add rate, increase and delta

Signed-off-by: Ruihang Xia <waynestxia@gmail.com>
disable recursion limit
2025-12-25 15:40:02 +00:00 · 2025-04-29 10:54:32 +08:00 · 2025-04-28 16:58:30 +08:00 · 2025-04-27 19:23:42 +08:00 · 2025-04-27 17:40:24 +08:00 · 2025-04-27 17:35:50 +08:00
168 changed files with 2576 additions and 4185 deletions
--- a/.github/scripts/create-version.sh
+++ b/.github/scripts/create-version.sh
@@ -10,17 +10,17 @@ set -e
 function create_version() {
  # Read from envrionment variables.
  if [ -z "$GITHUB_EVENT_NAME" ]; then
-      echo "GITHUB_EVENT_NAME is empty"
+      echo "GITHUB_EVENT_NAME is empty" >&2
      exit 1
  fi

  if [ -z "$NEXT_RELEASE_VERSION" ]; then
-      echo "NEXT_RELEASE_VERSION is empty"
-      exit 1
+      echo "NEXT_RELEASE_VERSION is empty, use version from Cargo.toml" >&2
+      export NEXT_RELEASE_VERSION=$(grep '^version = ' Cargo.toml | cut -d '"' -f 2 | head -n 1)
  fi

  if [ -z "$NIGHTLY_RELEASE_PREFIX" ]; then
-      echo "NIGHTLY_RELEASE_PREFIX is empty"
+      echo "NIGHTLY_RELEASE_PREFIX is empty" >&2
      exit 1
  fi

@@ -35,7 +35,7 @@ function create_version() {
  # It will be like 'dev-2023080819-f0e7216c'.
  if [ "$NEXT_RELEASE_VERSION" = dev ]; then
    if [ -z "$COMMIT_SHA" ]; then
-      echo "COMMIT_SHA is empty in dev build"
+      echo "COMMIT_SHA is empty in dev build" >&2
      exit 1
    fi
    echo "dev-$(date "+%Y%m%d-%s")-$(echo "$COMMIT_SHA" | cut -c1-8)"
@@ -45,7 +45,7 @@ function create_version() {
  # Note: Only output 'version=xxx' to stdout when everything is ok, so that it can be used in GitHub Actions Outputs.
  if [ "$GITHUB_EVENT_NAME" = push ]; then
    if [ -z "$GITHUB_REF_NAME" ]; then
-      echo "GITHUB_REF_NAME is empty in push event"
+      echo "GITHUB_REF_NAME is empty in push event" >&2
      exit 1
    fi
    echo "$GITHUB_REF_NAME"
@@ -54,7 +54,7 @@ function create_version() {
  elif [ "$GITHUB_EVENT_NAME" = schedule ]; then
    echo "$NEXT_RELEASE_VERSION-$NIGHTLY_RELEASE_PREFIX-$(date "+%Y%m%d")"
  else
-    echo "Unsupported GITHUB_EVENT_NAME: $GITHUB_EVENT_NAME"
+    echo "Unsupported GITHUB_EVENT_NAME: $GITHUB_EVENT_NAME" >&2
    exit 1
  fi
 }
--- a/.github/scripts/update-dev-builder-version.sh
+++ b/.github/scripts/update-dev-builder-version.sh
@@ -1,37 +0,0 @@
-#!/bin/bash
-
-DEV_BUILDER_IMAGE_TAG=$1
-
-update_dev_builder_version() {
-  if [ -z "$DEV_BUILDER_IMAGE_TAG" ]; then
-    echo "Error: Should specify the dev-builder image tag" 
-    exit 1
-  fi
-
-  # Configure Git configs.
-  git config --global user.email greptimedb-ci@greptime.com
-  git config --global user.name greptimedb-ci
-
-  # Checkout a new branch.
-  BRANCH_NAME="ci/update-dev-builder-$(date +%Y%m%d%H%M%S)"
-  git checkout -b $BRANCH_NAME
-
-  # Update the dev-builder image tag in the Makefile.
-  gsed -i "s/DEV_BUILDER_IMAGE_TAG ?=.*/DEV_BUILDER_IMAGE_TAG ?= ${DEV_BUILDER_IMAGE_TAG}/g" Makefile
-
-  # Commit the changes.
-  git add Makefile
-  git commit -m "ci: update dev-builder image tag"
-  git push origin $BRANCH_NAME
-
-  # Create a Pull Request.
-  gh pr create \
-      --title "ci: update dev-builder image tag" \
-      --body "This PR updates the dev-builder image tag" \
-      --base main \
-      --head $BRANCH_NAME \
-      --reviewer zyy17 \
-      --reviewer daviderli614
-}
-
-update_dev_builder_version
--- a/.github/workflows/release-dev-builder-images.yaml
+++ b/.github/workflows/release-dev-builder-images.yaml
@@ -24,19 +24,11 @@ on:
        description: Release dev-builder-android image
        required: false
        default: false
-      update_dev_builder_image_tag:
-        type: boolean
-        description: Update the DEV_BUILDER_IMAGE_TAG in Makefile and create a PR
-        required: false
-        default: false

 jobs:
  release-dev-builder-images:
    name: Release dev builder images
-    # The jobs are triggered by the following events:
-    # 1. Manually triggered workflow_dispatch event
-    # 2. Push event when the PR that modifies the `rust-toolchain.toml` or `docker/dev-builder/**` is merged to main
-    if: ${{ github.event_name == 'push' || inputs.release_dev_builder_ubuntu_image || inputs.release_dev_builder_centos_image || inputs.release_dev_builder_android_image }}
+    if: ${{ inputs.release_dev_builder_ubuntu_image || inputs.release_dev_builder_centos_image || inputs.release_dev_builder_android_image }} # Only manually trigger this job.
    runs-on: ubuntu-latest
    outputs:
      version: ${{ steps.set-version.outputs.version }}
@@ -65,9 +57,9 @@ jobs:
          version: ${{ env.VERSION }}
          dockerhub-image-registry-username: ${{ secrets.DOCKERHUB_USERNAME }}
          dockerhub-image-registry-token: ${{ secrets.DOCKERHUB_TOKEN }}
-          build-dev-builder-ubuntu: ${{ inputs.release_dev_builder_ubuntu_image || github.event_name == 'push' }}
-          build-dev-builder-centos: ${{ inputs.release_dev_builder_centos_image || github.event_name == 'push' }}
-          build-dev-builder-android: ${{ inputs.release_dev_builder_android_image || github.event_name == 'push' }}
+          build-dev-builder-ubuntu: ${{ inputs.release_dev_builder_ubuntu_image }}
+          build-dev-builder-centos: ${{ inputs.release_dev_builder_centos_image }}
+          build-dev-builder-android: ${{ inputs.release_dev_builder_android_image }}

  release-dev-builder-images-ecr:
    name: Release dev builder images to AWS ECR
@@ -93,7 +85,7 @@ jobs:

      - name: Push dev-builder-ubuntu image
        shell: bash
-        if: ${{ inputs.release_dev_builder_ubuntu_image || github.event_name == 'push' }}
+        if: ${{ inputs.release_dev_builder_ubuntu_image }}
        env:
          IMAGE_VERSION: ${{ needs.release-dev-builder-images.outputs.version }}
          IMAGE_NAMESPACE: ${{ vars.IMAGE_NAMESPACE }}
@@ -114,7 +106,7 @@ jobs:

      - name: Push dev-builder-centos image
        shell: bash
-        if: ${{ inputs.release_dev_builder_centos_image || github.event_name == 'push' }}
+        if: ${{ inputs.release_dev_builder_centos_image }}
        env:
          IMAGE_VERSION: ${{ needs.release-dev-builder-images.outputs.version }}
          IMAGE_NAMESPACE: ${{ vars.IMAGE_NAMESPACE }}
@@ -135,7 +127,7 @@ jobs:

      - name: Push dev-builder-android image
        shell: bash
-        if: ${{ inputs.release_dev_builder_android_image || github.event_name == 'push' }}
+        if: ${{ inputs.release_dev_builder_android_image }}
        env:
          IMAGE_VERSION: ${{ needs.release-dev-builder-images.outputs.version }}
          IMAGE_NAMESPACE: ${{ vars.IMAGE_NAMESPACE }}
@@ -170,7 +162,7 @@ jobs:

      - name: Push dev-builder-ubuntu image
        shell: bash
-        if: ${{ inputs.release_dev_builder_ubuntu_image || github.event_name == 'push' }}
+        if: ${{ inputs.release_dev_builder_ubuntu_image }}
        env:
          IMAGE_VERSION: ${{ needs.release-dev-builder-images.outputs.version }}
          IMAGE_NAMESPACE: ${{ vars.IMAGE_NAMESPACE }}
@@ -184,7 +176,7 @@ jobs:

      - name: Push dev-builder-centos image
        shell: bash
-        if: ${{ inputs.release_dev_builder_centos_image || github.event_name == 'push' }}
+        if: ${{ inputs.release_dev_builder_centos_image }}
        env:
          IMAGE_VERSION: ${{ needs.release-dev-builder-images.outputs.version }}
          IMAGE_NAMESPACE: ${{ vars.IMAGE_NAMESPACE }}
@@ -198,7 +190,7 @@ jobs:

      - name: Push dev-builder-android image
        shell: bash
-        if: ${{ inputs.release_dev_builder_android_image || github.event_name == 'push' }}
+        if: ${{ inputs.release_dev_builder_android_image }}
        env:
          IMAGE_VERSION: ${{ needs.release-dev-builder-images.outputs.version }}
          IMAGE_NAMESPACE: ${{ vars.IMAGE_NAMESPACE }}
@@ -209,24 +201,3 @@ jobs:
            quay.io/skopeo/stable:latest \
            copy -a docker://docker.io/$IMAGE_NAMESPACE/dev-builder-android:$IMAGE_VERSION \
            docker://$ACR_IMAGE_REGISTRY/$IMAGE_NAMESPACE/dev-builder-android:$IMAGE_VERSION
-  
-  update-dev-builder-image-tag:
-    name: Update dev-builder image tag
-    runs-on: ubuntu-latest
-    permissions:
-      contents: write
-      pull-requests: write
-    if: ${{ github.event_name == 'push' || inputs.update_dev_builder_image_tag }}
-    needs: [
-      release-dev-builder-images
-    ]
-    steps:
-      - name: Checkout repository
-        uses: actions/checkout@v4
-
-      - name: Update dev-builder image tag
-        shell: bash
-        env:
-          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
-        run: |
-          ./.github/scripts/update-dev-builder-version.sh ${{ needs.release-dev-builder-images.outputs.version }}
--- a/.github/workflows/release.yml
+++ b/.github/workflows/release.yml
@@ -90,8 +90,6 @@ env:

  # The scheduled version is '${{ env.NEXT_RELEASE_VERSION }}-nightly-YYYYMMDD', like v0.2.0-nigthly-20230313;
  NIGHTLY_RELEASE_PREFIX: nightly
-  # Note: The NEXT_RELEASE_VERSION should be modified manually by every formal release.
-  NEXT_RELEASE_VERSION: v0.14.0

 jobs:
  allocate-runners:
@@ -135,7 +133,6 @@ jobs:
        env:
          GITHUB_EVENT_NAME: ${{ github.event_name }}
          GITHUB_REF_NAME: ${{ github.ref_name }}
-          NEXT_RELEASE_VERSION: ${{ env.NEXT_RELEASE_VERSION }}
          NIGHTLY_RELEASE_PREFIX: ${{ env.NIGHTLY_RELEASE_PREFIX }}

      - name: Allocate linux-amd64 runner
--- a/Cargo.lock
+++ b/Cargo.lock
@@ -173,9 +173,9 @@ dependencies = [

 [[package]]
 name = "anyhow"
-version = "1.0.89"
+version = "1.0.98"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "86fdf8605db99b54d3cd748a44c6d04df638eb5dafb219b135d0149bd0db01f6"
+checksum = "e16d2d3311acee920a9eb8d33b8cbc1787ce4a264e85f964c2404b969bdcd487"

 [[package]]
 name = "anymap2"
@@ -185,7 +185,7 @@ checksum = "d301b3b94cb4b2f23d7917810addbbaff90738e0ca2be692bd027e70d7e0330c"

 [[package]]
 name = "api"
-version = "0.14.3"
+version = "0.15.0"
 dependencies = [
 "common-base",
 "common-decimal",
@@ -915,7 +915,7 @@ dependencies = [

 [[package]]
 name = "auth"
-version = "0.14.3"
+version = "0.15.0"
 dependencies = [
 "api",
 "async-trait",
@@ -1537,7 +1537,7 @@ dependencies = [

 [[package]]
 name = "cache"
-version = "0.14.3"
+version = "0.15.0"
 dependencies = [
 "catalog",
 "common-error",
@@ -1561,7 +1561,7 @@ checksum = "37b2a672a2cb129a2e41c10b1224bb368f9f37a2b16b612598138befd7b37eb5"

