e3145d8ad11ba0995949bbcf48b9df2be62772b3d351ef017dff6ecb853"

+[[package]]
+name = "float_extras"
+version = "0.1.6"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "b22b70f8649ea2315955f1a36d964b0e4da482dfaa5f0d04df0d1fb7c338ab7a"
+dependencies = [
+ "libc",
+]
+
 [[package]]
 name = "flow"
-version = "0.9.3"
+version = "0.9.5"
 dependencies = [
 "api",
 "arrow",
@@ -4018,7 +4060,7 @@ dependencies = [
 "snafu 0.8.5",
 "store-api",
 "strum 0.25.0",
- "substrait 0.9.3",
+ "substrait 0.9.5",
 "table",
 "tokio",
 "tonic 0.11.0",
@@ -4080,7 +4122,7 @@ checksum = "6c2141d6d6c8512188a7891b4b01590a45f6dac67afb4f255c4124dbb86d4eaa"

 [[package]]
 name = "frontend"
-version = "0.9.3"
+version = "0.9.5"
 dependencies = [
 "api",
 "arc-swap",
@@ -4389,7 +4431,7 @@ version = "0.7.13"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "9ff16065e5720f376fbced200a5ae0f47ace85fd70b7e54269790281353b6d61"
 dependencies = [
- "approx",
+ "approx 0.5.1",
 "num-traits",
 "serde",
 ]
@@ -4476,7 +4518,7 @@ dependencies = [
 [[package]]
 name = "greptime-proto"
 version = "0.1.0"
-source = "git+https://github.com/GreptimeTeam/greptime-proto.git?rev=0b4f7c8ab06399f6b90e1626e8d5b9697cb33bb9#0b4f7c8ab06399f6b90e1626e8d5b9697cb33bb9"
+source = "git+https://github.com/GreptimeTeam/greptime-proto.git?rev=b4d301184eb0d01fd4d1042fcc7c5dfb54f3c1e3#b4d301184eb0d01fd4d1042fcc7c5dfb54f3c1e3"
 dependencies = [
 "prost 0.12.6",
 "serde",
@@ -5128,7 +5170,7 @@ dependencies = [

 [[package]]
 name = "index"
-version = "0.9.3"
+version = "0.9.5"
 dependencies = [
 "async-trait",
 "asynchronous-codec",
@@ -5959,7 +6001,7 @@ checksum = "a7a70ba024b9dc04c27ea2f0c0548feb474ec5c54bba33a7f72f873a39d07b24"

 [[package]]
 name = "log-store"
-version = "0.9.3"
+version = "0.9.5"
 dependencies = [
 "async-stream",
 "async-trait",
@@ -6279,7 +6321,7 @@ dependencies = [

 [[package]]
 name = "meta-client"
-version = "0.9.3"
+version = "0.9.5"
 dependencies = [
 "api",
 "async-trait",
@@ -6305,7 +6347,7 @@ dependencies = [

 [[package]]
 name = "meta-srv"
-version = "0.9.3"
+version = "0.9.5"
 dependencies = [
 "api",
 "async-trait",
@@ -6383,7 +6425,7 @@ dependencies = [

 [[package]]
 name = "metric-engine"
-version = "0.9.3"
+version = "0.9.5"
 dependencies = [
 "api",
 "aquamarine",
@@ -6486,7 +6528,7 @@ dependencies = [

 [[package]]
 name = "mito2"
-version = "0.9.3"
+version = "0.9.5"
 dependencies = [
 "api",
 "aquamarine",
@@ -6871,7 +6913,7 @@ version = "0.29.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "d506eb7e08d6329505faa8a3a00a5dcc6de9f76e0c77e4b75763ae3c770831ff"
 dependencies = [
- "approx",
+ "approx 0.5.1",
 "matrixmultiply",
 "nalgebra-macros",
 "num-complex",
@@ -7222,7 +7264,7 @@ dependencies = [

 [[package]]
 name = "object-store"
-version = "0.9.3"
+version = "0.9.5"
 dependencies = [
 "anyhow",
 "bytes",
@@ -7507,12 +7549,13 @@ dependencies = [
 "ordered-float 4.3.0",
 "percent-encoding",
 "rand",
+ "serde_json",
 "thiserror",
 ]

 [[package]]
 name = "operator"
-version = "0.9.3"
+version = "0.9.5"
 dependencies = [
 "api",
 "async-trait",
@@ -7557,7 +7600,7 @@ dependencies = [
 "sql",
 "sqlparser 0.45.0 (git+https://github.com/GreptimeTeam/sqlparser-rs.git?rev=54a267ac89c09b11c0c88934690530807185d3e7)",
 "store-api",
- "substrait 0.9.3",
+ "substrait 0.9.5",
 "table",
 "tokio",
 "tokio-util",
@@ -7807,7 +7850,7 @@ dependencies = [

 [[package]]
 name = "partition"
-version = "0.9.3"
+version = "0.9.5"
 dependencies = [
 "api",
 "async-trait",
@@ -8108,7 +8151,7 @@ checksum = "8b870d8c151b6f2fb93e84a13146138f05d02ed11c7e7c54f8826aaaf7c9f184"

 [[package]]
 name = "pipeline"
-version = "0.9.3"
+version = "0.9.5"
 dependencies = [
 "ahash 0.8.11",
 "api",
@@ -8270,7 +8313,7 @@ dependencies = [

 [[package]]
 name = "plugins"
-version = "0.9.3"
+version = "0.9.5"
 dependencies = [
 "auth",
 "common-base",
@@ -8544,7 +8587,7 @@ dependencies = [

 [[package]]
 name = "promql"
-version = "0.9.3"
+version = "0.9.5"
 dependencies = [
 "ahash 0.8.11",
 "async-trait",
@@ -8570,11 +8613,12 @@ dependencies = [

 [[package]]
 name = "promql-parser"
-version = "0.4.0"
+version = "0.4.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "007a331efb31f6ddb644590ef22359c9469784931162aad92599e34bcfa66583"
+checksum = "0c1ad4a4cfa84ec4aa5831c82e57af0a3faf3f0af83bee13fa1390b2d0a32dc9"
 dependencies = [
 "cfgrammar",
+ "chrono",
 "lazy_static",
 "lrlex",
 "lrpar",
@@ -8779,7 +8823,7 @@ dependencies = [

 [[package]]
 name = "puffin"
-version = "0.9.3"
+version = "0.9.5"
 dependencies = [
 "async-compression 0.4.13",
 "async-trait",
@@ -8901,7 +8945,7 @@ dependencies = [

 [[package]]
 name = "query"
-version = "0.9.3"
+version = "0.9.5"
 dependencies = [
 "ahash 0.8.11",
 "api",
@@ -8936,6 +8980,7 @@ dependencies = [
 "datafusion-physical-expr",
 "datafusion-sql",
 "datatypes",
+ "fastrand",
 "format_num",
 "futures",
 "futures-util",
@@ -8950,12 +8995,15 @@ dependencies = [
 "object-store",
 "once_cell",
 "paste",
+ "pretty_assertions",
 "prometheus",
 "promql",
 "promql-parser",
 "prost 0.12.6",
 "rand",
 "regex",
+ "serde",
+ "serde_json",
 "session",
 "snafu 0.8.5",
 "sql",
@@ -8964,7 +9012,7 @@ dependencies = [
 "stats-cli",
 "store-api",
 "streaming-stats",
- "substrait 0.9.3",
+ "substrait 0.9.5",
 "table",
 "tokio",
 "tokio-stream",
@@ -10262,6 +10310,20 @@ version = "1.0.18"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "f3cb5ba0dc43242ce17de99c180e96db90b235b8a9fdc9543c96d2209116bd9f"

+[[package]]
+name = "s2"
+version = "0.0.12"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "cc7fbc04bb52c40b5f48c9bb2d2961375301916e0c25d9f373750654d588cd5c"
+dependencies = [
+ "bigdecimal 0.3.1",
+ "cgmath",
+ "float_extras",
+ "lazy_static",
+ "libm",
+ "serde",
+]
+
 [[package]]
 name = "safe-proc-macro2"
 version = "1.0.67"
@@ -10384,7 +10446,7 @@ checksum = "94143f37725109f92c262ed2cf5e59bce7498c01bcc1502d7b9afe439a4e9f49"

 [[package]]
 name = "script"
-version = "0.9.3"
+version = "0.9.5"
 dependencies = [
 "api",
 "arc-swap",
@@ -10678,8 +10740,9 @@ dependencies = [

 [[package]]
 name = "servers"
-version = "0.9.3"
+version = "0.9.5"
 dependencies = [
+ "ahash 0.8.11",
 "aide",
 "api",
 "arrow",
@@ -10705,6 +10768,7 @@ dependencies = [
 "common-mem-prof",
 "common-meta",
 "common-plugins",
+ "common-pprof",
 "common-query",
 "common-recordbatch",
 "common-runtime",
@@ -10787,7 +10851,7 @@ dependencies = [

 [[package]]
 name = "session"
-version = "0.9.3"
+version = "0.9.5"
 dependencies = [
 "api",
 "arc-swap",
@@ -10899,7 +10963,7 @@ version = "0.6.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "f0b7840f121a46d63066ee7a99fc81dcabbc6105e437cae43528cea199b5a05f"
 dependencies = [
- "approx",
+ "approx 0.5.1",
 "num-complex",
 "num-traits",
 "paste",
@@ -11108,7 +11172,7 @@ dependencies = [

 [[package]]
 name = "sql"
-version = "0.9.3"
+version = "0.9.5"
 dependencies = [
 "api",
 "chrono",
@@ -11169,7 +11233,7 @@ dependencies = [

 [[package]]
 name = "sqlness-runner"
-version = "0.9.3"
+version = "0.9.5"
 dependencies = [
 "async-trait",
 "clap 4.5.19",
@@ -11370,7 +11434,7 @@ version = "0.16.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "b35a062dbadac17a42e0fc64c27f419b25d6fae98572eb43c8814c9e873d7721"
 dependencies = [
- "approx",
+ "approx 0.5.1",
 "lazy_static",
 "nalgebra",
 "num-traits",
@@ -11389,7 +11453,7 @@ dependencies = [

 [[package]]
 name = "store-api"
-version = "0.9.3"
+version = "0.9.5"
 dependencies = [
 "api",
 "aquamarine",
@@ -11558,7 +11622,7 @@ dependencies = [

 [[package]]
 name = "substrait"
-version = "0.9.3"
+version = "0.9.5"
 dependencies = [
 "async-trait",
 "bytes",
@@ -11757,7 +11821,7 @@ dependencies = [

 [[package]]
 name = "table"
-version = "0.9.3"
+version = "0.9.5"
 dependencies = [
 "api",
 "async-trait",
@@ -12023,7 +12087,7 @@ checksum = "3369f5ac52d5eb6ab48c6b4ffdc8efbcad6b89c765749064ba298f2c68a16a76"

 [[package]]
 name = "tests-fuzz"
-version = "0.9.3"
+version = "0.9.5"
 dependencies = [
 "arbitrary",
 "async-trait",
@@ -12065,7 +12129,7 @@ dependencies = [

 [[package]]
 name = "tests-integration"
-version = "0.9.3"
+version = "0.9.5"
 dependencies = [
 "api",
 "arrow-flight",
@@ -12127,7 +12191,7 @@ dependencies = [
 "sql",
 "sqlx",
 "store-api",
- "substrait 0.9.3",
+ "substrait 0.9.5",
 "table",
 "tempfile",
 "time",
--- a/Cargo.toml
+++ b/Cargo.toml
@@ -20,6 +20,7 @@ members = [
    "src/common/mem-prof",
    "src/common/meta",
    "src/common/plugins",
+    "src/common/pprof",
    "src/common/procedure",
    "src/common/procedure-test",
    "src/common/query",
@@ -64,7 +65,7 @@ members = [
 resolver = "2"

 [workspace.package]
-version = "0.9.3"
+version = "0.9.5"
 edition = "2021"
 license = "Apache-2.0"

@@ -120,7 +121,7 @@ etcd-client = { version = "0.13" }
 fst = "0.4.7"
 futures = "0.3"
 futures-util = "0.3"
-greptime-proto = { git = "https://github.com/GreptimeTeam/greptime-proto.git", rev = "0b4f7c8ab06399f6b90e1626e8d5b9697cb33bb9" }
+greptime-proto = { git = "https://github.com/GreptimeTeam/greptime-proto.git", rev = "b4d301184eb0d01fd4d1042fcc7c5dfb54f3c1e3" }
 humantime = "2.1"
 humantime-serde = "1.1"
 itertools = "0.10"
@@ -137,12 +138,13 @@ opentelemetry-proto = { version = "0.5", features = [
    "metrics",
    "trace",
    "with-serde",
+    "logs",
 ] }
 parquet = { version = "51.0.0", default-features = false, features = ["arrow", "async", "object_store"] }
 paste = "1.0"
 pin-project = "1.0"
 prometheus = { version = "0.13.3", features = ["process"] }
-promql-parser = { version = "0.4" }
+promql-parser = { version = "0.4.1" }
 prost = "0.12"
 raft-engine = { version = "0.4.1", default-features = false }
 rand = "0.8"
@@ -208,6 +210,7 @@ common-macro = { path = "src/common/macro" }
 common-mem-prof = { path = "src/common/mem-prof" }
 common-meta = { path = "src/common/meta" }
 common-plugins = { path = "src/common/plugins" }
+common-pprof = { path = "src/common/pprof" }
 common-procedure = { path = "src/common/procedure" }
 common-procedure-test = { path = "src/common/procedure-test" }
 common-query = { path = "src/common/query" }
--- a/config/config.md
+++ b/config/config.md
@@ -83,6 +83,7 @@
 | `wal.backoff_max` | String | `10s` | The maximum backoff delay.<br/>**It's only used when the provider is `kafka`**. |
 | `wal.backoff_base` | Integer | `2` | The exponential backoff rate, i.e. next backoff = base * current backoff.<br/>**It's only used when the provider is `kafka`**. |
 | `wal.backoff_deadline` | String | `5mins` | The deadline of retries.<br/>**It's only used when the provider is `kafka`**. |
+| `wal.overwrite_entry_start_id` | Bool | `false` | Ignore missing entries during read WAL.<br/>**It's only used when the provider is `kafka`**.<br/><br/>This option ensures that when Kafka messages are deleted, the system <br/>can still successfully replay memtable data without throwing an <br/>out-of-range error. <br/>However, enabling this option might lead to unexpected data loss, <br/>as the system will skip over missing entries instead of treating <br/>them as critical errors. |
 | `metadata_store` | -- | -- | Metadata storage options. |
 | `metadata_store.file_size` | String | `256MB` | Kv file size in bytes. |
 | `metadata_store.purge_threshold` | String | `4GB` | Kv purge threshold. |
@@ -409,6 +410,7 @@
 | `wal.backoff_deadline` | String | `5mins` | The deadline of retries.<br/>**It's only used when the provider is `kafka`**. |
 | `wal.create_index` | Bool | `true` | Whether to enable WAL index creation.<br/>**It's only used when the provider is `kafka`**. |
 | `wal.dump_index_interval` | String | `60s` | The interval for dumping WAL indexes.<br/>**It's only used when the provider is `kafka`**. |
+| `wal.overwrite_entry_start_id` | Bool | `false` | Ignore missing entries during read WAL.<br/>**It's only used when the provider is `kafka`**.<br/><br/>This option ensures that when Kafka messages are deleted, the system <br/>can still successfully replay memtable data without throwing an <br/>out-of-range error. <br/>However, enabling this option might lead to unexpected data loss, <br/>as the system will skip over missing entries instead of treating <br/>them as critical errors. |
 | `storage` | -- | -- | The data storage options. |
 | `storage.data_home` | String | `/tmp/greptimedb/` | The working home directory. |
 | `storage.type` | String | `File` | The storage type used to store the data.<br/>- `File`: the data is stored in the local file system.<br/>- `S3`: the data is stored in the S3 object storage.<br/>- `Gcs`: the data is stored in the Google Cloud Storage.<br/>- `Azblob`: the data is stored in the Azure Blob Storage.<br/>- `Oss`: the data is stored in the Aliyun OSS. |
--- a/config/datanode.example.toml
+++ b/config/datanode.example.toml
@@ -213,6 +213,17 @@ create_index = true
 ## **It's only used when the provider is `kafka`**.
 dump_index_interval = "60s"

+## Ignore missing entries during read WAL.
+## **It's only used when the provider is `kafka`**.
+## 
+## This option ensures that when Kafka messages are deleted, the system 
+## can still successfully replay memtable data without throwing an 
+## out-of-range error. 
+## However, enabling this option might lead to unexpected data loss, 
+## as the system will skip over missing entries instead of treating 
+## them as critical errors.
+overwrite_entry_start_id = false
+
 # The Kafka SASL configuration.
 # **It's only used when the provider is `kafka`**.
 # Available SASL mechanisms:
--- a/config/standalone.example.toml
+++ b/config/standalone.example.toml
@@ -237,6 +237,17 @@ backoff_base = 2
 ## **It's only used when the provider is `kafka`**.
 backoff_deadline = "5mins"

+## Ignore missing entries during read WAL.
+## **It's only used when the provider is `kafka`**.
+## 
+## This option ensures that when Kafka messages are deleted, the system 
+## can still successfully replay memtable data without throwing an 
+## out-of-range error. 
+## However, enabling this option might lead to unexpected data loss, 
+## as the system will skip over missing entries instead of treating 
+## them as critical errors.
+overwrite_entry_start_id = false
+
 # The Kafka SASL configuration.
 # **It's only used when the provider is `kafka`**.
 # Available SASL mechanisms:
--- a/docs/benchmarks/log/README.md
+++ b/docs/benchmarks/log/README.md
@@ -48,4 +48,4 @@ Please refer to [SQL query](./query.sql) for GreptimeDB and Clickhouse, and [que

 ## Addition
 - You can tune GreptimeDB's configuration to get better performance.
- You can setup GreptimeDB to use S3 as storage, see [here](https://docs.greptime.com/user-guide/operations/configuration/#storage-options).
+- You can setup GreptimeDB to use S3 as storage, see [here](https://docs.greptime.com/user-guide/deployments/configuration#storage-options).
--- a/docs/how-to/how-to-profile-cpu.md
+++ b/docs/how-to/how-to-profile-cpu.md
@@ -1,11 +1,5 @@
 # Profiling CPU

-## Build GreptimeDB with `pprof` feature
-
-```bash
-cargo build --features=pprof
-```
-
 ## HTTP API
 Sample at 99 Hertz, for 5 seconds, output report in [protobuf format](https://github.com/google/pprof/blob/master/proto/profile.proto).
 ```bash
--- a/docs/how-to/how-to-profile-memory.md
+++ b/docs/how-to/how-to-profile-memory.md
@@ -18,12 +18,6 @@ sudo apt install libjemalloc-dev
 curl https://raw.githubusercontent.com/brendangregg/FlameGraph/master/flamegraph.pl > ./flamegraph.pl
 ```

-### Build GreptimeDB with `mem-prof` feature.
-
-```bash
-cargo build --features=mem-prof
-```
-
 ## Profiling

 Start GreptimeDB instance with environment variables:
--- a/src/api/src/helper.rs
+++ b/src/api/src/helper.rs
@@ -17,10 +17,11 @@ use std::sync::Arc;
 use common_base::BitVec;
 use common_decimal::decimal128::{DECIMAL128_DEFAULT_SCALE, DECIMAL128_MAX_PRECISION};
 use common_decimal::Decimal128;
-use common_time::interval::IntervalUnit;
 use common_time::time::Time;
 use common_time::timestamp::TimeUnit;
-use common_time::{Date, DateTime, Interval, Timestamp};
+use common_time::{
+    Date, DateTime, IntervalDayTime, IntervalMonthDayNano, IntervalYearMonth, Timestamp,
+};
 use datatypes::prelude::{ConcreteDataType, ValueRef};
 use datatypes::scalars::ScalarVector;
 use datatypes::types::{
@@ -456,13 +457,11 @@ pub fn push_vals(column: &mut Column, origin_count: usize, vector: VectorRef) {
            TimeUnit::Microsecond => values.time_microsecond_values.push(val.value()),
            TimeUnit::Nanosecond => values.time_nanosecond_values.push(val.value()),
        },
-        Value::Interval(val) => match val.unit() {
-            IntervalUnit::YearMonth => values.interval_year_month_values.push(val.to_i32()),
-            IntervalUnit::DayTime => values.interval_day_time_values.push(val.to_i64()),
-            IntervalUnit::MonthDayNano => values
-                .interval_month_day_nano_values
-                .push(convert_i128_to_interval(val.to_i128())),
-        },
+        Value::IntervalYearMonth(val) => values.interval_year_month_values.push(val.to_i32()),
+        Value::IntervalDayTime(val) => values.interval_day_time_values.push(val.to_i64()),
+        Value::IntervalMonthDayNano(val) => values
+            .interval_month_day_nano_values
+            .push(convert_month_day_nano_to_pb(val)),
        Value::Decimal128(val) => values.decimal128_values.push(convert_to_pb_decimal128(val)),
        Value::List(_) | Value::Duration(_) => unreachable!(),
    });
@@ -507,14 +506,12 @@ fn ddl_request_type(request: &DdlRequest) -> &'static str {
    }
 }

-/// Converts an i128 value to google protobuf type [IntervalMonthDayNano].
-pub fn convert_i128_to_interval(v: i128) -> v1::IntervalMonthDayNano {
-    let interval = Interval::from_i128(v);
-    let (months, days, nanoseconds) = interval.to_month_day_nano();
+/// Converts an interval to google protobuf type [IntervalMonthDayNano].
+pub fn convert_month_day_nano_to_pb(v: IntervalMonthDayNano) -> v1::IntervalMonthDayNano {
    v1::IntervalMonthDayNano {
-        months,
-        days,
-        nanoseconds,
+        months: v.months,
+        days: v.days,
+        nanoseconds: v.nanoseconds,
    }
 }

@@ -562,11 +559,15 @@ pub fn pb_value_to_value_ref<'a>(
        ValueData::TimeMillisecondValue(t) => ValueRef::Time(Time::new_millisecond(*t)),
        ValueData::TimeMicrosecondValue(t) => ValueRef::Time(Time::new_microsecond(*t)),
        ValueData::TimeNanosecondValue(t) => ValueRef::Time(Time::new_nanosecond(*t)),
-        ValueData::IntervalYearMonthValue(v) => ValueRef::Interval(Interval::from_i32(*v)),
-        ValueData::IntervalDayTimeValue(v) => ValueRef::Interval(Interval::from_i64(*v)),
+        ValueData::IntervalYearMonthValue(v) => {
+            ValueRef::IntervalYearMonth(IntervalYearMonth::from_i32(*v))
+        }
+        ValueData::IntervalDayTimeValue(v) => {
+            ValueRef::IntervalDayTime(IntervalDayTime::from_i64(*v))
+        }
        ValueData::IntervalMonthDayNanoValue(v) => {
-            let interval = Interval::from_month_day_nano(v.months, v.days, v.nanoseconds);
-            ValueRef::Interval(interval)
+            let interval = IntervalMonthDayNano::new(v.months, v.days, v.nanoseconds);
+            ValueRef::IntervalMonthDayNano(interval)
        }
        ValueData::Decimal128Value(v) => {
            // get precision and scale from datatype_extension
@@ -657,7 +658,7 @@ pub fn pb_values_to_vector_ref(data_type: &ConcreteDataType, values: Values) ->
            IntervalType::MonthDayNano(_) => {
                Arc::new(IntervalMonthDayNanoVector::from_iter_values(
                    values.interval_month_day_nano_values.iter().map(|x| {
-                        Interval::from_month_day_nano(x.months, x.days, x.nanoseconds).to_i128()
+                        IntervalMonthDayNano::new(x.months, x.days, x.nanoseconds).to_i128()
                    }),
                ))
            }
@@ -802,18 +803,18 @@ pub fn pb_values_to_values(data_type: &ConcreteDataType, values: Values) -> Vec<
        ConcreteDataType::Interval(IntervalType::YearMonth(_)) => values
            .interval_year_month_values
            .into_iter()
-            .map(|v| Value::Interval(Interval::from_i32(v)))
+            .map(|v| Value::IntervalYearMonth(IntervalYearMonth::from_i32(v)))
            .collect(),
        ConcreteDataType::Interval(IntervalType::DayTime(_)) => values
            .interval_day_time_values
            .into_iter()
-            .map(|v| Value::Interval(Interval::from_i64(v)))
+            .map(|v| Value::IntervalDayTime(IntervalDayTime::from_i64(v)))
            .collect(),
        ConcreteDataType::Interval(IntervalType::MonthDayNano(_)) => values
            .interval_month_day_nano_values
            .into_iter()
            .map(|v| {
-                Value::Interval(Interval::from_month_day_nano(
+                Value::IntervalMonthDayNano(IntervalMonthDayNano::new(
                    v.months,
                    v.days,
                    v.nanoseconds,
@@ -941,18 +942,16 @@ pub fn to_proto_value(value: Value) -> Option<v1::Value> {
                value_data: Some(ValueData::TimeNanosecondValue(v.value())),
            },
        },
-        Value::Interval(v) => match v.unit() {
-            IntervalUnit::YearMonth => v1::Value {
-                value_data: Some(ValueData::IntervalYearMonthValue(v.to_i32())),
-            },
-            IntervalUnit::DayTime => v1::Value {
-                value_data: Some(ValueData::IntervalDayTimeValue(v.to_i64())),
-            },
-            IntervalUnit::MonthDayNano => v1::Value {
-                value_data: Some(ValueData::IntervalMonthDayNanoValue(
-                    convert_i128_to_interval(v.to_i128()),
-                )),
-            },
+        Value::IntervalYearMonth(v) => v1::Value {
+            value_data: Some(ValueData::IntervalYearMonthValue(v.to_i32())),
+        },
+        Value::IntervalDayTime(v) => v1::Value {
+            value_data: Some(ValueData::IntervalDayTimeValue(v.to_i64())),
+        },
+        Value::IntervalMonthDayNano(v) => v1::Value {
+            value_data: Some(ValueData::IntervalMonthDayNanoValue(
+                convert_month_day_nano_to_pb(v),
+            )),
        },
        Value::Decimal128(v) => v1::Value {
            value_data: Some(ValueData::Decimal128Value(convert_to_pb_decimal128(v))),
@@ -1044,13 +1043,11 @@ pub fn value_to_grpc_value(value: Value) -> GrpcValue {
                TimeUnit::Microsecond => ValueData::TimeMicrosecondValue(v.value()),
                TimeUnit::Nanosecond => ValueData::TimeNanosecondValue(v.value()),
            }),
-            Value::Interval(v) => Some(match v.unit() {
-                IntervalUnit::YearMonth => ValueData::IntervalYearMonthValue(v.to_i32()),
-                IntervalUnit::DayTime => ValueData::IntervalDayTimeValue(v.to_i64()),
-                IntervalUnit::MonthDayNano => {
-                    ValueData::IntervalMonthDayNanoValue(convert_i128_to_interval(v.to_i128()))
-                }
-            }),
+            Value::IntervalYearMonth(v) => Some(ValueData::IntervalYearMonthValue(v.to_i32())),
+            Value::IntervalDayTime(v) => Some(ValueData::IntervalDayTimeValue(v.to_i64())),
+            Value::IntervalMonthDayNano(v) => Some(ValueData::IntervalMonthDayNanoValue(
+                convert_month_day_nano_to_pb(v),
+            )),
            Value::Decimal128(v) => Some(ValueData::Decimal128Value(convert_to_pb_decimal128(v))),
            Value::List(_) | Value::Duration(_) => unreachable!(),
        },
@@ -1061,6 +1058,7 @@ pub fn value_to_grpc_value(value: Value) -> GrpcValue {
 mod tests {
    use std::sync::Arc;

+    use common_time::interval::IntervalUnit;
    use datatypes::types::{
        Int32Type, IntervalDayTimeType, IntervalMonthDayNanoType, IntervalYearMonthType,
        TimeMillisecondType, TimeSecondType, TimestampMillisecondType, TimestampSecondType,
@@ -1506,11 +1504,11 @@ mod tests {

    #[test]
    fn test_convert_i128_to_interval() {
-        let i128_val = 3000;
-        let interval = convert_i128_to_interval(i128_val);
+        let i128_val = 3;
+        let interval = convert_month_day_nano_to_pb(IntervalMonthDayNano::from_i128(i128_val));
        assert_eq!(interval.months, 0);
        assert_eq!(interval.days, 0);
-        assert_eq!(interval.nanoseconds, 3000);
+        assert_eq!(interval.nanoseconds, 3);
    }

    #[test]
@@ -1590,9 +1588,9 @@ mod tests {
            },
        );
        let expect = vec![
-            Value::Interval(Interval::from_year_month(1_i32)),
-            Value::Interval(Interval::from_year_month(2_i32)),
-            Value::Interval(Interval::from_year_month(3_i32)),
+            Value::IntervalYearMonth(IntervalYearMonth::new(1_i32)),
+            Value::IntervalYearMonth(IntervalYearMonth::new(2_i32)),
+            Value::IntervalYearMonth(IntervalYearMonth::new(3_i32)),
        ];
        assert_eq!(expect, actual);

@@ -1605,9 +1603,9 @@ mod tests {
            },
        );
        let expect = vec![
-            Value::Interval(Interval::from_i64(1_i64)),
-            Value::Interval(Interval::from_i64(2_i64)),
-            Value::Interval(Interval::from_i64(3_i64)),
+            Value::IntervalDayTime(IntervalDayTime::from_i64(1_i64)),
+            Value::IntervalDayTime(IntervalDayTime::from_i64(2_i64)),
+            Value::IntervalDayTime(IntervalDayTime::from_i64(3_i64)),
        ];
        assert_eq!(expect, actual);

@@ -1636,9 +1634,9 @@ mod tests {
            },
        );
        let expect = vec![
-            Value::Interval(Interval::from_month_day_nano(1, 2, 3)),
-            Value::Interval(Interval::from_month_day_nano(5, 6, 7)),
-            Value::Interval(Interval::from_month_day_nano(9, 10, 11)),
+            Value::IntervalMonthDayNano(IntervalMonthDayNano::new(1, 2, 3)),
+            Value::IntervalMonthDayNano(IntervalMonthDayNano::new(5, 6, 7)),
+            Value::IntervalMonthDayNano(IntervalMonthDayNano::new(9, 10, 11)),
        ];
        assert_eq!(expect, actual);
    }
--- a/src/cmd/Cargo.toml
+++ b/src/cmd/Cargo.toml
@@ -10,7 +10,7 @@ name = "greptime"
 path = "src/bin/greptime.rs"

 [features]
-default = ["python"]
+default = ["python", "servers/pprof", "servers/mem-prof"]
 tokio-console = ["common-telemetry/tokio-console"]
 python = ["frontend/python"]

--- a/src/cmd/src/metasrv.rs
+++ b/src/cmd/src/metasrv.rs
@@ -48,6 +48,10 @@ impl Instance {
            _guard: guard,
        }
    }
+
+    pub fn get_inner(&self) -> &MetasrvInstance {
+        &self.instance
+    }
 }

 #[async_trait]
@@ -86,6 +90,14 @@ impl Command {
    pub fn load_options(&self, global_options: &GlobalOptions) -> Result<MetasrvOptions> {
        self.subcmd.load_options(global_options)
    }
+
+    pub fn config_file(&self) -> &Option<String> {
+        self.subcmd.config_file()
+    }
+
+    pub fn env_prefix(&self) -> &String {
+        self.subcmd.env_prefix()
+    }
 }

 #[derive(Parser)]
@@ -105,6 +117,18 @@ impl SubCommand {
            SubCommand::Start(cmd) => cmd.load_options(global_options),
        }
    }
+
+    fn config_file(&self) -> &Option<String> {
+        match self {
+            SubCommand::Start(cmd) => &cmd.config_file,
+        }
+    }
+
+    fn env_prefix(&self) -> &String {
+        match self {
+            SubCommand::Start(cmd) => &cmd.env_prefix,
+        }
+    }
 }

 #[derive(Debug, Default, Parser)]
--- a/src/cmd/src/standalone.rs
+++ b/src/cmd/src/standalone.rs
@@ -653,7 +653,7 @@ impl StartCommand {
    }
 }

-struct StandaloneInformationExtension {
+pub struct StandaloneInformationExtension {
    region_server: RegionServer,
    procedure_manager: ProcedureManagerRef,
    start_time_ms: u64,
--- a/src/common/base/src/plugins.rs
+++ b/src/common/base/src/plugins.rs
@@ -38,6 +38,18 @@ impl Plugins {
        self.read().get::<T>().cloned()
    }

+    pub fn get_or_insert<T, F>(&self, f: F) -> T
+    where
+        T: 'static + Send + Sync + Clone,
+        F: FnOnce() -> T,
+    {
+        let mut binding = self.write();
+        if !binding.contains::<T>() {
+            binding.insert(f());
+        }
+        binding.get::<T>().cloned().unwrap()
+    }
+
    pub fn map_mut<T: 'static + Send + Sync, F, R>(&self, mapper: F) -> R
    where
        F: FnOnce(Option<&mut T>) -> R,
--- a/src/common/function/Cargo.toml
+++ b/src/common/function/Cargo.toml
@@ -9,7 +9,7 @@ workspace = true

 [features]
 default = ["geo"]
-geo = ["geohash", "h3o"]
+geo = ["geohash", "h3o", "s2"]

 [dependencies]
 api.workspace = true
@@ -35,6 +35,7 @@ num = "0.4"
 num-traits = "0.2"
 once_cell.workspace = true
 paste = "1.0"
+s2 = { version = "0.0.12", optional = true }
 serde.workspace = true
 serde_json.workspace = true
 session.workspace = true
--- a/src/common/function/src/scalars/aggregate.rs
+++ b/src/common/function/src/scalars/aggregate.rs
@@ -31,7 +31,6 @@ pub use polyval::PolyvalAccumulatorCreator;
 pub use scipy_stats_norm_cdf::ScipyStatsNormCdfAccumulatorCreator;
 pub use scipy_stats_norm_pdf::ScipyStatsNormPdfAccumulatorCreator;

-use super::geo::encoding::JsonPathEncodeFunctionCreator;
 use crate::function_registry::FunctionRegistry;

 /// A function creates `AggregateFunctionCreator`.
@@ -93,6 +92,11 @@ impl AggregateFunctions {
        register_aggr_func!("scipystatsnormcdf", 2, ScipyStatsNormCdfAccumulatorCreator);
        register_aggr_func!("scipystatsnormpdf", 2, ScipyStatsNormPdfAccumulatorCreator);

-        register_aggr_func!("json_encode_path", 3, JsonPathEncodeFunctionCreator);
+        #[cfg(feature = "geo")]
+        register_aggr_func!(
+            "json_encode_path",
+            3,
+            super::geo::encoding::JsonPathEncodeFunctionCreator
+        );
    }
 }
--- a/src/common/function/src/scalars/date/date_add.rs
+++ b/src/common/function/src/scalars/date/date_add.rs
@@ -14,18 +14,19 @@

 use std::fmt;

-use common_query::error::{InvalidFuncArgsSnafu, Result, UnsupportedInputDataTypeSnafu};
+use common_query::error::{ArrowComputeSnafu, IntoVectorSnafu, InvalidFuncArgsSnafu, Result};
 use common_query::prelude::Signature;
-use datatypes::data_type::DataType;
+use datatypes::arrow::compute::kernels::numeric;
 use datatypes::prelude::ConcreteDataType;
-use datatypes::value::ValueRef;
-use datatypes::vectors::VectorRef;
-use snafu::ensure;
+use datatypes::vectors::{Helper, VectorRef};
+use snafu::{ensure, ResultExt};

 use crate::function::{Function, FunctionContext};
 use crate::helper;

-/// A function adds an interval value to Timestamp, Date or DateTime, and return the result.
+/// A function adds an interval value to Timestamp, Date, and return the result.
+/// The implementation of datetime type is based on Date64 which is incorrect so this function
+/// doesn't support the datetime type.
 #[derive(Clone, Debug, Default)]
 pub struct DateAddFunction;

@@ -44,7 +45,6 @@ impl Function for DateAddFunction {
        helper::one_of_sigs2(
            vec![
                ConcreteDataType::date_datatype(),
-                ConcreteDataType::datetime_datatype(),
                ConcreteDataType::timestamp_second_datatype(),
                ConcreteDataType::timestamp_millisecond_datatype(),
                ConcreteDataType::timestamp_microsecond_datatype(),
@@ -69,64 +69,14 @@ impl Function for DateAddFunction {
            }
        );

-        let left = &columns[0];
-        let right = &columns[1];
+        let left = columns[0].to_arrow_array();
+        let right = columns[1].to_arrow_array();

-        let size = left.len();
-        let left_datatype = columns[0].data_type();
-        match left_datatype {
-            ConcreteDataType::Timestamp(_) => {
-                let mut result = left_datatype.create_mutable_vector(size);
-                for i in 0..size {
-                    let ts = left.get(i).as_timestamp();
-                    let interval = right.get(i).as_interval();
-
-                    let new_ts = match (ts, interval) {
-                        (Some(ts), Some(interval)) => ts.add_interval(interval),
-                        _ => ts,
-                    };
-
-                    result.push_value_ref(ValueRef::from(new_ts));
-                }
-
-                Ok(result.to_vector())
-            }
-            ConcreteDataType::Date(_) => {
-                let mut result = left_datatype.create_mutable_vector(size);
-                for i in 0..size {
-                    let date = left.get(i).as_date();
-                    let interval = right.get(i).as_interval();
-                    let new_date = match (date, interval) {
-                        (Some(date), Some(interval)) => date.add_interval(interval),
-                        _ => date,
-                    };
-
-                    result.push_value_ref(ValueRef::from(new_date));
-                }
-
-                Ok(result.to_vector())
-            }
-            ConcreteDataType::DateTime(_) => {
-                let mut result = left_datatype.create_mutable_vector(size);
-                for i in 0..size {
-                    let datetime = left.get(i).as_datetime();
-                    let interval = right.get(i).as_interval();
-                    let new_datetime = match (datetime, interval) {
-                        (Some(datetime), Some(interval)) => datetime.add_interval(interval),
-                        _ => datetime,
-                    };
-
-                    result.push_value_ref(ValueRef::from(new_datetime));
-                }
-
-                Ok(result.to_vector())
-            }
-            _ => UnsupportedInputDataTypeSnafu {
-                function: NAME,
-                datatypes: columns.iter().map(|c| c.data_type()).collect::<Vec<_>>(),
-            }
-            .fail(),
-        }
+        let result = numeric::add(&left, &right).context(ArrowComputeSnafu)?;
+        let arrow_type = result.data_type().clone();
+        Helper::try_into_vector(result).context(IntoVectorSnafu {
+            data_type: arrow_type,
+        })
    }
 }

@@ -144,8 +94,7 @@ mod tests {
    use datatypes::prelude::ConcreteDataType;
    use datatypes::value::Value;
    use datatypes::vectors::{
-        DateTimeVector, DateVector, IntervalDayTimeVector, IntervalYearMonthVector,
-        TimestampSecondVector,
+        DateVector, IntervalDayTimeVector, IntervalYearMonthVector, TimestampSecondVector,
    };

    use super::{DateAddFunction, *};
@@ -168,16 +117,15 @@ mod tests {
            ConcreteDataType::date_datatype(),
            f.return_type(&[ConcreteDataType::date_datatype()]).unwrap()
        );
-        assert_eq!(
-            ConcreteDataType::datetime_datatype(),
-            f.return_type(&[ConcreteDataType::datetime_datatype()])
-                .unwrap()
-        );
-        assert!(matches!(f.signature(),
+        assert!(
+            matches!(f.signature(),
                         Signature {
                             type_signature: TypeSignature::OneOf(sigs),
                             volatility: Volatility::Immutable
-                         } if  sigs.len() == 18));
+                         } if  sigs.len() == 15),
+            "{:?}",
+            f.signature()
+        );
    }

    #[test]
@@ -243,36 +191,4 @@ mod tests {
            }
        }
    }
-
-    #[test]
-    fn test_datetime_date_add() {
-        let f = DateAddFunction;
-
-        let dates = vec![Some(123), None, Some(42), None];
-        // Intervals in months
-        let intervals = vec![1, 2, 3, 1];
-        let results = [Some(2678400123), None, Some(7776000042), None];
-
-        let date_vector = DateTimeVector::from(dates.clone());
-        let interval_vector = IntervalYearMonthVector::from_vec(intervals);
-        let args: Vec<VectorRef> = vec![Arc::new(date_vector), Arc::new(interval_vector)];
-        let vector = f.eval(FunctionContext::default(), &args).unwrap();
-
-        assert_eq!(4, vector.len());
-        for (i, _t) in dates.iter().enumerate() {
-            let v = vector.get(i);
-            let result = results.get(i).unwrap();
-
-            if result.is_none() {
-                assert_eq!(Value::Null, v);
-                continue;
-            }
-            match v {
-                Value::DateTime(date) => {
-                    assert_eq!(date.val(), result.unwrap());
-                }
-                _ => unreachable!(),
-            }
-        }
-    }
 }
--- a/src/common/function/src/scalars/date/date_sub.rs
+++ b/src/common/function/src/scalars/date/date_sub.rs
@@ -14,18 +14,19 @@

 use std::fmt;

-use common_query::error::{InvalidFuncArgsSnafu, Result, UnsupportedInputDataTypeSnafu};
+use common_query::error::{ArrowComputeSnafu, IntoVectorSnafu, InvalidFuncArgsSnafu, Result};
 use common_query::prelude::Signature;
-use datatypes::data_type::DataType;
+use datatypes::arrow::compute::kernels::numeric;
 use datatypes::prelude::ConcreteDataType;
-use datatypes::value::ValueRef;
-use datatypes::vectors::VectorRef;
-use snafu::ensure;
+use datatypes::vectors::{Helper, VectorRef};
+use snafu::{ensure, ResultExt};

 use crate::function::{Function, FunctionContext};
 use crate::helper;

-/// A function subtracts an interval value to Timestamp, Date or DateTime, and return the result.
+/// A function subtracts an interval value to Timestamp, Date, and return the result.
+/// The implementation of datetime type is based on Date64 which is incorrect so this function
+/// doesn't support the datetime type.
 #[derive(Clone, Debug, Default)]
 pub struct DateSubFunction;

@@ -44,7 +45,6 @@ impl Function for DateSubFunction {
        helper::one_of_sigs2(
            vec![
                ConcreteDataType::date_datatype(),
-                ConcreteDataType::datetime_datatype(),
                ConcreteDataType::timestamp_second_datatype(),
                ConcreteDataType::timestamp_millisecond_datatype(),
                ConcreteDataType::timestamp_microsecond_datatype(),
@@ -69,65 +69,14 @@ impl Function for DateSubFunction {
            }
        );

-        let left = &columns[0];
-        let right = &columns[1];
+        let left = columns[0].to_arrow_array();
+        let right = columns[1].to_arrow_array();

-        let size = left.len();
-        let left_datatype = columns[0].data_type();
-
-        match left_datatype {
-            ConcreteDataType::Timestamp(_) => {
-                let mut result = left_datatype.create_mutable_vector(size);
-                for i in 0..size {
-                    let ts = left.get(i).as_timestamp();
-                    let interval = right.get(i).as_interval();
-
-                    let new_ts = match (ts, interval) {
-                        (Some(ts), Some(interval)) => ts.sub_interval(interval),
-                        _ => ts,
-                    };
-
-                    result.push_value_ref(ValueRef::from(new_ts));
-                }
-
-                Ok(result.to_vector())
-            }
-            ConcreteDataType::Date(_) => {
-                let mut result = left_datatype.create_mutable_vector(size);
-                for i in 0..size {
-                    let date = left.get(i).as_date();
-                    let interval = right.get(i).as_interval();
-                    let new_date = match (date, interval) {
-                        (Some(date), Some(interval)) => date.sub_interval(interval),
-                        _ => date,
-                    };
-
-                    result.push_value_ref(ValueRef::from(new_date));
-                }
-
-                Ok(result.to_vector())
-            }
-            ConcreteDataType::DateTime(_) => {
-                let mut result = left_datatype.create_mutable_vector(size);
-                for i in 0..size {
-                    let datetime = left.get(i).as_datetime();
-                    let interval = right.get(i).as_interval();
-                    let new_datetime = match (datetime, interval) {
-                        (Some(datetime), Some(interval)) => datetime.sub_interval(interval),
-                        _ => datetime,
-                    };
-
-                    result.push_value_ref(ValueRef::from(new_datetime));
-                }
-
-                Ok(result.to_vector())
-            }
-            _ => UnsupportedInputDataTypeSnafu {
-                function: NAME,
-                datatypes: columns.iter().map(|c| c.data_type()).collect::<Vec<_>>(),
-            }
-            .fail(),
-        }
+        let result = numeric::sub(&left, &right).context(ArrowComputeSnafu)?;
+        let arrow_type = result.data_type().clone();
+        Helper::try_into_vector(result).context(IntoVectorSnafu {
+            data_type: arrow_type,
+        })
    }
 }

@@ -145,8 +94,7 @@ mod tests {
    use datatypes::prelude::ConcreteDataType;
    use datatypes::value::Value;
    use datatypes::vectors::{
-        DateTimeVector, DateVector, IntervalDayTimeVector, IntervalYearMonthVector,
-        TimestampSecondVector,
+        DateVector, IntervalDayTimeVector, IntervalYearMonthVector, TimestampSecondVector,
    };

    use super::{DateSubFunction, *};
@@ -174,11 +122,15 @@ mod tests {
            f.return_type(&[ConcreteDataType::datetime_datatype()])
                .unwrap()
        );
-        assert!(matches!(f.signature(),
+        assert!(
+            matches!(f.signature(),
                         Signature {
                             type_signature: TypeSignature::OneOf(sigs),
                             volatility: Volatility::Immutable
-                         } if  sigs.len() == 18));
+                         } if  sigs.len() == 15),
+            "{:?}",
+            f.signature()
+        );
    }

    #[test]
@@ -250,42 +202,4 @@ mod tests {
            }
        }
    }
-
-    #[test]
-    fn test_datetime_date_sub() {
-        let f = DateSubFunction;
-        let millis_per_month = 3600 * 24 * 30 * 1000;
-
-        let dates = vec![
-            Some(123 * millis_per_month),
-            None,
-            Some(42 * millis_per_month),
-            None,
-        ];
-        // Intervals in months
-        let intervals = vec![1, 2, 3, 1];
-        let results = [Some(316137600000), None, Some(100915200000), None];
-
-        let date_vector = DateTimeVector::from(dates.clone());
-        let interval_vector = IntervalYearMonthVector::from_vec(intervals);
-        let args: Vec<VectorRef> = vec![Arc::new(date_vector), Arc::new(interval_vector)];
-        let vector = f.eval(FunctionContext::default(), &args).unwrap();
-
-        assert_eq!(4, vector.len());
-        for (i, _t) in dates.iter().enumerate() {
-            let v = vector.get(i);
-            let result = results.get(i).unwrap();
-
-            if result.is_none() {
-                assert_eq!(Value::Null, v);
-                continue;
-            }
-            match v {
-                Value::DateTime(date) => {
-                    assert_eq!(date.val(), result.unwrap());
-                }
-                _ => unreachable!(),
-            }
-        }
-    }
 }
--- a/src/common/function/src/scalars/geo.rs
+++ b/src/common/function/src/scalars/geo.rs
@@ -17,8 +17,7 @@ pub(crate) mod encoding;
 mod geohash;
 mod h3;
 mod helpers;
-
-use geohash::{GeohashFunction, GeohashNeighboursFunction};
+mod s2;

 use crate::function_registry::FunctionRegistry;

@@ -27,8 +26,8 @@ pub(crate) struct GeoFunctions;
 impl GeoFunctions {
    pub fn register(registry: &FunctionRegistry) {
        // geohash
-        registry.register(Arc::new(GeohashFunction));
-        registry.register(Arc::new(GeohashNeighboursFunction));
+        registry.register(Arc::new(geohash::GeohashFunction));
+        registry.register(Arc::new(geohash::GeohashNeighboursFunction));

        // h3 index
        registry.register(Arc::new(h3::H3LatLngToCell));
@@ -55,5 +54,11 @@ impl GeoFunctions {
        registry.register(Arc::new(h3::H3GridDiskDistances));
        registry.register(Arc::new(h3::H3GridDistance));
        registry.register(Arc::new(h3::H3GridPathCells));
+
+        // s2
+        registry.register(Arc::new(s2::S2LatLngToCell));
+        registry.register(Arc::new(s2::S2CellLevel));
+        registry.register(Arc::new(s2::S2CellToToken));
+        registry.register(Arc::new(s2::S2CellParent));
    }
 }
--- a/src/common/function/src/scalars/geo/encoding.rs
+++ b/src/common/function/src/scalars/geo/encoding.rs
@@ -17,7 +17,7 @@ use std::sync::Arc;
 use common_error::ext::{BoxedError, PlainError};
 use common_error::status_code::StatusCode;
 use common_macro::{as_aggr_func_creator, AggrFuncTypeStore};
-use common_query::error::{self, InvalidFuncArgsSnafu, InvalidInputStateSnafu, Result};
+use common_query::error::{self, InvalidInputStateSnafu, Result};
 use common_query::logical_plan::accumulator::AggrFuncTypeStore;
 use common_query::logical_plan::{Accumulator, AggregateFunctionCreator};
 use common_query::prelude::AccumulatorCreatorFunction;
--- a/src/common/function/src/scalars/geo/h3.rs
+++ b/src/common/function/src/scalars/geo/h3.rs
@@ -16,7 +16,7 @@ use std::str::FromStr;

 use common_error::ext::{BoxedError, PlainError};
 use common_error::status_code::StatusCode;
-use common_query::error::{self, InvalidFuncArgsSnafu, Result};
+use common_query::error::{self, Result};
 use common_query::prelude::{Signature, TypeSignature};
 use datafusion::logical_expr::Volatility;
 use datatypes::prelude::ConcreteDataType;
@@ -29,9 +29,9 @@ use datatypes::vectors::{
 use derive_more::Display;
 use h3o::{CellIndex, LatLng, Resolution};
 use once_cell::sync::Lazy;
-use snafu::{ensure, ResultExt};
+use snafu::ResultExt;

-use super::helpers::{ensure_columns_len, ensure_columns_n};
+use super::helpers::{ensure_and_coerce, ensure_columns_len, ensure_columns_n};
 use crate::function::{Function, FunctionContext};

 static CELL_TYPES: Lazy<Vec<ConcreteDataType>> = Lazy::new(|| {
@@ -382,15 +382,7 @@ impl Function for H3CellResolution {
    }

    fn eval(&self, _func_ctx: FunctionContext, columns: &[VectorRef]) -> Result<VectorRef> {
-        ensure!(
-            columns.len() == 1,
-            InvalidFuncArgsSnafu {
-                err_msg: format!(
-                    "The length of the args is not correct, expect 1, provided : {}",
-                    columns.len()
-                ),
-            }
-        );
+        ensure_columns_n!(columns, 1);

        let cell_vec = &columns[0];
        let size = cell_vec.len();
@@ -982,18 +974,6 @@ fn value_to_resolution(v: Value) -> Result<Resolution> {
        .context(error::ExecuteSnafu)
 }

-macro_rules! ensure_and_coerce {
-    ($compare:expr, $coerce:expr) => {{
-        ensure!(
-            $compare,
-            InvalidFuncArgsSnafu {
-                err_msg: "Argument was outside of acceptable range "
-            }
-        );
-        Ok($coerce)
-    }};
-}
-
 fn value_to_position(v: Value) -> Result<u64> {
    match v {
        Value::Int8(v) => ensure_and_coerce!(v >= 0, v as u64),
--- a/src/common/function/src/scalars/geo/helpers.rs
+++ b/src/common/function/src/scalars/geo/helpers.rs
@@ -14,15 +14,15 @@

 macro_rules! ensure_columns_len {
    ($columns:ident) => {
-        ensure!(
+        snafu::ensure!(
            $columns.windows(2).all(|c| c[0].len() == c[1].len()),
-            InvalidFuncArgsSnafu {
+            common_query::error::InvalidFuncArgsSnafu {
                err_msg: "The length of input columns are in different size"
            }
        )
    };
    ($column_a:ident, $column_b:ident, $($column_n:ident),*) => {
-        ensure!(
+        snafu::ensure!(
            {
                let mut result = $column_a.len() == $column_b.len();
                $(
@@ -30,7 +30,7 @@ macro_rules! ensure_columns_len {
                )*
                result
            }
-            InvalidFuncArgsSnafu {
+            common_query::error::InvalidFuncArgsSnafu {
                err_msg: "The length of input columns are in different size"
            }
        )
@@ -41,9 +41,9 @@ pub(super) use ensure_columns_len;

 macro_rules! ensure_columns_n {
    ($columns:ident, $n:literal) => {
-        ensure!(
+        snafu::ensure!(
            $columns.len() == $n,
-            InvalidFuncArgsSnafu {
+            common_query::error::InvalidFuncArgsSnafu {
                err_msg: format!(
                    "The length of arguments is not correct, expect {}, provided : {}",
                    stringify!($n),
@@ -59,3 +59,17 @@ macro_rules! ensure_columns_n {
 }

 pub(super) use ensure_columns_n;
+
+macro_rules! ensure_and_coerce {
+    ($compare:expr, $coerce:expr) => {{
+        snafu::ensure!(
+            $compare,
+            common_query::error::InvalidFuncArgsSnafu {
+                err_msg: "Argument was outside of acceptable range "
+            }
+        );
+        Ok($coerce)
+    }};
+}
+
+pub(super) use ensure_and_coerce;
--- a/src/common/function/src/scalars/geo/s2.rs
+++ b/src/common/function/src/scalars/geo/s2.rs
@@ -0,0 +1,275 @@
+// 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_query::error::{InvalidFuncArgsSnafu, Result};
+use common_query::prelude::{Signature, TypeSignature};
+use datafusion::logical_expr::Volatility;
+use datatypes::prelude::ConcreteDataType;
+use datatypes::scalars::ScalarVectorBuilder;
+use datatypes::value::Value;
+use datatypes::vectors::{MutableVector, StringVectorBuilder, UInt64VectorBuilder, VectorRef};
+use derive_more::Display;
+use once_cell::sync::Lazy;
+use s2::cellid::{CellID, MAX_LEVEL};
+use s2::latlng::LatLng;
+use snafu::ensure;
+
+use crate::function::{Function, FunctionContext};
+use crate::scalars::geo::helpers::{ensure_and_coerce, ensure_columns_len, ensure_columns_n};
+
+static CELL_TYPES: Lazy<Vec<ConcreteDataType>> = Lazy::new(|| {
+    vec![
+        ConcreteDataType::int64_datatype(),
+        ConcreteDataType::uint64_datatype(),
+    ]
+});
+
+static COORDINATE_TYPES: Lazy<Vec<ConcreteDataType>> = Lazy::new(|| {
+    vec![
+        ConcreteDataType::float32_datatype(),
+        ConcreteDataType::float64_datatype(),
+    ]
+});
+
+static LEVEL_TYPES: Lazy<Vec<ConcreteDataType>> = Lazy::new(|| {
+    vec![
+        ConcreteDataType::int8_datatype(),
+        ConcreteDataType::int16_datatype(),
+        ConcreteDataType::int32_datatype(),
+        ConcreteDataType::int64_datatype(),
+        ConcreteDataType::uint8_datatype(),
+        ConcreteDataType::uint16_datatype(),
+        ConcreteDataType::uint32_datatype(),
+        ConcreteDataType::uint64_datatype(),
+    ]
+});
+
+/// Function that returns [s2] encoding cellid for a given geospatial coordinate.
+///
+/// [s2]: http://s2geometry.io
+#[derive(Clone, Debug, Default, Display)]
+#[display("{}", self.name())]
+pub struct S2LatLngToCell;
+
+impl Function for S2LatLngToCell {
+    fn name(&self) -> &str {
+        "s2_latlng_to_cell"
+    }
+
+    fn return_type(&self, _input_types: &[ConcreteDataType]) -> Result<ConcreteDataType> {
+        Ok(ConcreteDataType::uint64_datatype())
+    }
+
+    fn signature(&self) -> Signature {
+        let mut signatures = Vec::with_capacity(COORDINATE_TYPES.len());
+        for coord_type in COORDINATE_TYPES.as_slice() {
+            signatures.push(TypeSignature::Exact(vec![
+                // latitude
+                coord_type.clone(),
+                // longitude
+                coord_type.clone(),
+            ]));
+        }
+        Signature::one_of(signatures, Volatility::Stable)
+    }
+
+    fn eval(&self, _func_ctx: FunctionContext, columns: &[VectorRef]) -> Result<VectorRef> {
+        ensure_columns_n!(columns, 2);
+
+        let lat_vec = &columns[0];
+        let lon_vec = &columns[1];
+
+        let size = lat_vec.len();
+        let mut results = UInt64VectorBuilder::with_capacity(size);
+
+        for i in 0..size {
+            let lat = lat_vec.get(i).as_f64_lossy();
+            let lon = lon_vec.get(i).as_f64_lossy();
+
+            let result = match (lat, lon) {
+                (Some(lat), Some(lon)) => {
+                    let coord = LatLng::from_degrees(lat, lon);
+                    ensure!(
+                        coord.is_valid(),
+                        InvalidFuncArgsSnafu {
+                            err_msg: "The input coordinates are invalid",
+                        }
+                    );
+                    let cellid = CellID::from(coord);
+                    let encoded: u64 = cellid.0;
+                    Some(encoded)
+                }
+                _ => None,
+            };
+
+            results.push(result);
+        }
+
+        Ok(results.to_vector())
+    }
+}
+
+/// Return the level of current s2 cell
+#[derive(Clone, Debug, Default, Display)]
+#[display("{}", self.name())]
+pub struct S2CellLevel;
+
+impl Function for S2CellLevel {
+    fn name(&self) -> &str {
+        "s2_cell_level"
+    }
+
+    fn return_type(&self, _input_types: &[ConcreteDataType]) -> Result<ConcreteDataType> {
+        Ok(ConcreteDataType::uint64_datatype())
+    }
+
+    fn signature(&self) -> Signature {
+        signature_of_cell()
+    }
+
+    fn eval(&self, _func_ctx: FunctionContext, columns: &[VectorRef]) -> Result<VectorRef> {
+        ensure_columns_n!(columns, 1);
+
+        let cell_vec = &columns[0];
+        let size = cell_vec.len();
+        let mut results = UInt64VectorBuilder::with_capacity(size);
+
+        for i in 0..size {
+            let cell = cell_from_value(cell_vec.get(i));
+            let res = cell.map(|cell| cell.level());
+
+            results.push(res);
+        }
+
+        Ok(results.to_vector())
+    }
+}
+
+/// Return the string presentation of the cell
+#[derive(Clone, Debug, Default, Display)]
+#[display("{}", self.name())]
+pub struct S2CellToToken;
+
+impl Function for S2CellToToken {
+    fn name(&self) -> &str {
+        "s2_cell_to_token"
+    }
+
+    fn return_type(&self, _input_types: &[ConcreteDataType]) -> Result<ConcreteDataType> {
+        Ok(ConcreteDataType::string_datatype())
+    }
+
+    fn signature(&self) -> Signature {
+        signature_of_cell()
+    }
+
+    fn eval(&self, _func_ctx: FunctionContext, columns: &[VectorRef]) -> Result<VectorRef> {
+        ensure_columns_n!(columns, 1);
+
+        let cell_vec = &columns[0];
+        let size = cell_vec.len();
+        let mut results = StringVectorBuilder::with_capacity(size);
+
+        for i in 0..size {
+            let cell = cell_from_value(cell_vec.get(i));
+            let res = cell.map(|cell| cell.to_token());
+
+            results.push(res.as_deref());
+        }
+
+        Ok(results.to_vector())
+    }
+}
+
+/// Return parent at given level of current s2 cell
+#[derive(Clone, Debug, Default, Display)]
+#[display("{}", self.name())]
+pub struct S2CellParent;
+
+impl Function for S2CellParent {
+    fn name(&self) -> &str {
+        "s2_cell_parent"
+    }
+
+    fn return_type(&self, _input_types: &[ConcreteDataType]) -> Result<ConcreteDataType> {
+        Ok(ConcreteDataType::uint64_datatype())
+    }
+
+    fn signature(&self) -> Signature {
+        signature_of_cell_and_level()
+    }
+
+    fn eval(&self, _func_ctx: FunctionContext, columns: &[VectorRef]) -> Result<VectorRef> {
+        ensure_columns_n!(columns, 2);
+
+        let cell_vec = &columns[0];
+        let level_vec = &columns[1];
+        let size = cell_vec.len();
+        let mut results = UInt64VectorBuilder::with_capacity(size);
+
+        for i in 0..size {
+            let cell = cell_from_value(cell_vec.get(i));
+            let level = value_to_level(level_vec.get(i))?;
+            let result = cell.map(|cell| cell.parent(level).0);
+
+            results.push(result);
+        }
+
+        Ok(results.to_vector())
+    }
+}
+
+fn signature_of_cell() -> Signature {
+    let mut signatures = Vec::with_capacity(CELL_TYPES.len());
+    for cell_type in CELL_TYPES.as_slice() {
+        signatures.push(TypeSignature::Exact(vec![cell_type.clone()]));
+    }
+
+    Signature::one_of(signatures, Volatility::Stable)
+}
+
+fn signature_of_cell_and_level() -> Signature {
+    let mut signatures = Vec::with_capacity(CELL_TYPES.len() * LEVEL_TYPES.len());
+    for cell_type in CELL_TYPES.as_slice() {
+        for level_type in LEVEL_TYPES.as_slice() {
+            signatures.push(TypeSignature::Exact(vec![
+                cell_type.clone(),
+                level_type.clone(),
+            ]));
+        }
+    }
+    Signature::one_of(signatures, Volatility::Stable)
+}
+
+fn cell_from_value(v: Value) -> Option<CellID> {
+    match v {
+        Value::Int64(v) => Some(CellID(v as u64)),
+        Value::UInt64(v) => Some(CellID(v)),
+        _ => None,
+    }
+}
+
+fn value_to_level(v: Value) -> Result<u64> {
+    match v {
+        Value::Int8(v) => ensure_and_coerce!(v >= 0 && v <= MAX_LEVEL as i8, v as u64),
+        Value::Int16(v) => ensure_and_coerce!(v >= 0 && v <= MAX_LEVEL as i16, v as u64),
+        Value::Int32(v) => ensure_and_coerce!(v >= 0 && v <= MAX_LEVEL as i32, v as u64),
+        Value::Int64(v) => ensure_and_coerce!(v >= 0 && v <= MAX_LEVEL as i64, v as u64),
+        Value::UInt8(v) => ensure_and_coerce!(v <= MAX_LEVEL as u8, v as u64),
+        Value::UInt16(v) => ensure_and_coerce!(v <= MAX_LEVEL as u16, v as u64),
+        Value::UInt32(v) => ensure_and_coerce!(v <= MAX_LEVEL as u32, v as u64),
+        Value::UInt64(v) => ensure_and_coerce!(v <= MAX_LEVEL, v),
+        _ => unreachable!(),
+    }
+}
--- a/src/common/grpc/src/select.rs
+++ b/src/common/grpc/src/select.rs
@@ -12,7 +12,7 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.

-use api::helper::{convert_i128_to_interval, convert_to_pb_decimal128};
+use api::helper::{convert_month_day_nano_to_pb, convert_to_pb_decimal128};
 use api::v1::column::Values;
 use common_base::BitVec;
 use datatypes::types::{IntervalType, TimeType, TimestampType, WrapperType};
@@ -211,7 +211,7 @@ pub fn values(arrays: &[VectorRef]) -> Result<Values> {
            ConcreteDataType::Interval(IntervalType::MonthDayNano(_)),
            IntervalMonthDayNanoVector,
            interval_month_day_nano_values,
-            |x| { convert_i128_to_interval(x.into_native()) }
+            |x| { convert_month_day_nano_to_pb(x) }
        ),
        (
            ConcreteDataType::Decimal128(_),
--- a/src/common/meta/src/leadership_notifier.rs
+++ b/src/common/meta/src/leadership_notifier.rs
@@ -12,10 +12,10 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.

-use std::sync::Arc;
+use std::sync::{Arc, Mutex};

 use async_trait::async_trait;
-use common_telemetry::error;
+use common_telemetry::{error, info};

 use crate::error::Result;

@@ -24,6 +24,8 @@ pub type LeadershipChangeNotifierCustomizerRef = Arc<dyn LeadershipChangeNotifie
 /// A trait for customizing the leadership change notifier.
 pub trait LeadershipChangeNotifierCustomizer: Send + Sync {
    fn customize(&self, notifier: &mut LeadershipChangeNotifier);
+
+    fn add_listener(&self, listener: Arc<dyn LeadershipChangeListener>);
 }

 /// A trait for handling leadership change events in a distributed system.
@@ -45,6 +47,31 @@ pub struct LeadershipChangeNotifier {
    listeners: Vec<Arc<dyn LeadershipChangeListener>>,
 }

+#[derive(Default)]
+pub struct DefaultLeadershipChangeNotifierCustomizer {
+    listeners: Mutex<Vec<Arc<dyn LeadershipChangeListener>>>,
+}
+
+impl DefaultLeadershipChangeNotifierCustomizer {
+    pub fn new() -> Self {
+        Self {
+            listeners: Mutex::new(Vec::new()),
+        }
+    }
+}
+
+impl LeadershipChangeNotifierCustomizer for DefaultLeadershipChangeNotifierCustomizer {
+    fn customize(&self, notifier: &mut LeadershipChangeNotifier) {
+        info!("Customizing leadership change notifier");
+        let listeners = self.listeners.lock().unwrap().clone();
+        notifier.listeners.extend(listeners);
+    }
+
+    fn add_listener(&self, listener: Arc<dyn LeadershipChangeListener>) {
+        self.listeners.lock().unwrap().push(listener);
+    }
+}
+
 impl LeadershipChangeNotifier {
    /// Adds a listener to the notifier.
    pub fn add_listener(&mut self, listener: Arc<dyn LeadershipChangeListener>) {
--- a/src/common/meta/src/rpc/router.rs
+++ b/src/common/meta/src/rpc/router.rs
@@ -289,6 +289,7 @@ pub enum LeaderState {
    ///
    /// - The [`Region`] may be unavailable (e.g., Crashed, Network disconnected).
    /// - The [`Region`] was planned to migrate to another [`Peer`].
+    #[serde(alias = "Downgraded")]
    Downgrading,
 }

@@ -516,6 +517,73 @@ mod tests {
        assert_eq!(decoded, region_route);
    }

+    #[test]
+    fn test_region_route_compatibility() {
+        let region_route = RegionRoute {
+            region: Region {
+                id: 2.into(),
+                name: "r2".to_string(),
+                partition: None,
+                attrs: BTreeMap::new(),
+            },
+            leader_peer: Some(Peer::new(1, "a1")),
+            follower_peers: vec![Peer::new(2, "a2"), Peer::new(3, "a3")],
+            leader_state: Some(LeaderState::Downgrading),
+            leader_down_since: None,
+        };
+        let input = r#"{"region":{"id":2,"name":"r2","partition":null,"attrs":{}},"leader_peer":{"id":1,"addr":"a1"},"follower_peers":[{"id":2,"addr":"a2"},{"id":3,"addr":"a3"}],"leader_state":"Downgraded","leader_down_since":null}"#;
+        let decoded: RegionRoute = serde_json::from_str(input).unwrap();
+        assert_eq!(decoded, region_route);
+
+        let region_route = RegionRoute {
+            region: Region {
+                id: 2.into(),
+                name: "r2".to_string(),
+                partition: None,
+                attrs: BTreeMap::new(),
+            },
+            leader_peer: Some(Peer::new(1, "a1")),
+            follower_peers: vec![Peer::new(2, "a2"), Peer::new(3, "a3")],
+            leader_state: Some(LeaderState::Downgrading),
+            leader_down_since: None,
+        };
+        let input = r#"{"region":{"id":2,"name":"r2","partition":null,"attrs":{}},"leader_peer":{"id":1,"addr":"a1"},"follower_peers":[{"id":2,"addr":"a2"},{"id":3,"addr":"a3"}],"leader_status":"Downgraded","leader_down_since":null}"#;
+        let decoded: RegionRoute = serde_json::from_str(input).unwrap();
+        assert_eq!(decoded, region_route);
+
+        let region_route = RegionRoute {
+            region: Region {
+                id: 2.into(),
+                name: "r2".to_string(),
+                partition: None,
+                attrs: BTreeMap::new(),
+            },
+            leader_peer: Some(Peer::new(1, "a1")),
+            follower_peers: vec![Peer::new(2, "a2"), Peer::new(3, "a3")],
+            leader_state: Some(LeaderState::Downgrading),
+            leader_down_since: None,
+        };
+        let input = r#"{"region":{"id":2,"name":"r2","partition":null,"attrs":{}},"leader_peer":{"id":1,"addr":"a1"},"follower_peers":[{"id":2,"addr":"a2"},{"id":3,"addr":"a3"}],"leader_state":"Downgrading","leader_down_since":null}"#;
+        let decoded: RegionRoute = serde_json::from_str(input).unwrap();
+        assert_eq!(decoded, region_route);
+
+        let region_route = RegionRoute {
+            region: Region {
+                id: 2.into(),
+                name: "r2".to_string(),
+                partition: None,
+                attrs: BTreeMap::new(),
+            },
+            leader_peer: Some(Peer::new(1, "a1")),
+            follower_peers: vec![Peer::new(2, "a2"), Peer::new(3, "a3")],
+            leader_state: Some(LeaderState::Downgrading),
+            leader_down_since: None,
+        };
+        let input = r#"{"region":{"id":2,"name":"r2","partition":null,"attrs":{}},"leader_peer":{"id":1,"addr":"a1"},"follower_peers":[{"id":2,"addr":"a2"},{"id":3,"addr":"a3"}],"leader_status":"Downgrading","leader_down_since":null}"#;
+        let decoded: RegionRoute = serde_json::from_str(input).unwrap();
+        assert_eq!(decoded, region_route);
+    }
+
    #[test]
    fn test_de_serialize_partition() {
        let p = Partition {
--- a/src/common/pprof/Cargo.toml
+++ b/src/common/pprof/Cargo.toml
@@ -0,0 +1,22 @@
+[package]
+name = "common-pprof"
+version.workspace = true
+edition.workspace = true
+license.workspace = true
+
+[dependencies]
+common-error.workspace = true
+common-macro.workspace = true
+prost.workspace = true
+snafu.workspace = true
+tokio.workspace = true
+
+[target.'cfg(unix)'.dependencies]
+pprof = { version = "0.13", features = [
+    "flamegraph",
+    "prost-codec",
+    "protobuf",
+] }
+
+[lints]
+workspace = true
--- a/src/common/pprof/src/lib.rs
+++ b/src/common/pprof/src/lib.rs
@@ -0,0 +1,99 @@
+// 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.
+
+#[cfg(unix)]
+pub mod nix;
+
+pub mod error {
+    use std::any::Any;
+
+    use common_error::ext::ErrorExt;
+    use common_error::status_code::StatusCode;
+    use common_macro::stack_trace_debug;
+    use snafu::{Location, Snafu};
+
+    #[derive(Snafu)]
+    #[stack_trace_debug]
+    #[snafu(visibility(pub(crate)))]
+    pub enum Error {
+        #[cfg(unix)]
+        #[snafu(display("Pprof error"))]
+        Pprof {
+            #[snafu(source)]
+            error: pprof::Error,
+            #[snafu(implicit)]
+            location: Location,
+        },
+
+        #[snafu(display("Pprof is unsupported on this platform"))]
+        Unsupported {
+            #[snafu(implicit)]
+            location: Location,
+        },
+    }
+
+    pub type Result<T> = std::result::Result<T, Error>;
+
+    impl ErrorExt for Error {
+        fn status_code(&self) -> StatusCode {
+            match self {
+                #[cfg(unix)]
+                Error::Pprof { .. } => StatusCode::Unexpected,
+                Error::Unsupported { .. } => StatusCode::Unsupported,
+            }
+        }
+
+        fn as_any(&self) -> &dyn Any {
+            self
+        }
+    }
+}
+
+#[cfg(not(unix))]
+pub mod dummy {
+    use std::time::Duration;
+
+    use crate::error::{Result, UnsupportedSnafu};
+
+    /// Dummpy CPU profiler utility.
+    #[derive(Debug)]
+    pub struct Profiling {}
+
+    impl Profiling {
+        /// Creates a new profiler.
+        pub fn new(_duration: Duration, _frequency: i32) -> Profiling {
+            Profiling {}
+        }
+
+        /// Profiles and returns a generated text.
+        pub async fn dump_text(&self) -> Result<String> {
+            UnsupportedSnafu {}.fail()
+        }
+
+        /// Profiles and returns a generated flamegraph.
+        pub async fn dump_flamegraph(&self) -> Result<Vec<u8>> {
+            UnsupportedSnafu {}.fail()
+        }
+
+        /// Profiles and returns a generated proto.
+        pub async fn dump_proto(&self) -> Result<Vec<u8>> {
+            UnsupportedSnafu {}.fail()
+        }
+    }
+}
+
+#[cfg(not(unix))]
+pub use dummy::Profiling;
+#[cfg(unix)]
+pub use nix::Profiling;
--- a/src/common/pprof/src/nix.rs
+++ b/src/common/pprof/src/nix.rs
@@ -0,0 +1,78 @@
+// 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::time::Duration;
+
+use pprof::protos::Message;
+use snafu::ResultExt;
+
+use crate::error::{PprofSnafu, Result};
+
+/// CPU profiler utility.
+// Inspired by https://github.com/datafuselabs/databend/blob/67f445e83cd4eceda98f6c1c114858929d564029/src/common/base/src/base/profiling.rs
+#[derive(Debug)]
+pub struct Profiling {
+    /// Sample duration.
+    duration: Duration,
+    /// Sample frequency.
+    frequency: i32,
+}
+
+impl Profiling {
+    /// Creates a new profiler.
+    pub fn new(duration: Duration, frequency: i32) -> Profiling {
+        Profiling {
+            duration,
+            frequency,
+        }
+    }
+
+    /// Profiles and returns a generated pprof report.
+    pub async fn report(&self) -> Result<pprof::Report> {
+        let guard = pprof::ProfilerGuardBuilder::default()
+            .frequency(self.frequency)
+            .blocklist(&["libc", "libgcc", "pthread", "vdso"])
+            .build()
+            .context(PprofSnafu)?;
+        tokio::time::sleep(self.duration).await;
+        guard.report().build().context(PprofSnafu)
+    }
+
+    /// Profiles and returns a generated text.
+    pub async fn dump_text(&self) -> Result<String> {
+        let report = self.report().await?;
+        let text = format!("{report:?}");
+        Ok(text)
+    }
+
+    /// Profiles and returns a generated flamegraph.
+    pub async fn dump_flamegraph(&self) -> Result<Vec<u8>> {
+        let mut body: Vec<u8> = Vec::new();
+
+        let report = self.report().await?;
+        report.flamegraph(&mut body).context(PprofSnafu)?;
+
+        Ok(body)
+    }
+
+    /// Profiles and returns a generated proto.
+    pub async fn dump_proto(&self) -> Result<Vec<u8>> {
+        let report = self.report().await?;
+        // Generate google’s pprof format report.
+        let profile = report.pprof().context(PprofSnafu)?;
+        let body = profile.encode_to_vec();
+
+        Ok(body)
+    }
+}
--- a/src/common/time/src/date.rs
+++ b/src/common/time/src/date.rs
@@ -14,13 +14,13 @@

 use std::fmt::{Display, Formatter, Write};

-use chrono::{Datelike, Days, LocalResult, Months, NaiveDate, NaiveTime, TimeZone};
+use chrono::{Datelike, Days, LocalResult, Months, NaiveDate, NaiveTime, TimeDelta, TimeZone};
 use serde::{Deserialize, Serialize};
 use serde_json::Value;
 use snafu::ResultExt;

 use crate::error::{InvalidDateStrSnafu, ParseDateStrSnafu, Result};
-use crate::interval::Interval;
+use crate::interval::{IntervalDayTime, IntervalMonthDayNano, IntervalYearMonth};
 use crate::timezone::get_timezone;
 use crate::util::datetime_to_utc;
 use crate::Timezone;
@@ -134,29 +134,64 @@ impl Date {
        (self.0 as i64) * 24 * 3600
    }

-    /// Adds given Interval to the current date.
-    /// Returns None if the resulting date would be out of range.
-    pub fn add_interval(&self, interval: Interval) -> Option<Date> {
+    // FIXME(yingwen): remove add/sub intervals later
+    /// Adds given [IntervalYearMonth] to the current date.
+    pub fn add_year_month(&self, interval: IntervalYearMonth) -> Option<Date> {
        let naive_date = self.to_chrono_date()?;

-        let (months, days, _) = interval.to_month_day_nano();
-
        naive_date
-            .checked_add_months(Months::new(months as u32))?
-            .checked_add_days(Days::new(days as u64))
+            .checked_add_months(Months::new(interval.months as u32))
            .map(Into::into)
    }

-    /// Subtracts given Interval to the current date.
-    /// Returns None if the resulting date would be out of range.
-    pub fn sub_interval(&self, interval: Interval) -> Option<Date> {
+    /// Adds given [IntervalDayTime] to the current date.
+    pub fn add_day_time(&self, interval: IntervalDayTime) -> Option<Date> {
        let naive_date = self.to_chrono_date()?;

-        let (months, days, _) = interval.to_month_day_nano();
+        naive_date
+            .checked_add_days(Days::new(interval.days as u64))?
+            .checked_add_signed(TimeDelta::milliseconds(interval.milliseconds as i64))
+            .map(Into::into)
+    }
+
+    /// Adds given [IntervalMonthDayNano] to the current date.
+    pub fn add_month_day_nano(&self, interval: IntervalMonthDayNano) -> Option<Date> {
+        let naive_date = self.to_chrono_date()?;

        naive_date
-            .checked_sub_months(Months::new(months as u32))?
-            .checked_sub_days(Days::new(days as u64))
+            .checked_add_months(Months::new(interval.months as u32))?
+            .checked_add_days(Days::new(interval.days as u64))?
+            .checked_add_signed(TimeDelta::nanoseconds(interval.nanoseconds))
+            .map(Into::into)
+    }
+
+    /// Subtracts given [IntervalYearMonth] to the current date.
+    pub fn sub_year_month(&self, interval: IntervalYearMonth) -> Option<Date> {
+        let naive_date = self.to_chrono_date()?;
+
+        naive_date
+            .checked_sub_months(Months::new(interval.months as u32))
+            .map(Into::into)
+    }
+
+    /// Subtracts given [IntervalDayTime] to the current date.
+    pub fn sub_day_time(&self, interval: IntervalDayTime) -> Option<Date> {
+        let naive_date = self.to_chrono_date()?;
+
+        naive_date
+            .checked_sub_days(Days::new(interval.days as u64))?
+            .checked_sub_signed(TimeDelta::milliseconds(interval.milliseconds as i64))
+            .map(Into::into)
+    }
+
+    /// Subtracts given [IntervalMonthDayNano] to the current date.
+    pub fn sub_month_day_nano(&self, interval: IntervalMonthDayNano) -> Option<Date> {
+        let naive_date = self.to_chrono_date()?;
+
+        naive_date
+            .checked_sub_months(Months::new(interval.months as u32))?
+            .checked_sub_days(Days::new(interval.days as u64))?
+            .checked_sub_signed(TimeDelta::nanoseconds(interval.nanoseconds))
            .map(Into::into)
    }

@@ -246,12 +281,12 @@ mod tests {
    fn test_add_sub_interval() {
        let date = Date::new(1000);

-        let interval = Interval::from_year_month(3);
+        let interval = IntervalYearMonth::new(3);

-        let new_date = date.add_interval(interval).unwrap();
+        let new_date = date.add_year_month(interval).unwrap();
        assert_eq!(new_date.val(), 1091);

-        assert_eq!(date, new_date.sub_interval(interval).unwrap());
+        assert_eq!(date, new_date.sub_year_month(interval).unwrap());
    }

    #[test]
--- a/src/common/time/src/datetime.rs
+++ b/src/common/time/src/datetime.rs
@@ -13,16 +13,18 @@
 // limitations under the License.

 use std::fmt::{Display, Formatter, Write};
-use std::time::Duration;

-use chrono::{Days, LocalResult, Months, NaiveDateTime, TimeZone as ChronoTimeZone, Utc};
+use chrono::{
+    Days, LocalResult, Months, NaiveDateTime, TimeDelta, TimeZone as ChronoTimeZone, Utc,
+};
 use serde::{Deserialize, Serialize};
 use snafu::ResultExt;

 use crate::error::{InvalidDateStrSnafu, Result};
+use crate::interval::{IntervalDayTime, IntervalMonthDayNano, IntervalYearMonth};
 use crate::timezone::{get_timezone, Timezone};
 use crate::util::{datetime_to_utc, format_utc_datetime};
-use crate::{Date, Interval};
+use crate::Date;

 const DATETIME_FORMAT: &str = "%F %H:%M:%S%.f";
 const DATETIME_FORMAT_WITH_TZ: &str = "%F %H:%M:%S%.f%z";
@@ -160,32 +162,66 @@ impl DateTime {
            None => Utc.from_utc_datetime(&v).naive_local(),
        })
    }
-    /// Adds given Interval to the current datetime.
-    /// Returns None if the resulting datetime would be out of range.
-    pub fn add_interval(&self, interval: Interval) -> Option<Self> {
+
+    // FIXME(yingwen): remove add/sub intervals later
+    /// Adds given [IntervalYearMonth] to the current datetime.
+    pub fn add_year_month(&self, interval: IntervalYearMonth) -> Option<Self> {
        let naive_datetime = self.to_chrono_datetime()?;
-        let (months, days, nsecs) = interval.to_month_day_nano();

-        let naive_datetime = naive_datetime
-            .checked_add_months(Months::new(months as u32))?
-            .checked_add_days(Days::new(days as u64))?
-            + Duration::from_nanos(nsecs as u64);
-
-        Some(naive_datetime.into())
+        naive_datetime
+            .checked_add_months(Months::new(interval.months as u32))
+            .map(Into::into)
    }

-    /// Subtracts given Interval to the current datetime.
-    /// Returns None if the resulting datetime would be out of range.
-    pub fn sub_interval(&self, interval: Interval) -> Option<Self> {
+    /// Adds given [IntervalDayTime] to the current datetime.
+    pub fn add_day_time(&self, interval: IntervalDayTime) -> Option<Self> {
        let naive_datetime = self.to_chrono_datetime()?;
-        let (months, days, nsecs) = interval.to_month_day_nano();

-        let naive_datetime = naive_datetime
-            .checked_sub_months(Months::new(months as u32))?
-            .checked_sub_days(Days::new(days as u64))?
-            - Duration::from_nanos(nsecs as u64);
+        naive_datetime
+            .checked_add_days(Days::new(interval.days as u64))?
+            .checked_add_signed(TimeDelta::milliseconds(interval.milliseconds as i64))
+            .map(Into::into)
+    }

-        Some(naive_datetime.into())
+    /// Adds given [IntervalMonthDayNano] to the current datetime.
+    pub fn add_month_day_nano(&self, interval: IntervalMonthDayNano) -> Option<Self> {
+        let naive_datetime = self.to_chrono_datetime()?;
+
+        naive_datetime
+            .checked_add_months(Months::new(interval.months as u32))?
+            .checked_add_days(Days::new(interval.days as u64))?
+            .checked_add_signed(TimeDelta::nanoseconds(interval.nanoseconds))
+            .map(Into::into)
+    }
+
+    /// Subtracts given [IntervalYearMonth] to the current datetime.
+    pub fn sub_year_month(&self, interval: IntervalYearMonth) -> Option<Self> {
+        let naive_datetime = self.to_chrono_datetime()?;
+
+        naive_datetime
+            .checked_sub_months(Months::new(interval.months as u32))
+            .map(Into::into)
+    }
+
+    /// Subtracts given [IntervalDayTime] to the current datetime.
+    pub fn sub_day_time(&self, interval: IntervalDayTime) -> Option<Self> {
+        let naive_datetime = self.to_chrono_datetime()?;
+
+        naive_datetime
+            .checked_sub_days(Days::new(interval.days as u64))?
+            .checked_sub_signed(TimeDelta::milliseconds(interval.milliseconds as i64))
+            .map(Into::into)
+    }
+
+    /// Subtracts given [IntervalMonthDayNano] to the current datetime.
+    pub fn sub_month_day_nano(&self, interval: IntervalMonthDayNano) -> Option<Self> {
+        let naive_datetime = self.to_chrono_datetime()?;
+
+        naive_datetime
+            .checked_sub_months(Months::new(interval.months as u32))?
+            .checked_sub_days(Days::new(interval.days as u64))?
+            .checked_sub_signed(TimeDelta::nanoseconds(interval.nanoseconds))
+            .map(Into::into)
    }

    /// Convert to [common_time::date].
@@ -231,12 +267,12 @@ mod tests {
    fn test_add_sub_interval() {
        let datetime = DateTime::new(1000);

-        let interval = Interval::from_day_time(1, 200);
+        let interval = IntervalDayTime::new(1, 200);

-        let new_datetime = datetime.add_interval(interval).unwrap();
+        let new_datetime = datetime.add_day_time(interval).unwrap();
        assert_eq!(new_datetime.val(), 1000 + 3600 * 24 * 1000 + 200);

-        assert_eq!(datetime, new_datetime.sub_interval(interval).unwrap());
+        assert_eq!(datetime, new_datetime.sub_day_time(interval).unwrap());
    }

    #[test]
--- a/src/common/time/src/interval.rs
+++ b/src/common/time/src/interval.rs
@@ -12,18 +12,10 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.

-use std::cmp::Ordering;
-use std::default::Default;
-use std::fmt::{self, Display, Formatter, Write};
-use std::hash::{Hash, Hasher};
+use std::hash::Hash;

 use arrow::datatypes::IntervalUnit as ArrowIntervalUnit;
 use serde::{Deserialize, Serialize};
-use serde_json::Value;
-use snafu::ResultExt;
-
-use crate::duration::Duration;
-use crate::error::{Result, TimestampOverflowSnafu};

 #[derive(
    Debug, Default, Copy, Clone, PartialEq, Eq, Hash, PartialOrd, Ord, Serialize, Deserialize,
@@ -61,268 +53,269 @@ impl From<ArrowIntervalUnit> for IntervalUnit {
    }
 }

-/// Interval Type represents a period of time.
-/// It is composed of months, days and nanoseconds.
-/// 3 kinds of interval are supported: year-month, day-time and
-/// month-day-nano, which will be stored in the following format.
-/// Interval data format:
-/// | months | days   | nsecs      |
-/// | 4bytes | 4bytes | 8bytes     |
-#[derive(Debug, Clone, Default, Copy, Serialize, Deserialize)]
-pub struct Interval {
-    months: i32,
-    days: i32,
-    nsecs: i64,
-    unit: IntervalUnit,
+// The `Value` type requires Serialize, Deserialize.
+#[derive(
+    Debug, Default, Copy, Clone, Eq, PartialEq, Hash, Ord, PartialOrd, Serialize, Deserialize,
+)]
+#[repr(C)]
+pub struct IntervalYearMonth {
+    /// Number of months
+    pub months: i32,
 }

-// Nanosecond convert to other time unit
-pub const NANOS_PER_SEC: i64 = 1_000_000_000;
-pub const NANOS_PER_MILLI: i64 = 1_000_000;
-pub const NANOS_PER_MICRO: i64 = 1_000;
-pub const NANOS_PER_HOUR: i64 = 60 * 60 * NANOS_PER_SEC;
-pub const NANOS_PER_DAY: i64 = 24 * NANOS_PER_HOUR;
-pub const NANOS_PER_MONTH: i64 = 30 * NANOS_PER_DAY;
-
-pub const DAYS_PER_MONTH: i64 = 30;
-
-impl Interval {
-    /// Creates a new interval from months, days and nanoseconds.
-    /// Precision is nanosecond.
-    pub fn from_month_day_nano(months: i32, days: i32, nsecs: i64) -> Self {
-        Interval {
-            months,
-            days,
-            nsecs,
-            unit: IntervalUnit::MonthDayNano,
-        }
-    }
-
-    /// Creates a new interval from months.
-    pub fn from_year_month(months: i32) -> Self {
-        Interval {
-            months,
-            days: 0,
-            nsecs: 0,
-            unit: IntervalUnit::YearMonth,
-        }
-    }
-
-    /// Creates a new interval from days and milliseconds.
-    pub fn from_day_time(days: i32, millis: i32) -> Self {
-        Interval {
-            months: 0,
-            days,
-            nsecs: (millis as i64) * NANOS_PER_MILLI,
-            unit: IntervalUnit::DayTime,
-        }
-    }
-
-    pub fn to_duration(&self) -> Result<Duration> {
-        Ok(Duration::new_nanosecond(
-            self.to_nanosecond()
-                .try_into()
-                .context(TimestampOverflowSnafu)?,
-        ))
-    }
-
-    /// Return a tuple(months, days, nanoseconds) from the interval.
-    pub fn to_month_day_nano(&self) -> (i32, i32, i64) {
-        (self.months, self.days, self.nsecs)
-    }
-
-    /// Converts the interval to nanoseconds.
-    pub fn to_nanosecond(&self) -> i128 {
-        let days = (self.days as i64) + DAYS_PER_MONTH * (self.months as i64);
-        (self.nsecs as i128) + (NANOS_PER_DAY as i128) * (days as i128)
-    }
-
-    /// Smallest interval value.
-    pub const MIN: Self = Self {
-        months: i32::MIN,
-        days: i32::MIN,
-        nsecs: i64::MIN,
-        unit: IntervalUnit::MonthDayNano,
-    };
-
-    /// Largest interval value.
-    pub const MAX: Self = Self {
-        months: i32::MAX,
-        days: i32::MAX,
-        nsecs: i64::MAX,
-        unit: IntervalUnit::MonthDayNano,
-    };
-
-    /// Returns the justified interval.
-    /// allows you to adjust the interval of 30-day as one month and the interval of 24-hour as one day
-    pub fn justified_interval(&self) -> Self {
-        let mut result = *self;
-        let extra_months_d = self.days as i64 / DAYS_PER_MONTH;
-        let extra_months_nsecs = self.nsecs / NANOS_PER_MONTH;
-        result.days -= (extra_months_d * DAYS_PER_MONTH) as i32;
-        result.nsecs -= extra_months_nsecs * NANOS_PER_MONTH;
-
-        let extra_days = self.nsecs / NANOS_PER_DAY;
-        result.nsecs -= extra_days * NANOS_PER_DAY;
-
-        result.months += extra_months_d as i32 + extra_months_nsecs as i32;
-        result.days += extra_days as i32;
-
-        result
-    }
-
-    /// Convert Interval to nanoseconds,
-    /// to check whether Interval is positive
-    pub fn is_positive(&self) -> bool {
-        self.to_nanosecond() > 0
-    }
-
-    /// is_zero
-    pub fn is_zero(&self) -> bool {
-        self.months == 0 && self.days == 0 && self.nsecs == 0
-    }
-
-    /// get unit
-    pub fn unit(&self) -> IntervalUnit {
-        self.unit
-    }
-
-    /// Multiple Interval by an integer with overflow check.
-    /// Returns justified Interval, or `None` if overflow occurred.
-    pub fn checked_mul_int<I>(&self, rhs: I) -> Option<Self>
-    where
-        I: TryInto<i32>,
-    {
-        let rhs = rhs.try_into().ok()?;
-        let months = self.months.checked_mul(rhs)?;
-        let days = self.days.checked_mul(rhs)?;
-        let nsecs = self.nsecs.checked_mul(rhs as i64)?;
-
-        Some(
-            Self {
-                months,
-                days,
-                nsecs,
-                unit: self.unit,
-            }
-            .justified_interval(),
-        )
-    }
-
-    /// Convert Interval to ISO 8601 string
-    pub fn to_iso8601_string(self) -> String {
-        IntervalFormat::from(self).to_iso8601_string()
-    }
-
-    /// Convert Interval to postgres verbose string
-    pub fn to_postgres_string(self) -> String {
-        IntervalFormat::from(self).to_postgres_string()
-    }
-
-    /// Convert Interval to sql_standard string
-    pub fn to_sql_standard_string(self) -> String {
-        IntervalFormat::from(self).to_sql_standard_string()
-    }
-
-    /// Interval Type and i128 [IntervalUnit::MonthDayNano] Convert
-    /// v consists of months(i32) | days(i32) | nsecs(i64)
-    pub fn from_i128(v: i128) -> Self {
-        Interval {
-            nsecs: v as i64,
-            days: (v >> 64) as i32,
-            months: (v >> 96) as i32,
-            unit: IntervalUnit::MonthDayNano,
-        }
-    }
-
-    /// `Interval` Type and i64 [IntervalUnit::DayTime] Convert
-    /// v consists of days(i32) | milliseconds(i32)
-    pub fn from_i64(v: i64) -> Self {
-        Interval {
-            nsecs: ((v as i32) as i64) * NANOS_PER_MILLI,
-            days: (v >> 32) as i32,
-            months: 0,
-            unit: IntervalUnit::DayTime,
-        }
-    }
-
-    /// `Interval` Type and i32 [IntervalUnit::YearMonth] Convert
-    /// v consists of months(i32)
-    pub fn from_i32(v: i32) -> Self {
-        Interval {
-            nsecs: 0,
-            days: 0,
-            months: v,
-            unit: IntervalUnit::YearMonth,
-        }
-    }
-
-    pub fn to_i128(&self) -> i128 {
-        // 128            96              64                               0
-        // +-------+-------+-------+-------+-------+-------+-------+-------+
-        // |     months    |      days     |          nanoseconds          |
-        // +-------+-------+-------+-------+-------+-------+-------+-------+
-        let months = (self.months as u128 & u32::MAX as u128) << 96;
-        let days = (self.days as u128 & u32::MAX as u128) << 64;
-        let nsecs = self.nsecs as u128 & u64::MAX as u128;
-        (months | days | nsecs) as i128
-    }
-
-    pub fn to_i64(&self) -> i64 {
-        // 64                              32                              0
-        // +-------+-------+-------+-------+-------+-------+-------+-------+
-        // |             days              |         milliseconds          |
-        // +-------+-------+-------+-------+-------+-------+-------+-------+
-        let days = (self.days as u64 & u32::MAX as u64) << 32;
-        let milliseconds = (self.nsecs / NANOS_PER_MILLI) as u64 & u32::MAX as u64;
-        (days | milliseconds) as i64
+impl IntervalYearMonth {
+    pub fn new(months: i32) -> Self {
+        Self { months }
    }

    pub fn to_i32(&self) -> i32 {
        self.months
    }

+    pub fn from_i32(months: i32) -> Self {
+        Self { months }
+    }
+
    pub fn negative(&self) -> Self {
-        Self {
-            months: -self.months,
-            days: -self.days,
-            nsecs: -self.nsecs,
-            unit: self.unit,
+        Self::new(-self.months)
+    }
+
+    pub fn to_iso8601_string(&self) -> String {
+        IntervalFormat::from(*self).to_iso8601_string()
+    }
+}
+
+impl From<IntervalYearMonth> for IntervalFormat {
+    fn from(interval: IntervalYearMonth) -> Self {
+        IntervalFormat {
+            years: interval.months / 12,
+            months: interval.months % 12,
+            ..Default::default()
        }
    }
 }