 [[package]]
 name = "catalog"
-version = "0.14.3"
+version = "0.15.0"
 dependencies = [
 "api",
 "arrow 54.2.1",
@@ -1597,7 +1597,7 @@ dependencies = [
 "partition",
 "paste",
 "prometheus",
- "rustc-hash 2.0.0",
+ "rustc-hash 2.1.1",
 "serde_json",
 "session",
 "snafu 0.8.5",
@@ -1874,7 +1874,7 @@ checksum = "1462739cb27611015575c0c11df5df7601141071f07518d56fcc1be504cbec97"

 [[package]]
 name = "cli"
-version = "0.14.3"
+version = "0.15.0"
 dependencies = [
 "async-trait",
 "auth",
@@ -1917,7 +1917,7 @@ dependencies = [
 "session",
 "snafu 0.8.5",
 "store-api",
- "substrait 0.14.3",
+ "substrait 0.15.0",
 "table",
 "tempfile",
 "tokio",
@@ -1926,7 +1926,7 @@ dependencies = [

 [[package]]
 name = "client"
-version = "0.14.3"
+version = "0.15.0"
 dependencies = [
 "api",
 "arc-swap",
@@ -1955,7 +1955,7 @@ dependencies = [
 "rand 0.9.0",
 "serde_json",
 "snafu 0.8.5",
- "substrait 0.14.3",
+ "substrait 0.15.0",
 "substrait 0.37.3",
 "tokio",
 "tokio-stream",
@@ -1996,7 +1996,7 @@ dependencies = [

 [[package]]
 name = "cmd"
-version = "0.14.3"
+version = "0.15.0"
 dependencies = [
 "async-trait",
 "auth",
@@ -2056,7 +2056,7 @@ dependencies = [
 "similar-asserts",
 "snafu 0.8.5",
 "store-api",
- "substrait 0.14.3",
+ "substrait 0.15.0",
 "table",
 "temp-env",
 "tempfile",
@@ -2102,7 +2102,7 @@ checksum = "55b672471b4e9f9e95499ea597ff64941a309b2cdbffcc46f2cc5e2d971fd335"

 [[package]]
 name = "common-base"
-version = "0.14.3"
+version = "0.15.0"
 dependencies = [
 "anymap2",
 "async-trait",
@@ -2124,11 +2124,11 @@ dependencies = [

 [[package]]
 name = "common-catalog"
-version = "0.14.3"
+version = "0.15.0"

 [[package]]
 name = "common-config"
-version = "0.14.3"
+version = "0.15.0"
 dependencies = [
 "common-base",
 "common-error",
@@ -2153,7 +2153,7 @@ dependencies = [

 [[package]]
 name = "common-datasource"
-version = "0.14.3"
+version = "0.15.0"
 dependencies = [
 "arrow 54.2.1",
 "arrow-schema 54.3.1",
@@ -2190,7 +2190,7 @@ dependencies = [

 [[package]]
 name = "common-decimal"
-version = "0.14.3"
+version = "0.15.0"
 dependencies = [
 "bigdecimal 0.4.8",
 "common-error",
@@ -2203,7 +2203,7 @@ dependencies = [

 [[package]]
 name = "common-error"
-version = "0.14.3"
+version = "0.15.0"
 dependencies = [
 "common-macro",
 "http 1.1.0",
@@ -2214,7 +2214,7 @@ dependencies = [

 [[package]]
 name = "common-frontend"
-version = "0.14.3"
+version = "0.15.0"
 dependencies = [
 "async-trait",
 "common-error",
@@ -2224,7 +2224,7 @@ dependencies = [

 [[package]]
 name = "common-function"
-version = "0.14.3"
+version = "0.15.0"
 dependencies = [
 "ahash 0.8.11",
 "api",
@@ -2277,7 +2277,7 @@ dependencies = [

 [[package]]
 name = "common-greptimedb-telemetry"
-version = "0.14.3"
+version = "0.15.0"
 dependencies = [
 "async-trait",
 "common-runtime",
@@ -2294,7 +2294,7 @@ dependencies = [

 [[package]]
 name = "common-grpc"
-version = "0.14.3"
+version = "0.15.0"
 dependencies = [
 "api",
 "arrow-flight",
@@ -2325,7 +2325,7 @@ dependencies = [

 [[package]]
 name = "common-grpc-expr"
-version = "0.14.3"
+version = "0.15.0"
 dependencies = [
 "api",
 "common-base",
@@ -2344,7 +2344,7 @@ dependencies = [

 [[package]]
 name = "common-macro"
-version = "0.14.3"
+version = "0.15.0"
 dependencies = [
 "arc-swap",
 "common-query",
@@ -2358,7 +2358,7 @@ dependencies = [

 [[package]]
 name = "common-mem-prof"
-version = "0.14.3"
+version = "0.15.0"
 dependencies = [
 "common-error",
 "common-macro",
@@ -2371,7 +2371,7 @@ dependencies = [

 [[package]]
 name = "common-meta"
-version = "0.14.3"
+version = "0.15.0"
 dependencies = [
 "anymap2",
 "api",
@@ -2432,7 +2432,7 @@ dependencies = [

 [[package]]
 name = "common-options"
-version = "0.14.3"
+version = "0.15.0"
 dependencies = [
 "common-grpc",
 "humantime-serde",
@@ -2441,11 +2441,11 @@ dependencies = [

 [[package]]
 name = "common-plugins"
-version = "0.14.3"
+version = "0.15.0"

 [[package]]
 name = "common-pprof"
-version = "0.14.3"
+version = "0.15.0"
 dependencies = [
 "common-error",
 "common-macro",
@@ -2457,7 +2457,7 @@ dependencies = [

 [[package]]
 name = "common-procedure"
-version = "0.14.3"
+version = "0.15.0"
 dependencies = [
 "async-stream",
 "async-trait",
@@ -2484,7 +2484,7 @@ dependencies = [

 [[package]]
 name = "common-procedure-test"
-version = "0.14.3"
+version = "0.15.0"
 dependencies = [
 "async-trait",
 "common-procedure",
@@ -2493,7 +2493,7 @@ dependencies = [

 [[package]]
 name = "common-query"
-version = "0.14.3"
+version = "0.15.0"
 dependencies = [
 "api",
 "async-trait",
@@ -2519,7 +2519,7 @@ dependencies = [

 [[package]]
 name = "common-recordbatch"
-version = "0.14.3"
+version = "0.15.0"
 dependencies = [
 "arc-swap",
 "common-error",
@@ -2539,7 +2539,7 @@ dependencies = [

 [[package]]
 name = "common-runtime"
-version = "0.14.3"
+version = "0.15.0"
 dependencies = [
 "async-trait",
 "clap 4.5.19",
@@ -2569,14 +2569,14 @@ dependencies = [

 [[package]]
 name = "common-session"
-version = "0.14.3"
+version = "0.15.0"
 dependencies = [
 "strum 0.27.1",
 ]

 [[package]]
 name = "common-telemetry"
-version = "0.14.3"
+version = "0.15.0"
 dependencies = [
 "atty",
 "backtrace",
@@ -2604,7 +2604,7 @@ dependencies = [

 [[package]]
 name = "common-test-util"
-version = "0.14.3"
+version = "0.15.0"
 dependencies = [
 "client",
 "common-query",
@@ -2616,7 +2616,7 @@ dependencies = [

 [[package]]
 name = "common-time"
-version = "0.14.3"
+version = "0.15.0"
 dependencies = [
 "arrow 54.2.1",
 "chrono",
@@ -2634,7 +2634,7 @@ dependencies = [

 [[package]]
 name = "common-version"
-version = "0.14.3"
+version = "0.15.0"
 dependencies = [
 "build-data",
 "const_format",
@@ -2644,7 +2644,7 @@ dependencies = [

 [[package]]
 name = "common-wal"
-version = "0.14.3"
+version = "0.15.0"
 dependencies = [
 "common-base",
 "common-error",
@@ -3110,9 +3110,9 @@ dependencies = [

 [[package]]
 name = "data-encoding"
-version = "2.6.0"
+version = "2.9.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "e8566979429cf69b49a5c740c60791108e86440e8be149bbea4fe54d2c32d6e2"
+checksum = "2a2330da5de22e8a3cb63252ce2abb30116bf5265e89c0e01bc17015ce30a476"

 [[package]]
 name = "datafusion"
@@ -3572,7 +3572,7 @@ dependencies = [

 [[package]]
 name = "datanode"
-version = "0.14.3"
+version = "0.15.0"
 dependencies = [
 "api",
 "arrow-flight",
@@ -3624,7 +3624,7 @@ dependencies = [
 "session",
 "snafu 0.8.5",
 "store-api",
- "substrait 0.14.3",
+ "substrait 0.15.0",
 "table",
 "tokio",
 "toml 0.8.19",
@@ -3633,7 +3633,7 @@ dependencies = [

 [[package]]
 name = "datatypes"
-version = "0.14.3"
+version = "0.15.0"
 dependencies = [
 "arrow 54.2.1",
 "arrow-array 54.2.1",
@@ -4259,7 +4259,7 @@ dependencies = [

 [[package]]
 name = "file-engine"
-version = "0.14.3"
+version = "0.15.0"
 dependencies = [
 "api",
 "async-trait",
@@ -4382,7 +4382,7 @@ checksum = "8bf7cc16383c4b8d58b9905a8509f02926ce3058053c056376248d958c9df1e8"

 [[package]]
 name = "flow"
-version = "0.14.3"
+version = "0.15.0"
 dependencies = [
 "api",
 "arrow 54.2.1",
@@ -4436,7 +4436,6 @@ dependencies = [
 "prometheus",
 "prost 0.13.5",
 "query",
- "rand 0.9.0",
 "serde",
 "serde_json",
 "servers",
@@ -4445,7 +4444,7 @@ dependencies = [
 "snafu 0.8.5",
 "store-api",
 "strum 0.27.1",
- "substrait 0.14.3",
+ "substrait 0.15.0",
 "table",
 "tokio",
 "tonic 0.12.3",
@@ -4500,7 +4499,7 @@ checksum = "6c2141d6d6c8512188a7891b4b01590a45f6dac67afb4f255c4124dbb86d4eaa"

 [[package]]
 name = "frontend"
-version = "0.14.3"
+version = "0.15.0"
 dependencies = [
 "api",
 "arc-swap",
@@ -4557,7 +4556,7 @@ dependencies = [
 "sqlparser 0.54.0 (git+https://github.com/GreptimeTeam/sqlparser-rs.git?rev=0cf6c04490d59435ee965edd2078e8855bd8471e)",
 "store-api",
 "strfmt",
- "substrait 0.14.3",
+ "substrait 0.15.0",
 "table",
 "tokio",
 "toml 0.8.19",
@@ -4945,7 +4944,7 @@ dependencies = [
 [[package]]
 name = "greptime-proto"
 version = "0.1.0"
-source = "git+https://github.com/GreptimeTeam/greptime-proto.git?rev=4d4136692fe7fbbd509ebc8c902f6afcc0ce61e4#4d4136692fe7fbbd509ebc8c902f6afcc0ce61e4"
+source = "git+https://github.com/GreptimeTeam/greptime-proto.git?rev=e82b0158cd38d4021edb4e4c0ae77f999051e62f#e82b0158cd38d4021edb4e4c0ae77f999051e62f"
 dependencies = [
 "prost 0.13.5",
 "serde",
@@ -5796,7 +5795,7 @@ dependencies = [

 [[package]]
 name = "index"
-version = "0.14.3"
+version = "0.15.0"
 dependencies = [
 "async-trait",
 "asynchronous-codec",
@@ -6510,7 +6509,7 @@ source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "4979f22fdb869068da03c9f7528f8297c6fd2606bc3a4affe42e6a823fdb8da4"
 dependencies = [
 "cfg-if",
- "windows-targets 0.48.5",
+ "windows-targets 0.52.6",
 ]

 [[package]]
@@ -6600,13 +6599,13 @@ dependencies = [

 [[package]]
 name = "log"
-version = "0.4.22"
+version = "0.4.27"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "a7a70ba024b9dc04c27ea2f0c0548feb474ec5c54bba33a7f72f873a39d07b24"
+checksum = "13dc2df351e3202783a1fe0d44375f7295ffb4049267b0f3018346dc122a1d94"

 [[package]]
 name = "log-query"
-version = "0.14.3"
+version = "0.15.0"
 dependencies = [
 "chrono",
 "common-error",
@@ -6618,7 +6617,7 @@ dependencies = [

 [[package]]
 name = "log-store"
-version = "0.14.3"
+version = "0.15.0"
 dependencies = [
 "async-stream",
 "async-trait",
@@ -6912,7 +6911,7 @@ dependencies = [