-impl From<i128> for Interval {
+impl From<i32> for IntervalYearMonth {
+    fn from(v: i32) -> Self {
+        Self::from_i32(v)
+    }
+}
+
+impl From<IntervalYearMonth> for i32 {
+    fn from(v: IntervalYearMonth) -> Self {
+        v.to_i32()
+    }
+}
+
+impl From<IntervalYearMonth> for serde_json::Value {
+    fn from(v: IntervalYearMonth) -> Self {
+        serde_json::Value::from(v.to_i32())
+    }
+}
+
+#[derive(
+    Debug, Default, Copy, Clone, Eq, PartialEq, Hash, Ord, PartialOrd, Serialize, Deserialize,
+)]
+#[repr(C)]
+pub struct IntervalDayTime {
+    /// Number of days
+    pub days: i32,
+    /// Number of milliseconds
+    pub milliseconds: i32,
+}
+
+impl IntervalDayTime {
+    /// The additive identity i.e. `0`.
+    pub const ZERO: Self = Self::new(0, 0);
+
+    /// The multiplicative inverse, i.e. `-1`.
+    pub const MINUS_ONE: Self = Self::new(-1, -1);
+
+    /// The maximum value that can be represented
+    pub const MAX: Self = Self::new(i32::MAX, i32::MAX);
+
+    /// The minimum value that can be represented
+    pub const MIN: Self = Self::new(i32::MIN, i32::MIN);
+
+    pub const fn new(days: i32, milliseconds: i32) -> Self {
+        Self { days, milliseconds }
+    }
+
+    pub fn to_i64(&self) -> i64 {
+        let d = (self.days as u64 & u32::MAX as u64) << 32;
+        let m = self.milliseconds as u64 & u32::MAX as u64;
+        (d | m) as i64
+    }
+
+    pub fn from_i64(value: i64) -> Self {
+        let days = (value >> 32) as i32;
+        let milliseconds = value as i32;
+        Self { days, milliseconds }
+    }
+
+    pub fn negative(&self) -> Self {
+        Self::new(-self.days, -self.milliseconds)
+    }
+
+    pub fn to_iso8601_string(&self) -> String {
+        IntervalFormat::from(*self).to_iso8601_string()
+    }
+
+    pub fn as_millis(&self) -> i64 {
+        self.days as i64 * MS_PER_DAY + self.milliseconds as i64
+    }
+}
+
+impl From<i64> for IntervalDayTime {
+    fn from(v: i64) -> Self {
+        Self::from_i64(v)
+    }
+}
+
+impl From<IntervalDayTime> for i64 {
+    fn from(v: IntervalDayTime) -> Self {
+        v.to_i64()
+    }
+}
+
+impl From<IntervalDayTime> for serde_json::Value {
+    fn from(v: IntervalDayTime) -> Self {
+        serde_json::Value::from(v.to_i64())
+    }
+}
+
+// Millisecond convert to other time unit
+pub const MS_PER_SEC: i64 = 1_000;
+pub const MS_PER_MINUTE: i64 = 60 * MS_PER_SEC;
+pub const MS_PER_HOUR: i64 = 60 * MS_PER_MINUTE;
+pub const MS_PER_DAY: i64 = 24 * MS_PER_HOUR;
+pub const NANOS_PER_MILLI: i64 = 1_000_000;
+
+impl From<IntervalDayTime> for IntervalFormat {
+    fn from(interval: IntervalDayTime) -> Self {
+        IntervalFormat {
+            days: interval.days,
+            hours: interval.milliseconds as i64 / MS_PER_HOUR,
+            minutes: (interval.milliseconds as i64 % MS_PER_HOUR) / MS_PER_MINUTE,
+            seconds: (interval.milliseconds as i64 % MS_PER_MINUTE) / MS_PER_SEC,
+            microseconds: (interval.milliseconds as i64 % MS_PER_SEC) * MS_PER_SEC,
+            ..Default::default()
+        }
+    }
+}
+
+#[derive(
+    Debug, Default, Copy, Clone, Eq, PartialEq, Hash, Ord, PartialOrd, Serialize, Deserialize,
+)]
+#[repr(C)]
+pub struct IntervalMonthDayNano {
+    /// Number of months
+    pub months: i32,
+    /// Number of days
+    pub days: i32,
+    /// Number of nanoseconds
+    pub nanoseconds: i64,
+}
+
+impl IntervalMonthDayNano {
+    /// The additive identity i.e. `0`.
+    pub const ZERO: Self = Self::new(0, 0, 0);
+
+    /// The multiplicative inverse, i.e. `-1`.
+    pub const MINUS_ONE: Self = Self::new(-1, -1, -1);
+
+    /// The maximum value that can be represented
+    pub const MAX: Self = Self::new(i32::MAX, i32::MAX, i64::MAX);
+
+    /// The minimum value that can be represented
+    pub const MIN: Self = Self::new(i32::MIN, i32::MIN, i64::MIN);
+
+    pub const fn new(months: i32, days: i32, nanoseconds: i64) -> Self {
+        Self {
+            months,
+            days,
+            nanoseconds,
+        }
+    }
+
+    pub fn to_i128(&self) -> i128 {
+        let m = (self.months as u128 & u32::MAX as u128) << 96;
+        let d = (self.days as u128 & u32::MAX as u128) << 64;
+        let n = self.nanoseconds as u128 & u64::MAX as u128;
+        (m | d | n) as i128
+    }
+
+    pub fn from_i128(value: i128) -> Self {
+        let months = (value >> 96) as i32;
+        let days = (value >> 64) as i32;
+        let nanoseconds = value as i64;
+        Self {
+            months,
+            days,
+            nanoseconds,
+        }
+    }
+
+    pub fn negative(&self) -> Self {
+        Self::new(-self.months, -self.days, -self.nanoseconds)
+    }
+
+    pub fn to_iso8601_string(&self) -> String {
+        IntervalFormat::from(*self).to_iso8601_string()
+    }
+}
+
+impl From<i128> for IntervalMonthDayNano {
    fn from(v: i128) -> Self {
        Self::from_i128(v)
    }
 }

-impl From<Interval> for i128 {
-    fn from(v: Interval) -> Self {
+impl From<IntervalMonthDayNano> for i128 {
+    fn from(v: IntervalMonthDayNano) -> Self {
        v.to_i128()
    }
 }

-impl From<Interval> for serde_json::Value {
-    fn from(v: Interval) -> Self {
-        Value::String(v.to_string())
+impl From<IntervalMonthDayNano> for serde_json::Value {
+    fn from(v: IntervalMonthDayNano) -> Self {
+        serde_json::Value::from(v.to_i128().to_string())
    }
 }

-impl Display for Interval {
-    fn fmt(&self, f: &mut Formatter<'_>) -> fmt::Result {
-        let mut s = String::new();
-        if self.months != 0 {
-            write!(s, "{} months ", self.months)?;
+// Nanosecond convert to other time unit
+pub const NS_PER_SEC: i64 = 1_000_000_000;
+pub const NS_PER_MINUTE: i64 = 60 * NS_PER_SEC;
+pub const NS_PER_HOUR: i64 = 60 * NS_PER_MINUTE;
+pub const NS_PER_DAY: i64 = 24 * NS_PER_HOUR;
+
+impl From<IntervalMonthDayNano> for IntervalFormat {
+    fn from(interval: IntervalMonthDayNano) -> Self {
+        IntervalFormat {
+            years: interval.months / 12,
+            months: interval.months % 12,
+            days: interval.days,
+            hours: interval.nanoseconds / NS_PER_HOUR,
+            minutes: (interval.nanoseconds % NS_PER_HOUR) / NS_PER_MINUTE,
+            seconds: (interval.nanoseconds % NS_PER_MINUTE) / NS_PER_SEC,
+            microseconds: (interval.nanoseconds % NS_PER_SEC) / 1_000,
        }
-        if self.days != 0 {
-            write!(s, "{} days ", self.days)?;
-        }
-        if self.nsecs != 0 {
-            write!(s, "{} nsecs", self.nsecs)?;
-        }
-        write!(f, "{}", s.trim())
+    }
+}
+
+pub fn interval_year_month_to_month_day_nano(interval: IntervalYearMonth) -> IntervalMonthDayNano {
+    IntervalMonthDayNano {
+        months: interval.months,
+        days: 0,
+        nanoseconds: 0,
+    }
+}
+
+pub fn interval_day_time_to_month_day_nano(interval: IntervalDayTime) -> IntervalMonthDayNano {
+    IntervalMonthDayNano {
+        months: 0,
+        days: interval.days,
+        nanoseconds: interval.milliseconds as i64 * NANOS_PER_MILLI,
    }
 }

@@ -339,31 +332,6 @@ pub struct IntervalFormat {
    pub microseconds: i64,
 }

-impl From<Interval> for IntervalFormat {
-    fn from(val: Interval) -> IntervalFormat {
-        let months = val.months;
-        let days = val.days;
-        let microseconds = val.nsecs / NANOS_PER_MICRO;
-        let years = (months - (months % 12)) / 12;
-        let months = months - years * 12;
-        let hours = (microseconds - (microseconds % 3_600_000_000)) / 3_600_000_000;
-        let microseconds = microseconds - hours * 3_600_000_000;
-        let minutes = (microseconds - (microseconds % 60_000_000)) / 60_000_000;
-        let microseconds = microseconds - minutes * 60_000_000;
-        let seconds = (microseconds - (microseconds % 1_000_000)) / 1_000_000;
-        let microseconds = microseconds - seconds * 1_000_000;
-        IntervalFormat {
-            years,
-            months,
-            days,
-            hours,
-            minutes,
-            seconds,
-            microseconds,
-        }
-    }
-}
-
 impl IntervalFormat {
    /// All the field in the interval is 0
    pub fn is_zero(&self) -> bool {
@@ -540,117 +508,37 @@ fn get_time_part(
    interval
 }

-/// IntervalCompare is used to compare two intervals
-/// It makes interval into nanoseconds style.
-#[derive(PartialEq, Eq, Hash, PartialOrd, Ord)]
-struct IntervalCompare(i128);
-
-impl From<Interval> for IntervalCompare {
-    fn from(interval: Interval) -> Self {
-        Self(interval.to_nanosecond())
-    }
-}
-
-impl Ord for Interval {
-    fn cmp(&self, other: &Self) -> Ordering {
-        IntervalCompare::from(*self).cmp(&IntervalCompare::from(*other))
-    }
-}
-
-impl PartialOrd for Interval {
-    fn partial_cmp(&self, other: &Self) -> Option<Ordering> {
-        Some(self.cmp(other))
-    }
-}
-
-impl Eq for Interval {}
-
-impl PartialEq for Interval {
-    fn eq(&self, other: &Self) -> bool {
-        self.cmp(other).is_eq()
-    }
-}
-
-impl Hash for Interval {
-    fn hash<H: Hasher>(&self, state: &mut H) {
-        IntervalCompare::from(*self).hash(state)
-    }
-}
-
 #[cfg(test)]
 mod tests {
-    use std::collections::HashMap;
-
    use super::*;
-    use crate::timestamp::TimeUnit;

    #[test]
    fn test_from_year_month() {
-        let interval = Interval::from_year_month(1);
+        let interval = IntervalYearMonth::new(1);
        assert_eq!(interval.months, 1);
    }

    #[test]
    fn test_from_date_time() {
-        let interval = Interval::from_day_time(1, 2);
+        let interval = IntervalDayTime::new(1, 2);
        assert_eq!(interval.days, 1);
-        assert_eq!(interval.nsecs, 2_000_000);
+        assert_eq!(interval.milliseconds, 2);
    }

    #[test]
-    fn test_to_duration() {
-        let interval = Interval::from_day_time(1, 2);
-
-        let duration = interval.to_duration().unwrap();
-        assert_eq!(86400002000000, duration.value());
-        assert_eq!(TimeUnit::Nanosecond, duration.unit());
-
-        let interval = Interval::from_year_month(12);
-
-        let duration = interval.to_duration().unwrap();
-        assert_eq!(31104000000000000, duration.value());
-        assert_eq!(TimeUnit::Nanosecond, duration.unit());
-    }
-
-    #[test]
-    fn test_interval_is_positive() {
-        let interval = Interval::from_year_month(1);
-        assert!(interval.is_positive());
-        let interval = Interval::from_year_month(-1);
-        assert!(!interval.is_positive());
-
-        let interval = Interval::from_day_time(1, i32::MIN);
-        assert!(!interval.is_positive());
-    }
-
-    #[test]
-    fn test_to_nanosecond() {
-        let interval = Interval::from_year_month(1);
-        assert_eq!(interval.to_nanosecond(), 2592000000000000);
-        let interval = Interval::from_day_time(1, 2);
-        assert_eq!(interval.to_nanosecond(), 86400002000000);
-
-        let max_interval = Interval::from_month_day_nano(i32::MAX, i32::MAX, i64::MAX);
-        assert_eq!(max_interval.to_nanosecond(), 5751829423496836854775807);
-
-        let min_interval = Interval::from_month_day_nano(i32::MIN, i32::MIN, i64::MIN);
-        assert_eq!(min_interval.to_nanosecond(), -5751829426175236854775808);
-    }
-
-    #[test]
-    fn test_interval_is_zero() {
-        let interval = Interval::from_month_day_nano(1, 1, 1);
-        assert!(!interval.is_zero());
-        let interval = Interval::from_month_day_nano(0, 0, 0);
-        assert!(interval.is_zero());
+    fn test_from_month_day_nano() {
+        let interval = IntervalMonthDayNano::new(1, 2, 3);
+        assert_eq!(interval.months, 1);
+        assert_eq!(interval.days, 2);
+        assert_eq!(interval.nanoseconds, 3);
    }

    #[test]
    fn test_interval_i128_convert() {
        let test_interval_eq = |month, day, nano| {
-            let interval = Interval::from_month_day_nano(month, day, nano);
+            let interval = IntervalMonthDayNano::new(month, day, nano);
            let interval_i128 = interval.to_i128();
-            let interval2 = Interval::from_i128(interval_i128);
+            let interval2 = IntervalMonthDayNano::from_i128(interval_i128);
            assert_eq!(interval, interval2);
        };

@@ -666,11 +554,26 @@ mod tests {
        test_interval_eq(i32::MAX, i32::MIN, i64::MIN);
        test_interval_eq(i32::MIN, i32::MAX, i64::MIN);
        test_interval_eq(i32::MIN, i32::MIN, i64::MIN);
+
+        let interval = IntervalMonthDayNano::from_i128(1);
+        assert_eq!(interval, IntervalMonthDayNano::new(0, 0, 1));
+        assert_eq!(1, IntervalMonthDayNano::new(0, 0, 1).to_i128());
+    }
+
+    #[test]
+    fn test_interval_i64_convert() {
+        let interval = IntervalDayTime::from_i64(1);
+        assert_eq!(interval, IntervalDayTime::new(0, 1));
+        assert_eq!(1, IntervalDayTime::new(0, 1).to_i64());
    }

    #[test]
    fn test_convert_interval_format() {
-        let interval = Interval::from_month_day_nano(14, 160, 1000000);
+        let interval = IntervalMonthDayNano {
+            months: 14,
+            days: 160,
+            nanoseconds: 1000000,
+        };
        let interval_format = IntervalFormat::from(interval);
        assert_eq!(interval_format.years, 1);
        assert_eq!(interval_format.months, 2);
@@ -681,94 +584,34 @@ mod tests {
        assert_eq!(interval_format.microseconds, 1000);
    }

-    #[test]
-    fn test_interval_hash() {
-        let interval = Interval::from_month_day_nano(1, 31, 1);
-        let interval2 = Interval::from_month_day_nano(2, 1, 1);
-        let mut map = HashMap::new();
-        map.insert(interval, 1);
-        assert_eq!(map.get(&interval2), Some(&1));
-    }
-
-    #[test]
-    fn test_interval_mul_int() {
-        let interval = Interval::from_month_day_nano(1, 1, 1);
-        let interval2 = interval.checked_mul_int(2).unwrap();
-        assert_eq!(interval2.months, 2);
-        assert_eq!(interval2.days, 2);
-        assert_eq!(interval2.nsecs, 2);
-
-        // test justified interval
-        let interval = Interval::from_month_day_nano(1, 31, 1);
-        let interval2 = interval.checked_mul_int(2).unwrap();
-        assert_eq!(interval2.months, 4);
-        assert_eq!(interval2.days, 2);
-        assert_eq!(interval2.nsecs, 2);
-
-        // test overflow situation
-        let interval = Interval::from_month_day_nano(i32::MAX, 1, 1);
-        let interval2 = interval.checked_mul_int(2);
-        assert!(interval2.is_none());
-    }
-
-    #[test]
-    fn test_display() {
-        let interval = Interval::from_month_day_nano(1, 1, 1);
-        assert_eq!(interval.to_string(), "1 months 1 days 1 nsecs");
-
-        let interval = Interval::from_month_day_nano(14, 31, 10000000000);
-        assert_eq!(interval.to_string(), "14 months 31 days 10000000000 nsecs");
-    }
-
-    #[test]
-    fn test_interval_justified() {
-        let interval = Interval::from_month_day_nano(1, 131, 1).justified_interval();
-        let interval2 = Interval::from_month_day_nano(5, 11, 1);
-        assert_eq!(interval, interval2);
-
-        let interval = Interval::from_month_day_nano(1, 1, NANOS_PER_MONTH + 2 * NANOS_PER_DAY)
-            .justified_interval();
-        let interval2 = Interval::from_month_day_nano(2, 3, 0);
-        assert_eq!(interval, interval2);
-    }
-
-    #[test]
-    fn test_serde_json() {
-        let interval = Interval::from_month_day_nano(1, 1, 1);
-        let json = serde_json::to_string(&interval).unwrap();
-        assert_eq!(
-            json,
-            "{\"months\":1,\"days\":1,\"nsecs\":1,\"unit\":\"MonthDayNano\"}"
-        );
-        let interval2: Interval = serde_json::from_str(&json).unwrap();
-        assert_eq!(interval, interval2);
-    }
-
    #[test]
    fn test_to_iso8601_string() {
        // Test interval zero
-        let interval = Interval::from_month_day_nano(0, 0, 0);
+        let interval = IntervalMonthDayNano::new(0, 0, 0);
        assert_eq!(interval.to_iso8601_string(), "PT0S");

-        let interval = Interval::from_month_day_nano(1, 1, 1);
+        let interval = IntervalMonthDayNano::new(1, 1, 1);
        assert_eq!(interval.to_iso8601_string(), "P0Y1M1DT0H0M0S");

-        let interval = Interval::from_month_day_nano(14, 31, 10000000000);
+        let interval = IntervalMonthDayNano::new(14, 31, 10000000000);
        assert_eq!(interval.to_iso8601_string(), "P1Y2M31DT0H0M10S");

-        let interval = Interval::from_month_day_nano(14, 31, 23210200000000);
+        let interval = IntervalMonthDayNano::new(14, 31, 23210200000000);
        assert_eq!(interval.to_iso8601_string(), "P1Y2M31DT6H26M50.2S");
    }

    #[test]
    fn test_to_postgres_string() {
        // Test interval zero
-        let interval = Interval::from_month_day_nano(0, 0, 0);
-        assert_eq!(interval.to_postgres_string(), "00:00:00");
-
-        let interval = Interval::from_month_day_nano(23, 100, 23210200000000);
+        let interval = IntervalMonthDayNano::new(0, 0, 0);
        assert_eq!(
-            interval.to_postgres_string(),
+            IntervalFormat::from(interval).to_postgres_string(),
+            "00:00:00"
+        );
+
+        let interval = IntervalMonthDayNano::new(23, 100, 23210200000000);
+        assert_eq!(
+            IntervalFormat::from(interval).to_postgres_string(),
            "1 year 11 mons 100 days 06:26:50.200000"
        );
    }
@@ -776,18 +619,21 @@ mod tests {
    #[test]
    fn test_to_sql_standard_string() {
        // Test zero interval
-        let interval = Interval::from_month_day_nano(0, 0, 0);
-        assert_eq!(interval.to_sql_standard_string(), "0");
+        let interval = IntervalMonthDayNano::new(0, 0, 0);
+        assert_eq!(IntervalFormat::from(interval).to_sql_standard_string(), "0");

-        let interval = Interval::from_month_day_nano(23, 100, 23210200000000);
+        let interval = IntervalMonthDayNano::new(23, 100, 23210200000000);
        assert_eq!(
-            interval.to_sql_standard_string(),
+            IntervalFormat::from(interval).to_sql_standard_string(),
            "+1-11 +100 +6:26:50.200000"
        );

        // Test interval without year, month, day
-        let interval = Interval::from_month_day_nano(0, 0, 23210200000000);
-        assert_eq!(interval.to_sql_standard_string(), "6:26:50.200000");
+        let interval = IntervalMonthDayNano::new(0, 0, 23210200000000);
+        assert_eq!(
+            IntervalFormat::from(interval).to_sql_standard_string(),
+            "6:26:50.200000"
+        );
    }

    #[test]
--- a/src/common/time/src/lib.rs
+++ b/src/common/time/src/lib.rs
@@ -27,7 +27,7 @@ pub mod util;
 pub use date::Date;
 pub use datetime::DateTime;
 pub use duration::Duration;
-pub use interval::Interval;
+pub use interval::{IntervalDayTime, IntervalMonthDayNano, IntervalYearMonth};
 pub use range::RangeMillis;
 pub use timestamp::Timestamp;
 pub use timestamp_millis::TimestampMillis;
--- a/src/common/time/src/timestamp.rs
+++ b/src/common/time/src/timestamp.rs
@@ -20,16 +20,17 @@ use std::time::Duration;

 use arrow::datatypes::TimeUnit as ArrowTimeUnit;
 use chrono::{
-    DateTime, Days, LocalResult, Months, NaiveDate, NaiveDateTime, NaiveTime,
+    DateTime, Days, LocalResult, Months, NaiveDate, NaiveDateTime, NaiveTime, TimeDelta,
    TimeZone as ChronoTimeZone, Utc,
 };
 use serde::{Deserialize, Serialize};
 use snafu::{OptionExt, ResultExt};

+use crate::error;
 use crate::error::{ArithmeticOverflowSnafu, ParseTimestampSnafu, Result, TimestampOverflowSnafu};
+use crate::interval::{IntervalDayTime, IntervalMonthDayNano, IntervalYearMonth};
 use crate::timezone::{get_timezone, Timezone};
 use crate::util::{datetime_to_utc, div_ceil};
-use crate::{error, Interval};

 /// Timestamp represents the value of units(seconds/milliseconds/microseconds/nanoseconds) elapsed
 /// since UNIX epoch. The valid value range of [Timestamp] depends on it's unit (all in UTC timezone):
@@ -140,40 +141,77 @@ impl Timestamp {
        })
    }

-    /// Adds given Interval to the current timestamp.
-    /// Returns None if the resulting timestamp would be out of range.
-    pub fn add_interval(&self, interval: Interval) -> Option<Timestamp> {
+    // FIXME(yingwen): remove add/sub intervals later
+    /// Adds given [IntervalYearMonth] to the current timestamp.
+    pub fn add_year_month(&self, interval: IntervalYearMonth) -> Option<Timestamp> {
        let naive_datetime = self.to_chrono_datetime()?;
-        let (months, days, nsecs) = interval.to_month_day_nano();

-        let naive_datetime = naive_datetime
-            .checked_add_months(Months::new(months as u32))?
-            .checked_add_days(Days::new(days as u64))?
-            + Duration::from_nanos(nsecs as u64);
+        let naive_datetime =
+            naive_datetime.checked_add_months(Months::new(interval.months as u32))?;

-        match Timestamp::from_chrono_datetime(naive_datetime) {
-            // Have to convert the new timestamp by the current unit.
-            Some(ts) => ts.convert_to(self.unit),
-            None => None,
-        }
+        // Have to convert the new timestamp by the current unit.
+        Timestamp::from_chrono_datetime(naive_datetime).and_then(|ts| ts.convert_to(self.unit))
    }

-    /// Subtracts given Interval to the current timestamp.
-    /// Returns None if the resulting timestamp would be out of range.
-    pub fn sub_interval(&self, interval: Interval) -> Option<Timestamp> {
+    /// Adds given [IntervalDayTime] to the current timestamp.
+    pub fn add_day_time(&self, interval: IntervalDayTime) -> Option<Timestamp> {
        let naive_datetime = self.to_chrono_datetime()?;
-        let (months, days, nsecs) = interval.to_month_day_nano();

        let naive_datetime = naive_datetime
-            .checked_sub_months(Months::new(months as u32))?
-            .checked_sub_days(Days::new(days as u64))?
-            - Duration::from_nanos(nsecs as u64);
+            .checked_add_days(Days::new(interval.days as u64))?
+            .checked_add_signed(TimeDelta::milliseconds(interval.milliseconds as i64))?;

-        match Timestamp::from_chrono_datetime(naive_datetime) {
-            // Have to convert the new timestamp by the current unit.
-            Some(ts) => ts.convert_to(self.unit),
-            None => None,
-        }
+        // Have to convert the new timestamp by the current unit.
+        Timestamp::from_chrono_datetime(naive_datetime).and_then(|ts| ts.convert_to(self.unit))
+    }
+
+    /// Adds given [IntervalMonthDayNano] to the current timestamp.
+    pub fn add_month_day_nano(&self, interval: IntervalMonthDayNano) -> Option<Timestamp> {
+        let naive_datetime = self.to_chrono_datetime()?;
+
+        let naive_datetime = naive_datetime
+            .checked_add_months(Months::new(interval.months as u32))?
+            .checked_add_days(Days::new(interval.days as u64))?
+            .checked_add_signed(TimeDelta::nanoseconds(interval.nanoseconds))?;
+
+        // Have to convert the new timestamp by the current unit.
+        Timestamp::from_chrono_datetime(naive_datetime).and_then(|ts| ts.convert_to(self.unit))
+    }
+
+    /// Subtracts given [IntervalYearMonth] to the current timestamp.
+    pub fn sub_year_month(&self, interval: IntervalYearMonth) -> Option<Timestamp> {
+        let naive_datetime = self.to_chrono_datetime()?;
+
+        let naive_datetime =
+            naive_datetime.checked_sub_months(Months::new(interval.months as u32))?;
+
+        // Have to convert the new timestamp by the current unit.
+        Timestamp::from_chrono_datetime(naive_datetime).and_then(|ts| ts.convert_to(self.unit))
+    }
+
+    /// Subtracts given [IntervalDayTime] to the current timestamp.
+    pub fn sub_day_time(&self, interval: IntervalDayTime) -> Option<Timestamp> {
+        let naive_datetime = self.to_chrono_datetime()?;
+
+        let naive_datetime = naive_datetime
+            .checked_sub_days(Days::new(interval.days as u64))?
+            .checked_sub_signed(TimeDelta::milliseconds(interval.milliseconds as i64))?;
+
+        // Have to convert the new timestamp by the current unit.
+        Timestamp::from_chrono_datetime(naive_datetime).and_then(|ts| ts.convert_to(self.unit))
+    }
+
+    /// Subtracts given [IntervalMonthDayNano] to the current timestamp.
+    pub fn sub_month_day_nano(&self, interval: IntervalMonthDayNano) -> Option<Timestamp> {
+        let naive_datetime = self.to_chrono_datetime()?;
+
+        let naive_datetime = naive_datetime
+            .checked_sub_months(Months::new(interval.months as u32))?
+            .checked_sub_days(Days::new(interval.days as u64))?
+            .checked_sub_signed(TimeDelta::nanoseconds(interval.nanoseconds))?;
+
+        // Have to convert the new timestamp by the current unit.
+        Timestamp::from_chrono_datetime(naive_datetime).and_then(|ts| ts.convert_to(self.unit))
    }

    /// Subtracts current timestamp with another timestamp, yielding a duration.
@@ -688,13 +726,13 @@ mod tests {
    fn test_add_sub_interval() {
        let ts = Timestamp::new(1000, TimeUnit::Millisecond);

-        let interval = Interval::from_day_time(1, 200);
+        let interval = IntervalDayTime::new(1, 200);

-        let new_ts = ts.add_interval(interval).unwrap();
+        let new_ts = ts.add_day_time(interval).unwrap();
        assert_eq!(new_ts.unit(), TimeUnit::Millisecond);
        assert_eq!(new_ts.value(), 1000 + 3600 * 24 * 1000 + 200);

-        assert_eq!(ts, new_ts.sub_interval(interval).unwrap());
+        assert_eq!(ts, new_ts.sub_day_time(interval).unwrap());
    }

    #[test]
--- a/src/common/wal/src/config/kafka/datanode.rs
+++ b/src/common/wal/src/config/kafka/datanode.rs
@@ -46,6 +46,8 @@ pub struct DatanodeKafkaConfig {
    pub create_index: bool,
    #[serde(with = "humantime_serde")]
    pub dump_index_interval: Duration,
+    /// Ignore missing entries during read WAL.
+    pub overwrite_entry_start_id: bool,
 }

 impl Default for DatanodeKafkaConfig {
@@ -60,6 +62,7 @@ impl Default for DatanodeKafkaConfig {
            auto_create_topics: true,
            create_index: true,
            dump_index_interval: Duration::from_secs(60),
+            overwrite_entry_start_id: false,
        }
    }
 }
--- a/src/datanode/src/region_server.rs
+++ b/src/datanode/src/region_server.rs
@@ -32,7 +32,7 @@ use common_recordbatch::SendableRecordBatchStream;
 use common_runtime::Runtime;
 use common_telemetry::tracing::{self, info_span};
 use common_telemetry::tracing_context::{FutureExt, TracingContext};
-use common_telemetry::{error, info, warn};
+use common_telemetry::{debug, error, info, warn};
 use dashmap::DashMap;
 use datafusion::datasource::{provider_as_source, TableProvider};
 use datafusion::error::Result as DfResult;
@@ -893,7 +893,7 @@ impl RegionServerInner {
        for region in logical_regions {
            self.region_map
                .insert(region, RegionEngineWithStatus::Ready(engine.clone()));
-            info!("Logical region {} is registered!", region);
+            debug!("Logical region {} is registered!", region);
        }
        Ok(())
    }
@@ -935,17 +935,19 @@ impl RegionServerInner {
            .iter()
            .map(|x| (*x.key(), x.value().clone()))
            .collect::<Vec<_>>();
+        let num_regions = regions.len();

        for (region_id, engine) in regions {
            let closed = engine
                .handle_request(region_id, RegionRequest::Close(RegionCloseRequest {}))
                .await;
            match closed {
-                Ok(_) => info!("Region {region_id} is closed"),
+                Ok(_) => debug!("Region {region_id} is closed"),
                Err(e) => warn!("Failed to close region {region_id}, err: {e}"),
            }
        }
        self.region_map.clear();
+        info!("closed {num_regions} regions");

        let engines = self.engines.write().unwrap().drain().collect::<Vec<_>>();
        for (engine_name, engine) in engines {
--- a/src/datatypes/src/interval.rs
+++ b/src/datatypes/src/interval.rs
@@ -12,11 +12,10 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.

-use common_time::interval::Interval;
+use common_time::{IntervalDayTime, IntervalMonthDayNano, IntervalYearMonth};
 use paste::paste;
-use serde::{Deserialize, Serialize};

-use crate::prelude::{Scalar, Value, ValueRef};
+use crate::prelude::Scalar;
 use crate::scalars::ScalarRef;
 use crate::types::{
    IntervalDayTimeType, IntervalMonthDayNanoType, IntervalYearMonthType, WrapperType,
@@ -26,39 +25,6 @@ use crate::vectors::{IntervalDayTimeVector, IntervalMonthDayNanoVector, Interval
 macro_rules! define_interval_with_unit {
    ($unit: ident, $native_ty: ty) => {
        paste! {
-            #[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize, Deserialize)]
-            pub struct [<Interval $unit>](pub Interval);
-
-            impl [<Interval $unit>] {
-                pub fn new(val: $native_ty) -> Self {
-                    Self(Interval:: [<from_ $native_ty>](val))
-                }
-            }
-
-            impl Default for [<Interval $unit>] {
-                fn default() -> Self {
-                    Self::new(0)
-                }
-            }
-
-            impl From<[<Interval $unit>]> for Value {
-                fn from(t: [<Interval $unit>]) -> Value {
-                    Value::Interval(t.0)
-                }
-            }
-
-            impl From<[<Interval $unit>]> for serde_json::Value {
-                fn from(t: [<Interval $unit>]) -> Self {
-                    t.0.into()
-                }
-            }
-
-            impl From<[<Interval $unit>]> for ValueRef<'static> {
-                fn from(t: [<Interval $unit>]) -> Self {
-                    ValueRef::Interval(t.0)
-                }
-            }
-
            impl Scalar for [<Interval $unit>] {
                type VectorType = [<Interval $unit Vector>];
                type RefType<'a> = [<Interval $unit>];
@@ -87,41 +53,11 @@ macro_rules! define_interval_with_unit {
                type Native = $native_ty;

                fn from_native(value: Self::Native) -> Self {
-                    Self::new(value)
+                    Self::[<from_ $native_ty>](value)
                }

                fn into_native(self) -> Self::Native {
-                    self.0.[<to_ $native_ty>]()
-                }
-            }
-
-            impl From<$native_ty> for [<Interval $unit>] {
-                fn from(val: $native_ty) -> Self {
-                    [<Interval $unit>]::from_native(val as $native_ty)
-                }
-            }
-
-            impl From<[<Interval $unit>]> for $native_ty {
-                fn from(val: [<Interval $unit>]) -> Self {
-                    val.0.[<to_ $native_ty>]()
-                }
-            }
-
-            impl TryFrom<Value> for Option<[<Interval $unit>]> {
-                type Error = $crate::error::Error;
-
-                #[inline]
-                fn try_from(from: Value) -> std::result::Result<Self, Self::Error> {
-                    match from {
-                        Value::Interval(v) if v.unit() == common_time::interval::IntervalUnit::$unit => {
-                            Ok(Some([<Interval $unit>](v)))
-                        },
-                        Value::Null => Ok(None),
-                        _ => $crate::error::TryFromValueSnafu {
-                            reason: format!("{:?} is not a {}", from, stringify!([<Interval $unit>])),
-                        }
-                        .fail(),
-                    }
+                    self.[<to_ $native_ty>]()
                }
            }
        }
@@ -138,17 +74,17 @@ mod tests {

    #[test]
    fn test_interval_scalar() {
-        let interval = IntervalYearMonth::new(1000);
+        let interval = IntervalYearMonth::from(1000);
        assert_eq!(interval, interval.as_scalar_ref());
        assert_eq!(interval, interval.to_owned_scalar());
        assert_eq!(1000, interval.into_native());

-        let interval = IntervalDayTime::new(1000);
+        let interval = IntervalDayTime::from(1000);
        assert_eq!(interval, interval.as_scalar_ref());
        assert_eq!(interval, interval.to_owned_scalar());
        assert_eq!(1000, interval.into_native());

-        let interval = IntervalMonthDayNano::new(1000);
+        let interval = IntervalMonthDayNano::from(1000);
        assert_eq!(interval, interval.as_scalar_ref());
        assert_eq!(interval, interval.to_owned_scalar());
        assert_eq!(1000, interval.into_native());
@@ -156,15 +92,15 @@ mod tests {

    #[test]
    fn test_interval_convert_to_native_type() {
-        let interval = IntervalMonthDayNano::new(1000);
+        let interval = IntervalMonthDayNano::from(1000);
        let native_value: i128 = interval.into();
        assert_eq!(native_value, 1000);

-        let interval = IntervalDayTime::new(1000);
+        let interval = IntervalDayTime::from(1000);
        let native_interval: i64 = interval.into();
        assert_eq!(native_interval, 1000);

-        let interval = IntervalYearMonth::new(1000);
+        let interval = IntervalYearMonth::from(1000);
        let native_interval: i32 = interval.into();
        assert_eq!(native_interval, 1000);
    }
--- a/src/datatypes/src/types/interval_type.rs
+++ b/src/datatypes/src/types/interval_type.rs
@@ -17,8 +17,9 @@ use arrow::datatypes::{
    IntervalMonthDayNanoType as ArrowIntervalMonthDayNanoType, IntervalUnit as ArrowIntervalUnit,
    IntervalYearMonthType as ArrowIntervalYearMonthType,
 };
-use common_time::interval::IntervalUnit;
-use common_time::Interval;
+use common_time::interval::{
+    IntervalDayTime, IntervalMonthDayNano, IntervalUnit, IntervalYearMonth,
+};
 use enum_dispatch::enum_dispatch;
 use paste::paste;
 use serde::{Deserialize, Serialize};
@@ -26,7 +27,6 @@ use snafu::OptionExt;

 use crate::data_type::ConcreteDataType;
 use crate::error;
-use crate::interval::{IntervalDayTime, IntervalMonthDayNano, IntervalYearMonth};
 use crate::prelude::{
    DataType, LogicalTypeId, MutableVector, ScalarVectorBuilder, Value, ValueRef, Vector,
 };
@@ -75,7 +75,7 @@ macro_rules! impl_data_type_for_interval {
                }

                fn default_value(&self) -> Value {
-                    Value::Interval(Interval::from_i128(0))
+                    Value::[<Interval $unit>]([<Interval $unit>]::default())
                }

                fn as_arrow_type(&self) -> ArrowDataType {
@@ -124,7 +124,7 @@ macro_rules! impl_data_type_for_interval {
                fn cast_value_ref(value: ValueRef) -> crate::Result<Option<Self::Wrapper>> {
                    match value {
                        ValueRef::Null => Ok(None),
-                        ValueRef::Interval(t) => Ok(Some([<Interval $unit>](t))),
+                        ValueRef::[<Interval $unit>](t) => Ok(Some(t)),
                        other => error::CastTypeSnafu {
                            msg: format!("Failed to cast value {:?} to {}", other, stringify!([<Interval $unit>])),
                        }
--- a/src/datatypes/src/types/primitive_type.rs
+++ b/src/datatypes/src/types/primitive_type.rs
@@ -16,7 +16,6 @@ use std::cmp::Ordering;
 use std::fmt;

 use arrow::datatypes::{ArrowNativeType, ArrowPrimitiveType, DataType as ArrowDataType};
-use common_time::interval::IntervalUnit;
 use common_time::{Date, DateTime};
 use serde::{Deserialize, Serialize};
 use snafu::OptionExt;
@@ -30,6 +29,7 @@ use crate::types::{DateTimeType, DateType};
 use crate::value::{Value, ValueRef};
 use crate::vectors::{MutableVector, PrimitiveVector, PrimitiveVectorBuilder, Vector};

+// TODO(yingwen): Can we remove `Into<serde_json::Value>`?
 /// Represents the wrapper type that wraps a native type using the `newtype pattern`,
 /// such as [Date](`common_time::Date`) is a wrapper type for the underlying native
 /// type `i32`.
@@ -364,11 +364,7 @@ impl DataType for Int64Type {
            Value::DateTime(v) => Some(Value::Int64(v.val())),
            Value::Timestamp(v) => Some(Value::Int64(v.value())),
            Value::Time(v) => Some(Value::Int64(v.value())),
-            Value::Interval(v) => match v.unit() {
-                IntervalUnit::DayTime => Some(Value::Int64(v.to_i64())),
-                IntervalUnit::YearMonth => None,
-                IntervalUnit::MonthDayNano => None,
-            },
+            // We don't allow casting interval type to int.
            _ => None,
        }
    }
@@ -410,11 +406,7 @@ impl DataType for Int32Type {
            Value::Float64(v) => num::cast::cast(v).map(Value::Int32),
            Value::String(v) => v.as_utf8().parse::<i32>().map(Value::Int32).ok(),
            Value::Date(v) => Some(Value::Int32(v.val())),
-            Value::Interval(v) => match v.unit() {
-                IntervalUnit::YearMonth => Some(Value::Int32(v.to_i32())),
-                IntervalUnit::DayTime => None,
-                IntervalUnit::MonthDayNano => None,
-            },
+            // We don't allow casting interval type to int.
            _ => None,
        }
    }
--- a/src/datatypes/src/types/string_type.rs
+++ b/src/datatypes/src/types/string_type.rs
@@ -78,7 +78,15 @@ impl DataType for StringType {
            Value::DateTime(v) => Some(Value::String(StringBytes::from(v.to_string()))),
            Value::Timestamp(v) => Some(Value::String(StringBytes::from(v.to_iso8601_string()))),
            Value::Time(v) => Some(Value::String(StringBytes::from(v.to_iso8601_string()))),
-            Value::Interval(v) => Some(Value::String(StringBytes::from(v.to_iso8601_string()))),
+            Value::IntervalYearMonth(v) => {
+                Some(Value::String(StringBytes::from(v.to_iso8601_string())))
+            }
+            Value::IntervalDayTime(v) => {
+                Some(Value::String(StringBytes::from(v.to_iso8601_string())))
+            }
+            Value::IntervalMonthDayNano(v) => {
+                Some(Value::String(StringBytes::from(v.to_iso8601_string())))
+            }
            Value::Duration(v) => Some(Value::String(StringBytes::from(v.to_string()))),
            Value::Decimal128(v) => Some(Value::String(StringBytes::from(v.to_string()))),

--- a/src/datatypes/src/value.rs
+++ b/src/datatypes/src/value.rs
@@ -28,7 +28,7 @@ use common_time::datetime::DateTime;
 use common_time::interval::IntervalUnit;
 use common_time::time::Time;
 use common_time::timestamp::{TimeUnit, Timestamp};
-use common_time::{Duration, Interval, Timezone};
+use common_time::{Duration, IntervalDayTime, IntervalMonthDayNano, IntervalYearMonth, Timezone};
 use datafusion_common::ScalarValue;
 use greptime_proto::v1::value::ValueData;
 pub use ordered_float::OrderedFloat;
@@ -38,6 +38,7 @@ use snafu::{ensure, ResultExt};

 use crate::error::{self, ConvertArrowArrayToScalarsSnafu, Error, Result, TryFromValueSnafu};
 use crate::prelude::*;
+use crate::schema::ColumnSchema;
 use crate::type_id::LogicalTypeId;
 use crate::types::{IntervalType, ListType};
 use crate::vectors::ListVector;
@@ -78,7 +79,10 @@ pub enum Value {
    Timestamp(Timestamp),
    Time(Time),
    Duration(Duration),
-    Interval(Interval),
+    // Interval types:
+    IntervalYearMonth(IntervalYearMonth),
+    IntervalDayTime(IntervalDayTime),
+    IntervalMonthDayNano(IntervalMonthDayNano),

    List(ListValue),
 }
@@ -111,7 +115,15 @@ impl Display for Value {
            Value::DateTime(v) => write!(f, "{v}"),
            Value::Timestamp(v) => write!(f, "{}", v.to_iso8601_string()),
            Value::Time(t) => write!(f, "{}", t.to_iso8601_string()),
-            Value::Interval(v) => write!(f, "{}", v.to_iso8601_string()),
+            Value::IntervalYearMonth(v) => {
+                write!(f, "{}", v.to_iso8601_string())
+            }
+            Value::IntervalDayTime(v) => {
+                write!(f, "{}", v.to_iso8601_string())
+            }
+            Value::IntervalMonthDayNano(v) => {
+                write!(f, "{}", v.to_iso8601_string())
+            }
            Value::Duration(d) => write!(f, "{d}"),
            Value::List(v) => {
                let items = v
@@ -153,7 +165,15 @@ macro_rules! define_data_type_func {
                $struct::DateTime(_) => ConcreteDataType::datetime_datatype(),
                $struct::Time(t) => ConcreteDataType::time_datatype(*t.unit()),
                $struct::Timestamp(v) => ConcreteDataType::timestamp_datatype(v.unit()),
-                $struct::Interval(v) => ConcreteDataType::interval_datatype(v.unit()),
+                $struct::IntervalYearMonth(_) => {
+                    ConcreteDataType::interval_datatype(IntervalUnit::YearMonth)
+                }
+                $struct::IntervalDayTime(_) => {
+                    ConcreteDataType::interval_datatype(IntervalUnit::DayTime)
+                }
+                $struct::IntervalMonthDayNano(_) => {
+                    ConcreteDataType::interval_datatype(IntervalUnit::MonthDayNano)
+                }
                $struct::List(list) => ConcreteDataType::list_datatype(list.datatype().clone()),
                $struct::Duration(d) => ConcreteDataType::duration_datatype(d.unit()),
                $struct::Decimal128(d) => {
@@ -206,7 +226,9 @@ impl Value {
            Value::List(v) => ValueRef::List(ListValueRef::Ref { val: v }),
            Value::Timestamp(v) => ValueRef::Timestamp(*v),
            Value::Time(v) => ValueRef::Time(*v),
-            Value::Interval(v) => ValueRef::Interval(*v),
+            Value::IntervalYearMonth(v) => ValueRef::IntervalYearMonth(*v),
+            Value::IntervalDayTime(v) => ValueRef::IntervalDayTime(*v),
+            Value::IntervalMonthDayNano(v) => ValueRef::IntervalMonthDayNano(*v),
            Value::Duration(v) => ValueRef::Duration(*v),
            Value::Decimal128(v) => ValueRef::Decimal128(*v),
        }
@@ -220,14 +242,6 @@ impl Value {
        }
    }

-    /// Cast Value to Interval. Return None if value is not a valid interval data type.
-    pub fn as_interval(&self) -> Option<Interval> {
-        match self {
-            Value::Interval(i) => Some(*i),
-            _ => None,
-        }
-    }
-
    /// Cast Value to utf8 String. Return None if value is not a valid string data type.
    pub fn as_string(&self) -> Option<String> {
        match self {
@@ -255,12 +269,35 @@ impl Value {
    /// Cast Value to [Time]. Return None if value is not a valid time data type.
    pub fn as_time(&self) -> Option<Time> {
        match self {
-            Value::Int64(v) => Some(Time::new_millisecond(*v)),
            Value::Time(t) => Some(*t),
            _ => None,
        }
    }