 [[package]]
 name = "meta-client"
-version = "0.14.3"
+version = "0.15.0"
 dependencies = [
 "api",
 "async-trait",
@@ -6940,7 +6939,7 @@ dependencies = [

 [[package]]
 name = "meta-srv"
-version = "0.14.3"
+version = "0.15.0"
 dependencies = [
 "api",
 "async-trait",
@@ -7030,7 +7029,7 @@ dependencies = [

 [[package]]
 name = "metric-engine"
-version = "0.14.3"
+version = "0.15.0"
 dependencies = [
 "api",
 "aquamarine",
@@ -7119,7 +7118,7 @@ dependencies = [

 [[package]]
 name = "mito2"
-version = "0.14.3"
+version = "0.15.0"
 dependencies = [
 "api",
 "aquamarine",
@@ -7781,7 +7780,7 @@ version = "0.7.3"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "af1844ef2428cc3e1cb900be36181049ef3d3193c63e43026cfe202983b27a56"
 dependencies = [
- "proc-macro-crate 1.3.1",
+ "proc-macro-crate 3.2.0",
 "proc-macro2",
 "quote",
 "syn 2.0.100",
@@ -7825,7 +7824,7 @@ dependencies = [

 [[package]]
 name = "object-store"
-version = "0.14.3"
+version = "0.15.0"
 dependencies = [
 "anyhow",
 "bytes",
@@ -8120,7 +8119,7 @@ dependencies = [

 [[package]]
 name = "operator"
-version = "0.14.3"
+version = "0.15.0"
 dependencies = [
 "ahash 0.8.11",
 "api",
@@ -8169,7 +8168,7 @@ dependencies = [
 "sql",
 "sqlparser 0.54.0 (git+https://github.com/GreptimeTeam/sqlparser-rs.git?rev=0cf6c04490d59435ee965edd2078e8855bd8471e)",
 "store-api",
- "substrait 0.14.3",
+ "substrait 0.15.0",
 "table",
 "tokio",
 "tokio-util",
@@ -8424,7 +8423,7 @@ dependencies = [

 [[package]]
 name = "partition"
-version = "0.14.3"
+version = "0.15.0"
 dependencies = [
 "api",
 "async-trait",
@@ -8706,7 +8705,7 @@ checksum = "8b870d8c151b6f2fb93e84a13146138f05d02ed11c7e7c54f8826aaaf7c9f184"

 [[package]]
 name = "pipeline"
-version = "0.14.3"
+version = "0.15.0"
 dependencies = [
 "ahash 0.8.11",
 "api",
@@ -8848,7 +8847,7 @@ dependencies = [

 [[package]]
 name = "plugins"
-version = "0.14.3"
+version = "0.15.0"
 dependencies = [
 "auth",
 "clap 4.5.19",
@@ -9128,7 +9127,7 @@ dependencies = [

 [[package]]
 name = "promql"
-version = "0.14.3"
+version = "0.15.0"
 dependencies = [
 "ahash 0.8.11",
 "async-trait",
@@ -9153,7 +9152,8 @@ dependencies = [
 [[package]]
 name = "promql-parser"
 version = "0.5.1"
-source = "git+https://github.com/GreptimeTeam/promql-parser.git?rev=0410e8b459dda7cb222ce9596f8bf3971bd07bd2#0410e8b459dda7cb222ce9596f8bf3971bd07bd2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "60d851f6523a8215e2fbf86b6cef4548433f8b76092e9ffb607105de52ae63fd"
 dependencies = [
 "cfgrammar",
 "chrono",
@@ -9163,7 +9163,6 @@ dependencies = [
 "regex",
 "serde",
 "serde_json",
- "unescaper",
 ]

 [[package]]
@@ -9374,7 +9373,7 @@ dependencies = [

 [[package]]
 name = "puffin"
-version = "0.14.3"
+version = "0.15.0"
 dependencies = [
 "async-compression 0.4.13",
 "async-trait",
@@ -9415,7 +9414,7 @@ dependencies = [

 [[package]]
 name = "query"
-version = "0.14.3"
+version = "0.15.0"
 dependencies = [
 "ahash 0.8.11",
 "api",
@@ -9481,7 +9480,7 @@ dependencies = [
 "sqlparser 0.54.0 (git+https://github.com/GreptimeTeam/sqlparser-rs.git?rev=0cf6c04490d59435ee965edd2078e8855bd8471e)",
 "statrs",
 "store-api",
- "substrait 0.14.3",
+ "substrait 0.15.0",
 "table",
 "tokio",
 "tokio-stream",
@@ -9528,7 +9527,7 @@ dependencies = [
 "pin-project-lite",
 "quinn-proto",
 "quinn-udp",
- "rustc-hash 2.0.0",
+ "rustc-hash 2.1.1",
 "rustls",
 "socket2",
 "thiserror 1.0.64",
@@ -9545,7 +9544,7 @@ dependencies = [
 "bytes",
 "rand 0.8.5",
 "ring",
- "rustc-hash 2.0.0",
+ "rustc-hash 2.1.1",
 "rustls",
 "slab",
 "thiserror 1.0.64",
@@ -9822,9 +9821,9 @@ dependencies = [

 [[package]]
 name = "regex"
-version = "1.11.0"
+version = "1.11.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "38200e5ee88914975b69f657f0801b6f6dccafd44fd9326302a4aaeecfacb1d8"
+checksum = "b544ef1b4eac5dc2db33ea63606ae9ffcfac26c1416a2806ae0bf5f56b201191"
 dependencies = [
 "aho-corasick",
 "memchr",
@@ -10334,9 +10333,9 @@ checksum = "08d43f7aa6b08d49f382cde6a7982047c3426db949b1424bc4b7ec9ae12c6ce2"

 [[package]]
 name = "rustc-hash"
-version = "2.0.0"
+version = "2.1.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "583034fd73374156e66797ed8e5b0d5690409c9226b22d87cb7f19821c05d152"
+checksum = "357703d41365b4b27c590e3ed91eabb1b663f07c4c084095e60cbed4362dff0d"

 [[package]]
 name = "rustc_version"
@@ -10831,7 +10830,7 @@ dependencies = [

 [[package]]
 name = "servers"
-version = "0.14.3"
+version = "0.15.0"
 dependencies = [
 "ahash 0.8.11",
 "api",
@@ -10951,7 +10950,7 @@ dependencies = [

 [[package]]
 name = "session"
-version = "0.14.3"
+version = "0.15.0"
 dependencies = [
 "api",
 "arc-swap",
@@ -11159,9 +11158,9 @@ dependencies = [

 [[package]]
 name = "smallbitvec"
-version = "2.5.3"
+version = "2.6.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "fcc3fc564a4b53fd1e8589628efafe57602d91bde78be18186b5f61e8faea470"
+checksum = "d31d263dd118560e1a492922182ab6ca6dc1d03a3bf54e7699993f31a4150e3f"

 [[package]]
 name = "smallvec"
@@ -11276,7 +11275,7 @@ dependencies = [

 [[package]]
 name = "sql"
-version = "0.14.3"
+version = "0.15.0"
 dependencies = [
 "api",
 "chrono",
@@ -11331,7 +11330,7 @@ dependencies = [

 [[package]]
 name = "sqlness-runner"
-version = "0.14.3"
+version = "0.15.0"
 dependencies = [
 "async-trait",
 "clap 4.5.19",
@@ -11650,7 +11649,7 @@ dependencies = [

 [[package]]
 name = "store-api"
-version = "0.14.3"
+version = "0.15.0"
 dependencies = [
 "api",
 "aquamarine",
@@ -11658,7 +11657,6 @@ dependencies = [
 "async-trait",
 "common-base",
 "common-error",
- "common-grpc",
 "common-macro",
 "common-meta",
 "common-recordbatch",
@@ -11800,7 +11798,7 @@ dependencies = [

 [[package]]
 name = "substrait"
-version = "0.14.3"
+version = "0.15.0"
 dependencies = [
 "async-trait",
 "bytes",
@@ -11980,7 +11978,7 @@ dependencies = [

 [[package]]
 name = "table"
-version = "0.14.3"
+version = "0.15.0"
 dependencies = [
 "api",
 "async-trait",
@@ -12231,7 +12229,7 @@ checksum = "3369f5ac52d5eb6ab48c6b4ffdc8efbcad6b89c765749064ba298f2c68a16a76"

 [[package]]
 name = "tests-fuzz"
-version = "0.14.3"
+version = "0.15.0"
 dependencies = [
 "arbitrary",
 "async-trait",
@@ -12275,7 +12273,7 @@ dependencies = [

 [[package]]
 name = "tests-integration"
-version = "0.14.3"
+version = "0.15.0"
 dependencies = [
 "api",
 "arrow-flight",
@@ -12342,7 +12340,7 @@ dependencies = [
 "sql",
 "sqlx",
 "store-api",
- "substrait 0.14.3",
+ "substrait 0.15.0",
 "table",
 "tempfile",
 "time",
--- a/Cargo.toml
+++ b/Cargo.toml
@@ -68,15 +68,16 @@ members = [
 resolver = "2"

 [workspace.package]
-version = "0.14.3"
+version = "0.15.0"
 edition = "2021"
 license = "Apache-2.0"

 [workspace.lints]
-clippy.print_stdout = "warn"
-clippy.print_stderr = "warn"
 clippy.dbg_macro = "warn"
 clippy.implicit_clone = "warn"
+clippy.result_large_err = "allow"
+clippy.large_enum_variant = "allow"
+clippy.doc_overindented_list_items = "allow"
 rust.unknown_lints = "deny"
 rust.unexpected_cfgs = { level = "warn", check-cfg = ['cfg(tokio_unstable)'] }

@@ -129,7 +130,7 @@ etcd-client = "0.14"
 fst = "0.4.7"
 futures = "0.3"
 futures-util = "0.3"
-greptime-proto = { git = "https://github.com/GreptimeTeam/greptime-proto.git", rev = "4d4136692fe7fbbd509ebc8c902f6afcc0ce61e4" }
+greptime-proto = { git = "https://github.com/GreptimeTeam/greptime-proto.git", rev = "e82b0158cd38d4021edb4e4c0ae77f999051e62f" }
 hex = "0.4"
 http = "1"
 humantime = "2.1"
@@ -161,9 +162,7 @@ parquet = { version = "54.2", default-features = false, features = ["arrow", "as
 paste = "1.0"
 pin-project = "1.0"
 prometheus = { version = "0.13.3", features = ["process"] }
-promql-parser = { git = "https://github.com/GreptimeTeam/promql-parser.git", rev = "0410e8b459dda7cb222ce9596f8bf3971bd07bd2", features = [
-    "ser",
-] }
+promql-parser = { version = "0.5.1", features = ["ser"] }
 prost = { version = "0.13", features = ["no-recursion-limit"] }
 raft-engine = { version = "0.4.1", default-features = false }
 rand = "0.9"
--- a/flake.lock
+++ b/flake.lock
@@ -8,11 +8,11 @@
        "rust-analyzer-src": "rust-analyzer-src"
      },
      "locked": {
-        "lastModified": 1737613896,
-        "narHash": "sha256-ldqXIglq74C7yKMFUzrS9xMT/EVs26vZpOD68Sh7OcU=",
+        "lastModified": 1742452566,
+        "narHash": "sha256-sVuLDQ2UIWfXUBbctzrZrXM2X05YjX08K7XHMztt36E=",
        "owner": "nix-community",
        "repo": "fenix",
-        "rev": "303a062fdd8e89f233db05868468975d17855d80",
+        "rev": "7d9ba794daf5e8cc7ee728859bc688d8e26d5f06",
        "type": "github"
      },
      "original": {
@@ -41,11 +41,11 @@
    },
    "nixpkgs": {
      "locked": {
-        "lastModified": 1737569578,
-        "narHash": "sha256-6qY0pk2QmUtBT9Mywdvif0i/CLVgpCjMUn6g9vB+f3M=",
+        "lastModified": 1743576891,
+        "narHash": "sha256-vXiKURtntURybE6FMNFAVpRPr8+e8KoLPrYs9TGuAKc=",
        "owner": "NixOS",
        "repo": "nixpkgs",
-        "rev": "47addd76727f42d351590c905d9d1905ca895b82",
+        "rev": "44a69ed688786e98a101f02b712c313f1ade37ab",
        "type": "github"
      },
      "original": {
@@ -65,11 +65,11 @@
    "rust-analyzer-src": {
      "flake": false,
      "locked": {
-        "lastModified": 1737581772,
-        "narHash": "sha256-t1P2Pe3FAX9TlJsCZbmJ3wn+C4qr6aSMypAOu8WNsN0=",
+        "lastModified": 1742296961,
+        "narHash": "sha256-gCpvEQOrugHWLimD1wTFOJHagnSEP6VYBDspq96Idu0=",
        "owner": "rust-lang",
        "repo": "rust-analyzer",
-        "rev": "582af7ee9c8d84f5d534272fc7de9f292bd849be",
+        "rev": "15d87419f1a123d8f888d608129c3ce3ff8f13d4",
        "type": "github"
      },
      "original": {
--- a/flake.nix
+++ b/flake.nix
@@ -21,7 +21,7 @@
        lib = nixpkgs.lib;
        rustToolchain = fenix.packages.${system}.fromToolchainName {
          name = (lib.importTOML ./rust-toolchain.toml).toolchain.channel;
-          sha256 = "sha256-f/CVA1EC61EWbh0SjaRNhLL0Ypx2ObupbzigZp8NmL4=";
+          sha256 = "sha256-i0Sh/ZFFsHlZ3oFZFc24qdk6Cd8Do8OPU4HJQsrKOeM=";
        };
      in
      {
--- a/rust-toolchain.toml
+++ b/rust-toolchain.toml
@@ -1,2 +1,2 @@
 [toolchain]
-channel = "nightly-2024-12-25"
+channel = "nightly-2025-04-15"
--- a/src/catalog/src/system_schema/pg_catalog/pg_namespace/oid_map.rs
+++ b/src/catalog/src/system_schema/pg_catalog/pg_namespace/oid_map.rs
@@ -84,12 +84,6 @@ mod tests {
        let key1 = "3178510";
        let key2 = "4215648";