+    /// Cast Value to [IntervalYearMonth]. Return None if value is not a valid interval year month data type.
+    pub fn as_interval_year_month(&self) -> Option<IntervalYearMonth> {
+        match self {
+            Value::IntervalYearMonth(v) => Some(*v),
+            _ => None,
+        }
+    }
+
+    /// Cast Value to [IntervalDayTime]. Return None if value is not a valid interval day time data type.
+    pub fn as_interval_day_time(&self) -> Option<IntervalDayTime> {
+        match self {
+            Value::IntervalDayTime(v) => Some(*v),
+            _ => None,
+        }
+    }
+
+    /// Cast Value to [IntervalMonthDayNano]. Return None if value is not a valid interval month day nano data type.
+    pub fn as_interval_month_day_nano(&self) -> Option<IntervalMonthDayNano> {
+        match self {
+            Value::IntervalMonthDayNano(v) => Some(*v),
+            _ => None,
+        }
+    }
+
    /// Cast Value to u64. Return None if value is not a valid uint64 data type.
    pub fn as_u64(&self) -> Option<u64> {
        match self {
@@ -321,11 +358,9 @@ impl Value {
                TimeUnit::Microsecond => LogicalTypeId::TimeMicrosecond,
                TimeUnit::Nanosecond => LogicalTypeId::TimeNanosecond,
            },
-            Value::Interval(v) => match v.unit() {
-                IntervalUnit::YearMonth => LogicalTypeId::IntervalYearMonth,
-                IntervalUnit::DayTime => LogicalTypeId::IntervalDayTime,
-                IntervalUnit::MonthDayNano => LogicalTypeId::IntervalMonthDayNano,
-            },
+            Value::IntervalYearMonth(_) => LogicalTypeId::IntervalYearMonth,
+            Value::IntervalDayTime(_) => LogicalTypeId::IntervalDayTime,
+            Value::IntervalMonthDayNano(_) => LogicalTypeId::IntervalMonthDayNano,
            Value::Duration(d) => match d.unit() {
                TimeUnit::Second => LogicalTypeId::DurationSecond,
                TimeUnit::Millisecond => LogicalTypeId::DurationMillisecond,
@@ -375,11 +410,9 @@ impl Value {
            }
            Value::Timestamp(t) => timestamp_to_scalar_value(t.unit(), Some(t.value())),
            Value::Time(t) => time_to_scalar_value(*t.unit(), Some(t.value()))?,
-            Value::Interval(v) => match v.unit() {
-                IntervalUnit::YearMonth => ScalarValue::IntervalYearMonth(Some(v.to_i32())),
-                IntervalUnit::DayTime => ScalarValue::IntervalDayTime(Some(v.to_i64())),
-                IntervalUnit::MonthDayNano => ScalarValue::IntervalMonthDayNano(Some(v.to_i128())),
-            },
+            Value::IntervalYearMonth(v) => ScalarValue::IntervalYearMonth(Some(v.to_i32())),
+            Value::IntervalDayTime(v) => ScalarValue::IntervalDayTime(Some(v.to_i64())),
+            Value::IntervalMonthDayNano(v) => ScalarValue::IntervalMonthDayNano(Some(v.to_i128())),
            Value::Duration(d) => duration_to_scalar_value(d.unit(), Some(d.value())),
            Value::Decimal128(d) => {
                let (v, p, s) = d.to_scalar_value();
@@ -434,7 +467,9 @@ impl Value {
            Value::Timestamp(x) => Some(Value::Timestamp(x.negative())),
            Value::Time(x) => Some(Value::Time(x.negative())),
            Value::Duration(x) => Some(Value::Duration(x.negative())),
-            Value::Interval(x) => Some(Value::Interval(x.negative())),
+            Value::IntervalYearMonth(x) => Some(Value::IntervalYearMonth(x.negative())),
+            Value::IntervalDayTime(x) => Some(Value::IntervalDayTime(x.negative())),
+            Value::IntervalMonthDayNano(x) => Some(Value::IntervalMonthDayNano(x.negative())),

            Value::Binary(_) | Value::String(_) | Value::Boolean(_) | Value::List(_) => None,
        }
@@ -571,16 +606,6 @@ pub fn scalar_value_to_timestamp(
    }
 }

-/// Convert [ScalarValue] to [Interval].
-pub fn scalar_value_to_interval(scalar: &ScalarValue) -> Option<Interval> {
-    match scalar {
-        ScalarValue::IntervalYearMonth(v) => v.map(Interval::from_i32),
-        ScalarValue::IntervalDayTime(v) => v.map(Interval::from_i64),
-        ScalarValue::IntervalMonthDayNano(v) => v.map(Interval::from_i128),
-        _ => None,
-    }
-}
-
 macro_rules! impl_ord_for_value_like {
    ($Type: ident, $left: ident, $right: ident) => {
        if $left.is_null() && !$right.is_null() {
@@ -607,7 +632,9 @@ macro_rules! impl_ord_for_value_like {
                ($Type::DateTime(v1), $Type::DateTime(v2)) => v1.cmp(v2),
                ($Type::Timestamp(v1), $Type::Timestamp(v2)) => v1.cmp(v2),
                ($Type::Time(v1), $Type::Time(v2)) => v1.cmp(v2),
-                ($Type::Interval(v1), $Type::Interval(v2)) => v1.cmp(v2),
+                ($Type::IntervalYearMonth(v1), $Type::IntervalYearMonth(v2)) => v1.cmp(v2),
+                ($Type::IntervalDayTime(v1), $Type::IntervalDayTime(v2)) => v1.cmp(v2),
+                ($Type::IntervalMonthDayNano(v1), $Type::IntervalMonthDayNano(v2)) => v1.cmp(v2),
                ($Type::Duration(v1), $Type::Duration(v2)) => v1.cmp(v2),
                ($Type::List(v1), $Type::List(v2)) => v1.cmp(v2),
                _ => panic!(
@@ -685,7 +712,9 @@ impl_try_from_value!(Date, Date);
 impl_try_from_value!(Time, Time);
 impl_try_from_value!(DateTime, DateTime);
 impl_try_from_value!(Timestamp, Timestamp);
-impl_try_from_value!(Interval, Interval);
+impl_try_from_value!(IntervalYearMonth, IntervalYearMonth);
+impl_try_from_value!(IntervalDayTime, IntervalDayTime);
+impl_try_from_value!(IntervalMonthDayNano, IntervalMonthDayNano);
 impl_try_from_value!(Duration, Duration);
 impl_try_from_value!(Decimal128, Decimal128);

@@ -727,7 +756,9 @@ impl_value_from!(Date, Date);
 impl_value_from!(Time, Time);
 impl_value_from!(DateTime, DateTime);
 impl_value_from!(Timestamp, Timestamp);
-impl_value_from!(Interval, Interval);
+impl_value_from!(IntervalYearMonth, IntervalYearMonth);
+impl_value_from!(IntervalDayTime, IntervalDayTime);
+impl_value_from!(IntervalMonthDayNano, IntervalMonthDayNano);
 impl_value_from!(Duration, Duration);
 impl_value_from!(String, String);
 impl_value_from!(Decimal128, Decimal128);
@@ -774,7 +805,9 @@ impl TryFrom<Value> for serde_json::Value {
            Value::List(v) => serde_json::to_value(v)?,
            Value::Timestamp(v) => serde_json::to_value(v.value())?,
            Value::Time(v) => serde_json::to_value(v.value())?,
-            Value::Interval(v) => serde_json::to_value(v.to_i128())?,
+            Value::IntervalYearMonth(v) => serde_json::to_value(v.to_i32())?,
+            Value::IntervalDayTime(v) => serde_json::to_value(v.to_i64())?,
+            Value::IntervalMonthDayNano(v) => serde_json::to_value(v.to_i128())?,
            Value::Duration(v) => serde_json::to_value(v.value())?,
            Value::Decimal128(v) => serde_json::to_value(v.to_string())?,
        };
@@ -926,13 +959,13 @@ impl TryFrom<ScalarValue> for Value {
                .unwrap_or(Value::Null),

            ScalarValue::IntervalYearMonth(t) => t
-                .map(|x| Value::Interval(Interval::from_i32(x)))
+                .map(|x| Value::IntervalYearMonth(IntervalYearMonth::from_i32(x)))
                .unwrap_or(Value::Null),
            ScalarValue::IntervalDayTime(t) => t
-                .map(|x| Value::Interval(Interval::from_i64(x)))
+                .map(|x| Value::IntervalDayTime(IntervalDayTime::from_i64(x)))
                .unwrap_or(Value::Null),
            ScalarValue::IntervalMonthDayNano(t) => t
-                .map(|x| Value::Interval(Interval::from_i128(x)))
+                .map(|x| Value::IntervalMonthDayNano(IntervalMonthDayNano::from_i128(x)))
                .unwrap_or(Value::Null),
            ScalarValue::DurationSecond(d) => d
                .map(|x| Value::Duration(Duration::new(x, TimeUnit::Second)))
@@ -987,7 +1020,9 @@ impl From<ValueRef<'_>> for Value {
            ValueRef::DateTime(v) => Value::DateTime(v),
            ValueRef::Timestamp(v) => Value::Timestamp(v),
            ValueRef::Time(v) => Value::Time(v),
-            ValueRef::Interval(v) => Value::Interval(v),
+            ValueRef::IntervalYearMonth(v) => Value::IntervalYearMonth(v),
+            ValueRef::IntervalDayTime(v) => Value::IntervalDayTime(v),
+            ValueRef::IntervalMonthDayNano(v) => Value::IntervalMonthDayNano(v),
            ValueRef::Duration(v) => Value::Duration(v),
            ValueRef::List(v) => v.to_value(),
            ValueRef::Decimal128(v) => Value::Decimal128(v),
@@ -1026,7 +1061,10 @@ pub enum ValueRef<'a> {
    Timestamp(Timestamp),
    Time(Time),
    Duration(Duration),
-    Interval(Interval),
+    // Interval types:
+    IntervalYearMonth(IntervalYearMonth),
+    IntervalDayTime(IntervalDayTime),
+    IntervalMonthDayNano(IntervalMonthDayNano),

    // Compound types:
    List(ListValueRef<'a>),
@@ -1150,9 +1188,19 @@ impl<'a> ValueRef<'a> {
        impl_as_for_value_ref!(self, Duration)
    }

-    /// Cast itself to [Interval].
-    pub fn as_interval(&self) -> Result<Option<Interval>> {
-        impl_as_for_value_ref!(self, Interval)
+    /// Cast itself to [IntervalYearMonth].
+    pub fn as_interval_year_month(&self) -> Result<Option<IntervalYearMonth>> {
+        impl_as_for_value_ref!(self, IntervalYearMonth)
+    }
+
+    /// Cast itself to [IntervalDayTime].
+    pub fn as_interval_day_time(&self) -> Result<Option<IntervalDayTime>> {
+        impl_as_for_value_ref!(self, IntervalDayTime)
+    }
+
+    /// Cast itself to [IntervalMonthDayNano].
+    pub fn as_interval_month_day_nano(&self) -> Result<Option<IntervalMonthDayNano>> {
+        impl_as_for_value_ref!(self, IntervalMonthDayNano)
    }

    /// Cast itself to [ListValueRef].
@@ -1212,7 +1260,9 @@ impl_value_ref_from!(Date, Date);
 impl_value_ref_from!(DateTime, DateTime);
 impl_value_ref_from!(Timestamp, Timestamp);
 impl_value_ref_from!(Time, Time);
-impl_value_ref_from!(Interval, Interval);
+impl_value_ref_from!(IntervalYearMonth, IntervalYearMonth);
+impl_value_ref_from!(IntervalDayTime, IntervalDayTime);
+impl_value_ref_from!(IntervalMonthDayNano, IntervalMonthDayNano);
 impl_value_ref_from!(Duration, Duration);
 impl_value_ref_from!(Decimal128, Decimal128);

@@ -1237,37 +1287,52 @@ impl<'a> From<Option<ListValueRef<'a>>> for ValueRef<'a> {
    }
 }

-impl<'a> TryFrom<ValueRef<'a>> for serde_json::Value {
-    type Error = serde_json::Error;
+/// transform a [ValueRef] to a [serde_json::Value].
+/// The json type will be handled specially
+pub fn transform_value_ref_to_json_value<'a>(
+    value: ValueRef<'a>,
+    schema: &'a ColumnSchema,
+) -> serde_json::Result<serde_json::Value> {
+    let json_value = match value {
+        ValueRef::Null => serde_json::Value::Null,
+        ValueRef::Boolean(v) => serde_json::Value::Bool(v),
+        ValueRef::UInt8(v) => serde_json::Value::from(v),
+        ValueRef::UInt16(v) => serde_json::Value::from(v),
+        ValueRef::UInt32(v) => serde_json::Value::from(v),
+        ValueRef::UInt64(v) => serde_json::Value::from(v),
+        ValueRef::Int8(v) => serde_json::Value::from(v),
+        ValueRef::Int16(v) => serde_json::Value::from(v),
+        ValueRef::Int32(v) => serde_json::Value::from(v),
+        ValueRef::Int64(v) => serde_json::Value::from(v),
+        ValueRef::Float32(v) => serde_json::Value::from(v.0),
+        ValueRef::Float64(v) => serde_json::Value::from(v.0),
+        ValueRef::String(bytes) => serde_json::Value::String(bytes.to_string()),
+        ValueRef::Binary(bytes) => {
+            if let ConcreteDataType::Json(_) = schema.data_type {
+                match jsonb::from_slice(bytes) {
+                    Ok(json) => json.into(),
+                    Err(e) => {
+                        error!(e; "Failed to parse jsonb");
+                        serde_json::Value::Null
+                    }
+                }
+            } else {
+                serde_json::to_value(bytes)?
+            }
+        }
+        ValueRef::Date(v) => serde_json::Value::Number(v.val().into()),
+        ValueRef::DateTime(v) => serde_json::Value::Number(v.val().into()),
+        ValueRef::List(v) => serde_json::to_value(v)?,
+        ValueRef::Timestamp(v) => serde_json::to_value(v.value())?,
+        ValueRef::Time(v) => serde_json::to_value(v.value())?,
+        ValueRef::IntervalYearMonth(v) => serde_json::Value::from(v),
+        ValueRef::IntervalDayTime(v) => serde_json::Value::from(v),
+        ValueRef::IntervalMonthDayNano(v) => serde_json::Value::from(v),
+        ValueRef::Duration(v) => serde_json::to_value(v.value())?,
+        ValueRef::Decimal128(v) => serde_json::to_value(v.to_string())?,
+    };

-    fn try_from(value: ValueRef<'a>) -> serde_json::Result<serde_json::Value> {
-        let json_value = match value {
-            ValueRef::Null => serde_json::Value::Null,
-            ValueRef::Boolean(v) => serde_json::Value::Bool(v),
-            ValueRef::UInt8(v) => serde_json::Value::from(v),
-            ValueRef::UInt16(v) => serde_json::Value::from(v),
-            ValueRef::UInt32(v) => serde_json::Value::from(v),
-            ValueRef::UInt64(v) => serde_json::Value::from(v),
-            ValueRef::Int8(v) => serde_json::Value::from(v),
-            ValueRef::Int16(v) => serde_json::Value::from(v),
-            ValueRef::Int32(v) => serde_json::Value::from(v),
-            ValueRef::Int64(v) => serde_json::Value::from(v),
-            ValueRef::Float32(v) => serde_json::Value::from(v.0),
-            ValueRef::Float64(v) => serde_json::Value::from(v.0),
-            ValueRef::String(bytes) => serde_json::Value::String(bytes.to_string()),
-            ValueRef::Binary(bytes) => serde_json::to_value(bytes)?,
-            ValueRef::Date(v) => serde_json::Value::Number(v.val().into()),
-            ValueRef::DateTime(v) => serde_json::Value::Number(v.val().into()),
-            ValueRef::List(v) => serde_json::to_value(v)?,
-            ValueRef::Timestamp(v) => serde_json::to_value(v.value())?,
-            ValueRef::Time(v) => serde_json::to_value(v.value())?,
-            ValueRef::Interval(v) => serde_json::to_value(v.to_i128())?,
-            ValueRef::Duration(v) => serde_json::to_value(v.value())?,
-            ValueRef::Decimal128(v) => serde_json::to_value(v.to_string())?,
-        };
-
-        Ok(json_value)
-    }
+    Ok(json_value)
 }

 /// Reference to a [ListValue].
@@ -1359,7 +1424,9 @@ impl<'a> ValueRef<'a> {
            ValueRef::Timestamp(_) => 16,
            ValueRef::Time(_) => 16,
            ValueRef::Duration(_) => 16,
-            ValueRef::Interval(_) => 24,
+            ValueRef::IntervalYearMonth(_) => 4,
+            ValueRef::IntervalDayTime(_) => 8,
+            ValueRef::IntervalMonthDayNano(_) => 16,
            ValueRef::Decimal128(_) => 32,
            ValueRef::List(v) => match v {
                ListValueRef::Indexed { vector, .. } => vector.memory_size() / vector.len(),
@@ -1428,7 +1495,9 @@ pub fn column_data_to_json(data: ValueData) -> JsonValue {
 mod tests {
    use arrow::datatypes::DataType as ArrowDataType;
    use common_time::timezone::set_default_timezone;
-    use greptime_proto::v1::{Decimal128 as ProtoDecimal128, IntervalMonthDayNano};
+    use greptime_proto::v1::{
+        Decimal128 as ProtoDecimal128, IntervalMonthDayNano as ProtoIntervalMonthDayNano,
+    };
    use num_traits::Float;

    use super::*;
@@ -1525,11 +1594,13 @@ mod tests {
            JsonValue::String("interval year [12]".to_string())
        );
        assert_eq!(
-            column_data_to_json(ValueData::IntervalMonthDayNanoValue(IntervalMonthDayNano {
-                months: 1,
-                days: 2,
-                nanoseconds: 3,
-            })),
+            column_data_to_json(ValueData::IntervalMonthDayNanoValue(
+                ProtoIntervalMonthDayNano {
+                    months: 1,
+                    days: 2,
+                    nanoseconds: 3,
+                }
+            )),
            JsonValue::String("interval month [1][2][3]".to_string())
        );
        assert_eq!(
@@ -1740,12 +1811,10 @@ mod tests {
            ScalarValue::IntervalMonthDayNano(None).try_into().unwrap()
        );
        assert_eq!(
-            Value::Interval(Interval::from_month_day_nano(1, 1, 1)),
-            ScalarValue::IntervalMonthDayNano(Some(
-                Interval::from_month_day_nano(1, 1, 1).to_i128()
-            ))
-            .try_into()
-            .unwrap()
+            Value::IntervalMonthDayNano(IntervalMonthDayNano::new(1, 1, 1)),
+            ScalarValue::IntervalMonthDayNano(Some(IntervalMonthDayNano::new(1, 1, 1).to_i128()))
+                .try_into()
+                .unwrap()
        );

        assert_eq!(
@@ -1975,9 +2044,17 @@ mod tests {
            &ConcreteDataType::time_nanosecond_datatype(),
            &Value::Time(Time::new_nanosecond(1)),
        );
+        check_type_and_value(
+            &ConcreteDataType::interval_year_month_datatype(),
+            &Value::IntervalYearMonth(IntervalYearMonth::new(1)),
+        );
+        check_type_and_value(
+            &ConcreteDataType::interval_day_time_datatype(),
+            &Value::IntervalDayTime(IntervalDayTime::new(1, 2)),
+        );
        check_type_and_value(
            &ConcreteDataType::interval_month_day_nano_datatype(),
-            &Value::Interval(Interval::from_month_day_nano(1, 2, 3)),
+            &Value::IntervalMonthDayNano(IntervalMonthDayNano::new(1, 2, 3)),
        );
        check_type_and_value(
            &ConcreteDataType::duration_second_datatype(),
@@ -2160,7 +2237,9 @@ mod tests {
        check_as_value_ref!(Float64, OrderedF64::from(16.0));
        check_as_value_ref!(Timestamp, Timestamp::new_millisecond(1));
        check_as_value_ref!(Time, Time::new_millisecond(1));
-        check_as_value_ref!(Interval, Interval::from_month_day_nano(1, 2, 3));
+        check_as_value_ref!(IntervalYearMonth, IntervalYearMonth::new(1));
+        check_as_value_ref!(IntervalDayTime, IntervalDayTime::new(1, 2));
+        check_as_value_ref!(IntervalMonthDayNano, IntervalMonthDayNano::new(1, 2, 3));
        check_as_value_ref!(Duration, Duration::new_millisecond(1));

        assert_eq!(
@@ -2672,9 +2751,11 @@ mod tests {
        check_value_ref_size_eq(&ValueRef::DateTime(DateTime::new(1)), 8);
        check_value_ref_size_eq(&ValueRef::Timestamp(Timestamp::new_millisecond(1)), 16);
        check_value_ref_size_eq(&ValueRef::Time(Time::new_millisecond(1)), 16);
+        check_value_ref_size_eq(&ValueRef::IntervalYearMonth(IntervalYearMonth::new(1)), 4);
+        check_value_ref_size_eq(&ValueRef::IntervalDayTime(IntervalDayTime::new(1, 2)), 8);
        check_value_ref_size_eq(
-            &ValueRef::Interval(Interval::from_month_day_nano(1, 2, 3)),
-            24,
+            &ValueRef::IntervalMonthDayNano(IntervalMonthDayNano::new(1, 2, 3)),
+            16,
        );
        check_value_ref_size_eq(&ValueRef::Duration(Duration::new_millisecond(1)), 16);
        check_value_ref_size_eq(
--- a/src/datatypes/src/vectors/helper.rs
+++ b/src/datatypes/src/vectors/helper.rs
@@ -421,7 +421,7 @@ mod tests {
    use common_decimal::Decimal128;
    use common_time::time::Time;
    use common_time::timestamp::TimeUnit;
-    use common_time::{Date, DateTime, Duration, Interval};
+    use common_time::{Date, DateTime, Duration, IntervalMonthDayNano};

    use super::*;
    use crate::value::Value;
@@ -689,7 +689,10 @@ mod tests {
        );
        assert_eq!(3, vector.len());
        for i in 0..vector.len() {
-            assert_eq!(Value::Interval(Interval::from_i128(2000)), vector.get(i));
+            assert_eq!(
+                Value::IntervalMonthDayNano(IntervalMonthDayNano::from_i128(2000)),
+                vector.get(i)
+            );
        }
    }

--- a/src/flow/src/compute/render/reduce.rs
+++ b/src/flow/src/compute/render/reduce.rs
@@ -560,7 +560,7 @@ fn reduce_batch_subgraph(
                    .get_mut(i)
                    .context(InternalSnafu{
                        reason: format!(
-                            "Output builder should have the same length as the row, expected at most {} but got {}", 
+                            "Output builder should have the same length as the row, expected at most {} but got {}",
                            column_cnt - 1,
                            i
                        )
@@ -1162,7 +1162,9 @@ fn from_val_to_slice_idx(
 #[cfg(test)]
 mod test {

-    use common_time::{DateTime, Interval, Timestamp};
+    use std::time::Duration;
+
+    use common_time::Timestamp;
    use datatypes::data_type::{ConcreteDataType, ConcreteDataType as CDT};
    use hydroflow::scheduled::graph::Hydroflow;

@@ -1214,8 +1216,8 @@ mod test {
        let expected = TypedPlan {
            schema: RelationType::new(vec![
                ColumnType::new(CDT::uint64_datatype(), true), // sum(number)
-                ColumnType::new(CDT::datetime_datatype(), false), // window start
-                ColumnType::new(CDT::datetime_datatype(), false), // window end
+                ColumnType::new(CDT::timestamp_millisecond_datatype(), false), // window start
+                ColumnType::new(CDT::timestamp_millisecond_datatype(), false), // window end
            ])
            .into_unnamed(),
            // TODO(discord9): mfp indirectly ref to key columns
@@ -1232,7 +1234,10 @@ mod test {
                            .with_types(
                                RelationType::new(vec![
                                    ColumnType::new(ConcreteDataType::uint32_datatype(), false),
-                                    ColumnType::new(ConcreteDataType::datetime_datatype(), false),
+                                    ColumnType::new(
+                                        ConcreteDataType::timestamp_millisecond_datatype(),
+                                        false,
+                                    ),
                                ])
                                .into_unnamed(),
                            ),
@@ -1242,22 +1247,18 @@ mod test {
                                .map(vec![
                                    ScalarExpr::Column(1).call_unary(
                                        UnaryFunc::TumbleWindowFloor {
-                                            window_size: Interval::from_month_day_nano(
-                                                0,
-                                                0,
-                                                1_000_000_000,
-                                            ),
-                                            start_time: Some(DateTime::new(1625097600000)),
+                                            window_size: Duration::from_nanos(1_000_000_000),
+                                            start_time: Some(Timestamp::new_millisecond(
+                                                1625097600000,
+                                            )),
                                        },
                                    ),
                                    ScalarExpr::Column(1).call_unary(
                                        UnaryFunc::TumbleWindowCeiling {
-                                            window_size: Interval::from_month_day_nano(
-                                                0,
-                                                0,
-                                                1_000_000_000,
-                                            ),
-                                            start_time: Some(DateTime::new(1625097600000)),
+                                            window_size: Duration::from_nanos(1_000_000_000),
+                                            start_time: Some(Timestamp::new_millisecond(
+                                                1625097600000,
+                                            )),
                                        },
                                    ),
                                ])
@@ -1278,9 +1279,9 @@ mod test {
                    }
                    .with_types(
                        RelationType::new(vec![
-                            ColumnType::new(CDT::datetime_datatype(), false), // window start
-                            ColumnType::new(CDT::datetime_datatype(), false), // window end
-                            ColumnType::new(CDT::uint64_datatype(), true),    //sum(number)
+                            ColumnType::new(CDT::timestamp_millisecond_datatype(), false), // window start
+                            ColumnType::new(CDT::timestamp_millisecond_datatype(), false), // window end
+                            ColumnType::new(CDT::uint64_datatype(), true), //sum(number)
                        ])
                        .with_key(vec![1])
                        .with_time_index(Some(0))
--- a/src/flow/src/expr/df_func.rs
+++ b/src/flow/src/expr/df_func.rs
@@ -171,9 +171,13 @@ impl DfScalarFunction {
    }
 }

-#[derive(Debug, Clone, PartialEq, Eq, PartialOrd, Ord, Hash)]
+#[derive(Clone, PartialEq, Eq, PartialOrd, Ord, Hash)]
 pub struct RawDfScalarFn {
-    /// The raw bytes encoded datafusion scalar function
+    /// The raw bytes encoded datafusion scalar function,
+    /// due to substrait have too many layers of nested struct and `ScalarFunction` 's derive is different
+    /// for simplicity's sake
+    /// so we store bytes instead of `ScalarFunction` here
+    /// but in unit test we will still compare decoded struct(using `f_decoded` field in Debug impl)
    pub(crate) f: bytes::BytesMut,
    /// The input schema of the function
    pub(crate) input_schema: RelationDesc,
@@ -181,6 +185,17 @@ pub struct RawDfScalarFn {
    pub(crate) extensions: FunctionExtensions,
 }

+impl std::fmt::Debug for RawDfScalarFn {
+    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
+        f.debug_struct("RawDfScalarFn")
+            .field("f", &self.f)
+            .field("f_decoded", &ScalarFunction::decode(&mut self.f.as_ref()))
+            .field("df_schema", &self.input_schema)
+            .field("extensions", &self.extensions)
+            .finish()
+    }
+}
+
 impl RawDfScalarFn {
    pub fn from_proto(
        f: &substrait::substrait_proto_df::proto::expression::ScalarFunction,
--- a/src/flow/src/expr/func.rs
+++ b/src/flow/src/expr/func.rs
@@ -16,19 +16,18 @@

 use std::collections::HashMap;
 use std::sync::{Arc, OnceLock};
+use std::time::Duration;

 use arrow::array::{ArrayRef, BooleanArray};
 use common_error::ext::BoxedError;
 use common_time::timestamp::TimeUnit;
-use common_time::{DateTime, Timestamp};
+use common_time::Timestamp;
 use datafusion_expr::Operator;
 use datatypes::data_type::ConcreteDataType;
 use datatypes::prelude::DataType;
 use datatypes::types::cast;
 use datatypes::value::Value;
-use datatypes::vectors::{
-    BooleanVector, DateTimeVector, Helper, TimestampMillisecondVector, VectorRef,
-};
+use datatypes::vectors::{BooleanVector, Helper, TimestampMillisecondVector, VectorRef};
 use serde::{Deserialize, Serialize};
 use smallvec::smallvec;
 use snafu::{ensure, OptionExt, ResultExt};
@@ -52,8 +51,8 @@ pub enum UnmaterializableFunc {
    CurrentSchema,
    TumbleWindow {
        ts: Box<TypedExpr>,
-        window_size: common_time::Interval,
-        start_time: Option<DateTime>,
+        window_size: Duration,
+        start_time: Option<Timestamp>,
    },
 }

@@ -63,7 +62,8 @@ impl UnmaterializableFunc {
        match self {
            Self::Now => Signature {
                input: smallvec![],
-                output: ConcreteDataType::datetime_datatype(),
+                // TODO(yingwen): Maybe return timestamp.
+                output: ConcreteDataType::timestamp_millisecond_datatype(),
                generic_fn: GenericFn::Now,
            },
            Self::CurrentSchema => Signature {
@@ -110,12 +110,12 @@ pub enum UnaryFunc {
    StepTimestamp,
    Cast(ConcreteDataType),
    TumbleWindowFloor {
-        window_size: common_time::Interval,
-        start_time: Option<DateTime>,
+        window_size: Duration,
+        start_time: Option<Timestamp>,
    },
    TumbleWindowCeiling {
-        window_size: common_time::Interval,
-        start_time: Option<DateTime>,
+        window_size: Duration,
+        start_time: Option<Timestamp>,
    },
 }

@@ -139,8 +139,8 @@ impl UnaryFunc {
                },
            },
            Self::StepTimestamp => Signature {
-                input: smallvec![ConcreteDataType::datetime_datatype()],
-                output: ConcreteDataType::datetime_datatype(),
+                input: smallvec![ConcreteDataType::timestamp_millisecond_datatype()],
+                output: ConcreteDataType::timestamp_millisecond_datatype(),
                generic_fn: GenericFn::StepTimestamp,
            },
            Self::Cast(to) => Signature {
@@ -238,19 +238,19 @@ impl UnaryFunc {
                }
            }
            Self::StepTimestamp => {
-                let datetime_array = get_datetime_array(&arg_col)?;
-                let date_array_ref = datetime_array
+                let timestamp_array = get_timestamp_array(&arg_col)?;
+                let timestamp_array_ref = timestamp_array
                    .as_any()
-                    .downcast_ref::<arrow::array::Date64Array>()
+                    .downcast_ref::<arrow::array::TimestampMillisecondArray>()
                    .context({
                        TypeMismatchSnafu {
                            expected: ConcreteDataType::boolean_datatype(),
-                            actual: ConcreteDataType::from_arrow_type(datetime_array.data_type()),
+                            actual: ConcreteDataType::from_arrow_type(timestamp_array.data_type()),
                        }
                    })?;

-                let ret = arrow::compute::unary(date_array_ref, |arr| arr + 1);
-                let ret = DateTimeVector::from(ret);
+                let ret = arrow::compute::unary(timestamp_array_ref, |arr| arr + 1);
+                let ret = TimestampMillisecondVector::from(ret);
                Ok(Arc::new(ret))
            }
            Self::Cast(to) => {
@@ -266,19 +266,19 @@ impl UnaryFunc {
                window_size,
                start_time,
            } => {
-                let datetime_array = get_datetime_array(&arg_col)?;
-                let date_array_ref = datetime_array
+                let timestamp_array = get_timestamp_array(&arg_col)?;
+                let date_array_ref = timestamp_array
                    .as_any()
-                    .downcast_ref::<arrow::array::Date64Array>()
+                    .downcast_ref::<arrow::array::TimestampMillisecondArray>()
                    .context({
                        TypeMismatchSnafu {
                            expected: ConcreteDataType::boolean_datatype(),
-                            actual: ConcreteDataType::from_arrow_type(datetime_array.data_type()),
+                            actual: ConcreteDataType::from_arrow_type(timestamp_array.data_type()),
                        }
                    })?;

-                let start_time = start_time.map(|t| t.val());
-                let window_size = (window_size.to_nanosecond() / 1_000_000) as repr::Duration; // nanosecond to millisecond
+                let start_time = start_time.map(|t| t.value());
+                let window_size = window_size.as_millis() as repr::Duration;

                let ret = arrow::compute::unary(date_array_ref, |ts| {
                    get_window_start(ts, window_size, start_time)
@@ -291,19 +291,19 @@ impl UnaryFunc {
                window_size,
                start_time,
            } => {
-                let datetime_array = get_datetime_array(&arg_col)?;
-                let date_array_ref = datetime_array
+                let timestamp_array = get_timestamp_array(&arg_col)?;
+                let date_array_ref = timestamp_array
                    .as_any()
-                    .downcast_ref::<arrow::array::Date64Array>()
+                    .downcast_ref::<arrow::array::TimestampMillisecondArray>()
                    .context({
                        TypeMismatchSnafu {
                            expected: ConcreteDataType::boolean_datatype(),
-                            actual: ConcreteDataType::from_arrow_type(datetime_array.data_type()),
+                            actual: ConcreteDataType::from_arrow_type(timestamp_array.data_type()),
                        }
                    })?;

-                let start_time = start_time.map(|t| t.val());
-                let window_size = (window_size.to_nanosecond() / 1_000_000) as repr::Duration; // nanosecond to millisecond
+                let start_time = start_time.map(|t| t.value());
+                let window_size = window_size.as_millis() as repr::Duration;

                let ret = arrow::compute::unary(date_array_ref, |ts| {
                    get_window_start(ts, window_size, start_time) + window_size
@@ -330,19 +330,20 @@ impl UnaryFunc {
                    })?;
                    if let Some(window_size) = window_size_untyped.as_string() {
                        // cast as interval
-                        cast(
+                        let interval = cast(
                            Value::from(window_size),
-                            &ConcreteDataType::interval_month_day_nano_datatype(),
+                            &ConcreteDataType::interval_day_time_datatype(),
                        )
                        .map_err(BoxedError::new)
                        .context(ExternalSnafu)?
-                        .as_interval()
+                        .as_interval_day_time()
                        .context(UnexpectedSnafu {
                            reason: "Expect window size arg to be interval after successful cast"
                                .to_string(),
-                        })?
-                    } else if let Some(interval) = window_size_untyped.as_interval() {
-                        interval
+                        })?;
+                        Duration::from_millis(interval.as_millis() as u64)
+                    } else if let Some(interval) = window_size_untyped.as_interval_day_time() {
+                        Duration::from_millis(interval.as_millis() as u64)
                    } else {
                        InvalidQuerySnafu {
                                reason: format!(
@@ -357,16 +358,19 @@ impl UnaryFunc {
                let start_time = match args.get(2) {
                    Some(start_time) => {
                        if let Some(value) = start_time.expr.as_literal() {
-                            // cast as DateTime
-                            let ret = cast(value, &ConcreteDataType::datetime_datatype())
-                                .map_err(BoxedError::new)
-                                .context(ExternalSnafu)?
-                                .as_datetime()
-                                .context(UnexpectedSnafu {
-                                    reason:
-                                        "Expect start time arg to be datetime after successful cast"
-                                            .to_string(),
-                                })?;
+                            // cast as timestamp
+                            let ret = cast(
+                                value,
+                                &ConcreteDataType::timestamp_millisecond_datatype(),
+                            )
+                            .map_err(BoxedError::new)
+                            .context(ExternalSnafu)?
+                            .as_timestamp()
+                            .context(UnexpectedSnafu {
+                                reason:
+                                    "Expect start time arg to be timestamp after successful cast"
+                                        .to_string(),
+                            })?;
                            Some(ret)
                        } else {
                            UnexpectedSnafu {
@@ -446,15 +450,15 @@ impl UnaryFunc {
            }
            Self::StepTimestamp => {
                let ty = arg.data_type();
-                if let Value::DateTime(datetime) = arg {
-                    let datetime = DateTime::from(datetime.val() + 1);
-                    Ok(Value::from(datetime))
+                if let Value::Timestamp(timestamp) = arg {
+                    let timestamp = Timestamp::new_millisecond(timestamp.value() + 1);
+                    Ok(Value::from(timestamp))
                } else if let Ok(v) = value_to_internal_ts(arg) {
-                    let datetime = DateTime::from(v + 1);
-                    Ok(Value::from(datetime))
+                    let timestamp = Timestamp::new_millisecond(v + 1);
+                    Ok(Value::from(timestamp))
                } else {
                    TypeMismatchSnafu {
-                        expected: ConcreteDataType::datetime_datatype(),
+                        expected: ConcreteDataType::timestamp_millisecond_datatype(),
                        actual: ty,
                    }
                    .fail()?
@@ -474,8 +478,8 @@ impl UnaryFunc {
                start_time,
            } => {
                let ts = get_ts_as_millisecond(arg)?;
-                let start_time = start_time.map(|t| t.val());
-                let window_size = (window_size.to_nanosecond() / 1_000_000) as repr::Duration; // nanosecond to millisecond
+                let start_time = start_time.map(|t| t.value());
+                let window_size = window_size.as_millis() as repr::Duration;
                let window_start = get_window_start(ts, window_size, start_time);

                let ret = Timestamp::new_millisecond(window_start);
@@ -486,8 +490,8 @@ impl UnaryFunc {
                start_time,
            } => {
                let ts = get_ts_as_millisecond(arg)?;
-                let start_time = start_time.map(|t| t.val());
-                let window_size = (window_size.to_nanosecond() / 1_000_000) as repr::Duration; // nanosecond to millisecond
+                let start_time = start_time.map(|t| t.value());
+                let window_size = window_size.as_millis() as repr::Duration;
                let window_start = get_window_start(ts, window_size, start_time);

                let window_end = window_start + window_size;
@@ -498,21 +502,22 @@ impl UnaryFunc {
    }
 }

-fn get_datetime_array(vector: &VectorRef) -> Result<arrow::array::ArrayRef, EvalError> {
+fn get_timestamp_array(vector: &VectorRef) -> Result<arrow::array::ArrayRef, EvalError> {
    let arrow_array = vector.to_arrow_array();
-    let datetime_array =
-        if *arrow_array.data_type() == ConcreteDataType::datetime_datatype().as_arrow_type() {
-            arrow_array
-        } else {
-            arrow::compute::cast(
-                &arrow_array,
-                &ConcreteDataType::datetime_datatype().as_arrow_type(),
-            )
-            .context(ArrowSnafu {
-                context: "Trying to cast to datetime in StepTimestamp",
-            })?
-        };
-    Ok(datetime_array)
+    let timestamp_array = if *arrow_array.data_type()
+        == ConcreteDataType::timestamp_millisecond_datatype().as_arrow_type()
+    {
+        arrow_array
+    } else {
+        arrow::compute::cast(
+            &arrow_array,
+            &ConcreteDataType::timestamp_millisecond_datatype().as_arrow_type(),
+        )
+        .context(ArrowSnafu {
+            context: "Trying to cast to timestamp in StepTimestamp",
+        })?
+    };
+    Ok(timestamp_array)
 }

 fn get_window_start(
@@ -1284,7 +1289,6 @@ where
 mod test {
    use std::sync::Arc;

-    use common_time::Interval;
    use datatypes::vectors::Vector;
    use pretty_assertions::assert_eq;

@@ -1292,18 +1296,18 @@ mod test {

    #[test]
    fn test_tumble_batch() {
-        let datetime_vector = DateTimeVector::from_vec(vec![1, 2, 10, 13, 14, 20, 25]);
+        let timestamp_vector = TimestampMillisecondVector::from_vec(vec![1, 2, 10, 13, 14, 20, 25]);
        let tumble_start = UnaryFunc::TumbleWindowFloor {
-            window_size: Interval::from_day_time(0, 10),
+            window_size: Duration::from_millis(10),
            start_time: None,
        };
        let tumble_end = UnaryFunc::TumbleWindowCeiling {
-            window_size: Interval::from_day_time(0, 10),
+            window_size: Duration::from_millis(10),
            start_time: None,
        };

-        let len = datetime_vector.len();
-        let batch = Batch::try_new(vec![Arc::new(datetime_vector)], len).unwrap();
+        let len = timestamp_vector.len();
+        let batch = Batch::try_new(vec![Arc::new(timestamp_vector)], len).unwrap();
        let arg = ScalarExpr::Column(0);

        let start = tumble_start.eval_batch(&batch, &arg).unwrap();
@@ -1459,4 +1463,17 @@ mod test {
            Err(Error::InvalidQuery { .. })
        );
    }
+
+    #[test]
+    fn test_cast_int() {
+        let interval = cast(
+            Value::from("1 second"),
+            &ConcreteDataType::interval_day_time_datatype(),
+        )
+        .unwrap();
+        assert_eq!(
+            interval,
+            Value::from(common_time::IntervalDayTime::new(0, 1000))
+        );
+    }
 }
--- a/src/flow/src/repr.rs
+++ b/src/flow/src/repr.rs
@@ -61,7 +61,7 @@ pub const BATCH_SIZE: usize = 32 * 16384;
 /// Convert a value that is or can be converted to Datetime to internal timestamp
 ///
 /// support types are: `Date`, `DateTime`, `TimeStamp`, `i64`
-pub fn value_to_internal_ts(value: Value) -> Result<Timestamp, EvalError> {
+pub fn value_to_internal_ts(value: Value) -> Result<i64, EvalError> {
    let is_supported_time_type = |arg: &Value| {
        let ty = arg.data_type();
        matches!(
@@ -76,14 +76,14 @@ pub fn value_to_internal_ts(value: Value) -> Result<Timestamp, EvalError> {
        Value::Int64(ts) => Ok(ts),
        arg if is_supported_time_type(&arg) => {
            let arg_ty = arg.data_type();
-            let res = cast(arg, &ConcreteDataType::datetime_datatype()).context({
+            let res = cast(arg, &ConcreteDataType::timestamp_millisecond_datatype()).context({
                CastValueSnafu {
                    from: arg_ty,
-                    to: ConcreteDataType::datetime_datatype(),
+                    to: ConcreteDataType::timestamp_millisecond_datatype(),
                }
            })?;
-            if let Value::DateTime(ts) = res {
-                Ok(ts.val())
+            if let Value::Timestamp(ts) = res {
+                Ok(ts.value())
            } else {
                unreachable!()
            }
--- a/src/flow/src/transform.rs
+++ b/src/flow/src/transform.rs
@@ -156,10 +156,10 @@ mod test {

    use catalog::RegisterTableRequest;
    use common_catalog::consts::{DEFAULT_CATALOG_NAME, DEFAULT_SCHEMA_NAME, NUMBERS_TABLE_ID};
-    use common_time::DateTime;
    use datatypes::prelude::*;
    use datatypes::schema::Schema;
-    use datatypes::vectors::VectorRef;
+    use datatypes::timestamp::TimestampMillisecond;
+    use datatypes::vectors::{TimestampMillisecondVectorBuilder, VectorRef};
    use itertools::Itertools;
    use prost::Message;
    use query::parser::QueryLanguageParser;
@@ -202,7 +202,7 @@ mod test {
            ];
            let schema = RelationType::new(vec![
                ColumnType::new(CDT::uint32_datatype(), false),
-                ColumnType::new(CDT::datetime_datatype(), false),
+                ColumnType::new(CDT::timestamp_millisecond_datatype(), false),
            ]);
            schemas.insert(
                gid,
@@ -232,7 +232,11 @@ mod test {

        let schema = vec![
            datatypes::schema::ColumnSchema::new("number", CDT::uint32_datatype(), false),
-            datatypes::schema::ColumnSchema::new("ts", CDT::datetime_datatype(), false),
+            datatypes::schema::ColumnSchema::new(
+                "ts",
+                CDT::timestamp_millisecond_datatype(),
+                false,
+            ),
        ];
        let mut columns = vec![];
        let numbers = (1..=10).collect_vec();
@@ -240,7 +244,11 @@ mod test {
        columns.push(column);

        let ts = (1..=10).collect_vec();
-        let column: VectorRef = Arc::new(<DateTime as Scalar>::VectorType::from_vec(ts));
+        let mut builder = TimestampMillisecondVectorBuilder::with_capacity(10);
+        ts.into_iter()
+            .map(|v| builder.push(Some(TimestampMillisecond::new(v))))
+            .count();
+        let column: VectorRef = builder.to_vector_cloned();
        columns.push(column);

        let schema = Arc::new(Schema::new(schema));
--- a/src/flow/src/transform/aggr.rs
+++ b/src/flow/src/transform/aggr.rs
@@ -345,9 +345,10 @@ impl TypedPlan {
 #[cfg(test)]
 mod test {
    use std::collections::BTreeMap;
+    use std::time::Duration;