-        // have collision
-        assert_eq!(
-            oid_map.hasher.hash_one(key1) as u32,
-            oid_map.hasher.hash_one(key2) as u32
-        );
-
        // insert them into oid_map
        let oid1 = oid_map.get_oid(key1);
        let oid2 = oid_map.get_oid(key2);
--- a/src/client/src/database.rs
+++ b/src/client/src/database.rs
@@ -36,8 +36,8 @@ use common_grpc::flight::{FlightDecoder, FlightMessage};
 use common_query::Output;
 use common_recordbatch::error::ExternalSnafu;
 use common_recordbatch::RecordBatchStreamWrapper;
+use common_telemetry::error;
 use common_telemetry::tracing_context::W3cTrace;
-use common_telemetry::{error, warn};
 use futures::future;
 use futures_util::{Stream, StreamExt, TryStreamExt};
 use prost::Message;
@@ -192,36 +192,6 @@ impl Database {
        from_grpc_response(response)
    }

-    /// Retry if connection fails, max_retries is the max number of retries, so the total wait time
-    /// is `max_retries * GRPC_CONN_TIMEOUT`
-    pub async fn handle_with_retry(&self, request: Request, max_retries: u32) -> Result<u32> {
-        let mut client = make_database_client(&self.client)?.inner;
-        let mut retries = 0;
-        let request = self.to_rpc_request(request);
-        loop {
-            let raw_response = client.handle(request.clone()).await;
-            match (raw_response, retries < max_retries) {
-                (Ok(resp), _) => return from_grpc_response(resp.into_inner()),
-                (Err(err), true) => {
-                    // determine if the error is retryable
-                    if is_grpc_retryable(&err) {
-                        // retry
-                        retries += 1;
-                        warn!("Retrying {} times with error = {:?}", retries, err);
-                        continue;
-                    }
-                }
-                (Err(err), false) => {
-                    error!(
-                        "Failed to send request to grpc handle after {} retries, error = {:?}",
-                        retries, err
-                    );
-                    return Err(err.into());
-                }
-            }
-        }
-    }
-
    #[inline]
    fn to_rpc_request(&self, request: Request) -> GreptimeRequest {
        GreptimeRequest {
@@ -398,11 +368,6 @@ impl Database {
    }
 }

-/// by grpc standard, only `Unavailable` is retryable, see: https://github.com/grpc/grpc/blob/master/doc/statuscodes.md#status-codes-and-their-use-in-grpc
-pub fn is_grpc_retryable(err: &tonic::Status) -> bool {
-    matches!(err.code(), tonic::Code::Unavailable)
-}
-
 #[derive(Default, Debug, Clone)]
 struct FlightContext {
    auth_header: Option<AuthHeader>,
--- a/src/cmd/src/metasrv.rs
+++ b/src/cmd/src/metasrv.rs
@@ -12,7 +12,6 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.

-use std::fmt;
 use std::time::Duration;

 use async_trait::async_trait;
@@ -132,7 +131,7 @@ impl SubCommand {
    }
 }

-#[derive(Default, Parser)]
+#[derive(Debug, Default, Parser)]
 pub struct StartCommand {
    /// The address to bind the gRPC server.
    #[clap(long, alias = "bind-addr")]
@@ -172,27 +171,6 @@ pub struct StartCommand {
    backend: Option<BackendImpl>,
 }

-impl fmt::Debug for StartCommand {
-    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
-        f.debug_struct("StartCommand")
-            .field("rpc_bind_addr", &self.rpc_bind_addr)
-            .field("rpc_server_addr", &self.rpc_server_addr)
-            .field("store_addrs", &self.sanitize_store_addrs())
-            .field("config_file", &self.config_file)
-            .field("selector", &self.selector)
-            .field("use_memory_store", &self.use_memory_store)
-            .field("enable_region_failover", &self.enable_region_failover)
-            .field("http_addr", &self.http_addr)
-            .field("http_timeout", &self.http_timeout)
-            .field("env_prefix", &self.env_prefix)
-            .field("data_home", &self.data_home)
-            .field("store_key_prefix", &self.store_key_prefix)
-            .field("max_txn_ops", &self.max_txn_ops)
-            .field("backend", &self.backend)
-            .finish()
-    }
-}
-
 impl StartCommand {
    pub fn load_options(&self, global_options: &GlobalOptions) -> Result<MetasrvOptions> {
        let mut opts = MetasrvOptions::load_layered_options(
@@ -206,15 +184,6 @@ impl StartCommand {
        Ok(opts)
    }

-    fn sanitize_store_addrs(&self) -> Option<Vec<String>> {
-        self.store_addrs.as_ref().map(|addrs| {
-            addrs
-                .iter()
-                .map(|addr| common_meta::kv_backend::util::sanitize_connection_string(addr))
-                .collect()
-        })
-    }
-
    // The precedence order is: cli > config file > environment variables > default values.
    fn merge_with_cli_options(
        &self,
--- a/src/common/function/src/scalars/matches_term.rs
+++ b/src/common/function/src/scalars/matches_term.rs
@@ -12,8 +12,9 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.

+use std::fmt;
+use std::iter::repeat_n;
 use std::sync::Arc;
-use std::{fmt, iter};

 use common_query::error::{InvalidFuncArgsSnafu, Result};
 use common_query::prelude::Volatility;
@@ -126,9 +127,10 @@ impl Function for MatchesTermFunction {
            let term = term_column.get_ref(0).as_string().unwrap();
            match term {
                None => {
-                    return Ok(Arc::new(BooleanVector::from_iter(
-                        iter::repeat(None).take(text_column.len()),
-                    )));
+                    return Ok(Arc::new(BooleanVector::from_iter(repeat_n(
+                        None,
+                        text_column.len(),
+                    ))));
                }
                Some(term) => Some(MatchesTermFinder::new(term)),
            }
@@ -217,7 +219,7 @@ impl MatchesTermFinder {
        }

        let mut pos = 0;
-        while let Some(found_pos) = self.finder.find(text[pos..].as_bytes()) {
+        while let Some(found_pos) = self.finder.find(&text.as_bytes()[pos..]) {
            let actual_pos = pos + found_pos;

            let prev_ok = self.starts_with_non_alnum
--- a/src/common/function/src/scalars/math/rate.rs
+++ b/src/common/function/src/scalars/math/rate.rs
@@ -37,7 +37,7 @@ impl fmt::Display for RateFunction {

 impl Function for RateFunction {
    fn name(&self) -> &str {
-        "prom_rate"
+        "rate"
    }

    fn return_type(&self, _input_types: &[ConcreteDataType]) -> Result<ConcreteDataType> {
@@ -82,7 +82,7 @@ mod tests {
    #[test]
    fn test_rate_function() {
        let rate = RateFunction;
-        assert_eq!("prom_rate", rate.name());
+        assert_eq!("rate", rate.name());
        assert_eq!(
            ConcreteDataType::float64_datatype(),
            rate.return_type(&[]).unwrap()
--- a/src/common/function/src/scalars/timestamp.rs
+++ b/src/common/function/src/scalars/timestamp.rs
@@ -13,8 +13,10 @@
 // limitations under the License.

 use std::sync::Arc;
+mod greatest;
 mod to_unixtime;

+use greatest::GreatestFunction;
 use to_unixtime::ToUnixtimeFunction;