    use bytes::BytesMut;
-    use common_time::{DateTime, Interval};
+    use common_time::{IntervalMonthDayNano, Timestamp};
    use datatypes::prelude::ConcreteDataType;
    use datatypes::value::Value;
    use pretty_assertions::assert_eq;
@@ -398,7 +399,10 @@ mod test {
                            .with_types(
                                RelationType::new(vec![
                                    ColumnType::new(ConcreteDataType::uint32_datatype(), false),
-                                    ColumnType::new(ConcreteDataType::datetime_datatype(), false),
+                                    ColumnType::new(
+                                        ConcreteDataType::timestamp_millisecond_datatype(),
+                                        false,
+                                    ),
                                ])
                                .into_named(vec![
                                    Some("number".to_string()),
@@ -413,22 +417,18 @@ mod test {
                                .map(vec![
                                    ScalarExpr::Column(1).call_unary(
                                        UnaryFunc::TumbleWindowFloor {
-                                            window_size: Interval::from_month_day_nano(
-                                                0,
-                                                0,
-                                                1_000_000_000,
-                                            ),
-                                            start_time: Some(DateTime::new(1625097600000)),
+                                            window_size: Duration::from_nanos(1_000_000_000),
+                                            start_time: Some(Timestamp::new_millisecond(
+                                                1625097600000,
+                                            )),
                                        },
                                    ),
                                    ScalarExpr::Column(1).call_unary(
                                        UnaryFunc::TumbleWindowCeiling {
-                                            window_size: Interval::from_month_day_nano(
-                                                0,
-                                                0,
-                                                1_000_000_000,
-                                            ),
-                                            start_time: Some(DateTime::new(1625097600000)),
+                                            window_size: Duration::from_nanos(1_000_000_000),
+                                            start_time: Some(Timestamp::new_millisecond(
+                                                1625097600000,
+                                            )),
                                        },
                                    ),
                                ])
@@ -539,7 +539,10 @@ mod test {
                            .with_types(
                                RelationType::new(vec![
                                    ColumnType::new(ConcreteDataType::uint32_datatype(), false),
-                                    ColumnType::new(ConcreteDataType::datetime_datatype(), false),
+                                    ColumnType::new(
+                                        ConcreteDataType::timestamp_millisecond_datatype(),
+                                        false,
+                                    ),
                                ])
                                .into_named(vec![
                                    Some("number".to_string()),
@@ -554,22 +557,18 @@ mod test {
                                .map(vec![
                                    ScalarExpr::Column(1).call_unary(
                                        UnaryFunc::TumbleWindowFloor {
-                                            window_size: Interval::from_month_day_nano(
-                                                0,
-                                                0,
-                                                1_000_000_000,
-                                            ),
-                                            start_time: Some(DateTime::new(1625097600000)),
+                                            window_size: Duration::from_nanos(1_000_000_000),
+                                            start_time: Some(Timestamp::new_millisecond(
+                                                1625097600000,
+                                            )),
                                        },
                                    ),
                                    ScalarExpr::Column(1).call_unary(
                                        UnaryFunc::TumbleWindowCeiling {
-                                            window_size: Interval::from_month_day_nano(
-                                                0,
-                                                0,
-                                                1_000_000_000,
-                                            ),
-                                            start_time: Some(DateTime::new(1625097600000)),
+                                            window_size: Duration::from_nanos(1_000_000_000),
+                                            start_time: Some(Timestamp::new_millisecond(
+                                                1625097600000,
+                                            )),
                                        },
                                    ),
                                ])
@@ -686,7 +685,10 @@ mod test {
                            .with_types(
                                RelationType::new(vec![
                                    ColumnType::new(ConcreteDataType::uint32_datatype(), false),
-                                    ColumnType::new(ConcreteDataType::datetime_datatype(), false),
+                                    ColumnType::new(
+                                        ConcreteDataType::timestamp_millisecond_datatype(),
+                                        false,
+                                    ),
                                ])
                                .into_named(vec![
                                    Some("number".to_string()),
@@ -701,21 +703,13 @@ mod test {
                                .map(vec![
                                    ScalarExpr::Column(1).call_unary(
                                        UnaryFunc::TumbleWindowFloor {
-                                            window_size: Interval::from_month_day_nano(
-                                                0,
-                                                0,
-                                                3_600_000_000_000,
-                                            ),
+                                            window_size: Duration::from_nanos(3_600_000_000_000),
                                            start_time: None,
                                        },
                                    ),
                                    ScalarExpr::Column(1).call_unary(
                                        UnaryFunc::TumbleWindowCeiling {
-                                            window_size: Interval::from_month_day_nano(
-                                                0,
-                                                0,
-                                                3_600_000_000_000,
-                                            ),
+                                            window_size: Duration::from_nanos(3_600_000_000_000),
                                            start_time: None,
                                        },
                                    ),
@@ -833,7 +827,10 @@ mod test {
                            .with_types(
                                RelationType::new(vec![
                                    ColumnType::new(ConcreteDataType::uint32_datatype(), false),
-                                    ColumnType::new(ConcreteDataType::datetime_datatype(), false),
+                                    ColumnType::new(
+                                        ConcreteDataType::timestamp_millisecond_datatype(),
+                                        false,
+                                    ),
                                ])
                                .into_named(vec![
                                    Some("number".to_string()),
@@ -848,21 +845,13 @@ mod test {
                                .map(vec![
                                    ScalarExpr::Column(1).call_unary(
                                        UnaryFunc::TumbleWindowFloor {
-                                            window_size: Interval::from_month_day_nano(
-                                                0,
-                                                0,
-                                                3_600_000_000_000,
-                                            ),
+                                            window_size: Duration::from_nanos(3_600_000_000_000),
                                            start_time: None,
                                        },
                                    ),
                                    ScalarExpr::Column(1).call_unary(
                                        UnaryFunc::TumbleWindowCeiling {
-                                            window_size: Interval::from_month_day_nano(
-                                                0,
-                                                0,
-                                                3_600_000_000_000,
-                                            ),
+                                            window_size: Duration::from_nanos(3_600_000_000_000),
                                            start_time: None,
                                        },
                                    ),
@@ -948,7 +937,10 @@ mod test {
                            .with_types(
                                RelationType::new(vec![
                                    ColumnType::new(ConcreteDataType::uint32_datatype(), false),
-                                    ColumnType::new(ConcreteDataType::datetime_datatype(), false),
+                                    ColumnType::new(
+                                        ConcreteDataType::timestamp_millisecond_datatype(),
+                                        false,
+                                    ),
                                ])
                                .into_named(vec![
                                    Some("number".to_string()),
@@ -963,22 +955,18 @@ mod test {
                                .map(vec![
                                    ScalarExpr::Column(1).call_unary(
                                        UnaryFunc::TumbleWindowFloor {
-                                            window_size: Interval::from_month_day_nano(
-                                                0,
-                                                0,
-                                                3_600_000_000_000,
-                                            ),
-                                            start_time: Some(DateTime::new(1625097600000)),
+                                            window_size: Duration::from_nanos(3_600_000_000_000),
+                                            start_time: Some(Timestamp::new_millisecond(
+                                                1625097600000,
+                                            )),
                                        },
                                    ),
                                    ScalarExpr::Column(1).call_unary(
                                        UnaryFunc::TumbleWindowCeiling {
-                                            window_size: Interval::from_month_day_nano(
-                                                0,
-                                                0,
-                                                3_600_000_000_000,
-                                            ),
-                                            start_time: Some(DateTime::new(1625097600000)),
+                                            window_size: Duration::from_nanos(3_600_000_000_000),
+                                            start_time: Some(Timestamp::new_millisecond(
+                                                1625097600000,
+                                            )),
                                        },
                                    ),
                                ])
@@ -1512,7 +1500,7 @@ mod test {
                            .with_types(
                                RelationType::new(vec![
                                    ColumnType::new(ConcreteDataType::uint32_datatype(), false),
-                                    ColumnType::new(ConcreteDataType::datetime_datatype(), false),
+                                    ColumnType::new(ConcreteDataType::timestamp_millisecond_datatype(), false),
                                ])
                                .into_named(vec![
                                    Some("number".to_string()),
@@ -1536,7 +1524,7 @@ mod test {
                                                true,
                                            ),ColumnType::new(
                                                ConcreteDataType::timestamp_millisecond_datatype(),
-                                                true,
+                                                false,
                                            )])
                                            .into_unnamed(),
                                            extensions: FunctionExtensions {
@@ -1554,10 +1542,10 @@ mod test {
                                    .unwrap(),
                                    exprs: vec![
                                        ScalarExpr::Literal(
-                                            Value::Interval(Interval::from_month_day_nano(0, 0, 30000000000)),
+                                            Value::IntervalMonthDayNano(IntervalMonthDayNano::new(0, 0, 30000000000)),
                                            CDT::interval_month_day_nano_datatype()
                                        ),
-                                        ScalarExpr::Column(1).cast(CDT::timestamp_millisecond_datatype())
+                                        ScalarExpr::Column(1)
                                        ],
                                }])
                                .unwrap()
--- a/src/flow/src/transform/literal.rs
+++ b/src/flow/src/transform/literal.rs
@@ -178,14 +178,14 @@ pub(crate) fn from_substrait_literal(lit: &Literal) -> Result<(Value, CDT), Erro
            let (days, seconds, microseconds) =
                (interval.days, interval.seconds, interval.microseconds);
            let millis = microseconds / 1000 + seconds * 1000;
-            let value_interval = common_time::Interval::from_day_time(days, millis);
+            let value_interval = common_time::IntervalDayTime::new(days, millis);
            (
-                Value::Interval(value_interval),
+                Value::IntervalDayTime(value_interval),
                CDT::interval_day_time_datatype(),
            )
        }
        Some(LiteralType::IntervalYearToMonth(interval)) => (
-            Value::Interval(common_time::Interval::from_year_month(
+            Value::IntervalYearMonth(common_time::IntervalYearMonth::new(
                interval.years * 12 + interval.months,
            )),
            CDT::interval_year_month_datatype(),
@@ -239,9 +239,9 @@ fn from_substrait_user_defined_type(user_defined: &UserDefined) -> Result<(Value
                    }
                );
                let i: i32 = from_bytes(&val.value)?;
-                let value_interval = common_time::Interval::from_year_month(i);
+                let value_interval = common_time::IntervalYearMonth::new(i);
                (
-                    Value::Interval(value_interval),
+                    Value::IntervalYearMonth(value_interval),
                    CDT::interval_year_month_datatype(),
                )
            }
@@ -255,12 +255,12 @@ fn from_substrait_user_defined_type(user_defined: &UserDefined) -> Result<(Value
                        )
                    }
                );
+                // TODO(yingwen): Datafusion may change the representation of the interval type.
                let i: i128 = from_bytes(&val.value)?;
                let (months, days, nsecs) = ((i >> 96) as i32, (i >> 64) as i32, i as i64);
-                let value_interval =
-                    common_time::Interval::from_month_day_nano(months, days, nsecs);
+                let value_interval = common_time::IntervalMonthDayNano::new(months, days, nsecs);
                (
-                    Value::Interval(value_interval),
+                    Value::IntervalMonthDayNano(value_interval),
                    CDT::interval_month_day_nano_datatype(),
                )
            }
@@ -274,11 +274,12 @@ fn from_substrait_user_defined_type(user_defined: &UserDefined) -> Result<(Value
                        )
                    }
                );
+                // TODO(yingwen): Datafusion may change the representation of the interval type.
                let i: i64 = from_bytes(&val.value)?;
                let (days, millis) = ((i >> 32) as i32, i as i32);
-                let value_interval = common_time::Interval::from_day_time(days, millis);
+                let value_interval = common_time::IntervalDayTime::new(days, millis);
                (
-                    Value::Interval(value_interval),
+                    Value::IntervalDayTime(value_interval),
                    CDT::interval_day_time_datatype(),
                )
            }
--- a/src/frontend/src/instance/otlp.rs
+++ b/src/frontend/src/instance/otlp.rs
@@ -17,8 +17,10 @@ use auth::{PermissionChecker, PermissionCheckerRef, PermissionReq};
 use client::Output;
 use common_error::ext::BoxedError;
 use common_telemetry::tracing;
+use opentelemetry_proto::tonic::collector::logs::v1::ExportLogsServiceRequest;
 use opentelemetry_proto::tonic::collector::metrics::v1::ExportMetricsServiceRequest;
 use opentelemetry_proto::tonic::collector::trace::v1::ExportTraceServiceRequest;
+use pipeline::PipelineWay;
 use servers::error::{self, AuthSnafu, Result as ServerResult};
 use servers::interceptor::{OpenTelemetryProtocolInterceptor, OpenTelemetryProtocolInterceptorRef};
 use servers::otlp;
@@ -28,7 +30,7 @@ use session::context::QueryContextRef;
 use snafu::ResultExt;

 use crate::instance::Instance;
-use crate::metrics::{OTLP_METRICS_ROWS, OTLP_TRACES_ROWS};
+use crate::metrics::{OTLP_LOGS_ROWS, OTLP_METRICS_ROWS, OTLP_TRACES_ROWS};

 #[async_trait]
 impl OpenTelemetryProtocolHandler for Instance {
@@ -92,4 +94,31 @@ impl OpenTelemetryProtocolHandler for Instance {
            .map_err(BoxedError::new)
            .context(error::ExecuteGrpcQuerySnafu)
    }
+
+    #[tracing::instrument(skip_all)]
+    async fn logs(
+        &self,
+        request: ExportLogsServiceRequest,
+        pipeline: PipelineWay,
+        table_name: String,
+        ctx: QueryContextRef,
+    ) -> ServerResult<Output> {
+        self.plugins
+            .get::<PermissionCheckerRef>()
+            .as_ref()
+            .check_permission(ctx.current_user(), PermissionReq::Otlp)
+            .context(AuthSnafu)?;
+
+        let interceptor_ref = self
+            .plugins
+            .get::<OpenTelemetryProtocolInterceptorRef<servers::error::Error>>();
+        interceptor_ref.pre_execute(ctx.clone())?;
+        let (requests, rows) = otlp::logs::to_grpc_insert_requests(request, pipeline, table_name)?;
+
+        self.handle_log_inserts(requests, ctx)
+            .await
+            .inspect(|_| OTLP_LOGS_ROWS.inc_by(rows as u64))
+            .map_err(BoxedError::new)
+            .context(error::ExecuteGrpcQuerySnafu)
+    }
 }
--- a/src/frontend/src/metrics.rs
+++ b/src/frontend/src/metrics.rs
@@ -41,16 +41,28 @@ lazy_static! {
        .with_label_values(&["insert"]);
    pub static ref EXECUTE_SCRIPT_ELAPSED: Histogram = HANDLE_SCRIPT_ELAPSED
        .with_label_values(&["execute"]);
+
+    /// The number of OpenTelemetry metrics send by frontend node.
    pub static ref OTLP_METRICS_ROWS: IntCounter = register_int_counter!(
        "greptime_frontend_otlp_metrics_rows",
        "frontend otlp metrics rows"
    )
    .unwrap();
+
+    /// The number of OpenTelemetry traces send by frontend node.
    pub static ref OTLP_TRACES_ROWS: IntCounter = register_int_counter!(
        "greptime_frontend_otlp_traces_rows",
        "frontend otlp traces rows"
    )
    .unwrap();
+
+    /// The number of OpenTelemetry logs send by frontend node.
+    pub static ref OTLP_LOGS_ROWS: IntCounter = register_int_counter!(
+        "greptime_frontend_otlp_logs_rows",
+        "frontend otlp logs rows"
+    )
+    .unwrap();
+
    /// The number of heartbeats send by frontend node.
    pub static ref HEARTBEAT_SENT_COUNT: IntCounter = register_int_counter!(
        "greptime_frontend_heartbeat_send_count",
--- a/src/log-store/src/kafka/log_store.rs
+++ b/src/log-store/src/kafka/log_store.rs
@@ -49,6 +49,8 @@ pub struct KafkaLogStore {
    max_batch_bytes: usize,
    /// The consumer wait timeout.
    consumer_wait_timeout: Duration,
+    /// Ignore missing entries during read WAL.
+    overwrite_entry_start_id: bool,
 }

 impl KafkaLogStore {
@@ -64,6 +66,7 @@ impl KafkaLogStore {
            client_manager,
            max_batch_bytes: config.max_batch_bytes.as_bytes() as usize,
            consumer_wait_timeout: config.consumer_wait_timeout,
+            overwrite_entry_start_id: config.overwrite_entry_start_id,
        })
    }
 }
@@ -205,7 +208,7 @@ impl LogStore for KafkaLogStore {
    async fn read(
        &self,
        provider: &Provider,
-        entry_id: EntryId,
+        mut entry_id: EntryId,
        index: Option<WalIndex>,
    ) -> Result<SendableEntryStream<'static, Entry, Self::Error>> {
        let provider = provider
@@ -225,6 +228,25 @@ impl LogStore for KafkaLogStore {
            .client()
            .clone();

+        if self.overwrite_entry_start_id {
+            let start_offset =
+                client
+                    .get_offset(OffsetAt::Earliest)
+                    .await
+                    .context(GetOffsetSnafu {
+                        topic: &provider.topic,
+                    })?;
+
+            if entry_id as i64 <= start_offset {
+                warn!(
+                "The entry_id: {} is less than start_offset: {}, topic: {}. Overwriting entry_id with start_offset",
+                entry_id, start_offset, &provider.topic
+            );
+
+                entry_id = start_offset as u64;
+            }
+        }
+
        // Gets the offset of the latest record in the topic. Actually, it's the latest record of the single partition in the topic.
        // The read operation terminates when this record is consumed.
        // Warning: the `get_offset` returns the end offset of the latest record. For our usage, it should be decremented.
--- a/src/meta-client/examples/lock.rs
+++ b/src/meta-client/examples/lock.rs
@@ -1,129 +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::sync::Arc;
-use std::time::Duration;
-
-use api::v1::meta::Role;
-use common_grpc::channel_manager::{ChannelConfig, ChannelManager};
-use common_meta::rpc::lock::{LockRequest, UnlockRequest};
-use meta_client::client::MetaClientBuilder;
-use meta_client::MetaClientRef;
-use tracing::{info, subscriber};
-use tracing_subscriber::FmtSubscriber;
-
-fn main() {
-    subscriber::set_global_default(FmtSubscriber::builder().finish()).unwrap();
-    run();
-}
-
-#[tokio::main]
-async fn run() {
-    let id = (1000u64, 2000u64);
-    let config = ChannelConfig::new()
-        .timeout(Duration::from_secs(30))
-        .connect_timeout(Duration::from_secs(5))
-        .tcp_nodelay(true);
-    let channel_manager = ChannelManager::with_config(config);
-    let mut meta_client = MetaClientBuilder::new(id.0, id.1, Role::Datanode)
-        .enable_lock()
-        .channel_manager(channel_manager)
-        .build();
-    meta_client.start(&["127.0.0.1:3002"]).await.unwrap();
-    let meta_client = Arc::new(meta_client);
-
-    run_normal(meta_client.clone()).await;
-
-    run_multi_thread(meta_client.clone()).await;
-
-    run_multi_thread_with_one_timeout(meta_client).await;
-}
-
-async fn run_normal(meta_client: MetaClientRef) {
-    let name = "lock_name".as_bytes().to_vec();
-    let expire_secs = 60;
-
-    let lock_req = LockRequest { name, expire_secs };
-
-    let lock_result = meta_client.lock(lock_req).await.unwrap();
-    let key = lock_result.key;
-    info!(
-        "lock success! Returned key: {}",
-        String::from_utf8(key.clone()).unwrap()
-    );
-
-    // It is recommended that time of holding lock is less than the timeout of the grpc channel
-    info!("do some work, take 3 seconds");
-    tokio::time::sleep(Duration::from_secs(3)).await;
-
-    let unlock_req = UnlockRequest { key };
-
-    meta_client.unlock(unlock_req).await.unwrap();
-    info!("unlock success!");
-}
-
-async fn run_multi_thread(meta_client: MetaClientRef) {
-    let meta_client_clone = meta_client.clone();
-    let join1 = tokio::spawn(async move {
-        run_normal(meta_client_clone.clone()).await;
-    });
-
-    tokio::time::sleep(Duration::from_secs(1)).await;
-
-    let join2 = tokio::spawn(async move {
-        run_normal(meta_client).await;
-    });
-
-    join1.await.unwrap();
-    join2.await.unwrap();
-}
-
-async fn run_multi_thread_with_one_timeout(meta_client: MetaClientRef) {
-    let meta_client_clone = meta_client.clone();
-    let join1 = tokio::spawn(async move {
-        run_with_timeout(meta_client_clone.clone()).await;
-    });
-
-    tokio::time::sleep(Duration::from_secs(1)).await;
-
-    let join2 = tokio::spawn(async move {
-        run_normal(meta_client).await;
-    });
-
-    join1.await.unwrap();
-    join2.await.unwrap();
-}
-
-async fn run_with_timeout(meta_client: MetaClientRef) {
-    let name = "lock_name".as_bytes().to_vec();
-    let expire_secs = 5;
-
-    let lock_req = LockRequest { name, expire_secs };
-
-    let lock_result = meta_client.lock(lock_req).await.unwrap();
-    let key = lock_result.key;
-    info!(
-        "lock success! Returned key: {}",
-        String::from_utf8(key.clone()).unwrap()
-    );
-
-    // It is recommended that time of holding lock is less than the timeout of the grpc channel
-    info!("do some work, take 20 seconds");
-    tokio::time::sleep(Duration::from_secs(20)).await;
-
-    let unlock_req = UnlockRequest { key };
-
-    meta_client.unlock(unlock_req).await.unwrap();
-    info!("unlock success!");
-}
--- a/src/meta-client/src/client.rs
+++ b/src/meta-client/src/client.rs
@@ -15,7 +15,6 @@
 mod ask_leader;
 mod heartbeat;
 mod load_balance;
-mod lock;
 mod procedure;

 mod cluster;
@@ -33,7 +32,6 @@ use common_meta::datanode::{DatanodeStatKey, DatanodeStatValue, RegionStat};
 use common_meta::ddl::{ExecutorContext, ProcedureExecutor};
 use common_meta::error::{self as meta_error, Result as MetaResult};
 use common_meta::rpc::ddl::{SubmitDdlTaskRequest, SubmitDdlTaskResponse};
-use common_meta::rpc::lock::{LockRequest, LockResponse, UnlockRequest};
 use common_meta::rpc::procedure::{
    MigrateRegionRequest, MigrateRegionResponse, ProcedureStateResponse,
 };
@@ -45,7 +43,6 @@ use common_meta::rpc::store::{
 use common_meta::ClusterId;
 use common_telemetry::info;
 use heartbeat::Client as HeartbeatClient;
-use lock::Client as LockClient;
 use procedure::Client as ProcedureClient;
 use snafu::{OptionExt, ResultExt};
 use store::Client as StoreClient;
@@ -67,7 +64,6 @@ pub struct MetaClientBuilder {
    role: Role,
    enable_heartbeat: bool,
    enable_store: bool,
-    enable_lock: bool,
    enable_procedure: bool,
    enable_access_cluster_info: bool,
    channel_manager: Option<ChannelManager>,
@@ -123,13 +119,6 @@ impl MetaClientBuilder {
        }
    }

-    pub fn enable_lock(self) -> Self {
-        Self {
-            enable_lock: true,
-            ..self
-        }
-    }
-
    pub fn enable_procedure(self) -> Self {
        Self {
            enable_procedure: true,
@@ -188,10 +177,6 @@ impl MetaClientBuilder {
            client.store = Some(StoreClient::new(self.id, self.role, mgr.clone()));
        }

-        if self.enable_lock {
-            client.lock = Some(LockClient::new(self.id, self.role, mgr.clone()));
-        }
-
        if self.enable_procedure {
            let mgr = self.ddl_channel_manager.unwrap_or(mgr.clone());
            client.procedure = Some(ProcedureClient::new(
@@ -221,7 +206,6 @@ pub struct MetaClient {
    channel_manager: ChannelManager,
    heartbeat: Option<HeartbeatClient>,
    store: Option<StoreClient>,
-    lock: Option<LockClient>,
    procedure: Option<ProcedureClient>,
    cluster: Option<ClusterClient>,
 }
@@ -383,10 +367,6 @@ impl MetaClient {
            client.start(urls.clone()).await?;
            info!("Store client started");
        }
-        if let Some(client) = &mut self.lock {
-            client.start(urls.clone()).await?;
-            info!("Lock client started");
-        }
        if let Some(client) = &mut self.procedure {
            client.start(urls.clone()).await?;
            info!("DDL client started");
@@ -482,15 +462,6 @@ impl MetaClient {
            .context(ConvertMetaResponseSnafu)
    }

-    pub async fn lock(&self, req: LockRequest) -> Result<LockResponse> {
-        self.lock_client()?.lock(req.into()).await.map(Into::into)
-    }
-
-    pub async fn unlock(&self, req: UnlockRequest) -> Result<()> {
-        let _ = self.lock_client()?.unlock(req.into()).await?;
-        Ok(())
-    }
-
    /// Query the procedure state by its id.
    pub async fn query_procedure_state(&self, pid: &str) -> Result<ProcedureStateResponse> {
        self.procedure_client()?.query_procedure_state(pid).await
@@ -538,12 +509,6 @@ impl MetaClient {
        })
    }

-    pub fn lock_client(&self) -> Result<LockClient> {
-        self.lock.clone().context(NotStartedSnafu {
-            name: "lock_client",
-        })
-    }
-
    pub fn procedure_client(&self) -> Result<ProcedureClient> {
        self.procedure.clone().context(NotStartedSnafu {
            name: "procedure_client",
--- a/src/meta-client/src/client/lock.rs
+++ b/src/meta-client/src/client/lock.rs
@@ -1,178 +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 std::sync::Arc;
-
-use api::v1::meta::lock_client::LockClient;
-use api::v1::meta::{LockRequest, LockResponse, Role, UnlockRequest, UnlockResponse};
-use common_grpc::channel_manager::ChannelManager;
-use common_telemetry::tracing_context::TracingContext;
-use snafu::{ensure, OptionExt, ResultExt};
-use tokio::sync::RwLock;
-use tonic::transport::Channel;
-
-use crate::client::{load_balance, Id};
-use crate::error;
-use crate::error::Result;
-
-#[derive(Clone, Debug)]
-pub struct Client {
-    inner: Arc<RwLock<Inner>>,
-}
-
-impl Client {
-    pub fn new(id: Id, role: Role, channel_manager: ChannelManager) -> Self {
-        let inner = Arc::new(RwLock::new(Inner {
-            id,
-            role,
-            channel_manager,
-            peers: vec![],
-        }));
-
-        Self { inner }
-    }
-
-    pub async fn start<U, A>(&mut self, urls: A) -> Result<()>
-    where
-        U: AsRef<str>,
-        A: AsRef<[U]>,
-    {
-        let mut inner = self.inner.write().await;
-        inner.start(urls).await
-    }
-
-    pub async fn lock(&self, req: LockRequest) -> Result<LockResponse> {
-        let inner = self.inner.read().await;
-        inner.lock(req).await
-    }
-
-    pub async fn unlock(&self, req: UnlockRequest) -> Result<UnlockResponse> {
-        let inner = self.inner.read().await;
-        inner.unlock(req).await
-    }
-}
-
-#[derive(Debug)]
-struct Inner {
-    id: Id,
-    role: Role,
-    channel_manager: ChannelManager,
-    peers: Vec<String>,
-}
-
-impl Inner {
-    async fn start<U, A>(&mut self, urls: A) -> Result<()>
-    where
-        U: AsRef<str>,
-        A: AsRef<[U]>,
-    {
-        ensure!(
-            !self.is_started(),
-            error::IllegalGrpcClientStateSnafu {
-                err_msg: "Lock client already started",
-            }
-        );
-
-        self.peers = urls
-            .as_ref()
-            .iter()
-            .map(|url| url.as_ref().to_string())
-            .collect::<HashSet<_>>()
-            .drain()
-            .collect::<Vec<_>>();
-
-        Ok(())
-    }
-
-    fn random_client(&self) -> Result<LockClient<Channel>> {
-        let len = self.peers.len();
-        let peer = load_balance::random_get(len, |i| Some(&self.peers[i])).context(
-            error::IllegalGrpcClientStateSnafu {
-                err_msg: "Empty peers, lock client may not start yet",
-            },
-        )?;
-
-        self.make_client(peer)
-    }
-
-    fn make_client(&self, addr: impl AsRef<str>) -> Result<LockClient<Channel>> {
-        let channel = self
-            .channel_manager
-            .get(addr)
-            .context(error::CreateChannelSnafu)?;
-
-        Ok(LockClient::new(channel))
-    }
-
-    #[inline]
-    fn is_started(&self) -> bool {
-        !self.peers.is_empty()
-    }
-
-    async fn lock(&self, mut req: LockRequest) -> Result<LockResponse> {
-        let mut client = self.random_client()?;
-        req.set_header(
-            self.id,
-            self.role,
-            TracingContext::from_current_span().to_w3c(),
-        );
-        let res = client.lock(req).await.map_err(error::Error::from)?;
-
-        Ok(res.into_inner())
-    }
-
-    async fn unlock(&self, mut req: UnlockRequest) -> Result<UnlockResponse> {
-        let mut client = self.random_client()?;
-        req.set_header(
-            self.id,
-            self.role,
-            TracingContext::from_current_span().to_w3c(),
-        );
-        let res = client.unlock(req).await.map_err(error::Error::from)?;
-
-        Ok(res.into_inner())
-    }
-}
-
-#[cfg(test)]
-mod tests {
-    use super::*;
-
-    #[tokio::test]
-    async fn test_already_start() {
-        let mut client = Client::new((0, 0), Role::Datanode, ChannelManager::default());
-        client
-            .start(&["127.0.0.1:1000", "127.0.0.1:1001"])
-            .await
-            .unwrap();
-        let res = client.start(&["127.0.0.1:1002"]).await;
-        assert!(res.is_err());
-        assert!(matches!(
-            res.err(),
-            Some(error::Error::IllegalGrpcClientState { .. })
-        ));
-    }
-
-    #[tokio::test]
-    async fn test_start_with_duplicate_peers() {
-        let mut client = Client::new((0, 0), Role::Datanode, ChannelManager::default());
-        client
-            .start(&["127.0.0.1:1000", "127.0.0.1:1000", "127.0.0.1:1000"])
-            .await
-            .unwrap();
-
-        assert_eq!(1, client.inner.write().await.peers.len());
-    }
-}
--- a/src/meta-srv/src/bootstrap.rs
+++ b/src/meta-srv/src/bootstrap.rs
@@ -16,7 +16,6 @@ use std::sync::Arc;

 use api::v1::meta::cluster_server::ClusterServer;
 use api::v1::meta::heartbeat_server::HeartbeatServer;
-use api::v1::meta::lock_server::LockServer;
 use api::v1::meta::procedure_service_server::ProcedureServiceServer;
 use api::v1::meta::store_server::StoreServer;
 use common_base::Plugins;
@@ -48,8 +47,6 @@ use crate::election::etcd::EtcdElection;
 #[cfg(feature = "pg_kvbackend")]
 use crate::error::InvalidArgumentsSnafu;
 use crate::error::{InitExportMetricsTaskSnafu, TomlFormatSnafu};
-use crate::lock::etcd::EtcdLock;
-use crate::lock::memory::MemLock;
 use crate::metasrv::builder::MetasrvBuilder;
 use crate::metasrv::{BackendImpl, Metasrv, MetasrvOptions, SelectorRef};
 use crate::selector::lease_based::LeaseBasedSelector;
@@ -59,9 +56,8 @@ use crate::selector::SelectorType;
 use crate::service::admin;
 use crate::{error, Result};

-#[derive(Clone)]
 pub struct MetasrvInstance {
-    metasrv: Metasrv,
+    metasrv: Arc<Metasrv>,

    httpsrv: Arc<HttpServer>,

@@ -86,8 +82,9 @@ impl MetasrvInstance {
                .with_greptime_config_options(opts.to_toml().context(TomlFormatSnafu)?)
                .build(),
        );
+        let metasrv = Arc::new(metasrv);
        // put metasrv into plugins for later use
-        plugins.insert::<Arc<Metasrv>>(Arc::new(metasrv.clone()));
+        plugins.insert::<Arc<Metasrv>>(metasrv.clone());
        let export_metrics_task = ExportMetricsTask::try_new(&opts.export_metrics, Some(&plugins))
            .context(InitExportMetricsTaskSnafu)?;
        Ok(MetasrvInstance {
@@ -151,6 +148,10 @@ impl MetasrvInstance {
    pub fn plugins(&self) -> Plugins {
        self.plugins.clone()
    }
+
+    pub fn get_inner(&self) -> &Metasrv {
+        &self.metasrv
+    }
 }

 pub async fn bootstrap_metasrv_with_router(
@@ -177,14 +178,13 @@ pub async fn bootstrap_metasrv_with_router(
    Ok(())
 }

-pub fn router(metasrv: Metasrv) -> Router {
+pub fn router(metasrv: Arc<Metasrv>) -> Router {
    tonic::transport::Server::builder()
        .accept_http1(true) // for admin services
-        .add_service(HeartbeatServer::new(metasrv.clone()))
-        .add_service(StoreServer::new(metasrv.clone()))
-        .add_service(ClusterServer::new(metasrv.clone()))
-        .add_service(LockServer::new(metasrv.clone()))
-        .add_service(ProcedureServiceServer::new(metasrv.clone()))
+        .add_service(HeartbeatServer::from_arc(metasrv.clone()))
+        .add_service(StoreServer::from_arc(metasrv.clone()))
+        .add_service(ClusterServer::from_arc(metasrv.clone()))
+        .add_service(ProcedureServiceServer::from_arc(metasrv.clone()))
        .add_service(admin::make_admin_service(metasrv))
 }

@@ -193,13 +193,9 @@ pub async fn metasrv_builder(
    plugins: Plugins,
    kv_backend: Option<KvBackendRef>,
 ) -> Result<MetasrvBuilder> {
-    let (kv_backend, election, lock) = match (kv_backend, &opts.backend) {
-        (Some(kv_backend), _) => (kv_backend, None, Some(Arc::new(MemLock::default()) as _)),
-        (None, BackendImpl::MemoryStore) => (
-            Arc::new(MemoryKvBackend::new()) as _,
-            None,
-            Some(Arc::new(MemLock::default()) as _),
-        ),
+    let (kv_backend, election) = match (kv_backend, &opts.backend) {
+        (Some(kv_backend), _) => (kv_backend, None),
+        (None, BackendImpl::MemoryStore) => (Arc::new(MemoryKvBackend::new()) as _, None),
        (None, BackendImpl::EtcdStore) => {
            let etcd_client = create_etcd_client(opts).await?;
            let kv_backend = {
@@ -224,18 +220,13 @@ pub async fn metasrv_builder(
                    )
                    .await?,
                ),
-                Some(EtcdLock::with_etcd_client(
-                    etcd_client,
-                    opts.store_key_prefix.clone(),
-                )?),
            )
        }
        #[cfg(feature = "pg_kvbackend")]
        (None, BackendImpl::PostgresStore) => {
            let pg_client = create_postgres_client(opts).await?;
            let kv_backend = PgStore::with_pg_client(pg_client).await.unwrap();
-            // TODO: implement locking and leader election for pg backend.
-            (kv_backend, None, None)
+            (kv_backend, None)
        }
    };

@@ -253,7 +244,6 @@ pub async fn metasrv_builder(
        .in_memory(in_memory)
        .selector(selector)
        .election(election)
-        .lock(lock)
        .plugins(plugins))
 }

--- a/src/meta-srv/src/error.rs
+++ b/src/meta-srv/src/error.rs
@@ -448,30 +448,6 @@ pub enum Error {
        location: Location,
    },

-    #[snafu(display("Failed to lock based on etcd"))]
-    Lock {
-        #[snafu(source)]
-        error: etcd_client::Error,
-        #[snafu(implicit)]
-        location: Location,
-    },
-
-    #[snafu(display("Failed to unlock based on etcd"))]
-    Unlock {
-        #[snafu(source)]
-        error: etcd_client::Error,
-        #[snafu(implicit)]
-        location: Location,
-    },
-
-    #[snafu(display("Failed to grant lease"))]
-    LeaseGrant {
-        #[snafu(source)]
-        error: etcd_client::Error,
-        #[snafu(implicit)]
-        location: Location,
-    },
-
    #[snafu(display("Invalid utf-8 value"))]
    InvalidUtf8Value {
        #[snafu(source)]
@@ -770,9 +746,6 @@ impl ErrorExt for Error {
            | Error::ResponseHeaderNotFound { .. }
            | Error::IsNotLeader { .. }
            | Error::InvalidHttpBody { .. }
-            | Error::Lock { .. }
-            | Error::Unlock { .. }
-            | Error::LeaseGrant { .. }
            | Error::ExceededRetryLimit { .. }
            | Error::SendShutdownSignal { .. }
            | Error::PusherNotFound { .. }
--- a/src/meta-srv/src/handler.rs
+++ b/src/meta-srv/src/handler.rs
@@ -14,7 +14,7 @@

 use std::collections::{BTreeMap, HashSet};
 use std::ops::Range;
-use std::sync::Arc;
+use std::sync::{Arc, Mutex};
 use std::time::{Duration, Instant};

 use api::v1::meta::mailbox_message::Payload;
@@ -209,15 +209,16 @@ impl Pushers {
    }
 }

-struct NameCachedHandler {
+#[derive(Clone)]
+pub struct NameCachedHandler {
    name: &'static str,
-    handler: Box<dyn HeartbeatHandler>,
+    handler: Arc<dyn HeartbeatHandler>,
 }

 impl NameCachedHandler {
    fn new(handler: impl HeartbeatHandler + 'static) -> Self {
        let name = handler.name();
-        let handler = Box::new(handler);
+        let handler = Arc::new(handler);
        Self { name, handler }
    }
 }
@@ -225,7 +226,7 @@ impl NameCachedHandler {
 pub type HeartbeatHandlerGroupRef = Arc<HeartbeatHandlerGroup>;

 /// The group of heartbeat handlers.
-#[derive(Default)]
+#[derive(Default, Clone)]
 pub struct HeartbeatHandlerGroup {
    handlers: Vec<NameCachedHandler>,
    pushers: Pushers,
@@ -540,20 +541,32 @@ impl HeartbeatHandlerGroupBuilder {
        }

        Ok(HeartbeatHandlerGroup {
-            handlers: self.handlers.into_iter().collect(),
+            handlers: self.handlers,
            pushers: self.pushers,
        })
    }

+    fn add_handler_after_inner(&mut self, target: &str, handler: NameCachedHandler) -> Result<()> {
+        if let Some(pos) = self.handlers.iter().position(|x| x.name == target) {
+            self.handlers.insert(pos + 1, handler);
+            return Ok(());
+        }
+
+        error::HandlerNotFoundSnafu { name: target }.fail()
+    }
+
    /// Adds the handler after the specified handler.
    pub fn add_handler_after(
        &mut self,
        target: &'static str,
        handler: impl HeartbeatHandler + 'static,
    ) -> Result<()> {
+        self.add_handler_after_inner(target, NameCachedHandler::new(handler))
+    }
+
+    fn add_handler_before_inner(&mut self, target: &str, handler: NameCachedHandler) -> Result<()> {
        if let Some(pos) = self.handlers.iter().position(|x| x.name == target) {
-            self.handlers
-                .insert(pos + 1, NameCachedHandler::new(handler));
+            self.handlers.insert(pos, handler);
            return Ok(());
        }

@@ -566,8 +579,12 @@ impl HeartbeatHandlerGroupBuilder {
        target: &'static str,
        handler: impl HeartbeatHandler + 'static,
    ) -> Result<()> {
+        self.add_handler_before_inner(target, NameCachedHandler::new(handler))
+    }
+
+    fn replace_handler_inner(&mut self, target: &str, handler: NameCachedHandler) -> Result<()> {
        if let Some(pos) = self.handlers.iter().position(|x| x.name == target) {
-            self.handlers.insert(pos, NameCachedHandler::new(handler));
+            self.handlers[pos] = handler;
            return Ok(());
        }

@@ -580,25 +597,115 @@ impl HeartbeatHandlerGroupBuilder {
        target: &'static str,
        handler: impl HeartbeatHandler + 'static,
    ) -> Result<()> {
-        if let Some(pos) = self.handlers.iter().position(|x| x.name == target) {
-            self.handlers[pos] = NameCachedHandler::new(handler);
-            return Ok(());
-        }
+        self.replace_handler_inner(target, NameCachedHandler::new(handler))
+    }

-        error::HandlerNotFoundSnafu { name: target }.fail()
+    fn add_handler_last_inner(&mut self, handler: NameCachedHandler) {
+        self.handlers.push(handler);
    }

    fn add_handler_last(&mut self, handler: impl HeartbeatHandler + 'static) {
-        self.handlers.push(NameCachedHandler::new(handler));
+        self.add_handler_last_inner(NameCachedHandler::new(handler));
    }
 }

 pub type HeartbeatHandlerGroupBuilderCustomizerRef =
    Arc<dyn HeartbeatHandlerGroupBuilderCustomizer>;

+pub enum CustomizeHeartbeatGroupAction {
+    AddHandlerAfter {
+        target: String,
+        handler: NameCachedHandler,
+    },
+    AddHandlerBefore {
+        target: String,
+        handler: NameCachedHandler,
+    },
+    ReplaceHandler {
+        target: String,
+        handler: NameCachedHandler,
+    },
+    AddHandlerLast {
+        handler: NameCachedHandler,
+    },
+}
+
+impl CustomizeHeartbeatGroupAction {
+    pub fn new_add_handler_after(
+        target: &'static str,
+        handler: impl HeartbeatHandler + 'static,
+    ) -> Self {
+        Self::AddHandlerAfter {
+            target: target.to_string(),
+            handler: NameCachedHandler::new(handler),
+        }
+    }
+
+    pub fn new_add_handler_before(
+        target: &'static str,
+        handler: impl HeartbeatHandler + 'static,
+    ) -> Self {
+        Self::AddHandlerBefore {
+            target: target.to_string(),
+            handler: NameCachedHandler::new(handler),
+        }
+    }
+
+    pub fn new_replace_handler(
+        target: &'static str,
+        handler: impl HeartbeatHandler + 'static,
+    ) -> Self {
+        Self::ReplaceHandler {
+            target: target.to_string(),
+            handler: NameCachedHandler::new(handler),
+        }
+    }
+
+    pub fn new_add_handler_last(handler: impl HeartbeatHandler + 'static) -> Self {
+        Self::AddHandlerLast {
+            handler: NameCachedHandler::new(handler),
+        }
+    }
+}
+
 /// The customizer of the [`HeartbeatHandlerGroupBuilder`].
 pub trait HeartbeatHandlerGroupBuilderCustomizer: Send + Sync {
    fn customize(&self, builder: &mut HeartbeatHandlerGroupBuilder) -> Result<()>;
+
+    fn add_action(&self, action: CustomizeHeartbeatGroupAction);
+}
+
+#[derive(Default)]
+pub struct DefaultHeartbeatHandlerGroupBuilderCustomizer {
+    actions: Mutex<Vec<CustomizeHeartbeatGroupAction>>,
+}
+
+impl HeartbeatHandlerGroupBuilderCustomizer for DefaultHeartbeatHandlerGroupBuilderCustomizer {
+    fn customize(&self, builder: &mut HeartbeatHandlerGroupBuilder) -> Result<()> {
+        info!("Customizing the heartbeat handler group builder");
+        let mut actions = self.actions.lock().unwrap();
+        for action in actions.drain(..) {
+            match action {
+                CustomizeHeartbeatGroupAction::AddHandlerAfter { target, handler } => {
+                    builder.add_handler_after_inner(&target, handler)?;
+                }
+                CustomizeHeartbeatGroupAction::AddHandlerBefore { target, handler } => {
+                    builder.add_handler_before_inner(&target, handler)?;
+                }
+                CustomizeHeartbeatGroupAction::ReplaceHandler { target, handler } => {
+                    builder.replace_handler_inner(&target, handler)?;
+                }
+                CustomizeHeartbeatGroupAction::AddHandlerLast { handler } => {
+                    builder.add_handler_last_inner(handler);
+                }
+            }
+        }
+        Ok(())
+    }
+
+    fn add_action(&self, action: CustomizeHeartbeatGroupAction) {
+        self.actions.lock().unwrap().push(action);
+    }
 }

 #[cfg(test)]
--- a/src/meta-srv/src/handler/filter_inactive_region_stats.rs
+++ b/src/meta-srv/src/handler/filter_inactive_region_stats.rs
@@ -20,6 +20,8 @@ use crate::error::Result;
 use crate::handler::{HandleControl, HeartbeatAccumulator, HeartbeatHandler};
 use crate::metasrv::Context;

+pub const NAME: &str = "FilterInactiveRegionStatsHandler";
+
 pub struct FilterInactiveRegionStatsHandler;

 #[async_trait]
--- a/src/meta-srv/src/lib.rs
+++ b/src/meta-srv/src/lib.rs
@@ -28,7 +28,6 @@ pub mod flow_meta_alloc;
 pub mod handler;
 pub mod key;
 pub mod lease;
-pub mod lock;
 pub mod metasrv;
 mod metrics;
 #[cfg(feature = "mock")]
--- a/src/meta-srv/src/lock.rs
+++ b/src/meta-srv/src/lock.rs
@@ -1,99 +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.
-
-pub mod etcd;
-pub(crate) mod memory;
-
-use std::sync::Arc;
-
-use common_telemetry::error;
-
-use crate::error::Result;
-
-pub type Key = Vec<u8>;
-
-pub const DEFAULT_EXPIRE_TIME_SECS: u64 = 10;
-
-pub struct Opts {
-    // If the expiration time is exceeded and currently holds the lock, the lock is
-    // automatically released.
-    pub expire_secs: Option<u64>,
-}
-
-impl Default for Opts {
-    fn default() -> Self {
-        Opts {
-            expire_secs: Some(DEFAULT_EXPIRE_TIME_SECS),
-        }
-    }
-}
-
-#[async_trait::async_trait]
-pub trait DistLock: Send + Sync {
-    // Lock acquires a distributed shared lock on a given named lock. On success, it
-    // will return a unique key that exists so long as the lock is held by the caller.
-    async fn lock(&self, name: Vec<u8>, opts: Opts) -> Result<Key>;
-
-    // Unlock takes a key returned by Lock and releases the hold on lock.
-    async fn unlock(&self, key: Vec<u8>) -> Result<()>;
-}
-
-pub type DistLockRef = Arc<dyn DistLock>;
-
-pub struct DistLockGuard<'a> {
-    lock: &'a DistLockRef,
-    name: Vec<u8>,
-    key: Option<Key>,
-}
-
-impl<'a> DistLockGuard<'a> {
-    pub fn new(lock: &'a DistLockRef, name: Vec<u8>) -> Self {
-        Self {
-            lock,
-            name,
-            key: None,
-        }
-    }
-
-    pub async fn lock(&mut self) -> Result<()> {
-        if self.key.is_some() {
-            return Ok(());
-        }
-        let key = self
-            .lock
-            .lock(
-                self.name.clone(),
-                Opts {
-                    expire_secs: Some(2),
-                },
-            )
-            .await?;
-        self.key = Some(key);
-        Ok(())
-    }
-}
-
-impl Drop for DistLockGuard<'_> {
-    fn drop(&mut self) {
-        if let Some(key) = self.key.take() {
-            let lock = self.lock.clone();
-            let name = self.name.clone();
-            let _handle = common_runtime::spawn_global(async move {
-                if let Err(e) = lock.unlock(key).await {
-                    error!(e; "Failed to unlock '{}'", String::from_utf8_lossy(&name));
-                }
-            });
-        }
-    }
-}
--- a/src/meta-srv/src/lock/etcd.rs
+++ b/src/meta-srv/src/lock/etcd.rs
@@ -1,93 +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::sync::Arc;
-
-use etcd_client::{Client, LockOptions};
-use snafu::ResultExt;
-
-use super::{DistLock, DistLockRef, Opts, DEFAULT_EXPIRE_TIME_SECS};
-use crate::error;
-use crate::error::Result;
-
-/// A implementation of distributed lock based on etcd. The Clone of EtcdLock is cheap.
-#[derive(Clone)]
-pub struct EtcdLock {
-    client: Client,
-    store_key_prefix: String,
-}
-
-impl EtcdLock {
-    pub async fn with_endpoints<E, S>(endpoints: S, store_key_prefix: String) -> Result<DistLockRef>
-    where
-        E: AsRef<str>,
-        S: AsRef<[E]>,
-    {
-        let client = Client::connect(endpoints, None)
-            .await
-            .context(error::ConnectEtcdSnafu)?;
-
-        Self::with_etcd_client(client, store_key_prefix)
-    }
-
-    pub fn with_etcd_client(client: Client, store_key_prefix: String) -> Result<DistLockRef> {
-        Ok(Arc::new(EtcdLock {
-            client,
-            store_key_prefix,
-        }))
-    }
-
-    fn lock_key(&self, key: Vec<u8>) -> Vec<u8> {
-        if self.store_key_prefix.is_empty() {
-            key
-        } else {
-            let mut prefix = self.store_key_prefix.as_bytes().to_vec();
-            prefix.extend_from_slice(&key);
-            prefix
-        }
-    }
-}
-
-#[async_trait::async_trait]
-impl DistLock for EtcdLock {
-    async fn lock(&self, key: Vec<u8>, opts: Opts) -> Result<Vec<u8>> {
-        let expire = opts.expire_secs.unwrap_or(DEFAULT_EXPIRE_TIME_SECS) as i64;
-
-        let mut client = self.client.clone();
-
-        let resp = client
-            .lease_grant(expire, None)
-            .await
-            .context(error::LeaseGrantSnafu)?;
-
-        let lease_id = resp.id();
-        let lock_opts = LockOptions::new().with_lease(lease_id);
-
-        let resp = client
-            .lock(self.lock_key(key), Some(lock_opts))
-            .await
-            .context(error::LockSnafu)?;
-
-        Ok(resp.key().to_vec())
-    }
-
-    async fn unlock(&self, key: Vec<u8>) -> Result<()> {
-        let mut client = self.client.clone();
-        let _ = client
-            .unlock(self.lock_key(key))
-            .await
-            .context(error::UnlockSnafu)?;
-        Ok(())
-    }
-}
--- a/src/meta-srv/src/lock/memory.rs
+++ b/src/meta-srv/src/lock/memory.rs
@@ -1,112 +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::sync::Arc;
-
-use async_trait::async_trait;
-use dashmap::DashMap;
-use tokio::sync::{Mutex, OwnedMutexGuard};
-
-use crate::error::Result;
-use crate::lock::{DistLock, Key, Opts};
-
-#[derive(Default)]
-pub(crate) struct MemLock {
-    mutexes: DashMap<Key, Arc<Mutex<()>>>,
-    guards: DashMap<Key, OwnedMutexGuard<()>>,
-}
-
-#[async_trait]
-impl DistLock for MemLock {
-    async fn lock(&self, key: Vec<u8>, _opts: Opts) -> Result<Key> {
-        let mutex = self
-            .mutexes
-            .entry(key.clone())
-            .or_insert_with(|| Arc::new(Mutex::new(())))
-            .clone();
-
-        let guard = mutex.lock_owned().await;
-
-        let _ = self.guards.insert(key.clone(), guard);
-        Ok(key)
-    }
-
-    async fn unlock(&self, key: Vec<u8>) -> Result<()> {
-        // drop the guard, so that the mutex can be unlocked,
-        // effectively make the `mutex.lock_owned` in `lock` method to proceed
-        let _ = self.guards.remove(&key);
-        Ok(())
-    }
-}
-
-#[cfg(test)]
-mod tests {
-    use std::collections::HashMap;
-    use std::sync::atomic::{AtomicU32, Ordering};
-
-    use rand::seq::SliceRandom;
-
-    use super::*;
-
-    #[tokio::test(flavor = "multi_thread")]
-    async fn test_mem_lock_concurrently() {
-        let lock = Arc::new(MemLock::default());
-
-        let keys = (0..10)
-            .map(|i| format!("my-lock-{i}").into_bytes())
-            .collect::<Vec<Key>>();
-        let counters: [(Key, AtomicU32); 10] = keys
-            .iter()
-            .map(|x| (x.clone(), AtomicU32::new(0)))
-            .collect::<Vec<_>>()
-            .try_into()
-            .unwrap();
-        let counters = Arc::new(HashMap::from(counters));
-
-        let tasks = (0..100)
-            .map(|_| {
-                let mut keys = keys.clone();
-                keys.shuffle(&mut rand::thread_rng());
-
-                let lock_clone = lock.clone();
-                let counters_clone = counters.clone();
-                tokio::spawn(async move {
-                    // every key counter will be added by 1 for 10 times
-                    for i in 0..100 {
-                        let key = &keys[i % keys.len()];
-                        assert!(lock_clone
-                            .lock(key.clone(), Opts { expire_secs: None })
-                            .await
-                            .is_ok());
-
-                        // Intentionally create a critical section:
-                        // if our MemLock is flawed, the resulting counter is wrong.
-                        //
-                        // Note that AtomicU32 is only used to enable the updates from multiple tasks,
-                        // does not make any guarantee about the correctness of the result.
-
-                        let counter = counters_clone.get(key).unwrap();
-                        let v = counter.load(Ordering::Relaxed);
-                        counter.store(v + 1, Ordering::Relaxed);
-
-                        lock_clone.unlock(key.clone()).await.unwrap();
-                    }
-                })
-            })
-            .collect::<Vec<_>>();
-        let _ = futures::future::join_all(tasks).await;
-
-        assert!(counters.values().all(|x| x.load(Ordering::Relaxed) == 1000));
-    }
-}
--- a/src/meta-srv/src/metasrv.rs
+++ b/src/meta-srv/src/metasrv.rs
@@ -16,7 +16,7 @@ pub mod builder;

 use std::fmt::Display;
 use std::sync::atomic::{AtomicBool, Ordering};
-use std::sync::Arc;
+use std::sync::{Arc, Mutex, RwLock};
 use std::time::Duration;

 use clap::ValueEnum;
@@ -44,20 +44,19 @@ use common_wal::config::MetasrvWalConfig;
 use serde::{Deserialize, Serialize};
 use servers::export_metrics::ExportMetricsOption;
 use servers::http::HttpOptions;
-use snafu::ResultExt;
+use snafu::{OptionExt, ResultExt};
 use table::metadata::TableId;
 use tokio::sync::broadcast::error::RecvError;

 use crate::cluster::MetaPeerClientRef;
 use crate::election::{Election, LeaderChangeMessage};
 use crate::error::{
-    InitMetadataSnafu, KvBackendSnafu, Result, StartProcedureManagerSnafu, StartTelemetryTaskSnafu,
-    StopProcedureManagerSnafu,
+    self, InitMetadataSnafu, KvBackendSnafu, Result, StartProcedureManagerSnafu,
+    StartTelemetryTaskSnafu, StopProcedureManagerSnafu,
 };
 use crate::failure_detector::PhiAccrualFailureDetectorOptions;
-use crate::handler::HeartbeatHandlerGroupRef;
+use crate::handler::{HeartbeatHandlerGroupBuilder, HeartbeatHandlerGroupRef};
 use crate::lease::lookup_datanode_peer;
-use crate::lock::DistLockRef;
 use crate::procedure::region_migration::manager::RegionMigrationManagerRef;
 use crate::procedure::ProcedureManagerListenerAdapter;
 use crate::pubsub::{PublisherRef, SubscriptionManagerRef};
@@ -338,7 +337,6 @@ impl MetaStateHandler {
    }
 }

-#[derive(Clone)]
 pub struct Metasrv {
    state: StateRef,
    started: Arc<AtomicBool>,
@@ -354,9 +352,9 @@ pub struct Metasrv {
    selector: SelectorRef,
    // The flow selector is used to select a target flownode.
    flow_selector: SelectorRef,
-    handler_group: HeartbeatHandlerGroupRef,
+    handler_group: RwLock<Option<HeartbeatHandlerGroupRef>>,
+    handler_group_builder: Mutex<Option<HeartbeatHandlerGroupBuilder>>,
    election: Option<ElectionRef>,
-    lock: DistLockRef,
    procedure_manager: ProcedureManagerRef,
    mailbox: MailboxRef,
    procedure_executor: ProcedureExecutorRef,
@@ -382,6 +380,16 @@ impl Metasrv {
            return Ok(());
        }

+        let handler_group_builder =
+            self.handler_group_builder
+                .lock()
+                .unwrap()
+                .take()
+                .context(error::UnexpectedSnafu {
+                    violated: "expected heartbeat handler group builder",
+                })?;
+        *self.handler_group.write().unwrap() = Some(Arc::new(handler_group_builder.build()?));
+
        // Creates default schema if not exists
        self.table_metadata_manager
            .init()
@@ -560,18 +568,14 @@ impl Metasrv {
        &self.flow_selector
    }

-    pub fn handler_group(&self) -> &HeartbeatHandlerGroupRef {
-        &self.handler_group
+    pub fn handler_group(&self) -> Option<HeartbeatHandlerGroupRef> {
+        self.handler_group.read().unwrap().clone()
    }

    pub fn election(&self) -> Option<&ElectionRef> {
        self.election.as_ref()
    }

-    pub fn lock(&self) -> &DistLockRef {
-        &self.lock
-    }
-
    pub fn mailbox(&self) -> &MailboxRef {
        &self.mailbox
    }
--- a/src/meta-srv/src/metasrv/builder.rs
+++ b/src/meta-srv/src/metasrv/builder.rs
@@ -13,7 +13,7 @@
 // limitations under the License.

 use std::sync::atomic::AtomicBool;
-use std::sync::{Arc, RwLock};
+use std::sync::{Arc, Mutex, RwLock};
 use std::time::Duration;

 use client::client_manager::NodeClients;
@@ -48,12 +48,8 @@ use crate::flow_meta_alloc::FlowPeerAllocator;
 use crate::greptimedb_telemetry::get_greptimedb_telemetry_task;
 use crate::handler::failure_handler::RegionFailureHandler;
 use crate::handler::region_lease_handler::RegionLeaseHandler;
-use crate::handler::{
-    HeartbeatHandlerGroup, HeartbeatHandlerGroupBuilder, HeartbeatMailbox, Pushers,
-};
+use crate::handler::{HeartbeatHandlerGroupBuilder, HeartbeatMailbox, Pushers};
 use crate::lease::MetaPeerLookupService;
-use crate::lock::memory::MemLock;
-use crate::lock::DistLockRef;
 use crate::metasrv::{
    ElectionRef, Metasrv, MetasrvInfo, MetasrvOptions, SelectorContext, SelectorRef, TABLE_ID_SEQ,
 };
@@ -76,10 +72,9 @@ pub struct MetasrvBuilder {
    kv_backend: Option<KvBackendRef>,
    in_memory: Option<ResettableKvBackendRef>,
    selector: Option<SelectorRef>,
-    handler_group: Option<HeartbeatHandlerGroup>,
+    handler_group_builder: Option<HeartbeatHandlerGroupBuilder>,
    election: Option<ElectionRef>,
    meta_peer_client: Option<MetaPeerClientRef>,
-    lock: Option<DistLockRef>,
    node_manager: Option<NodeManagerRef>,
    plugins: Option<Plugins>,
    table_metadata_allocator: Option<TableMetadataAllocatorRef>,
@@ -91,11 +86,10 @@ impl MetasrvBuilder {
            kv_backend: None,
            in_memory: None,
            selector: None,
-            handler_group: None,
+            handler_group_builder: None,
            meta_peer_client: None,
            election: None,
            options: None,
-            lock: None,
            node_manager: None,
            plugins: None,
            table_metadata_allocator: None,
@@ -122,8 +116,11 @@ impl MetasrvBuilder {
        self
    }

-    pub fn heartbeat_handler(mut self, handler_group: HeartbeatHandlerGroup) -> Self {
-        self.handler_group = Some(handler_group);
+    pub fn heartbeat_handler(
+        mut self,
+        handler_group_builder: HeartbeatHandlerGroupBuilder,
+    ) -> Self {
+        self.handler_group_builder = Some(handler_group_builder);
        self
    }

@@ -137,11 +134,6 @@ impl MetasrvBuilder {
        self
    }

-    pub fn lock(mut self, lock: Option<DistLockRef>) -> Self {
-        self.lock = lock;
-        self
-    }
-
    pub fn node_manager(mut self, node_manager: NodeManagerRef) -> Self {
        self.node_manager = Some(node_manager);
        self
@@ -170,8 +162,7 @@ impl MetasrvBuilder {
            kv_backend,
            in_memory,
            selector,
-            handler_group,
-            lock,
+            handler_group_builder,
            node_manager,
            plugins,
            table_metadata_allocator,
@@ -205,7 +196,6 @@ impl MetasrvBuilder {
        let flow_metadata_manager = Arc::new(FlowMetadataManager::new(
            leader_cached_kv_backend.clone() as _,
        ));
-        let lock = lock.unwrap_or_else(|| Arc::new(MemLock::default()));
        let selector_ctx = SelectorContext {
            server_addr: options.server_addr.clone(),
            datanode_lease_secs: distributed_time_constants::DATANODE_LEASE_SECS,
@@ -349,8 +339,8 @@ impl MetasrvBuilder {
            .context(error::InitDdlManagerSnafu)?,
        );

-        let handler_group = match handler_group {
-            Some(handler_group) => handler_group,
+        let handler_group_builder = match handler_group_builder {
+            Some(handler_group_builder) => handler_group_builder,
            None => {
                let region_lease_handler = RegionLeaseHandler::new(
                    distributed_time_constants::REGION_LEASE_SECS,
@@ -363,7 +353,6 @@ impl MetasrvBuilder {
                    .with_region_failure_handler(region_failover_handler)
                    .with_region_lease_handler(Some(region_lease_handler))
                    .add_default_handlers()
-                    .build()?
            }
        };

@@ -382,9 +371,9 @@ impl MetasrvBuilder {
            selector,
            // TODO(jeremy): We do not allow configuring the flow selector.
            flow_selector: Arc::new(RoundRobinSelector::new(SelectTarget::Flownode)),
-            handler_group: Arc::new(handler_group),
+            handler_group: RwLock::new(None),
+            handler_group_builder: Mutex::new(Some(handler_group_builder)),
            election,
-            lock,
            procedure_manager,
            mailbox,
            procedure_executor: ddl_manager,
--- a/src/meta-srv/src/metrics.rs
+++ b/src/meta-srv/src/metrics.rs
@@ -45,7 +45,20 @@ lazy_static! {
    /// Meta kv cache miss counter.
    pub static ref METRIC_META_KV_CACHE_MISS: IntCounterVec =
        register_int_counter_vec!("greptime_meta_kv_cache_miss", "meta kv cache miss", &["op"]).unwrap();
-    //  Heartbeat received by metasrv.
+    ///  Heartbeat received by metasrv.
    pub static ref METRIC_META_HEARTBEAT_RECV: IntCounterVec =
        register_int_counter_vec!("greptime_meta_heartbeat_recv", "heartbeats received by metasrv", &["pusher_key"]).unwrap();
+    /// The migration execute histogram.
+    pub static ref METRIC_META_REGION_MIGRATION_EXECUTE: HistogramVec =
+        register_histogram_vec!("greptime_meta_region_migration_execute", "meta region migration execute", &["state"]).unwrap();
+    /// The migration error counter.
+    pub static ref METRIC_META_REGION_MIGRATION_ERROR: IntCounterVec =
+        register_int_counter_vec!("greptime_meta_region_migration_error", "meta region migration abort", &["state", "error_type"]).unwrap();
+    /// The migration datanode counter.
+    pub static ref METRIC_META_REGION_MIGRATION_DATANODES: IntCounterVec =
+        register_int_counter_vec!("greptime_meta_region_migration_stat", "meta region migration stat", &["datanode_type", "datanode_id"]).unwrap();
+    /// The migration fail counter.
+    pub static ref METRIC_META_REGION_MIGRATION_FAIL: IntCounter =
+        register_int_counter!("greptime_meta_region_migration_fail", "meta region migration fail").unwrap();
+
 }
--- a/src/meta-srv/src/mocks.rs
+++ b/src/meta-srv/src/mocks.rs
@@ -33,7 +33,7 @@ use crate::metasrv::{Metasrv, MetasrvOptions, SelectorRef};
 pub struct MockInfo {
    pub server_addr: String,
    pub channel_manager: ChannelManager,
-    pub metasrv: Metasrv,
+    pub metasrv: Arc<Metasrv>,
 }

 pub async fn mock_with_memstore() -> MockInfo {
@@ -78,12 +78,13 @@ pub async fn mock(
    metasrv.try_start().await.unwrap();

    let (client, server) = tokio::io::duplex(1024);
+    let metasrv = Arc::new(metasrv);
    let service = metasrv.clone();
    let _handle = tokio::spawn(async move {
        tonic::transport::Server::builder()
-            .add_service(HeartbeatServer::new(service.clone()))
-            .add_service(StoreServer::new(service.clone()))
-            .add_service(ProcedureServiceServer::new(service.clone()))
+            .add_service(HeartbeatServer::from_arc(service.clone()))
+            .add_service(StoreServer::from_arc(service.clone()))
+            .add_service(ProcedureServiceServer::from_arc(service.clone()))
            .serve_with_incoming(futures::stream::iter(vec![Ok::<_, std::io::Error>(server)]))
            .await
    });
--- a/src/meta-srv/src/procedure/region_migration.rs
+++ b/src/meta-srv/src/procedure/region_migration.rs
@@ -54,6 +54,7 @@ use tokio::time::Instant;

 use self::migration_start::RegionMigrationStart;
 use crate::error::{self, Result};
+use crate::metrics::{METRIC_META_REGION_MIGRATION_ERROR, METRIC_META_REGION_MIGRATION_EXECUTE};
 use crate::service::mailbox::MailboxRef;

 /// It's shared in each step and available even after recovering.
@@ -390,6 +391,12 @@ impl Context {
 #[async_trait::async_trait]
 #[typetag::serde(tag = "region_migration_state")]
 pub(crate) trait State: Sync + Send + Debug {
+    fn name(&self) -> &'static str {
+        let type_name = std::any::type_name::<Self>();
+        // short name
+        type_name.split("::").last().unwrap_or(type_name)
+    }
+
    /// Yields the next [State] and [Status].
    async fn next(&mut self, ctx: &mut Context) -> Result<(Box<dyn State>, Status)>;

@@ -478,10 +485,20 @@ impl Procedure for RegionMigrationProcedure {
    async fn execute(&mut self, _ctx: &ProcedureContext) -> ProcedureResult<Status> {
        let state = &mut self.state;

+        let name = state.name();
+        let _timer = METRIC_META_REGION_MIGRATION_EXECUTE
+            .with_label_values(&[name])
+            .start_timer();
        let (next, status) = state.next(&mut self.context).await.map_err(|e| {
            if e.is_retryable() {
+                METRIC_META_REGION_MIGRATION_ERROR
+                    .with_label_values(&[name, "retryable"])
+                    .inc();
                ProcedureError::retry_later(e)
            } else {
+                METRIC_META_REGION_MIGRATION_ERROR
+                    .with_label_values(&[name, "external"])
+                    .inc();
                ProcedureError::external(e)
            }
        })?;
--- a/src/meta-srv/src/procedure/region_migration/manager.rs
+++ b/src/meta-srv/src/procedure/region_migration/manager.rs
@@ -30,6 +30,7 @@ use store_api::storage::RegionId;
 use table::table_name::TableName;

 use crate::error::{self, Result};
+use crate::metrics::{METRIC_META_REGION_MIGRATION_DATANODES, METRIC_META_REGION_MIGRATION_FAIL};
 use crate::procedure::region_migration::{
    DefaultContextFactory, PersistentContext, RegionMigrationProcedure,
 };
@@ -323,6 +324,12 @@ impl RegionMigrationManager {
            schema_name,
            ..
        } = table_info.table_name();
+        METRIC_META_REGION_MIGRATION_DATANODES
+            .with_label_values(&["src", &task.from_peer.id.to_string()])
+            .inc();
+        METRIC_META_REGION_MIGRATION_DATANODES
+            .with_label_values(&["desc", &task.to_peer.id.to_string()])
+            .inc();
        let RegionMigrationProcedureTask {
            cluster_id,
            region_id,
@@ -358,6 +365,7 @@ impl RegionMigrationManager {

            if let Err(e) = watcher::wait(watcher).await {
                error!(e; "Failed to wait region migration procedure {procedure_id} for {task}");
+                METRIC_META_REGION_MIGRATION_FAIL.inc();
                return;
            }

--- a/src/meta-srv/src/region/supervisor.rs
+++ b/src/meta-srv/src/region/supervisor.rs
@@ -258,6 +258,7 @@ impl RegionFailureDetectorController for RegionFailureDetectorControl {
 }

 /// [`HeartbeatAcceptor`] forwards heartbeats to [`RegionSupervisor`].
+#[derive(Clone)]
 pub(crate) struct HeartbeatAcceptor {
    sender: Sender<Event>,
 }
--- a/src/meta-srv/src/service.rs
+++ b/src/meta-srv/src/service.rs
@@ -20,7 +20,6 @@ use tonic::{Response, Status};
 pub mod admin;
 pub mod cluster;
 mod heartbeat;
-pub mod lock;
 pub mod mailbox;
 pub mod procedure;
 pub mod store;
--- a/src/meta-srv/src/service/admin.rs
+++ b/src/meta-srv/src/service/admin.rs
@@ -30,7 +30,7 @@ use tonic::server::NamedService;

 use crate::metasrv::Metasrv;

-pub fn make_admin_service(metasrv: Metasrv) -> Admin {
+pub fn make_admin_service(metasrv: Arc<Metasrv>) -> Admin {
    let router = Router::new().route("/health", health::HealthHandler);

    let router = router.route(
--- a/src/meta-srv/src/service/heartbeat.rs
+++ b/src/meta-srv/src/service/heartbeat.rs
@@ -23,6 +23,7 @@ use api::v1::meta::{
 use common_telemetry::{debug, error, info, warn};
 use futures::StreamExt;
 use once_cell::sync::OnceCell;
+use snafu::OptionExt;
 use tokio::sync::mpsc;
 use tokio::sync::mpsc::Sender;
 use tokio_stream::wrappers::ReceiverStream;
@@ -45,7 +46,10 @@ impl heartbeat_server::Heartbeat for Metasrv {
    ) -> GrpcResult<Self::HeartbeatStream> {
        let mut in_stream = req.into_inner();
        let (tx, rx) = mpsc::channel(128);
-        let handler_group = self.handler_group().clone();
+        let handler_group = self.handler_group().context(error::UnexpectedSnafu {
+            violated: "expected heartbeat handlers",
+        })?;
+
        let ctx = self.new_ctx();
        let _handle = common_runtime::spawn_global(async move {
            let mut pusher_key = None;
--- a/src/meta-srv/src/service/lock.rs
+++ b/src/meta-srv/src/service/lock.rs
@@ -1,51 +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::meta::{lock_server, LockRequest, LockResponse, UnlockRequest, UnlockResponse};
-use tonic::{Request, Response};
-
-use super::GrpcResult;
-use crate::lock::Opts;
-use crate::metasrv::Metasrv;
-
-#[async_trait::async_trait]
-impl lock_server::Lock for Metasrv {
-    async fn lock(&self, request: Request<LockRequest>) -> GrpcResult<LockResponse> {
-        let LockRequest {
-            name, expire_secs, ..
-        } = request.into_inner();
-        let expire_secs = Some(expire_secs as u64);
-
-        let key = self.lock().lock(name, Opts { expire_secs }).await?;
-
-        let resp = LockResponse {
-            key,
-            ..Default::default()
-        };
-
-        Ok(Response::new(resp))
-    }
-
-    async fn unlock(&self, request: Request<UnlockRequest>) -> GrpcResult<UnlockResponse> {
-        let UnlockRequest { key, .. } = request.into_inner();
-
-        let _ = self.lock().unlock(key).await?;
-
-        let resp = UnlockResponse {
-            ..Default::default()
-        };
-
-        Ok(Response::new(resp))
-    }
-}
--- a/src/metric-engine/src/engine.rs
+++ b/src/metric-engine/src/engine.rs
@@ -17,6 +17,7 @@ mod catchup;
 mod close;
 mod create;
 mod drop;
+mod flush;
 mod open;
 mod options;
 mod put;
@@ -145,7 +146,7 @@ impl RegionEngine for MetricEngine {
                    .alter_region(region_id, alter, &mut extension_return_value)
                    .await
            }
-            RegionRequest::Flush(_) | RegionRequest::Compact(_) => {
+            RegionRequest::Compact(_) => {
                if self.inner.is_physical_region(region_id) {
                    self.inner
                        .mito
@@ -157,10 +158,11 @@ impl RegionEngine for MetricEngine {
                    UnsupportedRegionRequestSnafu { request }.fail()
                }
            }
+            RegionRequest::Flush(req) => self.inner.flush_region(region_id, req).await,
            RegionRequest::Delete(_) | RegionRequest::Truncate(_) => {
                UnsupportedRegionRequestSnafu { request }.fail()
            }
-            RegionRequest::Catchup(ref req) => self.inner.catchup_region(region_id, *req).await,
+            RegionRequest::Catchup(req) => self.inner.catchup_region(region_id, req).await,
        };

        result.map_err(BoxedError::new).map(|rows| RegionResponse {
@@ -338,7 +340,7 @@ mod test {
            .await
            .unwrap();

-        // close nonexistent region
+        // close nonexistent region won't report error
        let nonexistent_region_id = RegionId::new(12313, 12);
        engine
            .handle_request(
@@ -346,7 +348,7 @@ mod test {
                RegionRequest::Close(RegionCloseRequest {}),
            )
            .await
-            .unwrap_err();
+            .unwrap();

        // open nonexistent region won't report error
        let invalid_open_request = RegionOpenRequest {
--- a/src/metric-engine/src/engine/alter.rs
+++ b/src/metric-engine/src/engine/alter.rs
@@ -84,6 +84,12 @@ impl MetricEngineInner {
            return Ok(physical_region_id);
        };

+        // lock metadata region for this logical region id
+        let _write_guard = self
+            .metadata_region
+            .write_lock_logical_region(logical_region_id)
+            .await;
+
        let metadata_region_id = to_metadata_region_id(physical_region_id);
        let mut columns_to_add = vec![];
        for col in &columns {
--- a/src/metric-engine/src/engine/catchup.rs
+++ b/src/metric-engine/src/engine/catchup.rs
@@ -47,9 +47,10 @@ impl MetricEngineInner {
            .await
            .context(MitoCatchupOperationSnafu)?;

+        let data_region_id = utils::to_data_region_id(region_id);
        self.mito
            .handle_request(
-                region_id,
+                data_region_id,
                RegionRequest::Catchup(RegionCatchupRequest {
                    set_writable: req.set_writable,
                    entry_id: req.entry_id,
--- a/src/metric-engine/src/engine/close.rs
+++ b/src/metric-engine/src/engine/close.rs
@@ -14,13 +14,14 @@

 //! Close a metric region

+use common_telemetry::debug;
 use snafu::ResultExt;
 use store_api::region_engine::RegionEngine;
 use store_api::region_request::{AffectedRows, RegionCloseRequest, RegionRequest};
 use store_api::storage::RegionId;

 use super::MetricEngineInner;
-use crate::error::{CloseMitoRegionSnafu, LogicalRegionNotFoundSnafu, Result};
+use crate::error::{CloseMitoRegionSnafu, Result};
 use crate::metrics::PHYSICAL_REGION_COUNT;
 use crate::utils;

@@ -54,7 +55,8 @@ impl MetricEngineInner {
        {
            Ok(0)
        } else {
-            Err(LogicalRegionNotFoundSnafu { region_id }.build())
+            debug!("Closing a non-existent logical region {}", region_id);
+            Ok(0)
        }
    }

--- a/src/metric-engine/src/engine/flush.rs
+++ b/src/metric-engine/src/engine/flush.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 snafu::ResultExt;
+use store_api::region_engine::RegionEngine;
+use store_api::region_request::{AffectedRows, RegionFlushRequest, RegionRequest};
+use store_api::storage::RegionId;
+
+use crate::engine::MetricEngineInner;
+use crate::error::{MitoFlushOperationSnafu, Result, UnsupportedRegionRequestSnafu};
+use crate::utils;
+
+impl MetricEngineInner {
+    pub async fn flush_region(
+        &self,
+        region_id: RegionId,
+        req: RegionFlushRequest,
+    ) -> Result<AffectedRows> {
+        if !self.is_physical_region(region_id) {
+            return UnsupportedRegionRequestSnafu {
+                request: RegionRequest::Flush(req),
+            }
+            .fail();
+        }
+
+        let metadata_region_id = utils::to_metadata_region_id(region_id);
+        // Flushes the metadata region as well
+        self.mito
+            .handle_request(metadata_region_id, RegionRequest::Flush(req.clone()))
+            .await
+            .context(MitoFlushOperationSnafu)
+            .map(|response| response.affected_rows)?;
+
+        let data_region_id = utils::to_data_region_id(region_id);
+        self.mito
+            .handle_request(data_region_id, RegionRequest::Flush(req.clone()))
+            .await
+            .context(MitoFlushOperationSnafu)
+            .map(|response| response.affected_rows)
+    }
+}
--- a/src/metric-engine/src/engine/open.rs
+++ b/src/metric-engine/src/engine/open.rs
@@ -134,17 +134,26 @@ impl MetricEngineInner {
            .await?;
        let logical_region_num = logical_regions.len();

-        let mut state = self.state.write().unwrap();
-        // recover physical column names
-        let physical_column_names = physical_columns
-            .into_iter()
-            .map(|col| col.column_schema.name)
-            .collect();
-        state.add_physical_region(physical_region_id, physical_column_names);
-        // recover logical regions
-        for logical_region_id in logical_regions {
-            state.add_logical_region(physical_region_id, logical_region_id);
+        {
+            let mut state = self.state.write().unwrap();
+            // recover physical column names
+            let physical_column_names = physical_columns
+                .into_iter()
+                .map(|col| col.column_schema.name)
+                .collect();
+            state.add_physical_region(physical_region_id, physical_column_names);
+            // recover logical regions
+            for logical_region_id in &logical_regions {
+                state.add_logical_region(physical_region_id, *logical_region_id);
+            }
        }
+
+        for logical_region_id in logical_regions {
+            self.metadata_region
+                .open_logical_region(logical_region_id)
+                .await;
+        }
+
        LOGICAL_REGION_COUNT.add(logical_region_num as i64);

        Ok(())
--- a/src/metric-engine/src/engine/region_metadata.rs
+++ b/src/metric-engine/src/engine/region_metadata.rs
@@ -41,6 +41,10 @@ impl MetricEngineInner {
        }

        // Else load from metadata region and update the cache.
+        let _read_guard = self
+            .metadata_region
+            .read_lock_logical_region(logical_region_id)
+            .await;
        // Load logical and physical columns, and intersect them to get logical column metadata.
        let mut logical_column_metadata = self
            .metadata_region
--- a/src/metric-engine/src/metadata_region.rs
+++ b/src/metric-engine/src/metadata_region.rs
@@ -13,6 +13,7 @@
 // limitations under the License.

 use std::collections::HashMap;
+use std::sync::Arc;

 use api::v1::value::ValueData;
 use api::v1::{ColumnDataType, ColumnSchema, Row, Rows, SemanticType, Value};
@@ -21,7 +22,7 @@ use base64::Engine;
 use common_recordbatch::util::collect;
 use datafusion::prelude::{col, lit};
 use mito2::engine::MitoEngine;
-use snafu::ResultExt;
+use snafu::{OptionExt, ResultExt};
 use store_api::metadata::ColumnMetadata;
 use store_api::metric_engine_consts::{
    METADATA_SCHEMA_KEY_COLUMN_INDEX, METADATA_SCHEMA_KEY_COLUMN_NAME,
@@ -31,11 +32,12 @@ use store_api::metric_engine_consts::{
 use store_api::region_engine::RegionEngine;
 use store_api::region_request::{RegionDeleteRequest, RegionPutRequest};
 use store_api::storage::{RegionId, ScanRequest};
+use tokio::sync::{OwnedRwLockReadGuard, OwnedRwLockWriteGuard, RwLock};

 use crate::error::{
    CollectRecordBatchStreamSnafu, DecodeColumnValueSnafu, DeserializeColumnMetadataSnafu,
-    MitoReadOperationSnafu, MitoWriteOperationSnafu, ParseRegionIdSnafu, RegionAlreadyExistsSnafu,
-    Result,
+    LogicalRegionNotFoundSnafu, MitoReadOperationSnafu, MitoWriteOperationSnafu,
+    ParseRegionIdSnafu, RegionAlreadyExistsSnafu, Result,
 };
 use crate::utils;

@@ -56,11 +58,19 @@ const COLUMN_PREFIX: &str = "__column_";
 /// itself.
 pub struct MetadataRegion {
    mito: MitoEngine,
+    /// Logical lock for operations that need to be serialized. Like update & read region columns.
+    ///
+    /// Region entry will be registered on creating and opening logical region, and deregistered on
+    /// removing logical region.
+    logical_region_lock: RwLock<HashMap<RegionId, Arc<RwLock<()>>>>,
 }

 impl MetadataRegion {
    pub fn new(mito: MitoEngine) -> Self {
-        Self { mito }
+        Self {
+            mito,
+            logical_region_lock: RwLock::new(HashMap::new()),
+        }
    }

    /// Add a new table key to metadata.
@@ -85,10 +95,21 @@ impl MetadataRegion {
            }
            .fail()
        } else {
+            self.logical_region_lock
+                .write()
+                .await
+                .insert(logical_region_id, Arc::new(RwLock::new(())));
            Ok(())
        }
    }

+    pub async fn open_logical_region(&self, logical_region_id: RegionId) {
+        self.logical_region_lock
+            .write()
+            .await
+            .insert(logical_region_id, Arc::new(RwLock::new(())));
+    }
+
    /// Add a new column key to metadata.
    ///
    /// This method won't check if the column already exists. But
@@ -111,6 +132,40 @@ impl MetadataRegion {
        .await
    }

+    /// Retrieve a read lock guard of given logical region id.
+    pub async fn read_lock_logical_region(
+        &self,
+        logical_region_id: RegionId,
+    ) -> Result<OwnedRwLockReadGuard<()>> {
+        let lock = self
+            .logical_region_lock
+            .read()
+            .await
+            .get(&logical_region_id)
+            .context(LogicalRegionNotFoundSnafu {
+                region_id: logical_region_id,
+            })?
+            .clone();
+        Ok(RwLock::read_owned(lock).await)
+    }
+
+    /// Retrieve a write lock guard of given logical region id.
+    pub async fn write_lock_logical_region(
+        &self,
+        logical_region_id: RegionId,
+    ) -> Result<OwnedRwLockWriteGuard<()>> {
+        let lock = self
+            .logical_region_lock
+            .read()
+            .await
+            .get(&logical_region_id)
+            .context(LogicalRegionNotFoundSnafu {
+                region_id: logical_region_id,
+            })?
+            .clone();
+        Ok(RwLock::write_owned(lock).await)
+    }
+
    /// Remove a registered logical region from metadata.
    ///
    /// This method doesn't check if the previous key exists.
@@ -136,6 +191,11 @@ impl MetadataRegion {
        column_keys.push(region_key);
        self.delete(region_id, &column_keys).await?;

+        self.logical_region_lock
+            .write()
+            .await
+            .remove(&logical_region_id);
+
        Ok(())
    }

--- a/src/mito2/src/read.rs
+++ b/src/mito2/src/read.rs
@@ -22,6 +22,7 @@ pub mod projection;
 pub(crate) mod prune;
 pub(crate) mod range;
 pub(crate) mod scan_region;
+pub(crate) mod scan_util;
 pub(crate) mod seq_scan;
 pub(crate) mod unordered_scan;

@@ -57,7 +58,6 @@ use crate::error::{
 use crate::memtable::BoxedBatchIterator;
 use crate::metrics::{READ_BATCHES_RETURN, READ_ROWS_RETURN, READ_STAGE_ELAPSED};
 use crate::read::prune::PruneReader;
-use crate::sst::parquet::reader::{ReaderFilterMetrics, ReaderMetrics};

 /// Storage internal representation of a batch of rows for a primary key (time series).
 ///
@@ -738,7 +738,7 @@ impl<T: BatchReader + ?Sized> BatchReader for Box<T> {
 pub(crate) struct ScannerMetrics {
    /// Duration to prepare the scan task.
    prepare_scan_cost: Duration,
-    /// Duration to build parts.
+    /// Duration to build file ranges.
    build_parts_cost: Duration,
    /// Duration to build the (merge) reader.
    build_reader_cost: Duration,
@@ -758,31 +758,17 @@ pub(crate) struct ScannerMetrics {
    num_mem_ranges: usize,
    /// Number of file ranges scanned.
    num_file_ranges: usize,
-    /// Filter related metrics for readers.
-    filter_metrics: ReaderFilterMetrics,
 }

 impl ScannerMetrics {
-    /// Sets and observes metrics on initializing parts.
-    fn observe_init_part(&mut self, build_parts_cost: Duration, reader_metrics: &ReaderMetrics) {
-        self.build_parts_cost = build_parts_cost;
-
-        // Observes metrics.
+    /// Observes metrics.
+    fn observe_metrics(&self) {
        READ_STAGE_ELAPSED
            .with_label_values(&["prepare_scan"])
            .observe(self.prepare_scan_cost.as_secs_f64());
        READ_STAGE_ELAPSED
            .with_label_values(&["build_parts"])
            .observe(self.build_parts_cost.as_secs_f64());
-
-        // We only call this once so we overwrite it directly.
-        self.filter_metrics = reader_metrics.filter_metrics;
-        // Observes filter metrics.
-        self.filter_metrics.observe();
-    }
-
-    /// Observes metrics on scanner finish.
-    fn observe_metrics_on_finish(&self) {
        READ_STAGE_ELAPSED
            .with_label_values(&["build_reader"])
            .observe(self.build_reader_cost.as_secs_f64());
@@ -801,6 +787,21 @@ impl ScannerMetrics {
        READ_ROWS_RETURN.observe(self.num_rows as f64);
        READ_BATCHES_RETURN.observe(self.num_batches as f64);
    }
+
+    /// Merges metrics from another [ScannerMetrics].
+    fn merge_from(&mut self, other: &ScannerMetrics) {
+        self.prepare_scan_cost += other.prepare_scan_cost;
+        self.build_parts_cost += other.build_parts_cost;
+        self.build_reader_cost += other.build_reader_cost;
+        self.scan_cost += other.scan_cost;
+        self.convert_cost += other.convert_cost;
+        self.yield_cost += other.yield_cost;
+        self.total_cost += other.total_cost;
+        self.num_batches += other.num_batches;
+        self.num_rows += other.num_rows;
+        self.num_mem_ranges += other.num_mem_ranges;
+        self.num_file_ranges += other.num_file_ranges;
+    }
 }

 #[cfg(test)]
--- a/src/mito2/src/read/range.rs
+++ b/src/mito2/src/read/range.rs
@@ -14,7 +14,9 @@

 //! Structs for partition ranges.

+use common_time::Timestamp;
 use smallvec::{smallvec, SmallVec};
+use store_api::region_engine::PartitionRange;

 use crate::memtable::MemtableRef;
 use crate::read::scan_region::ScanInput;
@@ -48,6 +50,26 @@ pub(crate) struct RangeMeta {
 }

 impl RangeMeta {
+    /// Creates a [PartitionRange] with specific identifier.
+    /// It converts the inclusive max timestamp to exclusive end timestamp.
+    pub(crate) fn new_partition_range(&self, identifier: usize) -> PartitionRange {
+        PartitionRange {
+            start: self.time_range.0,
+            end: Timestamp::new(
+                // The i64::MAX timestamp may be invisible but we don't guarantee to support this
+                // value now.
+                self.time_range
+                    .1
+                    .value()
+                    .checked_add(1)
+                    .unwrap_or(self.time_range.1.value()),
+                self.time_range.1.unit(),
+            ),
+            num_rows: self.num_rows,
+            identifier,
+        }
+    }
+
    /// Creates a list of ranges from the `input` for seq scan.
    pub(crate) fn seq_scan_ranges(input: &ScanInput) -> Vec<RangeMeta> {
        let mut ranges = Vec::with_capacity(input.memtables.len() + input.files.len());
@@ -177,7 +199,7 @@ impl RangeMeta {
    }

    fn push_seq_mem_ranges(memtables: &[MemtableRef], ranges: &mut Vec<RangeMeta>) {
-        // For non append-only mode, each range only contains one memtable.
+        // For non append-only mode, each range only contains one memtable by default.
        for (i, memtable) in memtables.iter().enumerate() {
            let stats = memtable.stats();
            let Some(time_range) = stats.time_range() else {
@@ -195,6 +217,7 @@ impl RangeMeta {
        }
    }

+    // TODO(yingwen): Support multiple row groups in a range so we can split them later.
    fn push_seq_file_ranges(
        num_memtables: usize,
        files: &[FileHandle],
@@ -264,3 +287,83 @@ fn maybe_split_ranges_for_seq_scan(ranges: Vec<RangeMeta>) -> Vec<RangeMeta> {

    new_ranges
 }
+
+#[cfg(test)]
+mod tests {
+    use common_time::timestamp::TimeUnit;
+    use common_time::Timestamp;
+
+    use super::*;
+
+    type Output = (Vec<usize>, i64, i64);
+
+    fn run_group_ranges_test(input: &[(usize, i64, i64)], expect: &[Output]) {
+        let ranges = input
+            .iter()
+            .map(|(idx, start, end)| {
+                let time_range = (
+                    Timestamp::new(*start, TimeUnit::Second),
+                    Timestamp::new(*end, TimeUnit::Second),
+                );
+                RangeMeta {
+                    time_range,
+                    indices: smallvec![*idx],
+                    row_group_indices: smallvec![RowGroupIndex {
+                        index: *idx,
+                        row_group_index: 0
+                    }],
+                    num_rows: 1,
+                }
+            })
+            .collect();
+        let output = group_ranges_for_seq_scan(ranges);
+        let actual: Vec<_> = output
+            .iter()
+            .map(|range| {
+                let indices = range.indices.to_vec();
+                let group_indices: Vec<_> = range
+                    .row_group_indices
+                    .iter()
+                    .map(|idx| idx.index)
+                    .collect();
+                assert_eq!(indices, group_indices);
+                let range = range.time_range;
+                (indices, range.0.value(), range.1.value())
+            })
+            .collect();
+        assert_eq!(expect, actual);
+    }
+
+    #[test]
+    fn test_group_ranges() {
+        // Group 1 part.
+        run_group_ranges_test(&[(1, 0, 2000)], &[(vec![1], 0, 2000)]);
+
+        // 1, 2, 3, 4 => [3, 1, 4], [2]
+        run_group_ranges_test(
+            &[
+                (1, 1000, 2000),
+                (2, 6000, 7000),
+                (3, 0, 1500),
+                (4, 1500, 3000),
+            ],
+            &[(vec![3, 1, 4], 0, 3000), (vec![2], 6000, 7000)],
+        );
+
+        // 1, 2, 3 => [3], [1], [2],
+        run_group_ranges_test(
+            &[(1, 3000, 4000), (2, 4001, 6000), (3, 0, 1000)],
+            &[
+                (vec![3], 0, 1000),
+                (vec![1], 3000, 4000),
+                (vec![2], 4001, 6000),
+            ],
+        );
+
+        // 1, 2, 3 => [3], [1, 2]
+        run_group_ranges_test(
+            &[(1, 3000, 4000), (2, 4000, 6000), (3, 0, 1000)],
+            &[(vec![3], 0, 1000), (vec![1, 2], 3000, 6000)],
+        );
+    }
+}
--- a/src/mito2/src/read/scan_region.rs
+++ b/src/mito2/src/read/scan_region.rs
@@ -17,14 +17,12 @@
 use std::collections::{BTreeMap, HashSet};
 use std::fmt;
 use std::sync::{Arc, Mutex as StdMutex};
-use std::time::{Duration, Instant};
+use std::time::Instant;

 use common_error::ext::BoxedError;
 use common_recordbatch::SendableRecordBatchStream;
 use common_telemetry::{debug, error, tracing, warn};
 use common_time::range::TimestampRange;
-use common_time::Timestamp;
-use datafusion::physical_plan::DisplayFormatType;
 use datafusion_expr::utils::expr_to_columns;
 use parquet::arrow::arrow_reader::RowSelection;
 use smallvec::SmallVec;
@@ -48,7 +46,7 @@ use crate::read::unordered_scan::UnorderedScan;
 use crate::read::{Batch, Source};
 use crate::region::options::MergeMode;
 use crate::region::version::VersionRef;
-use crate::sst::file::{overlaps, FileHandle, FileMeta};
+use crate::sst::file::FileHandle;
 use crate::sst::index::fulltext_index::applier::builder::FulltextIndexApplierBuilder;
 use crate::sst::index::fulltext_index::applier::FulltextIndexApplierRef;
 use crate::sst::index::inverted_index::applier::builder::InvertedIndexApplierBuilder;
@@ -700,73 +698,6 @@ impl ScanInput {
        })
    }

-    /// Prunes file ranges to scan and adds them to the `collector`.
-    pub(crate) async fn prune_file_ranges(
-        &self,
-        collector: &mut impl FileRangeCollector,
-    ) -> Result<ReaderMetrics> {
-        let mut file_prune_cost = Duration::ZERO;
-        let mut reader_metrics = ReaderMetrics::default();
-        for file in &self.files {
-            let prune_start = Instant::now();
-            let res = self
-                .access_layer
-                .read_sst(file.clone())
-                .predicate(self.predicate.clone())
-                .time_range(self.time_range)
-                .projection(Some(self.mapper.column_ids().to_vec()))
-                .cache(self.cache_manager.clone())
-                .inverted_index_applier(self.inverted_index_applier.clone())
-                .fulltext_index_applier(self.fulltext_index_applier.clone())
-                .expected_metadata(Some(self.mapper.metadata().clone()))
-                .build_reader_input(&mut reader_metrics)
-                .await;
-            file_prune_cost += prune_start.elapsed();
-            let (mut file_range_ctx, row_groups) = match res {
-                Ok(x) => x,
-                Err(e) => {
-                    if e.is_object_not_found() && self.ignore_file_not_found {
-                        error!(e; "File to scan does not exist, region_id: {}, file: {}", file.region_id(), file.file_id());
-                        continue;
-                    } else {
-                        return Err(e);
-                    }
-                }
-            };
-            if !compat::has_same_columns(
-                self.mapper.metadata(),
-                file_range_ctx.read_format().metadata(),
-            ) {
-                // They have different schema. We need to adapt the batch first so the
-                // mapper can convert it.
-                let compat = CompatBatch::new(
-                    &self.mapper,
-                    file_range_ctx.read_format().metadata().clone(),
-                )?;
-                file_range_ctx.set_compat_batch(Some(compat));
-            }
-            // Build ranges from row groups.
-            let file_range_ctx = Arc::new(file_range_ctx);
-            let file_ranges = row_groups
-                .into_iter()
-                .map(|(row_group_idx, row_selection)| {
-                    FileRange::new(file_range_ctx.clone(), row_group_idx, row_selection)
-                });
-            collector.append_file_ranges(file.meta_ref(), file_ranges);
-        }
-
-        READ_SST_COUNT.observe(self.files.len() as f64);
-
-        common_telemetry::debug!(
-            "Region {} prune {} files, cost is {:?}",
-            self.mapper.metadata().region_id,
-            self.files.len(),
-            file_prune_cost
-        );
-
-        Ok(reader_metrics)
-    }
-
    /// Scans the input source in another task and sends batches to the sender.
    pub(crate) fn spawn_scan_task(
        &self,
@@ -806,10 +737,7 @@ impl ScanInput {
    pub(crate) fn predicate(&self) -> Option<Predicate> {
        self.predicate.clone()
    }
-}

-#[cfg(test)]
-impl ScanInput {
    /// Returns number of memtables to scan.
    pub(crate) fn num_memtables(&self) -> usize {
        self.memtables.len()
@@ -819,166 +747,21 @@ impl ScanInput {
    pub(crate) fn num_files(&self) -> usize {
        self.files.len()
    }
+}

+#[cfg(test)]
+impl ScanInput {
    /// Returns SST file ids to scan.
    pub(crate) fn file_ids(&self) -> Vec<crate::sst::file::FileId> {
        self.files.iter().map(|file| file.file_id()).collect()
    }
 }

-/// Groups of file ranges. Each group in the list contains multiple file
-/// ranges to scan. File ranges in the same group may come from different files.
-pub(crate) type FileRangesGroup = SmallVec<[Vec<FileRange>; 4]>;
-
-/// A partition of a scanner to read.
-/// It contains memtables and file ranges to scan.
-#[derive(Clone, Default)]
-pub(crate) struct ScanPart {
-    /// Memtable ranges to scan.
-    pub(crate) memtable_ranges: Vec<MemtableRange>,
-    /// File ranges to scan.
-    pub(crate) file_ranges: FileRangesGroup,
-    /// Optional time range of the part (inclusive).
-    pub(crate) time_range: Option<(Timestamp, Timestamp)>,
-}
-
-impl fmt::Debug for ScanPart {
-    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
-        write!(
-            f,
-            "ScanPart({} memtable ranges, {} file ranges",
-            self.memtable_ranges.len(),
-            self.file_ranges
-                .iter()
-                .map(|ranges| ranges.len())
-                .sum::<usize>(),
-        )?;
-        if let Some(time_range) = &self.time_range {
-            write!(f, ", time range: {:?})", time_range)
-        } else {
-            write!(f, ")")
-        }
-    }
-}
-
-impl ScanPart {
-    /// Returns true if the time range given `part` overlaps with this part.
-    pub(crate) fn overlaps(&self, part: &ScanPart) -> bool {
-        let (Some(current_range), Some(part_range)) = (self.time_range, part.time_range) else {
-            return true;
-        };
-
-        overlaps(&current_range, &part_range)
-    }
-
-    /// Merges given `part` to this part.
-    pub(crate) fn merge(&mut self, mut part: ScanPart) {
-        self.memtable_ranges.append(&mut part.memtable_ranges);
-        self.file_ranges.append(&mut part.file_ranges);
-        let Some(part_range) = part.time_range else {
-            return;
-        };
-        let Some(current_range) = self.time_range else {
-            self.time_range = part.time_range;
-            return;
-        };
-        let start = current_range.0.min(part_range.0);
-        let end = current_range.1.max(part_range.1);
-        self.time_range = Some((start, end));
-    }
-
-    /// Returns true if the we can split the part into multiple parts
-    /// and preserving order.
-    pub(crate) fn can_split_preserve_order(&self) -> bool {
-        self.memtable_ranges.is_empty()
-            && self.file_ranges.len() == 1
-            && self.file_ranges[0].len() > 1
-    }
-}
-
-/// A trait to collect file ranges to scan.
-pub(crate) trait FileRangeCollector {
-    /// Appends file ranges from the **same file** to the collector.
-    fn append_file_ranges(
-        &mut self,
-        file_meta: &FileMeta,
-        file_ranges: impl Iterator<Item = FileRange>,
-    );
-}
-
-/// Optional list of [ScanPart]s.
-#[derive(Default)]
-pub(crate) struct ScanPartList(pub(crate) Option<Vec<ScanPart>>);
-
-impl fmt::Debug for ScanPartList {
-    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
-        match &self.0 {
-            Some(parts) => write!(f, "{:?}", parts),
-            None => write!(f, "[]"),
-        }
-    }
-}
-
-impl ScanPartList {
-    /// Returns true if the list is None.
-    pub(crate) fn is_none(&self) -> bool {
-        self.0.is_none()
-    }
-
-    /// Sets parts to the list.
-    pub(crate) fn set_parts(&mut self, parts: Vec<ScanPart>) {
-        self.0 = Some(parts);
-    }
-
-    /// Gets the part by index, returns None if the index is out of bound.
-    /// # Panics
-    /// Panics if parts are not initialized.
-    pub(crate) fn get_part(&mut self, index: usize) -> Option<&ScanPart> {
-        let parts = self.0.as_ref().unwrap();
-        parts.get(index)
-    }
-
-    /// Returns the number of parts.
-    pub(crate) fn len(&self) -> usize {
-        self.0.as_ref().map_or(0, |parts| parts.len())
-    }
-
-    /// Returns the number of memtable ranges.
-    pub(crate) fn num_mem_ranges(&self) -> usize {
-        self.0.as_ref().map_or(0, |parts| {
-            parts.iter().map(|part| part.memtable_ranges.len()).sum()
-        })
-    }
-
-    /// Returns the number of files.
-    pub(crate) fn num_files(&self) -> usize {
-        self.0.as_ref().map_or(0, |parts| {
-            parts.iter().map(|part| part.file_ranges.len()).sum()
-        })
-    }
-
-    /// Returns the number of file ranges.
-    pub(crate) fn num_file_ranges(&self) -> usize {
-        self.0.as_ref().map_or(0, |parts| {
-            parts
-                .iter()
-                .flat_map(|part| part.file_ranges.iter())
-                .map(|ranges| ranges.len())
-                .sum()
-        })
-    }
-}
-
 /// Context shared by different streams from a scanner.
-/// It contains the input and distributes input to multiple parts
-/// to scan.
+/// It contains the input and ranges to scan.
 pub(crate) struct StreamContext {
    /// Input memtables and files.
    pub(crate) input: ScanInput,
-    /// Parts to scan and the cost to build parts.
-    /// The scanner builds parts to scan from the input lazily.
-    /// The mutex is used to ensure the parts are only built once.
-    pub(crate) parts: Mutex<(ScanPartList, Duration)>,
    /// Metadata for partition ranges.
    pub(crate) ranges: Vec<RangeMeta>,
    /// Lists of range builders.
@@ -994,12 +777,11 @@ impl StreamContext {
    pub(crate) fn seq_scan_ctx(input: ScanInput) -> Self {
        let query_start = input.query_start.unwrap_or_else(Instant::now);
        let ranges = RangeMeta::seq_scan_ranges(&input);
-        READ_SST_COUNT.observe(input.files.len() as f64);
-        let range_builders = RangeBuilderList::new(input.memtables.len(), input.files.len());
+        READ_SST_COUNT.observe(input.num_files() as f64);
+        let range_builders = RangeBuilderList::new(input.num_memtables(), input.num_files());