 use crate::function_registry::FunctionRegistry;
@@ -24,5 +26,6 @@ pub(crate) struct TimestampFunction;
 impl TimestampFunction {
    pub fn register(registry: &FunctionRegistry) {
        registry.register(Arc::new(ToUnixtimeFunction));
+        registry.register(Arc::new(GreatestFunction));
    }
 }
--- a/src/common/function/src/scalars/timestamp/greatest.rs
+++ b/src/common/function/src/scalars/timestamp/greatest.rs
@@ -0,0 +1,328 @@
+// 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::fmt::{self};
+
+use common_query::error::{
+    self, ArrowComputeSnafu, InvalidFuncArgsSnafu, Result, UnsupportedInputDataTypeSnafu,
+};
+use common_query::prelude::{Signature, Volatility};
+use datafusion::arrow::compute::kernels::cmp::gt;
+use datatypes::arrow::array::AsArray;
+use datatypes::arrow::compute::cast;
+use datatypes::arrow::compute::kernels::zip;
+use datatypes::arrow::datatypes::{
+    DataType as ArrowDataType, Date32Type, TimeUnit, TimestampMicrosecondType,
+    TimestampMillisecondType, TimestampNanosecondType, TimestampSecondType,
+};
+use datatypes::prelude::ConcreteDataType;
+use datatypes::types::TimestampType;
+use datatypes::vectors::{Helper, VectorRef};
+use snafu::{ensure, ResultExt};
+
+use crate::function::{Function, FunctionContext};
+
+#[derive(Clone, Debug, Default)]
+pub struct GreatestFunction;
+
+const NAME: &str = "greatest";
+
+macro_rules! gt_time_types {
+    ($ty: ident, $columns:expr) => {{
+        let column1 = $columns[0].to_arrow_array();
+        let column2 = $columns[1].to_arrow_array();
+
+        let column1 = column1.as_primitive::<$ty>();
+        let column2 = column2.as_primitive::<$ty>();
+        let boolean_array = gt(&column1, &column2).context(ArrowComputeSnafu)?;
+
+        let result = zip::zip(&boolean_array, &column1, &column2).context(ArrowComputeSnafu)?;
+        Helper::try_into_vector(&result).context(error::FromArrowArraySnafu)
+    }};
+}
+
+impl Function for GreatestFunction {
+    fn name(&self) -> &str {
+        NAME
+    }
+
+    fn return_type(&self, input_types: &[ConcreteDataType]) -> Result<ConcreteDataType> {
+        ensure!(
+            input_types.len() == 2,
+            InvalidFuncArgsSnafu {
+                err_msg: format!(
+                    "The length of the args is not correct, expect exactly two, have: {}",
+                    input_types.len()
+                )
+            }
+        );
+
+        match &input_types[0] {
+            ConcreteDataType::String(_) => Ok(ConcreteDataType::timestamp_millisecond_datatype()),
+            ConcreteDataType::Date(_) => Ok(ConcreteDataType::date_datatype()),
+            ConcreteDataType::Timestamp(ts_type) => Ok(ConcreteDataType::Timestamp(*ts_type)),
+            _ => UnsupportedInputDataTypeSnafu {
+                function: NAME,
+                datatypes: input_types,
+            }
+            .fail(),
+        }
+    }
+
+    fn signature(&self) -> Signature {
+        Signature::uniform(
+            2,
+            vec![
+                ConcreteDataType::string_datatype(),
+                ConcreteDataType::date_datatype(),
+                ConcreteDataType::timestamp_nanosecond_datatype(),
+                ConcreteDataType::timestamp_microsecond_datatype(),
+                ConcreteDataType::timestamp_millisecond_datatype(),
+                ConcreteDataType::timestamp_second_datatype(),
+            ],
+            Volatility::Immutable,
+        )
+    }
+
+    fn eval(&self, _func_ctx: &FunctionContext, columns: &[VectorRef]) -> Result<VectorRef> {
+        ensure!(
+            columns.len() == 2,
+            InvalidFuncArgsSnafu {
+                err_msg: format!(
+                    "The length of the args is not correct, expect exactly two, have: {}",
+                    columns.len()
+                ),
+            }
+        );
+        match columns[0].data_type() {
+            ConcreteDataType::String(_) => {
+                let column1 = cast(
+                    &columns[0].to_arrow_array(),
+                    &ArrowDataType::Timestamp(TimeUnit::Millisecond, None),
+                )
+                .context(ArrowComputeSnafu)?;
+                let column1 = column1.as_primitive::<TimestampMillisecondType>();
+                let column2 = cast(
+                    &columns[1].to_arrow_array(),
+                    &ArrowDataType::Timestamp(TimeUnit::Millisecond, None),
+                )
+                .context(ArrowComputeSnafu)?;
+                let column2 = column2.as_primitive::<TimestampMillisecondType>();
+                let boolean_array = gt(&column1, &column2).context(ArrowComputeSnafu)?;
+                let result =
+                    zip::zip(&boolean_array, &column1, &column2).context(ArrowComputeSnafu)?;
+                Ok(Helper::try_into_vector(&result).context(error::FromArrowArraySnafu)?)
+            }
+            ConcreteDataType::Date(_) => gt_time_types!(Date32Type, columns),
+            ConcreteDataType::Timestamp(ts_type) => match ts_type {
+                TimestampType::Second(_) => gt_time_types!(TimestampSecondType, columns),
+                TimestampType::Millisecond(_) => {
+                    gt_time_types!(TimestampMillisecondType, columns)
+                }
+                TimestampType::Microsecond(_) => {
+                    gt_time_types!(TimestampMicrosecondType, columns)
+                }
+                TimestampType::Nanosecond(_) => {
+                    gt_time_types!(TimestampNanosecondType, columns)
+                }
+            },
+            _ => UnsupportedInputDataTypeSnafu {
+                function: NAME,
+                datatypes: columns.iter().map(|c| c.data_type()).collect::<Vec<_>>(),
+            }
+            .fail(),
+        }
+    }
+}
+
+impl fmt::Display for GreatestFunction {
+    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
+        write!(f, "GREATEST")
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use std::sync::Arc;
+
+    use common_time::timestamp::TimeUnit;
+    use common_time::{Date, Timestamp};
+    use datatypes::types::{
+        DateType, TimestampMicrosecondType, TimestampMillisecondType, TimestampNanosecondType,
+        TimestampSecondType,
+    };
+    use datatypes::value::Value;
+    use datatypes::vectors::{
+        DateVector, StringVector, TimestampMicrosecondVector, TimestampMillisecondVector,
+        TimestampNanosecondVector, TimestampSecondVector, Vector,
+    };
+    use paste::paste;
+
+    use super::*;
+    #[test]
+    fn test_greatest_takes_string_vector() {
+        let function = GreatestFunction;
+        assert_eq!(
+            function
+                .return_type(&[
+                    ConcreteDataType::string_datatype(),
+                    ConcreteDataType::string_datatype()
+                ])
+                .unwrap(),
+            ConcreteDataType::timestamp_millisecond_datatype()
+        );
+        let columns = vec![
+            Arc::new(StringVector::from(vec![
+                "1970-01-01".to_string(),
+                "2012-12-23".to_string(),
+            ])) as _,
+            Arc::new(StringVector::from(vec![
+                "2001-02-01".to_string(),
+                "1999-01-01".to_string(),
+            ])) as _,
+        ];
+
+        let result = function
+            .eval(&FunctionContext::default(), &columns)
+            .unwrap();
+        let result = result
+            .as_any()
+            .downcast_ref::<TimestampMillisecondVector>()
+            .unwrap();
+        assert_eq!(result.len(), 2);
+        assert_eq!(
+            result.get(0),
+            Value::Timestamp(Timestamp::from_str("2001-02-01 00:00:00", None).unwrap())
+        );
+        assert_eq!(
+            result.get(1),
+            Value::Timestamp(Timestamp::from_str("2012-12-23 00:00:00", None).unwrap())
+        );
+    }
+
+    #[test]
+    fn test_greatest_takes_date_vector() {
+        let function = GreatestFunction;
+        assert_eq!(
+            function
+                .return_type(&[
+                    ConcreteDataType::date_datatype(),
+                    ConcreteDataType::date_datatype()
+                ])
+                .unwrap(),
+            ConcreteDataType::Date(DateType)
+        );
+
+        let columns = vec![
+            Arc::new(DateVector::from_slice(vec![-1, 2])) as _,
+            Arc::new(DateVector::from_slice(vec![0, 1])) as _,
+        ];
+
+        let result = function
+            .eval(&FunctionContext::default(), &columns)
+            .unwrap();
+        let result = result.as_any().downcast_ref::<DateVector>().unwrap();
+        assert_eq!(result.len(), 2);
+        assert_eq!(
+            result.get(0),
+            Value::Date(Date::from_str_utc("1970-01-01").unwrap())
+        );
+        assert_eq!(
+            result.get(1),
+            Value::Date(Date::from_str_utc("1970-01-03").unwrap())
+        );
+    }
+
+    #[test]
+    fn test_greatest_takes_datetime_vector() {
+        let function = GreatestFunction;
+        assert_eq!(
+            function
+                .return_type(&[
+                    ConcreteDataType::timestamp_millisecond_datatype(),
+                    ConcreteDataType::timestamp_millisecond_datatype()
+                ])
+                .unwrap(),
+            ConcreteDataType::timestamp_millisecond_datatype()
+        );
+
+        let columns = vec![
+            Arc::new(TimestampMillisecondVector::from_slice(vec![-1, 2])) as _,
+            Arc::new(TimestampMillisecondVector::from_slice(vec![0, 1])) as _,
+        ];
+
+        let result = function
+            .eval(&FunctionContext::default(), &columns)
+            .unwrap();
+        let result = result
+            .as_any()
+            .downcast_ref::<TimestampMillisecondVector>()
+            .unwrap();
+        assert_eq!(result.len(), 2);
+        assert_eq!(
+            result.get(0),
+            Value::Timestamp(Timestamp::from_str("1970-01-01 00:00:00", None).unwrap())
+        );
+        assert_eq!(
+            result.get(1),
+            Value::Timestamp(Timestamp::from_str("1970-01-01 00:00:00.002", None).unwrap())
+        );
+    }
+
+    macro_rules! test_timestamp {
+        ($type: expr,$unit: ident) => {
+            paste! {
+                #[test]
+                fn [<test_greatest_takes_ $unit:lower _vector>]() {
+                    let function = GreatestFunction;
+                    assert_eq!(
+                        function.return_type(&[$type, $type]).unwrap(),
+                        ConcreteDataType::Timestamp(TimestampType::$unit([<Timestamp $unit Type>]))
+                    );
+
+                    let columns = vec![
+                        Arc::new([<Timestamp $unit Vector>]::from_slice(vec![-1, 2])) as _,
+                        Arc::new([<Timestamp $unit Vector>]::from_slice(vec![0, 1])) as _,
+                    ];
+
+                    let result = function.eval(&FunctionContext::default(), &columns).unwrap();
+                    let result = result.as_any().downcast_ref::<[<Timestamp $unit Vector>]>().unwrap();
+                    assert_eq!(result.len(), 2);
+                    assert_eq!(
+                        result.get(0),
+                        Value::Timestamp(Timestamp::new(0, TimeUnit::$unit))
+                    );
+                    assert_eq!(
+                        result.get(1),
+                        Value::Timestamp(Timestamp::new(2, TimeUnit::$unit))
+                    );
+                }
+            }
+        }
+    }
+
+    test_timestamp!(
+        ConcreteDataType::timestamp_nanosecond_datatype(),
+        Nanosecond
+    );
+    test_timestamp!(
+        ConcreteDataType::timestamp_microsecond_datatype(),
+        Microsecond
+    );
+    test_timestamp!(
+        ConcreteDataType::timestamp_millisecond_datatype(),
+        Millisecond
+    );
+    test_timestamp!(ConcreteDataType::timestamp_second_datatype(), Second);
+}
--- a/src/common/grpc/src/lib.rs
+++ b/src/common/grpc/src/lib.rs
@@ -18,5 +18,4 @@ pub mod flight;
 pub mod precision;
 pub mod select;

-pub use arrow_flight::FlightData;
 pub use error::Error;
--- a/src/common/meta/src/cache/flow/table_flownode.rs
+++ b/src/common/meta/src/cache/flow/table_flownode.rs
@@ -24,39 +24,21 @@ use crate::cache::{CacheContainer, Initializer};
 use crate::error::Result;
 use crate::instruction::{CacheIdent, CreateFlow, DropFlow};
 use crate::key::flow::{TableFlowManager, TableFlowManagerRef};
-use crate::key::{FlowId, FlowPartitionId};
 use crate::kv_backend::KvBackendRef;
 use crate::peer::Peer;
+use crate::FlownodeId;

-/// Flow id&flow partition key
-#[derive(Debug, Clone, PartialEq, Eq, Hash)]
-pub struct FlowIdent {
-    pub flow_id: FlowId,
-    pub partition_id: FlowPartitionId,
-}
-
-impl FlowIdent {
-    pub fn new(flow_id: FlowId, partition_id: FlowPartitionId) -> Self {
-        Self {
-            flow_id,
-            partition_id,
-        }
-    }
-}
-
-/// cache for TableFlowManager, the table_id part is in the outer cache
-/// include flownode_id, flow_id, partition_id mapping to Peer
-type FlownodeFlowSet = Arc<HashMap<FlowIdent, Peer>>;
+type FlownodeSet = Arc<HashMap<FlownodeId, Peer>>;

 pub type TableFlownodeSetCacheRef = Arc<TableFlownodeSetCache>;

 /// [TableFlownodeSetCache] caches the [TableId] to [FlownodeSet] mapping.
-pub type TableFlownodeSetCache = CacheContainer<TableId, FlownodeFlowSet, CacheIdent>;
+pub type TableFlownodeSetCache = CacheContainer<TableId, FlownodeSet, CacheIdent>;

 /// Constructs a [TableFlownodeSetCache].
 pub fn new_table_flownode_set_cache(
    name: String,
-    cache: Cache<TableId, FlownodeFlowSet>,
+    cache: Cache<TableId, FlownodeSet>,
    kv_backend: KvBackendRef,
 ) -> TableFlownodeSetCache {
    let table_flow_manager = Arc::new(TableFlowManager::new(kv_backend));
@@ -65,7 +47,7 @@ pub fn new_table_flownode_set_cache(
    CacheContainer::new(name, cache, Box::new(invalidator), init, filter)
 }

-fn init_factory(table_flow_manager: TableFlowManagerRef) -> Initializer<TableId, FlownodeFlowSet> {
+fn init_factory(table_flow_manager: TableFlowManagerRef) -> Initializer<TableId, FlownodeSet> {
    Arc::new(move |&table_id| {
        let table_flow_manager = table_flow_manager.clone();
        Box::pin(async move {
@@ -75,12 +57,7 @@ fn init_factory(table_flow_manager: TableFlowManagerRef) -> Initializer<TableId,
                .map(|flows| {
                    flows
                        .into_iter()
-                        .map(|(key, value)| {
-                            (
-                                FlowIdent::new(key.flow_id(), key.partition_id()),
-                                value.peer,
-                            )
-                        })
+                        .map(|(key, value)| (key.flownode_id(), value.peer))
                        .collect::<HashMap<_, _>>()
                })
                // We must cache the `HashSet` even if it's empty,
@@ -94,33 +71,26 @@ fn init_factory(table_flow_manager: TableFlowManagerRef) -> Initializer<TableId,
 }

 async fn handle_create_flow(
-    cache: &Cache<TableId, FlownodeFlowSet>,
+    cache: &Cache<TableId, FlownodeSet>,
    CreateFlow {
-        flow_id,
        source_table_ids,
-        partition_to_peer_mapping: flow_part2nodes,
+        flownodes: flownode_peers,
    }: &CreateFlow,
 ) {
    for table_id in source_table_ids {
        let entry = cache.entry(*table_id);
        entry
            .and_compute_with(
-                async |entry: Option<moka::Entry<u32, FlownodeFlowSet>>| match entry {
+                async |entry: Option<moka::Entry<u32, Arc<HashMap<u64, _>>>>| match entry {
                    Some(entry) => {
                        let mut map = entry.into_value().as_ref().clone();
-                        map.extend(
-                            flow_part2nodes.iter().map(|(part, peer)| {
-                                (FlowIdent::new(*flow_id, *part), peer.clone())
-                            }),
-                        );
+                        map.extend(flownode_peers.iter().map(|peer| (peer.id, peer.clone())));