        Self {
            input,
-            parts: Mutex::new((ScanPartList::default(), Duration::default())),
            ranges,
            range_builders,
            query_start,
@@ -1010,12 +792,11 @@ impl StreamContext {
    pub(crate) fn unordered_scan_ctx(input: ScanInput) -> Self {
        let query_start = input.query_start.unwrap_or_else(Instant::now);
        let ranges = RangeMeta::unordered_scan_ranges(&input);
-        READ_SST_COUNT.observe(input.files.len() as f64);
-        let range_builders = RangeBuilderList::new(input.memtables.len(), input.files.len());
+        READ_SST_COUNT.observe(input.num_files() as f64);
+        let range_builders = RangeBuilderList::new(input.num_memtables(), input.num_files());

        Self {
            input,
-            parts: Mutex::new((ScanPartList::default(), Duration::default())),
            ranges,
            range_builders,
            query_start,
@@ -1024,27 +805,28 @@ impl StreamContext {

    /// Returns true if the index refers to a memtable.
    pub(crate) fn is_mem_range_index(&self, index: RowGroupIndex) -> bool {
-        self.input.memtables.len() > index.index
+        self.input.num_memtables() > index.index
    }

    /// Creates file ranges to scan.
    pub(crate) async fn build_file_ranges(
        &self,
        index: RowGroupIndex,
-        ranges: &mut Vec<FileRange>,
        reader_metrics: &mut ReaderMetrics,
-    ) -> Result<()> {
-        ranges.clear();
+    ) -> Result<SmallVec<[FileRange; 2]>> {
+        let mut ranges = SmallVec::new();
        self.range_builders
-            .build_file_ranges(&self.input, index, ranges, reader_metrics)
-            .await
+            .build_file_ranges(&self.input, index, &mut ranges, reader_metrics)
+            .await?;
+        Ok(ranges)
    }

    /// Creates memtable ranges to scan.
-    pub(crate) fn build_mem_ranges(&self, index: RowGroupIndex, ranges: &mut Vec<MemtableRange>) {
-        ranges.clear();
+    pub(crate) fn build_mem_ranges(&self, index: RowGroupIndex) -> SmallVec<[MemtableRange; 2]> {
+        let mut ranges = SmallVec::new();
        self.range_builders
-            .build_mem_ranges(&self.input, index, ranges)
+            .build_mem_ranges(&self.input, index, &mut ranges);
+        ranges
    }

    /// Retrieves the partition ranges.
@@ -1052,35 +834,30 @@ impl StreamContext {
        self.ranges
            .iter()
            .enumerate()
-            .map(|(idx, range_meta)| PartitionRange {
-                start: range_meta.time_range.0,
-                end: range_meta.time_range.1,
-                num_rows: range_meta.num_rows,
-                identifier: idx,
-            })
+            .map(|(idx, range_meta)| range_meta.new_partition_range(idx))
            .collect()
    }

    /// Format the context for explain.
-    pub(crate) fn format_for_explain(
-        &self,
-        t: DisplayFormatType,
-        f: &mut fmt::Formatter,
-    ) -> fmt::Result {
-        match self.parts.try_lock() {
-            Ok(inner) => match t {
-                DisplayFormatType::Default => write!(
-                    f,
-                    "partition_count={} ({} memtable ranges, {} file {} ranges)",
-                    inner.0.len(),
-                    inner.0.num_mem_ranges(),
-                    inner.0.num_files(),
-                    inner.0.num_file_ranges()
-                )?,
-                DisplayFormatType::Verbose => write!(f, "{:?}", inner.0)?,
-            },
-            Err(_) => write!(f, "<locked>")?,
+    pub(crate) fn format_for_explain(&self, f: &mut fmt::Formatter) -> fmt::Result {
+        let (mut num_mem_ranges, mut num_file_ranges) = (0, 0);
+        for range_meta in &self.ranges {
+            for idx in &range_meta.row_group_indices {
+                if self.is_mem_range_index(*idx) {
+                    num_mem_ranges += 1;
+                } else {
+                    num_file_ranges += 1;
+                }
+            }
        }
+        write!(
+            f,
+            "partition_count={} ({} memtable ranges, {} file {} ranges)",
+            self.ranges.len(),
+            num_mem_ranges,
+            self.input.num_files(),
+            num_file_ranges,
+        )?;
        if let Some(selector) = &self.input.series_row_selector {
            write!(f, ", selector={}", selector)?;
        }
@@ -1110,7 +887,7 @@ impl RangeBuilderList {
        &self,
        input: &ScanInput,
        index: RowGroupIndex,
-        ranges: &mut Vec<FileRange>,
+        ranges: &mut SmallVec<[FileRange; 2]>,
        reader_metrics: &mut ReaderMetrics,
    ) -> Result<()> {
        let file_index = index.index - self.mem_builders.len();
@@ -1131,7 +908,7 @@ impl RangeBuilderList {
        &self,
        input: &ScanInput,
        index: RowGroupIndex,
-        ranges: &mut Vec<MemtableRange>,
+        ranges: &mut SmallVec<[MemtableRange; 2]>,
    ) {
        let mut builder_opt = self.mem_builders[index.index].lock().unwrap();
        match &mut *builder_opt {
@@ -1159,7 +936,7 @@ struct FileRangeBuilder {
 impl FileRangeBuilder {
    /// Builds file ranges to read.
    /// Negative `row_group_index` indicates all row groups.
-    fn build_ranges(&self, row_group_index: i64, ranges: &mut Vec<FileRange>) {
+    fn build_ranges(&self, row_group_index: i64, ranges: &mut SmallVec<[FileRange; 2]>) {
        let Some(context) = self.context.clone() else {
            return;
        };
@@ -1196,7 +973,7 @@ struct MemRangeBuilder {
 impl MemRangeBuilder {
    /// Builds mem ranges to read in the memtable.
    /// Negative `row_group_index` indicates all row groups.
-    fn build_ranges(&self, row_group_index: i64, ranges: &mut Vec<MemtableRange>) {
+    fn build_ranges(&self, row_group_index: i64, ranges: &mut SmallVec<[MemtableRange; 2]>) {
        if row_group_index >= 0 {
            let row_group_index = row_group_index as usize;
            // Scans one row group.
--- a/src/mito2/src/read/scan_util.rs
+++ b/src/mito2/src/read/scan_util.rs
@@ -0,0 +1,182 @@
+// 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.
+
+//! Utilities for scanners.
+
+use std::sync::{Arc, Mutex};
+use std::time::{Duration, Instant};
+
+use async_stream::try_stream;
+use common_telemetry::debug;
+use futures::Stream;
+use store_api::storage::RegionId;
+
+use crate::error::Result;
+use crate::read::range::RowGroupIndex;
+use crate::read::scan_region::StreamContext;
+use crate::read::{Batch, ScannerMetrics, Source};
+use crate::sst::parquet::reader::ReaderMetrics;
+
+struct PartitionMetricsInner {
+    region_id: RegionId,
+    /// Index of the partition to scan.
+    partition: usize,
+    /// Label to distinguish different scan operation.
+    scanner_type: &'static str,
+    /// Query start time.
+    query_start: Instant,
+    /// Elapsed time before the first poll operation.
+    first_poll: Duration,
+    metrics: ScannerMetrics,
+    reader_metrics: ReaderMetrics,
+}
+
+impl PartitionMetricsInner {
+    fn on_finish(&mut self) {
+        if self.metrics.total_cost.is_zero() {
+            self.metrics.total_cost = self.query_start.elapsed();
+        }
+        self.metrics.build_parts_cost = self.reader_metrics.build_cost;
+    }
+}
+
+impl Drop for PartitionMetricsInner {
+    fn drop(&mut self) {
+        self.on_finish();
+        self.metrics.observe_metrics();
+
+        debug!(
+            "{} finished, region_id: {}, partition: {}, first_poll: {:?}, metrics: {:?}, reader_metrics: {:?}",
+            self.scanner_type, self.region_id, self.partition, self.first_poll, self.metrics, self.reader_metrics
+        );
+    }
+}
+
+/// Metrics while reading a partition.
+#[derive(Clone)]
+pub(crate) struct PartitionMetrics(Arc<Mutex<PartitionMetricsInner>>);
+
+impl PartitionMetrics {
+    pub(crate) fn new(
+        region_id: RegionId,
+        partition: usize,
+        scanner_type: &'static str,
+        query_start: Instant,
+        metrics: ScannerMetrics,
+    ) -> Self {
+        let inner = PartitionMetricsInner {
+            region_id,
+            partition,
+            scanner_type,
+            query_start,
+            first_poll: Duration::default(),
+            metrics,
+            reader_metrics: ReaderMetrics::default(),
+        };
+        Self(Arc::new(Mutex::new(inner)))
+    }
+
+    pub(crate) fn on_first_poll(&self) {
+        let mut inner = self.0.lock().unwrap();
+        inner.first_poll = inner.query_start.elapsed();
+    }
+
+    pub(crate) fn inc_num_mem_ranges(&self, num: usize) {
+        let mut inner = self.0.lock().unwrap();
+        inner.metrics.num_mem_ranges += num;
+    }
+
+    pub(crate) fn inc_num_file_ranges(&self, num: usize) {
+        let mut inner = self.0.lock().unwrap();
+        inner.metrics.num_file_ranges += num;
+    }
+
+    pub(crate) fn inc_build_reader_cost(&self, cost: Duration) {
+        let mut inner = self.0.lock().unwrap();
+        inner.metrics.build_reader_cost += cost;
+    }
+
+    pub(crate) fn merge_metrics(&self, metrics: &ScannerMetrics) {
+        let mut inner = self.0.lock().unwrap();
+        inner.metrics.merge_from(metrics);
+    }
+
+    pub(crate) fn merge_reader_metrics(&self, metrics: &ReaderMetrics) {
+        let mut inner = self.0.lock().unwrap();
+        inner.reader_metrics.merge_from(metrics);
+    }
+
+    pub(crate) fn on_finish(&self) {
+        let mut inner = self.0.lock().unwrap();
+        inner.on_finish();
+    }
+}
+
+/// Scans memtable ranges at `index`.
+pub(crate) fn scan_mem_ranges(
+    stream_ctx: Arc<StreamContext>,
+    part_metrics: PartitionMetrics,
+    index: RowGroupIndex,
+) -> impl Stream<Item = Result<Batch>> {
+    try_stream! {
+        let ranges = stream_ctx.build_mem_ranges(index);
+        part_metrics.inc_num_mem_ranges(ranges.len());
+        for range in ranges {
+            let build_reader_start = Instant::now();
+            let iter = range.build_iter()?;
+            part_metrics.inc_build_reader_cost(build_reader_start.elapsed());
+
+            let mut source = Source::Iter(iter);
+            while let Some(batch) = source.next_batch().await? {
+                yield batch;
+            }
+        }
+    }
+}
+
+/// Scans file ranges at `index`.
+pub(crate) fn scan_file_ranges(
+    stream_ctx: Arc<StreamContext>,
+    part_metrics: PartitionMetrics,
+    index: RowGroupIndex,
+    read_type: &'static str,
+) -> impl Stream<Item = Result<Batch>> {
+    try_stream! {
+        let mut reader_metrics = ReaderMetrics::default();
+        let ranges = stream_ctx
+            .build_file_ranges(index, &mut reader_metrics)
+            .await?;
+        part_metrics.inc_num_file_ranges(ranges.len());
+        for range in ranges {
+            let build_reader_start = Instant::now();
+            let reader = range.reader(None).await?;
+            part_metrics.inc_build_reader_cost(build_reader_start.elapsed());
+            let compat_batch = range.compat_batch();
+            let mut source = Source::PruneReader(reader);
+            while let Some(mut batch) = source.next_batch().await? {
+                if let Some(compact_batch) = compat_batch {
+                    batch = compact_batch.compat_batch(batch)?;
+                }
+                yield batch;
+            }
+            if let Source::PruneReader(mut reader) = source {
+                reader_metrics.merge_from(reader.metrics());
+            }
+        }
+
+        // Reports metrics.
+        reader_metrics.observe_rows(read_type);
+        part_metrics.merge_reader_metrics(&reader_metrics);
+    }
+}
--- a/src/mito2/src/read/seq_scan.rs
+++ b/src/mito2/src/read/seq_scan.rs
@@ -16,36 +16,29 @@

 use std::fmt;
 use std::sync::Arc;
-use std::time::{Duration, Instant};
+use std::time::Instant;

 use async_stream::try_stream;
 use common_error::ext::BoxedError;
 use common_recordbatch::error::ExternalSnafu;
 use common_recordbatch::util::ChainedRecordBatchStream;
 use common_recordbatch::{RecordBatchStreamWrapper, SendableRecordBatchStream};
-use common_telemetry::{debug, tracing};
+use common_telemetry::tracing;
 use datafusion::physical_plan::{DisplayAs, DisplayFormatType};
 use datatypes::schema::SchemaRef;
-use smallvec::smallvec;
 use snafu::ResultExt;
 use store_api::region_engine::{PartitionRange, RegionScanner, ScannerProperties};
-use store_api::storage::{ColumnId, TimeSeriesRowSelector};
-use table::predicate::Predicate;
+use store_api::storage::TimeSeriesRowSelector;
 use tokio::sync::Semaphore;

 use crate::error::{PartitionOutOfRangeSnafu, Result};
-use crate::memtable::MemtableRef;
 use crate::read::dedup::{DedupReader, LastNonNull, LastRow};
 use crate::read::last_row::LastRowReader;
 use crate::read::merge::MergeReaderBuilder;
-use crate::read::scan_region::{
-    FileRangeCollector, ScanInput, ScanPart, ScanPartList, StreamContext,
-};
+use crate::read::scan_region::{ScanInput, StreamContext};
+use crate::read::scan_util::{scan_file_ranges, scan_mem_ranges, PartitionMetrics};
 use crate::read::{BatchReader, BoxedBatchReader, ScannerMetrics, Source};
 use crate::region::options::MergeMode;
-use crate::sst::file::FileMeta;
-use crate::sst::parquet::file_range::FileRange;
-use crate::sst::parquet::reader::ReaderMetrics;

 /// Scans a region and returns rows in a sorted sequence.
 ///
@@ -66,6 +59,8 @@ pub struct SeqScan {
 impl SeqScan {
    /// Creates a new [SeqScan].
    pub(crate) fn new(input: ScanInput) -> Self {
+        // TODO(yingwen): Set permits according to partition num. But we need to support file
+        // level parallelism.
        let parallelism = input.parallelism.parallelism.max(1);
        let mut properties = ScannerProperties::default()
            .with_append_mode(input.append_mode)
@@ -102,150 +97,49 @@ impl SeqScan {

    /// Builds a [BoxedBatchReader] from sequential scan for compaction.
    pub async fn build_reader(&self) -> Result<BoxedBatchReader> {
-        let mut metrics = ScannerMetrics {
-            prepare_scan_cost: self.stream_ctx.query_start.elapsed(),
-            ..Default::default()
-        };
-        let maybe_reader = Self::build_all_merge_reader(
+        let part_metrics = PartitionMetrics::new(
+            self.stream_ctx.input.mapper.metadata().region_id,
+            0,
+            get_scanner_type(self.compaction),
+            self.stream_ctx.query_start,
+            ScannerMetrics {
+                prepare_scan_cost: self.stream_ctx.query_start.elapsed(),
+                ..Default::default()
+            },
+        );
+        debug_assert_eq!(1, self.properties.partitions.len());
+        let partition_ranges = &self.properties.partitions[0];
+
+        let reader = Self::build_all_merge_reader(
            &self.stream_ctx,
+            partition_ranges,
            self.semaphore.clone(),
-            &mut metrics,
            self.compaction,
-            self.properties.num_partitions(),
+            &part_metrics,
        )
        .await?;
-        // Safety: `build_merge_reader()` always returns a reader if partition is None.
-        let reader = maybe_reader.unwrap();
        Ok(Box::new(reader))
    }

-    /// Builds sources from a [ScanPart].
-    fn build_part_sources(
-        part: &ScanPart,
-        sources: &mut Vec<Source>,
-        row_selector: Option<TimeSeriesRowSelector>,
-        compaction: bool,
-    ) -> Result<()> {
-        sources.reserve(part.memtable_ranges.len() + part.file_ranges.len());
-        // Read memtables.
-        for mem in &part.memtable_ranges {
-            let iter = mem.build_iter()?;
-            sources.push(Source::Iter(iter));
-        }
-        let read_type = if compaction {
-            "compaction"
-        } else {
-            "seq_scan_files"
-        };
-        // Read files.
-        for file in &part.file_ranges {
-            if file.is_empty() {
-                continue;
-            }
-
-            // Creates a stream to read the file.
-            let ranges = file.clone();
-            let stream = try_stream! {
-                let mut reader_metrics = ReaderMetrics::default();
-                // Safety: We checked whether it is empty before.
-                let file_id = ranges[0].file_handle().file_id();
-                let region_id = ranges[0].file_handle().region_id();
-                let range_num = ranges.len();
-                for range in ranges {
-                    let mut reader = range.reader(row_selector).await?;
-                    let compat_batch = range.compat_batch();
-                    while let Some(mut batch) = reader.next_batch().await? {
-                        if let Some(compat) = compat_batch {
-                            batch = compat
-                                .compat_batch(batch)?;
-                        }
-
-                        yield batch;
-                    }
-                    reader_metrics.merge_from(reader.metrics());
-                }
-                debug!(
-                    "Seq scan region {}, file {}, {} ranges finished, metrics: {:?}, compaction: {}",
-                    region_id, file_id, range_num, reader_metrics, compaction
-                );
-                // Reports metrics.
-                reader_metrics.observe_rows(read_type);
-            };
-            let stream = Box::pin(stream);
-            sources.push(Source::Stream(stream));
-        }
-
-        Ok(())
-    }
-
    /// Builds a merge reader that reads all data.
    async fn build_all_merge_reader(
-        stream_ctx: &StreamContext,
+        stream_ctx: &Arc<StreamContext>,
+        partition_ranges: &[PartitionRange],
        semaphore: Arc<Semaphore>,
-        metrics: &mut ScannerMetrics,
        compaction: bool,
-        parallelism: usize,
-    ) -> Result<Option<BoxedBatchReader>> {
-        // initialize parts list
-        let mut parts = stream_ctx.parts.lock().await;
-        Self::maybe_init_parts(&stream_ctx.input, &mut parts, metrics, parallelism).await?;
-        let parts_len = parts.0.len();
-
-        let mut sources = Vec::with_capacity(parts_len);
-        for id in 0..parts_len {
-            let Some(part) = parts.0.get_part(id) else {
-                return Ok(None);
-            };
-
-            Self::build_part_sources(part, &mut sources, None, compaction)?;
-        }
-
-        Self::build_reader_from_sources(stream_ctx, sources, semaphore).await
-    }
-
-    /// Builds a merge reader that reads data from one [`PartitionRange`].
-    ///
-    /// If the `range_id` is out of bound, returns None.
-    async fn build_merge_reader(
-        stream_ctx: &StreamContext,
-        range_id: usize,
-        semaphore: Arc<Semaphore>,
-        metrics: &mut ScannerMetrics,
-        compaction: bool,
-        parallelism: usize,
-    ) -> Result<Option<BoxedBatchReader>> {
+        part_metrics: &PartitionMetrics,
+    ) -> Result<BoxedBatchReader> {
        let mut sources = Vec::new();
-        let build_start = {
-            let mut parts = stream_ctx.parts.lock().await;
-            Self::maybe_init_parts(&stream_ctx.input, &mut parts, metrics, parallelism).await?;
-
-            let Some(part) = parts.0.get_part(range_id) else {
-                return Ok(None);
-            };
-
-            let build_start = Instant::now();
-            Self::build_part_sources(
-                part,
-                &mut sources,
-                stream_ctx.input.series_row_selector,
+        for part_range in partition_ranges {
+            build_sources(
+                stream_ctx,
+                part_range,
                compaction,
-            )?;
-
-            build_start
-        };
-
-        let maybe_reader = Self::build_reader_from_sources(stream_ctx, sources, semaphore).await;
-        let build_reader_cost = build_start.elapsed();
-        metrics.build_reader_cost += build_reader_cost;
-        debug!(
-            "Build reader region: {}, range_id: {}, from sources, build_reader_cost: {:?}, compaction: {}",
-            stream_ctx.input.mapper.metadata().region_id,
-            range_id,
-            build_reader_cost,
-            compaction,
-        );
-
-        maybe_reader
+                part_metrics,
+                &mut sources,
+            );
+        }
+        Self::build_reader_from_sources(stream_ctx, sources, semaphore).await
    }

    #[tracing::instrument(level = tracing::Level::DEBUG, skip_all)]
@@ -253,7 +147,7 @@ impl SeqScan {
        stream_ctx: &StreamContext,
        mut sources: Vec<Source>,
        semaphore: Arc<Semaphore>,
-    ) -> Result<Option<BoxedBatchReader>> {
+    ) -> Result<BoxedBatchReader> {
        if stream_ctx.input.parallelism.parallelism > 1 {
            // Read sources in parallel. We always spawn a task so we can control the parallelism
            // by the semaphore.
@@ -286,13 +180,11 @@ impl SeqScan {
            None => reader,
        };

-        Ok(Some(reader))
+        Ok(reader)
    }

    /// Scans the given partition when the part list is set properly.
    /// Otherwise the returned stream might not contains any data.
-    // TODO: refactor out `uncached_scan_part_impl`.
-    #[allow(dead_code)]
    fn scan_partition_impl(
        &self,
        partition: usize,
@@ -307,28 +199,36 @@ impl SeqScan {
            ));
        }

-        let mut metrics = ScannerMetrics {
-            prepare_scan_cost: self.stream_ctx.query_start.elapsed(),
-            ..Default::default()
-        };
        let stream_ctx = self.stream_ctx.clone();
        let semaphore = self.semaphore.clone();
        let partition_ranges = self.properties.partitions[partition].clone();
        let compaction = self.compaction;
-        let parallelism = self.properties.num_partitions();
-        let stream = try_stream! {
-            let first_poll = stream_ctx.query_start.elapsed();
+        let part_metrics = PartitionMetrics::new(
+            self.stream_ctx.input.mapper.metadata().region_id,
+            partition,
+            get_scanner_type(self.compaction),
+            stream_ctx.query_start,
+            ScannerMetrics {
+                prepare_scan_cost: self.stream_ctx.query_start.elapsed(),
+                ..Default::default()
+            },
+        );

-            for partition_range in partition_ranges {
-                let maybe_reader =
-                    Self::build_merge_reader(&stream_ctx, partition_range.identifier, semaphore.clone(), &mut metrics, compaction, parallelism)
+        let stream = try_stream! {
+            part_metrics.on_first_poll();
+
+            // Scans each part.
+            for part_range in partition_ranges {
+                let mut sources = Vec::new();
+                build_sources(&stream_ctx, &part_range, compaction, &part_metrics, &mut sources);
+
+                let mut reader =
+                    Self::build_reader_from_sources(&stream_ctx, sources, semaphore.clone())
                        .await
                        .map_err(BoxedError::new)
                        .context(ExternalSnafu)?;
-                let Some(mut reader) = maybe_reader else {
-                    return;
-                };
                let cache = stream_ctx.input.cache_manager.as_deref();
+                let mut metrics = ScannerMetrics::default();
                let mut fetch_start = Instant::now();
                while let Some(batch) = reader
                    .next_batch()
@@ -350,18 +250,10 @@ impl SeqScan {
                    fetch_start = Instant::now();
                }
                metrics.scan_cost += fetch_start.elapsed();
-                metrics.total_cost = stream_ctx.query_start.elapsed();
-                metrics.observe_metrics_on_finish();
-
-                debug!(
-                    "Seq scan finished, region_id: {:?}, partition: {}, metrics: {:?}, first_poll: {:?}, compaction: {}",
-                    stream_ctx.input.mapper.metadata().region_id,
-                    partition,
-                    metrics,
-                    first_poll,
-                    compaction,
-                );
+                part_metrics.merge_metrics(&metrics);
            }
+
+            part_metrics.on_finish();
        };

        let stream = Box::pin(RecordBatchStreamWrapper::new(
@@ -371,133 +263,6 @@ impl SeqScan {

        Ok(stream)
    }
-
-    /// Scans the given partition when the part list is not set.
-    /// This method will do a lazy initialize of part list and
-    /// ignores the partition settings in `properties`.
-    fn uncached_scan_part_impl(
-        &self,
-        partition: usize,
-    ) -> Result<SendableRecordBatchStream, BoxedError> {
-        let num_partitions = self.properties.partitions.len();
-        if partition >= num_partitions {
-            return Err(BoxedError::new(
-                PartitionOutOfRangeSnafu {
-                    given: partition,
-                    all: self.properties.partitions.len(),
-                }
-                .build(),
-            ));
-        }
-        let mut metrics = ScannerMetrics {
-            prepare_scan_cost: self.stream_ctx.query_start.elapsed(),
-            ..Default::default()
-        };
-        let stream_ctx = self.stream_ctx.clone();
-        let semaphore = self.semaphore.clone();
-        let compaction = self.compaction;
-        let parallelism = self.properties.num_partitions();
-
-        // build stream
-        let stream = try_stream! {
-            let first_poll = stream_ctx.query_start.elapsed();
-
-            // init parts
-            let parts_len = {
-                let mut parts = stream_ctx.parts.lock().await;
-                Self::maybe_init_parts(&stream_ctx.input, &mut parts, &mut metrics, parallelism).await
-                    .map_err(BoxedError::new)
-                    .context(ExternalSnafu)?;
-                parts.0.len()
-            };
-
-            for id in (0..parts_len).skip(partition).step_by(num_partitions) {
-                let maybe_reader = Self::build_merge_reader(
-                    &stream_ctx,
-                    id,
-                    semaphore.clone(),
-                    &mut metrics,
-                    compaction,
-                    parallelism
-                )
-                .await
-                .map_err(BoxedError::new)
-                .context(ExternalSnafu)?;
-                let Some(mut reader) = maybe_reader else {
-                    return;
-                };
-                let cache = stream_ctx.input.cache_manager.as_deref();
-                let mut fetch_start = Instant::now();
-                while let Some(batch) = reader
-                    .next_batch()
-                    .await
-                    .map_err(BoxedError::new)
-                    .context(ExternalSnafu)?
-                {
-                    metrics.scan_cost += fetch_start.elapsed();
-                    metrics.num_batches += 1;
-                    metrics.num_rows += batch.num_rows();
-
-                    let convert_start = Instant::now();
-                    let record_batch = stream_ctx.input.mapper.convert(&batch, cache)?;
-                    metrics.convert_cost += convert_start.elapsed();
-                    let yield_start = Instant::now();
-                    yield record_batch;
-                    metrics.yield_cost += yield_start.elapsed();
-
-                    fetch_start = Instant::now();
-                }
-                metrics.scan_cost += fetch_start.elapsed();
-                metrics.total_cost = stream_ctx.query_start.elapsed();
-                metrics.observe_metrics_on_finish();
-
-                debug!(
-                    "Seq scan finished, region_id: {}, partition: {}, id: {}, metrics: {:?}, first_poll: {:?}, compaction: {}",
-                    stream_ctx.input.mapper.metadata().region_id,
-                    partition,
-                    id,
-                    metrics,
-                    first_poll,
-                    compaction,
-                );
-            }
-        };
-
-        let stream = Box::pin(RecordBatchStreamWrapper::new(
-            self.stream_ctx.input.mapper.output_schema(),
-            Box::pin(stream),
-        ));
-
-        Ok(stream)
-    }
-
-    /// Initializes parts if they are not built yet.
-    async fn maybe_init_parts(
-        input: &ScanInput,
-        part_list: &mut (ScanPartList, Duration),
-        metrics: &mut ScannerMetrics,
-        parallelism: usize,
-    ) -> Result<()> {
-        if part_list.0.is_none() {
-            let now = Instant::now();
-            let mut distributor = SeqDistributor::default();
-            let reader_metrics = input.prune_file_ranges(&mut distributor).await?;
-            distributor.append_mem_ranges(
-                &input.memtables,
-                Some(input.mapper.column_ids()),
-                input.predicate.clone(),
-            );
-            part_list.0.set_parts(distributor.build_parts(parallelism));
-            let build_part_cost = now.elapsed();
-            part_list.1 = build_part_cost;
-
-            metrics.observe_init_part(build_part_cost, &reader_metrics);
-        } else {
-            // Updates the cost of building parts.
-            metrics.build_parts_cost = part_list.1;
-        }
-        Ok(())
-    }
 }

 impl RegionScanner for SeqScan {
@@ -510,7 +275,7 @@ impl RegionScanner for SeqScan {
    }

    fn scan_partition(&self, partition: usize) -> Result<SendableRecordBatchStream, BoxedError> {
-        self.uncached_scan_part_impl(partition)
+        self.scan_partition_impl(partition)
    }

    fn prepare(&mut self, ranges: Vec<Vec<PartitionRange>>) -> Result<(), BoxedError> {
@@ -525,24 +290,53 @@ impl RegionScanner for SeqScan {
 }

 impl DisplayAs for SeqScan {
-    fn fmt_as(&self, t: DisplayFormatType, f: &mut fmt::Formatter) -> fmt::Result {
+    fn fmt_as(&self, _t: DisplayFormatType, f: &mut fmt::Formatter) -> fmt::Result {
        write!(
            f,
            "SeqScan: region={}, ",
            self.stream_ctx.input.mapper.metadata().region_id
        )?;
-        self.stream_ctx.format_for_explain(t, f)
+        self.stream_ctx.format_for_explain(f)
    }
 }

 impl fmt::Debug for SeqScan {
    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
        f.debug_struct("SeqScan")
-            .field("parts", &self.stream_ctx.parts)
+            .field("num_ranges", &self.stream_ctx.ranges.len())
            .finish()
    }
 }

+/// Builds sources for the partition range.
+fn build_sources(
+    stream_ctx: &Arc<StreamContext>,
+    part_range: &PartitionRange,
+    compaction: bool,
+    part_metrics: &PartitionMetrics,
+    sources: &mut Vec<Source>,
+) {
+    // Gets range meta.
+    let range_meta = &stream_ctx.ranges[part_range.identifier];
+    sources.reserve(range_meta.row_group_indices.len());
+    for index in &range_meta.row_group_indices {
+        let stream = if stream_ctx.is_mem_range_index(*index) {
+            let stream = scan_mem_ranges(stream_ctx.clone(), part_metrics.clone(), *index);
+            Box::pin(stream) as _
+        } else {
+            let read_type = if compaction {
+                "compaction"
+            } else {
+                "seq_scan_files"
+            };
+            let stream =
+                scan_file_ranges(stream_ctx.clone(), part_metrics.clone(), *index, read_type);
+            Box::pin(stream) as _
+        };
+        sources.push(Source::Stream(stream));
+    }
+}
+
 #[cfg(test)]
 impl SeqScan {
    /// Returns the input.
@@ -551,266 +345,11 @@ impl SeqScan {
    }
 }

-/// Builds [ScanPart]s that preserves order.
-#[derive(Default)]
-pub(crate) struct SeqDistributor {
-    parts: Vec<ScanPart>,
-}
-
-impl FileRangeCollector for SeqDistributor {
-    fn append_file_ranges(
-        &mut self,
-        file_meta: &FileMeta,
-        file_ranges: impl Iterator<Item = FileRange>,
-    ) {
-        // Creates a [ScanPart] for each file.
-        let ranges: Vec<_> = file_ranges.collect();
-        if ranges.is_empty() {
-            // No ranges to read.
-            return;
-        }
-        let part = ScanPart {
-            memtable_ranges: Vec::new(),
-            file_ranges: smallvec![ranges],
-            time_range: Some(file_meta.time_range),
-        };
-        self.parts.push(part);
-    }
-}
-
-impl SeqDistributor {
-    /// Appends memtable ranges to the distributor.
-    fn append_mem_ranges(
-        &mut self,
-        memtables: &[MemtableRef],
-        projection: Option<&[ColumnId]>,
-        predicate: Option<Predicate>,
-    ) {
-        for mem in memtables {
-            let stats = mem.stats();
-            let mem_ranges = mem.ranges(projection, predicate.clone());
-            if mem_ranges.is_empty() {
-                continue;
-            }
-            let part = ScanPart {
-                memtable_ranges: mem_ranges.into_values().collect(),
-                file_ranges: smallvec![],
-                time_range: stats.time_range(),
-            };
-            self.parts.push(part);
-        }
-    }
-
-    /// Groups file ranges and memtable ranges by time ranges.
-    /// The output number of parts may be `<= parallelism`. If `parallelism` is 0, it will be set to 1.
-    ///
-    /// Output parts have non-overlapping time ranges.
-    fn build_parts(self, parallelism: usize) -> Vec<ScanPart> {
-        let parallelism = parallelism.max(1);
-        let parts = group_parts_by_range(self.parts);
-        let parts = maybe_split_parts(parts, parallelism);
-        // Ensures it doesn't returns parts more than `parallelism`.
-        maybe_merge_parts(parts, parallelism)
-    }
-}
-
-/// Groups parts by time range. It may generate parts more than parallelism.
-/// All time ranges are not None.
-fn group_parts_by_range(mut parts: Vec<ScanPart>) -> Vec<ScanPart> {
-    if parts.is_empty() {
-        return Vec::new();
-    }
-
-    // Sorts parts by time range.
-    parts.sort_unstable_by(|a, b| {
-        // Safety: time ranges of parts from [SeqPartBuilder] are not None.
-        let a = a.time_range.unwrap();
-        let b = b.time_range.unwrap();
-        a.0.cmp(&b.0).then_with(|| b.1.cmp(&a.1))
-    });
-    let mut part_in_range = None;
-    // Parts with exclusive time ranges.
-    let mut part_groups = Vec::new();
-    for part in parts {
-        let Some(mut prev_part) = part_in_range.take() else {
-            part_in_range = Some(part);
-            continue;
-        };
-
-        if prev_part.overlaps(&part) {
-            prev_part.merge(part);
-            part_in_range = Some(prev_part);
-        } else {
-            // A new group.
-            part_groups.push(prev_part);
-            part_in_range = Some(part);
-        }
-    }
-    if let Some(part) = part_in_range {
-        part_groups.push(part);
-    }
-
-    part_groups
-}
-
-/// Merges parts by parallelism.
-/// It merges parts if the number of parts is greater than `parallelism`.
-fn maybe_merge_parts(mut parts: Vec<ScanPart>, parallelism: usize) -> Vec<ScanPart> {
-    assert!(parallelism > 0);
-    if parts.len() <= parallelism {
-        // No need to merge parts.
-        return parts;
-    }
-
-    // Sort parts by number of memtables and ranges in reverse order.
-    parts.sort_unstable_by(|a, b| {
-        a.memtable_ranges
-            .len()
-            .cmp(&b.memtable_ranges.len())
-            .then_with(|| {
-                let a_ranges_len = a
-                    .file_ranges
-                    .iter()
-                    .map(|ranges| ranges.len())
-                    .sum::<usize>();
-                let b_ranges_len = b
-                    .file_ranges
-                    .iter()
-                    .map(|ranges| ranges.len())
-                    .sum::<usize>();
-                a_ranges_len.cmp(&b_ranges_len)
-            })
-            .reverse()
-    });
-
-    let parts_to_reduce = parts.len() - parallelism;
-    for _ in 0..parts_to_reduce {
-        // Safety: We ensure `parts.len() > parallelism`.
-        let part = parts.pop().unwrap();
-        parts.last_mut().unwrap().merge(part);
-    }
-
-    parts
-}
-
-/// Splits parts by parallelism.
-/// It splits a part if it only scans one file and doesn't scan any memtable.
-fn maybe_split_parts(mut parts: Vec<ScanPart>, parallelism: usize) -> Vec<ScanPart> {
-    assert!(parallelism > 0);
-    if parts.len() >= parallelism {
-        // No need to split parts.
-        return parts;
-    }
-
-    let has_part_to_split = parts.iter().any(|part| part.can_split_preserve_order());
-    if !has_part_to_split {
-        // No proper parts to scan.
-        return parts;
-    }
-
-    // Sorts parts by the number of ranges in the first file.
-    parts.sort_unstable_by(|a, b| {
-        let a_len = a.file_ranges.first().map(|file| file.len()).unwrap_or(0);
-        let b_len = b.file_ranges.first().map(|file| file.len()).unwrap_or(0);
-        a_len.cmp(&b_len).reverse()
-    });
-    let num_parts_to_split = parallelism - parts.len();
-    let mut output_parts = Vec::with_capacity(parallelism);
-    // Split parts up to num_parts_to_split.
-    for part in parts.iter_mut() {
-        if !part.can_split_preserve_order() {
-            continue;
-        }
-        // Safety: `can_split_preserve_order()` ensures file_ranges.len() == 1.
-        // Splits part into `num_parts_to_split + 1` new parts if possible.
-        let target_part_num = num_parts_to_split + 1;
-        let ranges_per_part = (part.file_ranges[0].len() + target_part_num - 1) / target_part_num;
-        // `can_split_preserve_order()` ensures part.file_ranges[0].len() > 1.
-        assert!(ranges_per_part > 0);
-        for ranges in part.file_ranges[0].chunks(ranges_per_part) {
-            let new_part = ScanPart {
-                memtable_ranges: Vec::new(),
-                file_ranges: smallvec![ranges.to_vec()],
-                time_range: part.time_range,
-            };
-            output_parts.push(new_part);
-        }
-        // Replace the current part with the last output part as we will put the current part
-        // into the output parts later.
-        *part = output_parts.pop().unwrap();
-        if output_parts.len() >= num_parts_to_split {
-            // We already split enough parts.
-            break;
-        }
-    }
-    // Put the remaining parts into the output parts.
-    output_parts.append(&mut parts);
-
-    output_parts
-}
-
-#[cfg(test)]
-mod tests {
-    use common_time::timestamp::TimeUnit;
-    use common_time::Timestamp;
-
-    use super::*;
-    use crate::memtable::MemtableId;
-    use crate::test_util::memtable_util::mem_range_for_test;
-
-    type Output = (Vec<MemtableId>, i64, i64);
-
-    fn run_group_parts_test(input: &[(MemtableId, i64, i64)], expect: &[Output]) {
-        let parts = input
-            .iter()
-            .map(|(id, start, end)| {
-                let range = (
-                    Timestamp::new(*start, TimeUnit::Second),
-                    Timestamp::new(*end, TimeUnit::Second),
-                );
-                ScanPart {
-                    memtable_ranges: vec![mem_range_for_test(*id)],
-                    file_ranges: smallvec![],
-                    time_range: Some(range),
-                }
-            })
-            .collect();
-        let output = group_parts_by_range(parts);
-        let actual: Vec<_> = output
-            .iter()
-            .map(|part| {
-                let ids: Vec<_> = part.memtable_ranges.iter().map(|mem| mem.id()).collect();
-                let range = part.time_range.unwrap();
-                (ids, range.0.value(), range.1.value())
-            })
-            .collect();
-        assert_eq!(expect, actual);
-    }
-
-    #[test]
-    fn test_group_parts() {
-        // Group 1 part.
-        run_group_parts_test(&[(1, 0, 2000)], &[(vec![1], 0, 2000)]);
-
-        // 1, 2, 3, 4 => [3, 1, 4], [2]
-        run_group_parts_test(
-            &[
-                (1, 1000, 2000),
-                (2, 6000, 7000),
-                (3, 0, 1500),
-                (4, 1500, 3000),
-            ],
-            &[(vec![3, 1, 4], 0, 3000), (vec![2], 6000, 7000)],
-        );
-
-        // 1, 2, 3 => [3], [1], [2],
-        run_group_parts_test(
-            &[(1, 3000, 4000), (2, 4001, 6000), (3, 0, 1000)],
-            &[
-                (vec![3], 0, 1000),
-                (vec![1], 3000, 4000),
-                (vec![2], 4001, 6000),
-            ],
-        );
+/// Returns the scanner type.
+fn get_scanner_type(compaction: bool) -> &'static str {
+    if compaction {
+        "SeqScan(compaction)"
+    } else {
+        "SeqScan"
    }
 }
--- a/src/mito2/src/read/unordered_scan.rs
+++ b/src/mito2/src/read/unordered_scan.rs
@@ -21,24 +21,17 @@ use std::time::Instant;
 use async_stream::{stream, try_stream};
 use common_error::ext::BoxedError;
 use common_recordbatch::error::ExternalSnafu;
-use common_recordbatch::{RecordBatch, RecordBatchStreamWrapper, SendableRecordBatchStream};
-use common_telemetry::debug;
+use common_recordbatch::{RecordBatchStreamWrapper, SendableRecordBatchStream};
 use datafusion::physical_plan::{DisplayAs, DisplayFormatType};
 use datatypes::schema::SchemaRef;
 use futures::{Stream, StreamExt};
 use snafu::ResultExt;
 use store_api::region_engine::{PartitionRange, RegionScanner, ScannerProperties};

-use crate::cache::CacheManager;
-use crate::error::Result;
-use crate::memtable::MemtableRange;
-use crate::read::compat::CompatBatch;
-use crate::read::projection::ProjectionMapper;
-use crate::read::range::RowGroupIndex;
+use crate::error::{PartitionOutOfRangeSnafu, Result};
 use crate::read::scan_region::{ScanInput, StreamContext};
-use crate::read::{ScannerMetrics, Source};
-use crate::sst::parquet::file_range::FileRange;
-use crate::sst::parquet::reader::ReaderMetrics;
+use crate::read::scan_util::{scan_file_ranges, scan_mem_ranges, PartitionMetrics};
+use crate::read::{Batch, ScannerMetrics};

 /// Scans a region without providing any output ordering guarantee.
 ///
@@ -85,62 +78,23 @@ impl UnorderedScan {
        Ok(stream)
    }

-    /// Fetch a batch from the source and convert it into a record batch.
-    async fn fetch_from_source(
-        source: &mut Source,
-        mapper: &ProjectionMapper,
-        cache: Option<&CacheManager>,
-        compat_batch: Option<&CompatBatch>,
-        metrics: &mut ScannerMetrics,
-    ) -> common_recordbatch::error::Result<Option<RecordBatch>> {
-        let start = Instant::now();
-
-        let Some(mut batch) = source
-            .next_batch()
-            .await
-            .map_err(BoxedError::new)
-            .context(ExternalSnafu)?
-        else {
-            metrics.scan_cost += start.elapsed();
-
-            return Ok(None);
-        };
-
-        if let Some(compat) = compat_batch {
-            batch = compat
-                .compat_batch(batch)
-                .map_err(BoxedError::new)
-                .context(ExternalSnafu)?;
-        }
-        metrics.scan_cost += start.elapsed();
-
-        let convert_start = Instant::now();
-        let record_batch = mapper.convert(&batch, cache)?;
-        metrics.convert_cost += convert_start.elapsed();
-
-        Ok(Some(record_batch))
-    }
-
-    /// Scans a [PartitionRange] and returns a stream.
-    fn scan_partition_range<'a>(
-        stream_ctx: &'a StreamContext,
-        part_range: &'a PartitionRange,
-        mem_ranges: &'a mut Vec<MemtableRange>,
-        file_ranges: &'a mut Vec<FileRange>,
-        reader_metrics: &'a mut ReaderMetrics,
-        metrics: &'a mut ScannerMetrics,
-    ) -> impl Stream<Item = common_recordbatch::error::Result<RecordBatch>> + 'a {
+    /// Scans a [PartitionRange] by its `identifier` and returns a stream.
+    fn scan_partition_range(
+        stream_ctx: Arc<StreamContext>,
+        part_range_id: usize,
+        part_metrics: PartitionMetrics,
+    ) -> impl Stream<Item = Result<Batch>> {
        stream! {
            // Gets range meta.
-            let range_meta = &stream_ctx.ranges[part_range.identifier];
+            let range_meta = &stream_ctx.ranges[part_range_id];
            for index in &range_meta.row_group_indices {
                if stream_ctx.is_mem_range_index(*index) {
-                    let stream = Self::scan_mem_ranges(stream_ctx, *index, mem_ranges, metrics);
+                    let stream = scan_mem_ranges(stream_ctx.clone(), part_metrics.clone(), *index);
                    for await batch in stream {
                        yield batch;
                    }
                } else {
-                    let stream = Self::scan_file_ranges(stream_ctx, *index, file_ranges, reader_metrics, metrics);
+                    let stream = scan_file_ranges(stream_ctx.clone(), part_metrics.clone(), *index, "unordered_scan_files");
                    for await batch in stream {
                        yield batch;
                    }
@@ -149,124 +103,68 @@ impl UnorderedScan {
        }
    }