                        Op::Put(Arc::new(map))
                    }
-                    None => {
-                        Op::Put(Arc::new(HashMap::from_iter(flow_part2nodes.iter().map(
-                            |(part, peer)| (FlowIdent::new(*flow_id, *part), peer.clone()),
-                        ))))
-                    }
+                    None => Op::Put(Arc::new(HashMap::from_iter(
+                        flownode_peers.iter().map(|peer| (peer.id, peer.clone())),
+                    ))),
                },
            )
            .await;
@@ -128,23 +98,21 @@ async fn handle_create_flow(
 }

 async fn handle_drop_flow(
-    cache: &Cache<TableId, FlownodeFlowSet>,
+    cache: &Cache<TableId, FlownodeSet>,
    DropFlow {
-        flow_id,
        source_table_ids,
-        flow_part2node_id,
+        flownode_ids,
    }: &DropFlow,
 ) {
    for table_id in source_table_ids {
        let entry = cache.entry(*table_id);
        entry
            .and_compute_with(
-                async |entry: Option<moka::Entry<u32, FlownodeFlowSet>>| match entry {
+                async |entry: Option<moka::Entry<u32, Arc<HashMap<u64, _>>>>| match entry {
                    Some(entry) => {
                        let mut set = entry.into_value().as_ref().clone();
-                        for (part, _node) in flow_part2node_id {
-                            let key = FlowIdent::new(*flow_id, *part);
-                            set.remove(&key);
+                        for flownode_id in flownode_ids {
+                            set.remove(flownode_id);
                        }

                        Op::Put(Arc::new(set))
@@ -160,7 +128,7 @@ async fn handle_drop_flow(
 }

 fn invalidator<'a>(
-    cache: &'a Cache<TableId, FlownodeFlowSet>,
+    cache: &'a Cache<TableId, FlownodeSet>,
    ident: &'a CacheIdent,
 ) -> BoxFuture<'a, Result<()>> {
    Box::pin(async move {
@@ -186,7 +154,7 @@ mod tests {
    use moka::future::CacheBuilder;
    use table::table_name::TableName;

-    use crate::cache::flow::table_flownode::{new_table_flownode_set_cache, FlowIdent};
+    use crate::cache::flow::table_flownode::new_table_flownode_set_cache;
    use crate::instruction::{CacheIdent, CreateFlow, DropFlow};
    use crate::key::flow::flow_info::FlowInfoValue;
    use crate::key::flow::flow_route::FlowRouteValue;
@@ -246,16 +214,12 @@ mod tests {
        let set = cache.get(1024).await.unwrap().unwrap();
        assert_eq!(
            set.as_ref().clone(),
-            HashMap::from_iter(
-                (1..=3).map(|i| { (FlowIdent::new(1024, (i - 1) as u32), Peer::empty(i),) })
-            )
+            HashMap::from_iter((1..=3).map(|i| { (i, Peer::empty(i),) }))
        );
        let set = cache.get(1025).await.unwrap().unwrap();
        assert_eq!(
            set.as_ref().clone(),
-            HashMap::from_iter(
-                (1..=3).map(|i| { (FlowIdent::new(1024, (i - 1) as u32), Peer::empty(i),) })
-            )
+            HashMap::from_iter((1..=3).map(|i| { (i, Peer::empty(i),) }))
        );
        let result = cache.get(1026).await.unwrap().unwrap();
        assert_eq!(result.len(), 0);
@@ -267,9 +231,8 @@ mod tests {
        let cache = CacheBuilder::new(128).build();
        let cache = new_table_flownode_set_cache("test".to_string(), cache, mem_kv);
        let ident = vec![CacheIdent::CreateFlow(CreateFlow {
-            flow_id: 2001,
            source_table_ids: vec![1024, 1025],
-            partition_to_peer_mapping: (1..=5).map(|i| (i as u32, Peer::empty(i + 1))).collect(),
+            flownodes: (1..=5).map(Peer::empty).collect(),
        })];
        cache.invalidate(&ident).await.unwrap();
        let set = cache.get(1024).await.unwrap().unwrap();
@@ -278,54 +241,6 @@ mod tests {
        assert_eq!(set.len(), 5);
    }

-    #[tokio::test]
-    async fn test_replace_flow() {
-        let mem_kv = Arc::new(MemoryKvBackend::default());
-        let cache = CacheBuilder::new(128).build();
-        let cache = new_table_flownode_set_cache("test".to_string(), cache, mem_kv);
-        let ident = vec![CacheIdent::CreateFlow(CreateFlow {
-            flow_id: 2001,
-            source_table_ids: vec![1024, 1025],
-            partition_to_peer_mapping: (1..=5).map(|i| (i as u32, Peer::empty(i + 1))).collect(),
-        })];
-        cache.invalidate(&ident).await.unwrap();
-        let set = cache.get(1024).await.unwrap().unwrap();
-        assert_eq!(set.len(), 5);
-        let set = cache.get(1025).await.unwrap().unwrap();
-        assert_eq!(set.len(), 5);
-
-        let drop_then_create_flow = vec![
-            CacheIdent::DropFlow(DropFlow {
-                flow_id: 2001,
-                source_table_ids: vec![1024, 1025],
-                flow_part2node_id: (1..=5).map(|i| (i as u32, i + 1)).collect(),
-            }),
-            CacheIdent::CreateFlow(CreateFlow {
-                flow_id: 2001,
-                source_table_ids: vec![1026, 1027],
-                partition_to_peer_mapping: (11..=15)
-                    .map(|i| (i as u32, Peer::empty(i + 1)))
-                    .collect(),
-            }),
-            CacheIdent::FlowId(2001),
-        ];
-        cache.invalidate(&drop_then_create_flow).await.unwrap();
-
-        let set = cache.get(1024).await.unwrap().unwrap();
-        assert!(set.is_empty());
-
-        let expected = HashMap::from_iter(
-            (11..=15).map(|i| (FlowIdent::new(2001, i as u32), Peer::empty(i + 1))),
-        );
-        let set = cache.get(1026).await.unwrap().unwrap();
-
-        assert_eq!(set.as_ref().clone(), expected);
-
-        let set = cache.get(1027).await.unwrap().unwrap();
-
-        assert_eq!(set.as_ref().clone(), expected);
-    }
-
    #[tokio::test]
    async fn test_drop_flow() {
        let mem_kv = Arc::new(MemoryKvBackend::default());
@@ -333,57 +248,34 @@ mod tests {
        let cache = new_table_flownode_set_cache("test".to_string(), cache, mem_kv);
        let ident = vec![
            CacheIdent::CreateFlow(CreateFlow {
-                flow_id: 2001,
                source_table_ids: vec![1024, 1025],
-                partition_to_peer_mapping: (1..=5)
-                    .map(|i| (i as u32, Peer::empty(i + 1)))
-                    .collect(),
+                flownodes: (1..=5).map(Peer::empty).collect(),
            }),
            CacheIdent::CreateFlow(CreateFlow {
-                flow_id: 2002,
                source_table_ids: vec![1024, 1025],
-                partition_to_peer_mapping: (11..=12)
-                    .map(|i| (i as u32, Peer::empty(i + 1)))
-                    .collect(),
-            }),
-            // same flownode that hold multiple flows
-            CacheIdent::CreateFlow(CreateFlow {
-                flow_id: 2003,
-                source_table_ids: vec![1024, 1025],
-                partition_to_peer_mapping: (1..=5)
-                    .map(|i| (i as u32, Peer::empty(i + 1)))
-                    .collect(),
+                flownodes: (11..=12).map(Peer::empty).collect(),
            }),
        ];
        cache.invalidate(&ident).await.unwrap();
        let set = cache.get(1024).await.unwrap().unwrap();
-        assert_eq!(set.len(), 12);
+        assert_eq!(set.len(), 7);
        let set = cache.get(1025).await.unwrap().unwrap();
-        assert_eq!(set.len(), 12);
+        assert_eq!(set.len(), 7);

        let ident = vec![CacheIdent::DropFlow(DropFlow {
-            flow_id: 2001,
            source_table_ids: vec![1024, 1025],
-            flow_part2node_id: (1..=5).map(|i| (i as u32, i + 1)).collect(),
+            flownode_ids: vec![1, 2, 3, 4, 5],
        })];
        cache.invalidate(&ident).await.unwrap();
        let set = cache.get(1024).await.unwrap().unwrap();
        assert_eq!(
            set.as_ref().clone(),
-            HashMap::from_iter(
-                (11..=12)
-                    .map(|i| (FlowIdent::new(2002, i as u32), Peer::empty(i + 1)))
-                    .chain((1..=5).map(|i| (FlowIdent::new(2003, i as u32), Peer::empty(i + 1))))
-            )
+            HashMap::from_iter((11..=12).map(|i| { (i, Peer::empty(i),) }))
        );
        let set = cache.get(1025).await.unwrap().unwrap();
        assert_eq!(
            set.as_ref().clone(),
-            HashMap::from_iter(
-                (11..=12)
-                    .map(|i| (FlowIdent::new(2002, i as u32), Peer::empty(i + 1)))
-                    .chain((1..=5).map(|i| (FlowIdent::new(2003, i as u32), Peer::empty(i + 1))))
-            )
+            HashMap::from_iter((11..=12).map(|i| { (i, Peer::empty(i),) }))
        );
    }
 }
--- a/src/common/meta/src/cache_invalidator.rs
+++ b/src/common/meta/src/cache_invalidator.rs
@@ -16,12 +16,9 @@ use std::sync::Arc;

 use crate::error::Result;
 use crate::flow_name::FlowName;
-use crate::instruction::{CacheIdent, DropFlow};
+use crate::instruction::CacheIdent;
 use crate::key::flow::flow_info::FlowInfoKey;
 use crate::key::flow::flow_name::FlowNameKey;
-use crate::key::flow::flow_route::FlowRouteKey;
-use crate::key::flow::flownode_flow::FlownodeFlowKey;
-use crate::key::flow::table_flow::TableFlowKey;
 use crate::key::schema_name::SchemaNameKey;
 use crate::key::table_info::TableInfoKey;
 use crate::key::table_name::TableNameKey;
@@ -92,40 +89,9 @@ where
                    let key: SchemaNameKey = schema_name.into();
                    self.invalidate_key(&key.to_bytes()).await;
                }
-                CacheIdent::CreateFlow(_) => {
+                CacheIdent::CreateFlow(_) | CacheIdent::DropFlow(_) => {
                    // Do nothing
                }
-                CacheIdent::DropFlow(DropFlow {
-                    flow_id,
-                    source_table_ids,
-                    flow_part2node_id,
-                }) => {
-                    // invalidate flow route/flownode flow/table flow
-                    let mut keys = Vec::with_capacity(
-                        source_table_ids.len() * flow_part2node_id.len()
-                            + flow_part2node_id.len() * 2,
-                    );
-                    for table_id in source_table_ids {
-                        for (partition_id, node_id) in flow_part2node_id {
-                            let key =
-                                TableFlowKey::new(*table_id, *node_id, *flow_id, *partition_id)
-                                    .to_bytes();
-                            keys.push(key);
-                        }
-                    }
-
-                    for (partition_id, node_id) in flow_part2node_id {
-                        let key =
-                            FlownodeFlowKey::new(*node_id, *flow_id, *partition_id).to_bytes();
-                        keys.push(key);
-                        let key = FlowRouteKey::new(*flow_id, *partition_id).to_bytes();
-                        keys.push(key);
-                    }
-
-                    for key in keys {
-                        self.invalidate_key(&key).await;
-                    }
-                }
                CacheIdent::FlowName(FlowName {
                    catalog_name,
                    flow_name,
--- a/src/common/meta/src/ddl/create_flow.rs
+++ b/src/common/meta/src/ddl/create_flow.rs
@@ -39,7 +39,7 @@ use crate::cache_invalidator::Context;
 use crate::ddl::utils::{add_peer_context_if_needed, handle_retry_error};
 use crate::ddl::DdlContext;
 use crate::error::{self, Result, UnexpectedSnafu};
-use crate::instruction::{CacheIdent, CreateFlow, DropFlow};
+use crate::instruction::{CacheIdent, CreateFlow};
 use crate::key::flow::flow_info::FlowInfoValue;
 use crate::key::flow::flow_route::FlowRouteValue;
 use crate::key::table_name::TableNameKey;
@@ -70,7 +70,6 @@ impl CreateFlowProcedure {
                query_context,
                state: CreateFlowState::Prepare,
                prev_flow_info_value: None,
-                did_replace: false,
                flow_type: None,
            },
        }
@@ -225,7 +224,6 @@ impl CreateFlowProcedure {
                .update_flow_metadata(flow_id, prev_flow_value, &flow_info, flow_routes)
                .await?;
            info!("Replaced flow metadata for flow {flow_id}");
-            self.data.did_replace = true;
        } else {
            self.context
                .flow_metadata_manager
@@ -242,43 +240,22 @@ impl CreateFlowProcedure {
        debug_assert!(self.data.state == CreateFlowState::InvalidateFlowCache);
        // Safety: The flow id must be allocated.
        let flow_id = self.data.flow_id.unwrap();
-        let did_replace = self.data.did_replace;
        let ctx = Context {
            subject: Some("Invalidate flow cache by creating flow".to_string()),
        };

-        let mut caches = vec![];
-
-        // if did replaced, invalidate the flow cache with drop the old flow
-        if did_replace {
-            let old_flow_info = self.data.prev_flow_info_value.as_ref().unwrap();
-
-            // only drop flow is needed, since flow name haven't changed, and flow id already invalidated below
-            caches.extend([CacheIdent::DropFlow(DropFlow {
-                flow_id,
-                source_table_ids: old_flow_info.source_table_ids.clone(),
-                flow_part2node_id: old_flow_info.flownode_ids().clone().into_iter().collect(),
-            })]);
-        }
-
-        let (_flow_info, flow_routes) = (&self.data).into();
-        let flow_part2peers = flow_routes
-            .into_iter()
-            .map(|(part_id, route)| (part_id, route.peer))
-            .collect();
-
-        caches.extend([
-            CacheIdent::CreateFlow(CreateFlow {
-                flow_id,
-                source_table_ids: self.data.source_table_ids.clone(),
-                partition_to_peer_mapping: flow_part2peers,
-            }),
-            CacheIdent::FlowId(flow_id),
-        ]);
-
        self.context
            .cache_invalidator
-            .invalidate(&ctx, &caches)
+            .invalidate(
+                &ctx,
+                &[
+                    CacheIdent::CreateFlow(CreateFlow {
+                        source_table_ids: self.data.source_table_ids.clone(),
+                        flownodes: self.data.peers.clone(),
+                    }),
+                    CacheIdent::FlowId(flow_id),
+                ],
+            )
            .await?;

        Ok(Status::done_with_output(flow_id))
@@ -400,10 +377,6 @@ pub struct CreateFlowData {
    /// For verify if prev value is consistent when need to update flow metadata.
    /// only set when `or_replace` is true.
    pub(crate) prev_flow_info_value: Option<DeserializedValueWithBytes<FlowInfoValue>>,
-    /// Only set to true when replace actually happened.
-    /// This is used to determine whether to invalidate the cache.
-    #[serde(default)]
-    pub(crate) did_replace: bool,
    pub(crate) flow_type: Option<FlowType>,
 }

--- a/src/common/meta/src/ddl/drop_flow.rs
+++ b/src/common/meta/src/ddl/drop_flow.rs
@@ -13,7 +13,6 @@
 // limitations under the License.

 mod metadata;
-
 use api::v1::flow::{flow_request, DropRequest, FlowRequest};
 use async_trait::async_trait;
 use common_catalog::format_full_flow_name;
@@ -154,12 +153,6 @@ impl DropFlowProcedure {
        };
        let flow_info_value = self.data.flow_info_value.as_ref().unwrap();

-        let flow_part2nodes = flow_info_value
-            .flownode_ids()
-            .clone()
-            .into_iter()
-            .collect::<Vec<_>>();
-
        self.context
            .cache_invalidator
            .invalidate(
@@ -171,9 +164,8 @@ impl DropFlowProcedure {
                        flow_name: flow_info_value.flow_name.to_string(),
                    }),
                    CacheIdent::DropFlow(DropFlow {
-                        flow_id,
                        source_table_ids: flow_info_value.source_table_ids.clone(),
-                        flow_part2node_id: flow_part2nodes,
+                        flownode_ids: flow_info_value.flownode_ids.values().cloned().collect(),
                    }),
                ],
            )
--- a/src/common/meta/src/error.rs
+++ b/src/common/meta/src/error.rs
@@ -514,25 +514,11 @@ pub enum Error {
    },

    #[snafu(display(
-        "Failed to get a Kafka partition client, topic: {}, partition: {}",
+        "Failed to build a Kafka partition client, topic: {}, partition: {}",
        topic,
        partition
    ))]
-    KafkaPartitionClient {
-        topic: String,
-        partition: i32,
-        #[snafu(implicit)]
-        location: Location,
-        #[snafu(source)]
-        error: rskafka::client::error::Error,
-    },
-
-    #[snafu(display(
-        "Failed to get offset from Kafka, topic: {}, partition: {}",
-        topic,
-        partition
-    ))]
-    KafkaGetOffset {
+    BuildKafkaPartitionClient {
        topic: String,
        partition: i32,
        #[snafu(implicit)]
@@ -857,7 +843,7 @@ impl ErrorExt for Error {
            | EncodeWalOptions { .. }
            | BuildKafkaClient { .. }
            | BuildKafkaCtrlClient { .. }
-            | KafkaPartitionClient { .. }
+            | BuildKafkaPartitionClient { .. }
            | ResolveKafkaEndpoint { .. }
            | ProduceRecord { .. }
            | CreateKafkaWalTopic { .. }
@@ -866,8 +852,7 @@ impl ErrorExt for Error {
            | ProcedureOutput { .. }
            | FromUtf8 { .. }
            | MetadataCorruption { .. }
-            | ParseWalOptions { .. }
-            | KafkaGetOffset { .. } => StatusCode::Unexpected,
+            | ParseWalOptions { .. } => StatusCode::Unexpected,

            SendMessage { .. } | GetKvCache { .. } | CacheNotGet { .. } => StatusCode::Internal,

--- a/src/common/meta/src/instruction.rs
+++ b/src/common/meta/src/instruction.rs
@@ -24,7 +24,7 @@ use table::table_name::TableName;

 use crate::flow_name::FlowName;
 use crate::key::schema_name::SchemaName;
-use crate::key::{FlowId, FlowPartitionId};
+use crate::key::FlowId;
 use crate::peer::Peer;
 use crate::{DatanodeId, FlownodeId};

@@ -184,19 +184,14 @@ pub enum CacheIdent {

 #[derive(Debug, Clone, Serialize, Deserialize, PartialEq, Eq)]
 pub struct CreateFlow {
-    /// The unique identifier for the flow.
-    pub flow_id: FlowId,
    pub source_table_ids: Vec<TableId>,
-    /// Mapping of flow partition to peer information
-    pub partition_to_peer_mapping: Vec<(FlowPartitionId, Peer)>,
+    pub flownodes: Vec<Peer>,
 }

 #[derive(Debug, Clone, Serialize, Deserialize, PartialEq, Eq)]
 pub struct DropFlow {
-    pub flow_id: FlowId,
    pub source_table_ids: Vec<TableId>,
-    /// Mapping of flow partition to flownode id
-    pub flow_part2node_id: Vec<(FlowPartitionId, FlownodeId)>,
+    pub flownode_ids: Vec<FlownodeId>,
 }

 /// Flushes a batch of regions.
--- a/src/common/meta/src/key/datanode_table.rs
+++ b/src/common/meta/src/key/datanode_table.rs
@@ -256,11 +256,6 @@ impl DatanodeTableManager {
            })?
            .and_then(|r| DatanodeTableValue::try_from_raw_value(&r.value))?
            .region_info;
-
-        // If the region options are the same, we don't need to update it.
-        if region_info.region_options == new_region_options {
-            return Ok(Txn::new());
-        }
        // substitute region options only.
        region_info.region_options = new_region_options;

--- a/src/common/meta/src/key/flow.rs
+++ b/src/common/meta/src/key/flow.rs
@@ -45,7 +45,7 @@ use crate::kv_backend::KvBackendRef;
 use crate::rpc::store::BatchDeleteRequest;

 /// The key of `__flow/` scope.
-#[derive(Debug, Clone, PartialEq)]
+#[derive(Debug, PartialEq)]
 pub struct FlowScoped<T> {
    inner: T,
 }
@@ -246,32 +246,27 @@ impl FlowMetadataManager {
        new_flow_info: &FlowInfoValue,
        flow_routes: Vec<(FlowPartitionId, FlowRouteValue)>,
    ) -> Result<()> {
-        let (update_flow_flow_name_txn, on_create_flow_flow_name_failure) =
+        let (create_flow_flow_name_txn, on_create_flow_flow_name_failure) =
            self.flow_name_manager.build_update_txn(
                &new_flow_info.catalog_name,
                &new_flow_info.flow_name,
                flow_id,
            )?;

-        let (update_flow_txn, on_create_flow_failure) =
+        let (create_flow_txn, on_create_flow_failure) =
            self.flow_info_manager
                .build_update_txn(flow_id, current_flow_info, new_flow_info)?;

-        let update_flow_routes_txn = self.flow_route_manager.build_update_txn(
-            flow_id,
-            current_flow_info,
-            flow_routes.clone(),
-        )?;
+        let create_flow_routes_txn = self
+            .flow_route_manager
+            .build_create_txn(flow_id, flow_routes.clone())?;

-        let update_flownode_flow_txn = self.flownode_flow_manager.build_update_txn(
-            flow_id,
-            current_flow_info,
-            new_flow_info.flownode_ids().clone(),
-        );
+        let create_flownode_flow_txn = self
+            .flownode_flow_manager
+            .build_create_txn(flow_id, new_flow_info.flownode_ids().clone());

-        let update_table_flow_txn = self.table_flow_manager.build_update_txn(
+        let create_table_flow_txn = self.table_flow_manager.build_create_txn(
            flow_id,
-            current_flow_info,
            flow_routes
                .into_iter()
                .map(|(partition_id, route)| (partition_id, TableFlowValue { peer: route.peer }))
@@ -280,11 +275,11 @@ impl FlowMetadataManager {
        )?;

        let txn = Txn::merge_all(vec![
-            update_flow_flow_name_txn,
-            update_flow_txn,
-            update_flow_routes_txn,
-            update_flownode_flow_txn,
-            update_table_flow_txn,
+            create_flow_flow_name_txn,
+            create_flow_txn,
+            create_flow_routes_txn,
+            create_flownode_flow_txn,
+            create_table_flow_txn,
        ]);
        info!(
            "Creating flow {}.{}({}), with {} txn operations",
@@ -788,141 +783,6 @@ mod tests {
        }
    }

-    #[tokio::test]
-    async fn test_update_flow_metadata_diff_flownode() {
-        let mem_kv = Arc::new(MemoryKvBackend::default());
-        let flow_metadata_manager = FlowMetadataManager::new(mem_kv.clone());
-        let flow_id = 10;
-        let flow_value = test_flow_info_value(
-            "flow",
-            [(0u32, 1u64), (1u32, 2u64)].into(),
-            vec![1024, 1025, 1026],
-        );
-        let flow_routes = vec![
-            (
-                0u32,
-                FlowRouteValue {
-                    peer: Peer::empty(1),
-                },
-            ),
-            (
-                1,
-                FlowRouteValue {
-                    peer: Peer::empty(2),
-                },
-            ),
-        ];
-        flow_metadata_manager
-            .create_flow_metadata(flow_id, flow_value.clone(), flow_routes.clone())
-            .await
-            .unwrap();
-
-        let new_flow_value = {
-            let mut tmp = flow_value.clone();
-            tmp.raw_sql = "new".to_string();
-            // move to different flownodes
-            tmp.flownode_ids = [(0, 3u64), (1, 4u64)].into();
-            tmp
-        };
-        let new_flow_routes = vec![
-            (
-                0u32,
-                FlowRouteValue {
-                    peer: Peer::empty(3),
-                },
-            ),
-            (
-                1,
-                FlowRouteValue {
-                    peer: Peer::empty(4),
-                },
-            ),
-        ];
-
-        // Update flow instead
-        flow_metadata_manager
-            .update_flow_metadata(
-                flow_id,
-                &DeserializedValueWithBytes::from_inner(flow_value.clone()),
-                &new_flow_value,
-                new_flow_routes.clone(),
-            )
-            .await
-            .unwrap();
-
-        let got = flow_metadata_manager
-            .flow_info_manager()
-            .get(flow_id)
-            .await
-            .unwrap()
-            .unwrap();
-        let routes = flow_metadata_manager
-            .flow_route_manager()
-            .routes(flow_id)
-            .await
-            .unwrap();
-        assert_eq!(
-            routes,
-            vec![
-                (
-                    FlowRouteKey::new(flow_id, 0),
-                    FlowRouteValue {
-                        peer: Peer::empty(3),
-                    },
-                ),
-                (
-                    FlowRouteKey::new(flow_id, 1),
-                    FlowRouteValue {
-                        peer: Peer::empty(4),
-                    },
-                ),
-            ]
-        );
-        assert_eq!(got, new_flow_value);
-
-        let flows = flow_metadata_manager
-            .flownode_flow_manager()
-            .flows(1)
-            .try_collect::<Vec<_>>()
-            .await
-            .unwrap();
-        // should moved to different flownode
-        assert_eq!(flows, vec![]);
-
-        let flows = flow_metadata_manager
-            .flownode_flow_manager()
-            .flows(3)
-            .try_collect::<Vec<_>>()
-            .await
-            .unwrap();
-        assert_eq!(flows, vec![(flow_id, 0)]);
-
-        for table_id in [1024, 1025, 1026] {
-            let nodes = flow_metadata_manager
-                .table_flow_manager()
-                .flows(table_id)
-                .await
-                .unwrap();
-            assert_eq!(
-                nodes,
-                vec![
-                    (
-                        TableFlowKey::new(table_id, 3, flow_id, 0),
-                        TableFlowValue {
-                            peer: Peer::empty(3)
-                        }
-                    ),
-                    (
-                        TableFlowKey::new(table_id, 4, flow_id, 1),
-                        TableFlowValue {
-                            peer: Peer::empty(4)
-                        }
-                    )
-                ]
-            );
-        }
-    }
-
    #[tokio::test]
    async fn test_update_flow_metadata_flow_replace_diff_id_err() {
        let mem_kv = Arc::new(MemoryKvBackend::default());
--- a/src/common/meta/src/key/flow/flow_info.rs
+++ b/src/common/meta/src/key/flow/flow_info.rs
@@ -153,15 +153,6 @@ impl FlowInfoValue {
        &self.flownode_ids
    }