-    /// Scans memtable ranges at `index`.
-    fn scan_mem_ranges<'a>(
-        stream_ctx: &'a StreamContext,
-        index: RowGroupIndex,
-        ranges: &'a mut Vec<MemtableRange>,
-        metrics: &'a mut ScannerMetrics,
-    ) -> impl Stream<Item = common_recordbatch::error::Result<RecordBatch>> + 'a {
-        try_stream! {
-            let mapper = &stream_ctx.input.mapper;
-            let cache = stream_ctx.input.cache_manager.as_deref();
-            stream_ctx.build_mem_ranges(index, ranges);
-            metrics.num_mem_ranges += ranges.len();
-            for range in ranges {
-                let build_reader_start = Instant::now();
-                let iter = range.build_iter().map_err(BoxedError::new).context(ExternalSnafu)?;
-                metrics.build_reader_cost = build_reader_start.elapsed();
-
-                let mut source = Source::Iter(iter);
-                while let Some(batch) =
-                    Self::fetch_from_source(&mut source, mapper, cache, None, metrics).await?
-                {
-                    metrics.num_batches += 1;
-                    metrics.num_rows += batch.num_rows();
-                    let yield_start = Instant::now();
-                    yield batch;
-                    metrics.yield_cost += yield_start.elapsed();
-                }
-            }
-        }
-    }
-
-    /// Scans file ranges at `index`.
-    fn scan_file_ranges<'a>(
-        stream_ctx: &'a StreamContext,
-        index: RowGroupIndex,
-        ranges: &'a mut Vec<FileRange>,
-        reader_metrics: &'a mut ReaderMetrics,
-        metrics: &'a mut ScannerMetrics,
-    ) -> impl Stream<Item = common_recordbatch::error::Result<RecordBatch>> + 'a {
-        try_stream! {
-            let mapper = &stream_ctx.input.mapper;
-            let cache = stream_ctx.input.cache_manager.as_deref();
-            stream_ctx
-                .build_file_ranges(index, ranges, reader_metrics)
-                .await
-                .map_err(BoxedError::new)
-                .context(ExternalSnafu)?;
-            metrics.num_file_ranges += ranges.len();
-            for range in ranges {
-                let build_reader_start = Instant::now();
-                let reader = range
-                    .reader(None)
-                    .await
-                    .map_err(BoxedError::new)
-                    .context(ExternalSnafu)?;
-                metrics.build_reader_cost += build_reader_start.elapsed();
-                let compat_batch = range.compat_batch();
-                let mut source = Source::PruneReader(reader);
-                while let Some(batch) =
-                    Self::fetch_from_source(&mut source, mapper, cache, compat_batch, metrics)
-                        .await?
-                {
-                    metrics.num_batches += 1;
-                    metrics.num_rows += batch.num_rows();
-                    let yield_start = Instant::now();
-                    yield batch;
-                    metrics.yield_cost += yield_start.elapsed();
-                }
-                if let Source::PruneReader(mut reader) = source {
-                    reader_metrics.merge_from(reader.metrics());
-                }
-            }
-        }
-    }
-
    fn scan_partition_impl(
        &self,
        partition: usize,
    ) -> Result<SendableRecordBatchStream, BoxedError> {
-        let mut metrics = ScannerMetrics {
-            prepare_scan_cost: self.stream_ctx.query_start.elapsed(),
-            ..Default::default()
-        };
+        if partition >= self.properties.partitions.len() {
+            return Err(BoxedError::new(
+                PartitionOutOfRangeSnafu {
+                    given: partition,
+                    all: self.properties.partitions.len(),
+                }
+                .build(),
+            ));
+        }
+
+        let part_metrics = PartitionMetrics::new(
+            self.stream_ctx.input.mapper.metadata().region_id,
+            partition,
+            "UnorderedScan",
+            self.stream_ctx.query_start,
+            ScannerMetrics {
+                prepare_scan_cost: self.stream_ctx.query_start.elapsed(),
+                ..Default::default()
+            },
+        );
        let stream_ctx = self.stream_ctx.clone();
-        let ranges_opt = self.properties.partitions.get(partition).cloned();
+        let part_ranges = self.properties.partitions[partition].clone();

-        let stream = stream! {
-            let first_poll = stream_ctx.query_start.elapsed();
-            let Some(part_ranges) = ranges_opt else {
-                return;
-            };
+        let stream = try_stream! {
+            part_metrics.on_first_poll();

-            let mut mem_ranges = Vec::new();
-            let mut file_ranges = Vec::new();
-            let mut reader_metrics = ReaderMetrics::default();
+            let cache = stream_ctx.input.cache_manager.as_deref();
            // Scans each part.
            for part_range in part_ranges {
+                let mut metrics = ScannerMetrics::default();
+                let mut fetch_start = Instant::now();
+
                let stream = Self::scan_partition_range(
-                    &stream_ctx,
-                    &part_range,
-                    &mut mem_ranges,
-                    &mut file_ranges,
-                    &mut reader_metrics,
-                    &mut metrics,
+                    stream_ctx.clone(),
+                    part_range.identifier,
+                    part_metrics.clone(),
                );
                for await batch in stream {
-                    yield batch;
+                    let batch = batch.map_err(BoxedError::new).context(ExternalSnafu)?;
+                    metrics.scan_cost += fetch_start.elapsed();
+                    metrics.num_batches += 1;
+                    metrics.num_rows += batch.num_rows();
+
+                    let convert_start = Instant::now();
+                    let record_batch = stream_ctx.input.mapper.convert(&batch, cache)?;
+                    metrics.convert_cost += convert_start.elapsed();
+                    let yield_start = Instant::now();
+                    yield record_batch;
+                    metrics.yield_cost += yield_start.elapsed();
+
+                    fetch_start = Instant::now();
                }
+
+                metrics.scan_cost += fetch_start.elapsed();
+                part_metrics.merge_metrics(&metrics);
            }

-            reader_metrics.observe_rows("unordered_scan_files");
-            metrics.total_cost = stream_ctx.query_start.elapsed();
-            metrics.observe_metrics_on_finish();
-            let mapper = &stream_ctx.input.mapper;
-            debug!(
-                "Unordered scan partition {} finished, region_id: {}, metrics: {:?}, reader_metrics: {:?}, first_poll: {:?}",
-                partition, mapper.metadata().region_id, metrics, reader_metrics, first_poll,
-            );
+            part_metrics.on_finish();
        };
        let stream = Box::pin(RecordBatchStreamWrapper::new(
            self.stream_ctx.input.mapper.output_schema(),
@@ -302,20 +200,20 @@ impl RegionScanner for UnorderedScan {
 }

 impl DisplayAs for UnorderedScan {
-    fn fmt_as(&self, t: DisplayFormatType, f: &mut fmt::Formatter) -> fmt::Result {
+    fn fmt_as(&self, _t: DisplayFormatType, f: &mut fmt::Formatter) -> fmt::Result {
        write!(
            f,
            "UnorderedScan: region={}, ",
            self.stream_ctx.input.mapper.metadata().region_id
        )?;
-        self.stream_ctx.format_for_explain(t, f)
+        self.stream_ctx.format_for_explain(f)
    }
 }

 impl fmt::Debug for UnorderedScan {
    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
        f.debug_struct("UnorderedScan")
-            .field("parts", &self.stream_ctx.parts)
+            .field("num_ranges", &self.stream_ctx.ranges.len())
            .finish()
    }
 }
--- a/src/mito2/src/row_converter.rs
+++ b/src/mito2/src/row_converter.rs
@@ -16,9 +16,10 @@ use bytes::Buf;
 use common_base::bytes::Bytes;
 use common_decimal::Decimal128;
 use common_time::time::Time;
-use common_time::{Date, Duration, Interval};
+use common_time::{Date, Duration, IntervalDayTime, IntervalMonthDayNano, IntervalYearMonth};
 use datatypes::data_type::ConcreteDataType;
 use datatypes::prelude::Value;
+use datatypes::types::IntervalType;
 use datatypes::value::ValueRef;
 use memcomparable::{Deserializer, Serializer};
 use paste::paste;
@@ -117,6 +118,24 @@ impl SortField {
                            .serialize($serializer)
                            .context(SerializeFieldSnafu)?;
                    }
+                    ConcreteDataType::Interval(IntervalType::YearMonth(_)) => {
+                        let interval = value.as_interval_year_month().context(FieldTypeMismatchSnafu)?;
+                        interval.map(|i| i.to_i32())
+                            .serialize($serializer)
+                            .context(SerializeFieldSnafu)?;
+                    }
+                    ConcreteDataType::Interval(IntervalType::DayTime(_)) => {
+                        let interval = value.as_interval_day_time().context(FieldTypeMismatchSnafu)?;
+                        interval.map(|i| i.to_i64())
+                            .serialize($serializer)
+                            .context(SerializeFieldSnafu)?;
+                    }
+                    ConcreteDataType::Interval(IntervalType::MonthDayNano(_)) => {
+                        let interval = value.as_interval_month_day_nano().context(FieldTypeMismatchSnafu)?;
+                        interval.map(|i| i.to_i128())
+                            .serialize($serializer)
+                            .context(SerializeFieldSnafu)?;
+                    }
                    ConcreteDataType::List(_) |
                    ConcreteDataType::Dictionary(_) |
                    ConcreteDataType::Null(_) => {
@@ -144,7 +163,6 @@ impl SortField {
            Date, date,
            DateTime, datetime,
            Time, time,
-            Interval, interval,
            Duration, duration,
            Decimal128, decimal128,
            Json, binary
@@ -181,6 +199,24 @@ impl SortField {
                            .map(|t|ty.create_timestamp(t));
                        Ok(Value::from(timestamp))
                    }
+                    ConcreteDataType::Interval(IntervalType::YearMonth(_)) => {
+                        let interval = Option::<i32>::deserialize(deserializer)
+                            .context(error::DeserializeFieldSnafu)?
+                            .map(IntervalYearMonth::from_i32);
+                        Ok(Value::from(interval))
+                    }
+                    ConcreteDataType::Interval(IntervalType::DayTime(_)) => {
+                        let interval = Option::<i64>::deserialize(deserializer)
+                            .context(error::DeserializeFieldSnafu)?
+                            .map(IntervalDayTime::from_i64);
+                        Ok(Value::from(interval))
+                    }
+                    ConcreteDataType::Interval(IntervalType::MonthDayNano(_)) => {
+                        let interval = Option::<i128>::deserialize(deserializer)
+                            .context(error::DeserializeFieldSnafu)?
+                            .map(IntervalMonthDayNano::from_i128);
+                        Ok(Value::from(interval))
+                    }
                    ConcreteDataType::List(l) => NotSupportedFieldSnafu {
                        data_type: ConcreteDataType::List(l.clone()),
                    }
@@ -212,7 +248,6 @@ impl SortField {
            Date, Date,
            Time, Time,
            DateTime, DateTime,
-            Interval, Interval,
            Duration, Duration,
            Decimal128, Decimal128
        )
@@ -263,7 +298,9 @@ impl SortField {
            ConcreteDataType::Timestamp(_) => 9, // We treat timestamp as Option<i64>
            ConcreteDataType::Time(_) => 10,     // i64 and 1 byte time unit
            ConcreteDataType::Duration(_) => 10,
-            ConcreteDataType::Interval(_) => 18,
+            ConcreteDataType::Interval(IntervalType::YearMonth(_)) => 5,
+            ConcreteDataType::Interval(IntervalType::DayTime(_)) => 9,
+            ConcreteDataType::Interval(IntervalType::MonthDayNano(_)) => 17,
            ConcreteDataType::Decimal128(_) => 19,
            ConcreteDataType::Null(_)
            | ConcreteDataType::List(_)
@@ -387,7 +424,9 @@ impl RowCodec for McmpRowCodec {
 #[cfg(test)]
 mod tests {
    use common_base::bytes::StringBytes;
-    use common_time::{DateTime, Timestamp};
+    use common_time::{
+        DateTime, IntervalDayTime, IntervalMonthDayNano, IntervalYearMonth, Timestamp,
+    };
    use datatypes::value::Value;

    use super::*;
@@ -563,6 +602,8 @@ mod tests {
                ConcreteDataType::timestamp_millisecond_datatype(),
                ConcreteDataType::time_millisecond_datatype(),
                ConcreteDataType::duration_millisecond_datatype(),
+                ConcreteDataType::interval_year_month_datatype(),
+                ConcreteDataType::interval_day_time_datatype(),
                ConcreteDataType::interval_month_day_nano_datatype(),
                ConcreteDataType::decimal128_default_datatype(),
            ],
@@ -585,7 +626,9 @@ mod tests {
                Value::Timestamp(Timestamp::new_millisecond(12)),
                Value::Time(Time::new_millisecond(13)),
                Value::Duration(Duration::new_millisecond(14)),
-                Value::Interval(Interval::from_month_day_nano(1, 1, 15)),
+                Value::IntervalYearMonth(IntervalYearMonth::new(1)),
+                Value::IntervalDayTime(IntervalDayTime::new(1, 15)),
+                Value::IntervalMonthDayNano(IntervalMonthDayNano::new(1, 1, 15)),
                Value::Decimal128(Decimal128::from(16)),
            ],
        );
--- a/src/mito2/src/sst/file.rs
+++ b/src/mito2/src/sst/file.rs
@@ -90,7 +90,8 @@ impl FromStr for FileId {
    }
 }

-/// Time range of a SST file.
+/// Time range (min and max timestamps) of a SST file.
+/// Both min and max are inclusive.
 pub type FileTimeRange = (Timestamp, Timestamp);

 /// Checks if two inclusive timestamp ranges overlap with each other.
--- a/src/mito2/src/sst/parquet/reader.rs
+++ b/src/mito2/src/sst/parquet/reader.rs
@@ -238,9 +238,6 @@ impl ParquetReaderBuilder {
            cache_manager: self.cache_manager.clone(),
        };

-        // TODO(yingwen): count the cost of the method.
-        metrics.build_cost = start.elapsed();
-
        let mut filters = if let Some(predicate) = &self.predicate {
            predicate
                .exprs()
@@ -270,6 +267,9 @@ impl ParquetReaderBuilder {
        );

        let context = FileRangeContext::new(reader_builder, filters, read_format, codec);
+
+        metrics.build_cost += start.elapsed();
+
        Ok((context, row_groups))
    }

--- a/src/mito2/src/test_util/memtable_util.rs
+++ b/src/mito2/src/test_util/memtable_util.rs
@@ -35,7 +35,7 @@ use crate::memtable::key_values::KeyValue;
 use crate::memtable::partition_tree::data::{timestamp_array_to_i64_slice, DataBatch, DataBuffer};
 use crate::memtable::{
    BoxedBatchIterator, BulkPart, IterBuilder, KeyValues, Memtable, MemtableBuilder, MemtableId,
-    MemtableRange, MemtableRangeContext, MemtableRef, MemtableStats,
+    MemtableRange, MemtableRef, MemtableStats,
 };
 use crate::row_converter::{McmpRowCodec, RowCodec, SortField};

@@ -361,11 +361,3 @@ pub(crate) fn collect_iter_timestamps(iter: BoxedBatchIterator) -> Vec<i64> {
    .map(|v| v.unwrap().0.value())
    .collect()
 }
-
-/// Builds a memtable range for test.
-pub(crate) fn mem_range_for_test(id: MemtableId) -> MemtableRange {
-    let builder = Box::new(EmptyIterBuilder::default());
-
-    let context = Arc::new(MemtableRangeContext::new(id, builder));
-    MemtableRange::new(context)
-}
--- a/src/pipeline/src/etl.rs
+++ b/src/pipeline/src/etl.rs
@@ -210,6 +210,37 @@ where
        self.transformer.transform_mut(val)
    }

+    pub fn prepare_pipeline_value(&self, val: Value, result: &mut [Value]) -> Result<()> {
+        match val {
+            Value::Map(map) => {
+                let mut search_from = 0;
+                // because of the key in the json map is ordered
+                for (payload_key, payload_value) in map.values.into_iter() {
+                    if search_from >= self.required_keys.len() {
+                        break;
+                    }
+
+                    // because of map key is ordered, required_keys is ordered too
+                    if let Some(pos) = self.required_keys[search_from..]
+                        .iter()
+                        .position(|k| k == &payload_key)
+                    {
+                        result[search_from + pos] = payload_value;
+                        // next search from is always after the current key
+                        search_from += pos;
+                    }
+                }
+            }
+            Value::String(_) => {
+                result[0] = val;
+            }
+            _ => {
+                return PrepareValueMustBeObjectSnafu.fail();
+            }
+        }
+        Ok(())
+    }
+
    pub fn prepare(&self, val: serde_json::Value, result: &mut [Value]) -> Result<()> {
        match val {
            serde_json::Value::Object(map) => {
@@ -286,6 +317,11 @@ pub(crate) fn find_key_index(intermediate_keys: &[String], key: &str, kind: &str
        .context(IntermediateKeyIndexSnafu { kind, key })
 }

+pub enum PipelineWay {
+    Identity,
+    Custom(std::sync::Arc<Pipeline<crate::GreptimeTransformer>>),
+}
+
 #[cfg(test)]
 mod tests {

--- a/src/pipeline/src/etl/error.rs
+++ b/src/pipeline/src/etl/error.rs
@@ -438,18 +438,26 @@ pub enum Error {
        #[snafu(implicit)]
        location: Location,
    },
-    #[snafu(display("failed to coerce complex value, not supported"))]
-    CoerceComplexType {
+    #[snafu(display("Can not coerce json type to {ty}"))]
+    CoerceJsonTypeTo {
+        ty: String,
        #[snafu(implicit)]
        location: Location,
    },
-    #[snafu(display("failed to coerce value: {msg}"))]
+    #[snafu(display(
+        "Can not coerce {ty} to json type. we only consider object and array to be json types."
+    ))]
+    CoerceTypeToJson {
+        ty: String,
+        #[snafu(implicit)]
+        location: Location,
+    },
+    #[snafu(display("Failed to coerce value: {msg}"))]
    CoerceIncompatibleTypes {
        msg: String,
        #[snafu(implicit)]
        location: Location,
    },
-
    #[snafu(display(
        "Invalid resolution: '{resolution}'. Available resolutions: {valid_resolution}"
    ))]
--- a/src/pipeline/src/etl/processor/cmcd.rs
+++ b/src/pipeline/src/etl/processor/cmcd.rs
@@ -402,7 +402,8 @@ impl Processor for CmcdProcessor {

 #[cfg(test)]
 mod tests {
-    use ahash::HashMap;
+    use std::collections::BTreeMap;
+
    use urlencoding::decode;

    use super::{CmcdProcessorBuilder, CMCD_KEYS};
@@ -563,14 +564,14 @@ mod tests {
            let values = vec
                .into_iter()
                .map(|(k, v)| (k.to_string(), v))
-                .collect::<HashMap<String, Value>>();
+                .collect::<BTreeMap<String, Value>>();
            let expected = Map { values };

            let actual = processor.parse(0, &decoded).unwrap();
            let actual = actual
                .into_iter()
                .map(|(index, value)| (intermediate_keys[index].clone(), value))
-                .collect::<HashMap<String, Value>>();
+                .collect::<BTreeMap<String, Value>>();
            let actual = Map { values: actual };
            assert_eq!(actual, expected);
        }
--- a/src/pipeline/src/etl/processor/regex.rs
+++ b/src/pipeline/src/etl/processor/regex.rs
@@ -383,6 +383,8 @@ impl Processor for RegexProcessor {
 }
 #[cfg(test)]
 mod tests {
+    use std::collections::BTreeMap;
+
    use ahash::{HashMap, HashMapExt};
    use itertools::Itertools;

@@ -475,14 +477,14 @@ ignore_missing: false"#;
            .map(|k| k.to_string())
            .collect_vec();
            let processor = builder.build(&intermediate_keys).unwrap();
-            let mut result = HashMap::new();
+            let mut result = BTreeMap::new();
            for (index, pattern) in processor.patterns.iter().enumerate() {
                let r = processor
                    .process(&breadcrumbs_str, pattern, (0, index))
                    .unwrap()
                    .into_iter()
                    .map(|(k, v)| (intermediate_keys[k].clone(), v))
-                    .collect::<HashMap<_, _>>();
+                    .collect::<BTreeMap<_, _>>();
                result.extend(r);
            }
            let map = Map { values: result };
--- a/src/pipeline/src/etl/transform/transformer/greptime/coerce.rs
+++ b/src/pipeline/src/etl/transform/transformer/greptime/coerce.rs
@@ -12,16 +12,17 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.

+use api::v1::column_data_type_extension::TypeExt;
 use api::v1::column_def::options_from_fulltext;
-use api::v1::ColumnOptions;
+use api::v1::{ColumnDataTypeExtension, ColumnOptions, JsonTypeExtension};
 use datatypes::schema::FulltextOptions;
 use greptime_proto::v1::value::ValueData;
 use greptime_proto::v1::{ColumnDataType, ColumnSchema, SemanticType};
 use snafu::ResultExt;

 use crate::etl::error::{
-    CoerceComplexTypeSnafu, CoerceIncompatibleTypesSnafu, CoerceStringToTypeSnafu,
-    CoerceUnsupportedEpochTypeSnafu, CoerceUnsupportedNullTypeSnafu,
+    CoerceIncompatibleTypesSnafu, CoerceJsonTypeToSnafu, CoerceStringToTypeSnafu,
+    CoerceTypeToJsonSnafu, CoerceUnsupportedEpochTypeSnafu, CoerceUnsupportedNullTypeSnafu,
    CoerceUnsupportedNullTypeToSnafu, ColumnOptionsSnafu, Error, Result,
 };
 use crate::etl::transform::index::Index;
@@ -62,7 +63,10 @@ impl TryFrom<Value> for ValueData {
            }
            Value::Timestamp(Timestamp::Second(s)) => Ok(ValueData::TimestampSecondValue(s)),

-            Value::Array(_) | Value::Map(_) => CoerceComplexTypeSnafu.fail(),
+            Value::Array(_) | Value::Map(_) => {
+                let data: jsonb::Value = value.into();
+                Ok(ValueData::BinaryValue(data.to_vec()))
+            }
        }
    }
 }
@@ -74,15 +78,15 @@ pub(crate) fn coerce_columns(transform: &Transform) -> Result<Vec<ColumnSchema>>
    for field in transform.real_fields.iter() {
        let column_name = field.output_name().to_string();

-        let datatype = coerce_type(transform)? as i32;
+        let (datatype, datatype_extension) = coerce_type(transform)?;

        let semantic_type = coerce_semantic_type(transform) as i32;

        let column = ColumnSchema {
            column_name,
-            datatype,
+            datatype: datatype as i32,
            semantic_type,
-            datatype_extension: None,
+            datatype_extension,
            options: coerce_options(transform)?,
        };
        columns.push(column);
@@ -111,30 +115,41 @@ fn coerce_options(transform: &Transform) -> Result<Option<ColumnOptions>> {
    }
 }

-fn coerce_type(transform: &Transform) -> Result<ColumnDataType> {
+fn coerce_type(transform: &Transform) -> Result<(ColumnDataType, Option<ColumnDataTypeExtension>)> {
    match transform.type_ {
-        Value::Int8(_) => Ok(ColumnDataType::Int8),
-        Value::Int16(_) => Ok(ColumnDataType::Int16),
-        Value::Int32(_) => Ok(ColumnDataType::Int32),
-        Value::Int64(_) => Ok(ColumnDataType::Int64),
+        Value::Int8(_) => Ok((ColumnDataType::Int8, None)),
+        Value::Int16(_) => Ok((ColumnDataType::Int16, None)),
+        Value::Int32(_) => Ok((ColumnDataType::Int32, None)),
+        Value::Int64(_) => Ok((ColumnDataType::Int64, None)),

-        Value::Uint8(_) => Ok(ColumnDataType::Uint8),
-        Value::Uint16(_) => Ok(ColumnDataType::Uint16),
-        Value::Uint32(_) => Ok(ColumnDataType::Uint32),
-        Value::Uint64(_) => Ok(ColumnDataType::Uint64),
+        Value::Uint8(_) => Ok((ColumnDataType::Uint8, None)),
+        Value::Uint16(_) => Ok((ColumnDataType::Uint16, None)),
+        Value::Uint32(_) => Ok((ColumnDataType::Uint32, None)),
+        Value::Uint64(_) => Ok((ColumnDataType::Uint64, None)),

-        Value::Float32(_) => Ok(ColumnDataType::Float32),
-        Value::Float64(_) => Ok(ColumnDataType::Float64),
+        Value::Float32(_) => Ok((ColumnDataType::Float32, None)),
+        Value::Float64(_) => Ok((ColumnDataType::Float64, None)),

-        Value::Boolean(_) => Ok(ColumnDataType::Boolean),
-        Value::String(_) => Ok(ColumnDataType::String),
+        Value::Boolean(_) => Ok((ColumnDataType::Boolean, None)),
+        Value::String(_) => Ok((ColumnDataType::String, None)),

-        Value::Timestamp(Timestamp::Nanosecond(_)) => Ok(ColumnDataType::TimestampNanosecond),
-        Value::Timestamp(Timestamp::Microsecond(_)) => Ok(ColumnDataType::TimestampMicrosecond),
-        Value::Timestamp(Timestamp::Millisecond(_)) => Ok(ColumnDataType::TimestampMillisecond),
-        Value::Timestamp(Timestamp::Second(_)) => Ok(ColumnDataType::TimestampSecond),
+        Value::Timestamp(Timestamp::Nanosecond(_)) => {
+            Ok((ColumnDataType::TimestampNanosecond, None))
+        }
+        Value::Timestamp(Timestamp::Microsecond(_)) => {
+            Ok((ColumnDataType::TimestampMicrosecond, None))
+        }
+        Value::Timestamp(Timestamp::Millisecond(_)) => {
+            Ok((ColumnDataType::TimestampMillisecond, None))
+        }
+        Value::Timestamp(Timestamp::Second(_)) => Ok((ColumnDataType::TimestampSecond, None)),

-        Value::Array(_) | Value::Map(_) => CoerceComplexTypeSnafu.fail(),
+        Value::Array(_) | Value::Map(_) => Ok((
+            ColumnDataType::Binary,
+            Some(ColumnDataTypeExtension {
+                type_ext: Some(TypeExt::JsonType(JsonTypeExtension::JsonBinary.into())),
+            }),
+        )),

        Value::Null => CoerceUnsupportedNullTypeToSnafu {
            ty: transform.type_.to_str_type(),
@@ -191,12 +206,12 @@ pub(crate) fn coerce_value(val: &Value, transform: &Transform) -> Result<Option<
                ))),
            },
            _ => CoerceIncompatibleTypesSnafu {
-                msg: "Timestamp can only be coerced to another timestamp",
+                msg: "Timestamp can only be coerced to another type",
            }
            .fail(),
        },

-        Value::Array(_) | Value::Map(_) => CoerceComplexTypeSnafu.fail(),
+        Value::Array(_) | Value::Map(_) => coerce_json_value(val, transform),
    }
 }

@@ -228,7 +243,12 @@ fn coerce_bool_value(b: bool, transform: &Transform) -> Result<Option<ValueData>
            }
        },

-        Value::Array(_) | Value::Map(_) => return CoerceComplexTypeSnafu.fail(),
+        Value::Array(_) | Value::Map(_) => {
+            return CoerceJsonTypeToSnafu {
+                ty: transform.type_.to_str_type(),
+            }
+            .fail()
+        }

        Value::Null => return Ok(None),
    };
@@ -264,7 +284,12 @@ fn coerce_i64_value(n: i64, transform: &Transform) -> Result<Option<ValueData>>
            }
        },

-        Value::Array(_) | Value::Map(_) => return CoerceComplexTypeSnafu.fail(),
+        Value::Array(_) | Value::Map(_) => {
+            return CoerceJsonTypeToSnafu {
+                ty: transform.type_.to_str_type(),
+            }
+            .fail()
+        }

        Value::Null => return Ok(None),
    };
@@ -300,7 +325,12 @@ fn coerce_u64_value(n: u64, transform: &Transform) -> Result<Option<ValueData>>
            }
        },

-        Value::Array(_) | Value::Map(_) => return CoerceComplexTypeSnafu.fail(),
+        Value::Array(_) | Value::Map(_) => {
+            return CoerceJsonTypeToSnafu {
+                ty: transform.type_.to_str_type(),
+            }
+            .fail()
+        }

        Value::Null => return Ok(None),
    };
@@ -336,7 +366,12 @@ fn coerce_f64_value(n: f64, transform: &Transform) -> Result<Option<ValueData>>
            }
        },

-        Value::Array(_) | Value::Map(_) => return CoerceComplexTypeSnafu.fail(),
+        Value::Array(_) | Value::Map(_) => {
+            return CoerceJsonTypeToSnafu {
+                ty: transform.type_.to_str_type(),
+            }
+            .fail()
+        }

        Value::Null => return Ok(None),
    };
@@ -411,12 +446,43 @@ fn coerce_string_value(s: &String, transform: &Transform) -> Result<Option<Value
            None => CoerceUnsupportedEpochTypeSnafu { ty: "String" }.fail(),
        },

-        Value::Array(_) | Value::Map(_) => CoerceComplexTypeSnafu.fail(),
+        Value::Array(_) | Value::Map(_) => {
+            return CoerceJsonTypeToSnafu {
+                ty: transform.type_.to_str_type(),
+            }
+            .fail()
+        }

        Value::Null => Ok(None),
    }
 }

+fn coerce_json_value(v: &Value, transform: &Transform) -> Result<Option<ValueData>> {
+    match &transform.type_ {
+        Value::Array(_) | Value::Map(_) => (),
+        t => {
+            return CoerceTypeToJsonSnafu {
+                ty: t.to_str_type(),
+            }
+            .fail();
+        }
+    }
+    match v {
+        Value::Map(_) => {
+            let data: jsonb::Value = v.into();
+            Ok(Some(ValueData::BinaryValue(data.to_vec())))
+        }
+        Value::Array(_) => {
+            let data: jsonb::Value = v.into();
+            Ok(Some(ValueData::BinaryValue(data.to_vec())))
+        }
+        _ => CoerceTypeToJsonSnafu {
+            ty: v.to_str_type(),
+        }
+        .fail(),
+    }
+}
+
 #[cfg(test)]
 mod tests {
    use super::*;
--- a/src/pipeline/src/etl/value.rs
+++ b/src/pipeline/src/etl/value.rs
@@ -16,8 +16,10 @@ pub mod array;
 pub mod map;
 pub mod time;

-use ahash::{HashMap, HashMapExt};
+use std::collections::BTreeMap;
+
 pub use array::Array;
+use jsonb::{Number as JsonbNumber, Object as JsonbObject, Value as JsonbValue};
 pub use map::Map;
 use snafu::{OptionExt, ResultExt};
 pub use time::Timestamp;
@@ -57,6 +59,7 @@ pub enum Value {

    Timestamp(Timestamp),

+    /// We only consider object and array to be json types.
    Array(Array),
    Map(Map),
 }
@@ -110,8 +113,9 @@ impl Value {
                _ => Ok(Value::Timestamp(Timestamp::Nanosecond(0))),
            },

-            "array" => Ok(Value::Array(Array::default())),
-            "map" => Ok(Value::Map(Map::default())),
+            // We only consider object and array to be json types. and use Map to represent json
+            // TODO(qtang): Needs to be defined with better semantics
+            "json" => Ok(Value::Map(Map::default())),

            _ => ValueParseTypeSnafu { t }.fail(),
        }
@@ -221,8 +225,7 @@ impl Value {

            Value::Timestamp(_) => "epoch",

-            Value::Array(_) => "array",
-            Value::Map(_) => "map",
+            Value::Array(_) | Value::Map(_) => "json",

            Value::Null => "null",
        }
@@ -287,7 +290,7 @@ impl TryFrom<serde_json::Value> for Value {
                Ok(Value::Array(Array { values }))
            }
            serde_json::Value::Object(v) => {
-                let mut values = HashMap::with_capacity(v.len());
+                let mut values = BTreeMap::new();
                for (k, v) in v {
                    values.insert(k, Value::try_from(v)?);
                }
@@ -318,7 +321,7 @@ impl TryFrom<&yaml_rust::Yaml> for Value {
                Ok(Value::Array(Array { values }))
            }
            yaml_rust::Yaml::Hash(v) => {
-                let mut values = HashMap::new();
+                let mut values = BTreeMap::new();
                for (k, v) in v {
                    let key = k
                        .as_str()
@@ -331,3 +334,79 @@ impl TryFrom<&yaml_rust::Yaml> for Value {
        }
    }
 }
+
+impl<'a> From<&Value> for JsonbValue<'a> {
+    fn from(value: &Value) -> Self {
+        match value {
+            Value::Null => JsonbValue::Null,
+            Value::Boolean(v) => JsonbValue::Bool(*v),
+
+            Value::Int8(v) => JsonbValue::Number(JsonbNumber::Int64(*v as i64)),
+            Value::Int16(v) => JsonbValue::Number(JsonbNumber::Int64(*v as i64)),
+            Value::Int32(v) => JsonbValue::Number(JsonbNumber::Int64(*v as i64)),
+            Value::Int64(v) => JsonbValue::Number(JsonbNumber::Int64(*v)),
+
+            Value::Uint8(v) => JsonbValue::Number(JsonbNumber::UInt64(*v as u64)),
+            Value::Uint16(v) => JsonbValue::Number(JsonbNumber::UInt64(*v as u64)),
+            Value::Uint32(v) => JsonbValue::Number(JsonbNumber::UInt64(*v as u64)),
+            Value::Uint64(v) => JsonbValue::Number(JsonbNumber::UInt64(*v)),
+            Value::Float32(v) => JsonbValue::Number(JsonbNumber::Float64(*v as f64)),
+            Value::Float64(v) => JsonbValue::Number(JsonbNumber::Float64(*v)),
+            Value::String(v) => JsonbValue::String(v.clone().into()),
+            Value::Timestamp(v) => JsonbValue::String(v.to_string().into()),
+            Value::Array(arr) => {
+                let mut vals: Vec<JsonbValue> = Vec::with_capacity(arr.len());
+                for val in arr.iter() {
+                    vals.push(val.into());
+                }
+                JsonbValue::Array(vals)
+            }
+            Value::Map(obj) => {
+                let mut map = JsonbObject::new();
+                for (k, v) in obj.iter() {
+                    let val: JsonbValue = v.into();
+                    map.insert(k.to_string(), val);
+                }
+                JsonbValue::Object(map)
+            }
+        }
+    }
+}
+
+impl<'a> From<Value> for JsonbValue<'a> {
+    fn from(value: Value) -> Self {
+        match value {
+            Value::Null => JsonbValue::Null,
+            Value::Boolean(v) => JsonbValue::Bool(v),
+
+            Value::Int8(v) => JsonbValue::Number(JsonbNumber::Int64(v as i64)),
+            Value::Int16(v) => JsonbValue::Number(JsonbNumber::Int64(v as i64)),
+            Value::Int32(v) => JsonbValue::Number(JsonbNumber::Int64(v as i64)),
+            Value::Int64(v) => JsonbValue::Number(JsonbNumber::Int64(v)),
+
+            Value::Uint8(v) => JsonbValue::Number(JsonbNumber::UInt64(v as u64)),
+            Value::Uint16(v) => JsonbValue::Number(JsonbNumber::UInt64(v as u64)),
+            Value::Uint32(v) => JsonbValue::Number(JsonbNumber::UInt64(v as u64)),
+            Value::Uint64(v) => JsonbValue::Number(JsonbNumber::UInt64(v)),
+            Value::Float32(v) => JsonbValue::Number(JsonbNumber::Float64(v as f64)),
+            Value::Float64(v) => JsonbValue::Number(JsonbNumber::Float64(v)),
+            Value::String(v) => JsonbValue::String(v.into()),
+            Value::Timestamp(v) => JsonbValue::String(v.to_string().into()),
+            Value::Array(arr) => {
+                let mut vals: Vec<JsonbValue> = Vec::with_capacity(arr.len());
+                for val in arr.into_iter() {
+                    vals.push(val.into());
+                }
+                JsonbValue::Array(vals)
+            }
+            Value::Map(obj) => {
+                let mut map = JsonbObject::new();
+                for (k, v) in obj.into_iter() {
+                    let val: JsonbValue = v.into();
+                    map.insert(k, val);
+                }
+                JsonbValue::Object(map)
+            }
+        }
+    }
+}
--- a/src/pipeline/src/etl/value/map.rs
+++ b/src/pipeline/src/etl/value/map.rs
@@ -12,21 +12,15 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.

-use ahash::{HashMap, HashMapExt};
+use std::collections::BTreeMap;
+
+use ahash::HashMap;

 use crate::etl::value::Value;

-#[derive(Debug, Clone, PartialEq)]
+#[derive(Debug, Clone, PartialEq, Default)]
 pub struct Map {
-    pub values: HashMap<String, Value>,
-}
-
-impl Default for Map {
-    fn default() -> Self {
-        Self {
-            values: HashMap::with_capacity(30),
-        }
-    }
+    pub values: BTreeMap<String, Value>,
 }

 impl Map {
@@ -47,12 +41,16 @@ impl Map {

 impl From<HashMap<String, Value>> for Map {
    fn from(values: HashMap<String, Value>) -> Self {
-        Map { values }
+        let mut map = Map::default();
+        for (k, v) in values.into_iter() {
+            map.insert(k, v);
+        }
+        map
    }
 }

 impl std::ops::Deref for Map {
-    type Target = HashMap<String, Value>;
+    type Target = BTreeMap<String, Value>;

    fn deref(&self) -> &Self::Target {
        &self.values
@@ -65,6 +63,16 @@ impl std::ops::DerefMut for Map {
    }
 }

+impl std::iter::IntoIterator for Map {
+    type Item = (String, Value);
+
+    type IntoIter = std::collections::btree_map::IntoIter<String, Value>;
+
+    fn into_iter(self) -> Self::IntoIter {
+        self.values.into_iter()
+    }
+}
+
 impl std::fmt::Display for Map {
    fn fmt(&self, f: &mut std::fmt::Formatter) -> std::fmt::Result {
        let values = self
--- a/src/pipeline/src/lib.rs
+++ b/src/pipeline/src/lib.rs
@@ -21,7 +21,7 @@ pub use etl::processor::Processor;
 pub use etl::transform::transformer::identity_pipeline;
 pub use etl::transform::{GreptimeTransformer, Transformer};
 pub use etl::value::{Array, Map, Value};
-pub use etl::{parse, Content, Pipeline};
+pub use etl::{error as etl_error, parse, Content, Pipeline, PipelineWay};
 pub use manager::{
    error, pipeline_operator, table, util, PipelineInfo, PipelineRef, PipelineTableRef,
    PipelineVersion,
--- a/src/pipeline/tests/pipeline.rs
+++ b/src/pipeline/tests/pipeline.rs
@@ -12,6 +12,7 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.

+use api::v1::value::ValueData;
 use api::v1::Rows;
 use common_telemetry::tracing::info;
 use greptime_proto::v1::value::ValueData::{
@@ -466,6 +467,57 @@ transform:
    }
 }

+#[test]
+fn test_json_type() {
+    let input_value_str = r#"
+{
+    "product_object": {"hello":"world"},
+    "product_array": ["hello", "world"]
+}
+"#;
+    let input_value = serde_json::from_str::<serde_json::Value>(input_value_str).unwrap();
+
+    let pipeline_yaml = r#"
+processors:
+
+transform:
+    - fields:
+        - product_object
+        - product_array
+      type: json
+"#;
+
+    let yaml_content = Content::Yaml(pipeline_yaml.into());
+    let pipeline: Pipeline<GreptimeTransformer> = parse(&yaml_content).unwrap();
+
+    let mut status = pipeline.init_intermediate_state();
+    pipeline.prepare(input_value, &mut status).unwrap();
+    let row = pipeline.exec_mut(&mut status).unwrap();
+    let r = row
+        .values
+        .into_iter()
+        .map(|v| v.value_data.unwrap())
+        .collect::<Vec<_>>();
+
+    let product_object = r[0].clone();
+    let product_array = r[1].clone();
+    match product_object {
+        ValueData::BinaryValue(data) => {
+            let jsonb = jsonb::from_slice(&data).unwrap().to_string();
+            assert_eq!(r#"{"hello":"world"}"#, jsonb);
+        }
+        _ => panic!("unexpected value"),
+    }
+
+    match product_array {
+        ValueData::BinaryValue(data) => {
+            let jsonb = jsonb::from_slice(&data).unwrap().to_string();
+            assert_eq!(r#"["hello","world"]"#, jsonb);
+        }
+        _ => panic!("unexpected value"),
+    }
+}
+
 #[test]
 fn test_simple_data() {
    let input_value_str = r#"
--- a/Show More
+++ b/Show More
Author	SHA1	Message	Date
Weny Xu	c02b5dae93	chore: bump version to 0.9.5 (#4853 )	2024-10-18 08:07:13 +00:00
Weny Xu	081c6d9e74	fix: flush metric metadata region (#4852 ) * fix: flush metric metadata region * chore: apply suggestions from CR	2024-10-18 07:21:35 +00:00
Weny Xu	ca6e02980e	fix: overwrite entry_id if entry id is less than start_offset (#4842 ) * fix: overwrite entry_id if entry id is less than start_offset * feat: add `overwrite_entry_start_id` to options * chore: update config.md	2024-10-18 06:31:02 +00:00
Weny Xu	74bdba4613	fix: fix metadata forward compatibility issue (#4846 )	2024-10-18 06:26:41 +00:00
Weny Xu	2e0e82ddc8	chore: update greptime-proto to `b4d3011` (#4850 )	2024-10-18 04:10:22 +00:00
Yingwen	e0c4157ad8	feat: Seq scanner scans data by time range (#4809 ) * feat: seq scan by partition * feat: part metrics * chore: remove unused codes * chore: fmt stream * feat: build ranges returns smallvec * feat: move scan mem/file ranges to util and reuse * feat: log metrics * chore: correct some metrics * feat: get explain info from ranges * test: group test and remove unused codes * chore: fix clippy * feat: change PartitionRange end to exclusive * test: add tests	2024-10-17 11:05:12 +00:00
discord9	613e07afb4	feat: window sort physical plan (#4814 ) * WIP * feat: range split& tests * WIP: split range * add sort exprs * chore: typo * WIP * feat: find successive runs * WIP * READY FOR REVIEW PART ONE: more tests * refactor: break into smaller functions * feat: precompute working range(need testing) * tests: on working range * tests: on working range * feat: support rev working range * feat(to be tested): core logic of merge sort * fix: poll results * fix: find_slice_from_range&test * chore: remove some unused util func&fields * chore: typos * chore: impl exec plan for WindowedSortExec * test(WIP): window sort stream * test: window sort stream * chore: remove unused * fix: fetch * fix: WIP intersection remaining * test: fix and test! * chore: remove outdated comments * chore: rename test * chore: remove dbg line * chore: sorted runs * feat: handling unexpected data * chore: unused * chore: remove a print in test * chore: per review * docs: wrong comment * chore: more test cases	2024-10-16 11:50:25 +00:00
Weny Xu	0ce93f0b88	chore: add more metrics for region migration (#4838 )	2024-10-16 09:36:57 +00:00
Ning Sun	c231eee7c1	fix: respect feature flags for geo function (#4836 )	2024-10-16 07:46:31 +00:00
Yiran	176f2df5b3	fix: dead links (#4837 )	2024-10-16 07:43:14 +00:00
localhost	4622412dfe	feat: add API to write OpenTelemetry logs to GreptimeDB (#4755 ) * chore: otlp logs api * feat: add API to write OpenTelemetry logs to GreptimeDB * chore: fix test data schema error * chore: modify the underlying data structure of the pipeline value map type from hashmap to btremap to keep key order * chore: fix by pr comment * chore: resolve conflicts and add some test * chore: remove useless error * chore: change otlp header name * chore: fmt code * chore: fix integration test for otlp log write api * chore: fix by pr comment * chore: set otlp body with fulltext default	2024-10-16 04:36:08 +00:00
jeremyhi	59ec90299b	refactor: metasrv cannot be cloned (#4834 ) * refactor: metasrv cannot be cloned * chore: remove MetasrvInstance's clone	2024-10-15 11:36:48 +00:00
discord9	16b8cdc3d5	chore: bump version v0.9.4 (#4833 )	2024-10-15 10:48:03 +00:00
Weny Xu	3197b8b535	feat: introduce default customizers (#4831 ) * feat: introduce `DefaultHeartbeatHandlerGroupBuilderCustomizer` and `DefaultLeadershipChangeNotifierCustomizer` * chore: code styling	2024-10-15 09:48:13 +00:00
zyy17	972c2441af	chore: bump promql-parser to v0.4.1 and use `to_string()` for EvalStmt (#4832 ) chore: bump promql-parser to v0.4.1 and use to_string() for EvalStmt	2024-10-15 08:50:37 +00:00
Ning Sun	bb8b54b5d3	feat: add some s2 geo functions (#4823 ) * feat: add first batch of s2 functions * refactor: update reusable code from main * test: add sqlness tests for s2 * feat: add tostring function for s2 * Update src/common/function/src/scalars/geo/s2.rs Co-authored-by: Lei, HUANG <6406592+v0y4g3r@users.noreply.github.com> * Apply suggestions from code review * one more change Signed-off-by: Ruihang Xia <waynestxia@gmail.com> --------- Signed-off-by: Ruihang Xia <waynestxia@gmail.com> Co-authored-by: Ruihang Xia <waynestxia@gmail.com> Co-authored-by: Lei, HUANG <6406592+v0y4g3r@users.noreply.github.com>	2024-10-15 06:47:29 +00:00
Weny Xu	b5233e500b	feat: defer `HeartbeatHandlerGroup` construction and enhance `LeadershipChangeNotifier` (#4826 ) * feat: enhance `HeartbeatHandlerGroup` * chore: apply suggestions from CR * chore: minor refactoring * chore: code styling * chore: apply suggestions from CR	2024-10-15 03:35:31 +00:00
Ruihang Xia	b61a388d04	refactor: replace info logs with debug logs in region server (#4829 ) * refactor: replace info logs with debug logs in region server Signed-off-by: Ruihang Xia <waynestxia@gmail.com> * fix: update error handling for closing and opening nonexistent regions Signed-off-by: Ruihang Xia <waynestxia@gmail.com> --------- Signed-off-by: Ruihang Xia <waynestxia@gmail.com>	2024-10-14 12:46:07 +00:00
Ruihang Xia	06e565d25a	feat: cache logical region's metadata (#4827 ) * feat: cache logical region's metadata Signed-off-by: Ruihang Xia <waynestxia@gmail.com> * feat: implement logical region locking for metadata operations Signed-off-by: Ruihang Xia <waynestxia@gmail.com> * fix: correct typo in comment for MetadataRegion struct Signed-off-by: Ruihang Xia <waynestxia@gmail.com> --------- Signed-off-by: Ruihang Xia <waynestxia@gmail.com>	2024-10-14 08:44:13 +00:00
Yingwen	3b2ce31a19	feat: enable prof features by default (#4815 ) * feat: enable prof by default * docs: don't need to build with features * feat: add common-pprof as optional dep for pprof feature * build: remove optional * feat: use dump_text	2024-10-14 03:32:47 +00:00
Ruihang Xia	a889ea88ca	fix: case sensitive for __field__ matcher (#4822 ) Signed-off-by: Ruihang Xia <waynestxia@gmail.com>	2024-10-14 03:18:59 +00:00
Yingwen	2f2b4b306c	feat!: implement interval type by multiple structs (#4772 ) * define structs and methods Signed-off-by: Ruihang Xia <waynestxia@gmail.com> * feat: re-implement interval types in time crate * feat: use new * feat: interval value * feat: query crate interval * feat: pg and mysql interval * chore: remove unused imports * chore: remove commented codes * feat: make flow compile but may not work * feat: flow datetime * test: fix some tests * test: fix some flow tests(WIP) * chore: some fix test&docs * fix: change interval order * chore: remove unused codes * chore: fix cilppy * chore: now signature change * chore: remove todo * feat: update error message --------- Signed-off-by: Ruihang Xia <waynestxia@gmail.com> Co-authored-by: Ruihang Xia <waynestxia@gmail.com> Co-authored-by: discord9 <discord9@163.com>	2024-10-14 03:09:03 +00:00
jeremyhi	856c0280f5	feat: remove the distributed lock (#4825 ) * feat: remove the distributed lock as we do not need it any more * chore: delete todo comment * chore: remove unused error	2024-10-12 09:04:22 +00:00