-    /// Insert a new flownode id for a partition.
-    pub fn insert_flownode_id(
-        &mut self,
-        partition: FlowPartitionId,
-        node: FlownodeId,
-    ) -> Option<FlownodeId> {
-        self.flownode_ids.insert(partition, node)
-    }
-
    /// Returns the `source_table`.
    pub fn source_table_ids(&self) -> &[TableId] {
        &self.source_table_ids
@@ -281,11 +272,10 @@ impl FlowInfoManager {
        let raw_value = new_flow_value.try_as_raw_value()?;
        let prev_value = current_flow_value.get_raw_bytes();
        let txn = Txn::new()
-            .when(vec![Compare::new(
-                key.clone(),
-                CompareOp::Equal,
-                Some(prev_value),
-            )])
+            .when(vec![
+                Compare::new(key.clone(), CompareOp::NotEqual, None),
+                Compare::new(key.clone(), CompareOp::Equal, Some(prev_value)),
+            ])
            .and_then(vec![TxnOp::Put(key.clone(), raw_value)])
            .or_else(vec![TxnOp::Get(key.clone())]);

--- a/src/common/meta/src/key/flow/flow_route.rs
+++ b/src/common/meta/src/key/flow/flow_route.rs
@@ -19,12 +19,9 @@ use serde::{Deserialize, Serialize};
 use snafu::OptionExt;

 use crate::error::{self, Result};
-use crate::key::flow::flow_info::FlowInfoValue;
 use crate::key::flow::{flownode_addr_helper, FlowScoped};
 use crate::key::node_address::NodeAddressKey;
-use crate::key::{
-    BytesAdapter, DeserializedValueWithBytes, FlowId, FlowPartitionId, MetadataKey, MetadataValue,
-};
+use crate::key::{BytesAdapter, FlowId, FlowPartitionId, MetadataKey, MetadataValue};
 use crate::kv_backend::txn::{Txn, TxnOp};
 use crate::kv_backend::KvBackendRef;
 use crate::peer::Peer;
@@ -42,7 +39,7 @@ lazy_static! {
 /// The key stores the route info of the flow.
 ///
 /// The layout: `__flow/route/{flow_id}/{partition_id}`.
-#[derive(Debug, Clone, PartialEq)]
+#[derive(Debug, PartialEq)]
 pub struct FlowRouteKey(FlowScoped<FlowRouteKeyInner>);

 impl FlowRouteKey {
@@ -145,12 +142,6 @@ pub struct FlowRouteValue {
    pub(crate) peer: Peer,
 }

-impl From<Peer> for FlowRouteValue {
-    fn from(peer: Peer) -> Self {
-        Self { peer }
-    }
-}
-
 impl FlowRouteValue {
    /// Returns the `peer`.
    pub fn peer(&self) -> &Peer {
@@ -213,33 +204,6 @@ impl FlowRouteManager {
        Ok(Txn::new().and_then(txns))
    }

-    /// Builds a update flow routes transaction.
-    ///
-    /// Puts `__flow/route/{flow_id}/{partition_id}` keys.
-    /// Also removes `__flow/route/{flow_id}/{old_partition_id}` keys.
-    pub(crate) fn build_update_txn<I: IntoIterator<Item = (FlowPartitionId, FlowRouteValue)>>(
-        &self,
-        flow_id: FlowId,
-        current_flow_info: &DeserializedValueWithBytes<FlowInfoValue>,
-        flow_routes: I,
-    ) -> Result<Txn> {
-        let del_txns = current_flow_info
-            .flownode_ids()
-            .iter()
-            .map(|(partition_id, _)| {
-                let key = FlowRouteKey::new(flow_id, *partition_id).to_bytes();
-                Ok(TxnOp::Delete(key))
-            });
-
-        let put_txns = flow_routes.into_iter().map(|(partition_id, route)| {
-            let key = FlowRouteKey::new(flow_id, partition_id).to_bytes();
-
-            Ok(TxnOp::Put(key, route.try_as_raw_value()?))
-        });
-        let txns = del_txns.chain(put_txns).collect::<Result<Vec<_>>>()?;
-        Ok(Txn::new().and_then(txns))
-    }
-
    async fn remap_flow_route_addresses(
        &self,
        flow_routes: &mut [(FlowRouteKey, FlowRouteValue)],
--- a/src/common/meta/src/key/flow/flownode_flow.rs
+++ b/src/common/meta/src/key/flow/flownode_flow.rs
@@ -19,9 +19,8 @@ use regex::Regex;
 use snafu::OptionExt;

 use crate::error::{self, Result};
-use crate::key::flow::flow_info::FlowInfoValue;
 use crate::key::flow::FlowScoped;
-use crate::key::{BytesAdapter, DeserializedValueWithBytes, FlowId, FlowPartitionId, MetadataKey};
+use crate::key::{BytesAdapter, FlowId, FlowPartitionId, MetadataKey};
 use crate::kv_backend::txn::{Txn, TxnOp};
 use crate::kv_backend::KvBackendRef;
 use crate::range_stream::{PaginationStream, DEFAULT_PAGE_SIZE};
@@ -166,17 +165,6 @@ impl FlownodeFlowManager {
        Self { kv_backend }
    }

-    /// Whether given flow exist on this flownode.
-    pub async fn exists(
-        &self,
-        flownode_id: FlownodeId,
-        flow_id: FlowId,
-        partition_id: FlowPartitionId,
-    ) -> Result<bool> {
-        let key = FlownodeFlowKey::new(flownode_id, flow_id, partition_id).to_bytes();
-        Ok(self.kv_backend.get(&key).await?.is_some())
-    }
-
    /// Retrieves all [FlowId] and [FlowPartitionId]s of the specified `flownode_id`.
    pub fn flows(
        &self,
@@ -214,33 +202,6 @@ impl FlownodeFlowManager {

        Txn::new().and_then(txns)
    }
-
-    /// Builds a update flownode flow transaction.
-    ///
-    /// Puts `__flownode_flow/{flownode_id}/{flow_id}/{partition_id}` keys.
-    /// Remove the old `__flownode_flow/{old_flownode_id}/{flow_id}/{old_partition_id}` keys.
-    pub(crate) fn build_update_txn<I: IntoIterator<Item = (FlowPartitionId, FlownodeId)>>(
-        &self,
-        flow_id: FlowId,
-        current_flow_info: &DeserializedValueWithBytes<FlowInfoValue>,
-        flownode_ids: I,
-    ) -> Txn {
-        let del_txns =
-            current_flow_info
-                .flownode_ids()
-                .iter()
-                .map(|(partition_id, flownode_id)| {
-                    let key = FlownodeFlowKey::new(*flownode_id, flow_id, *partition_id).to_bytes();
-                    TxnOp::Delete(key)
-                });
-        let put_txns = flownode_ids.into_iter().map(|(partition_id, flownode_id)| {
-            let key = FlownodeFlowKey::new(flownode_id, flow_id, partition_id).to_bytes();
-            TxnOp::Put(key, vec![])
-        });
-        let txns = del_txns.chain(put_txns).collect::<Vec<_>>();
-
-        Txn::new().and_then(txns)
-    }
 }

 #[cfg(test)]
--- a/src/common/meta/src/key/flow/table_flow.rs
+++ b/src/common/meta/src/key/flow/table_flow.rs
@@ -22,12 +22,9 @@ use snafu::OptionExt;
 use table::metadata::TableId;

 use crate::error::{self, Result};
-use crate::key::flow::flow_info::FlowInfoValue;
 use crate::key::flow::{flownode_addr_helper, FlowScoped};
 use crate::key::node_address::NodeAddressKey;
-use crate::key::{
-    BytesAdapter, DeserializedValueWithBytes, FlowId, FlowPartitionId, MetadataKey, MetadataValue,
-};
+use crate::key::{BytesAdapter, FlowId, FlowPartitionId, MetadataKey, MetadataValue};
 use crate::kv_backend::txn::{Txn, TxnOp};
 use crate::kv_backend::KvBackendRef;
 use crate::peer::Peer;
@@ -218,7 +215,7 @@ impl TableFlowManager {

    /// Builds a create table flow transaction.
    ///
-    /// Puts `__flow/source_table/{table_id}/{node_id}/{flow_id}/{partition_id}` keys.
+    /// Puts `__flow/source_table/{table_id}/{node_id}/{partition_id}` keys.
    pub fn build_create_txn(
        &self,
        flow_id: FlowId,
@@ -242,44 +239,6 @@ impl TableFlowManager {
        Ok(Txn::new().and_then(txns))
    }

-    /// Builds a update table flow transaction.
-    ///
-    /// Puts `__flow/source_table/{table_id}/{node_id}/{flow_id}/{partition_id}` keys,
-    /// Also remove previous
-    /// `__flow/source_table/{table_id}/{old_node_id}/{flow_id}/{partition_id}` keys.
-    pub fn build_update_txn(
-        &self,
-        flow_id: FlowId,
-        current_flow_info: &DeserializedValueWithBytes<FlowInfoValue>,
-        table_flow_values: Vec<(FlowPartitionId, TableFlowValue)>,
-        source_table_ids: &[TableId],
-    ) -> Result<Txn> {
-        let mut txns = Vec::with_capacity(2 * source_table_ids.len() * table_flow_values.len());
-
-        // first remove the old keys
-        for (part_id, node_id) in current_flow_info.flownode_ids() {
-            for source_table_id in current_flow_info.source_table_ids() {
-                txns.push(TxnOp::Delete(
-                    TableFlowKey::new(*source_table_id, *node_id, flow_id, *part_id).to_bytes(),
-                ));
-            }
-        }
-
-        for (partition_id, table_flow_value) in table_flow_values {
-            let flownode_id = table_flow_value.peer.id;
-            let value = table_flow_value.try_as_raw_value()?;
-            for source_table_id in source_table_ids {
-                txns.push(TxnOp::Put(
-                    TableFlowKey::new(*source_table_id, flownode_id, flow_id, partition_id)
-                        .to_bytes(),
-                    value.clone(),
-                ));
-            }
-        }
-
-        Ok(Txn::new().and_then(txns))
-    }
-
    async fn remap_table_flow_addresses(
        &self,
        table_flows: &mut [(TableFlowKey, TableFlowValue)],
--- a/src/common/meta/src/kv_backend.rs
+++ b/src/common/meta/src/kv_backend.rs
@@ -35,7 +35,7 @@ pub mod memory;
 pub mod rds;
 pub mod test;
 pub mod txn;
-pub mod util;
+
 pub type KvBackendRef<E = Error> = Arc<dyn KvBackend<Error = E> + Send + Sync>;

 #[async_trait]
--- a/src/common/meta/src/kv_backend/util.rs
+++ b/src/common/meta/src/kv_backend/util.rs
@@ -1,85 +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.
-
-/// Removes sensitive information like passwords from connection strings.
-///
-/// This function sanitizes connection strings by removing credentials:
-/// - For URL format (mysql://user:password@host:port/db): Removes everything before '@'
-/// - For parameter format (host=localhost password=secret): Removes the password parameter
-/// - For URL format without credentials (mysql://host:port/db): Removes the protocol prefix
-///
-/// # Arguments
-///
-/// * `conn_str` - The connection string to sanitize
-///
-/// # Returns
-///
-/// A sanitized version of the connection string with sensitive information removed
-pub fn sanitize_connection_string(conn_str: &str) -> String {
-    // Case 1: URL format with credentials (mysql://user:password@host:port/db)
-    // Extract everything after the '@' symbol
-    if let Some(at_pos) = conn_str.find('@') {
-        return conn_str[at_pos + 1..].to_string();
-    }
-
-    // Case 2: Parameter format with password (host=localhost password=secret dbname=mydb)
-    // Filter out any parameter that starts with "password="
-    if conn_str.contains("password=") {
-        return conn_str
-            .split_whitespace()
-            .filter(|param| !param.starts_with("password="))
-            .collect::<Vec<_>>()
-            .join(" ");
-    }
-
-    // Case 3: URL format without credentials (mysql://host:port/db)
-    // Extract everything after the protocol prefix
-    if let Some(host_part) = conn_str.split("://").nth(1) {
-        return host_part.to_string();
-    }
-
-    // Case 4: Already sanitized or unknown format
-    // Return as is
-    conn_str.to_string()
-}
-
-#[cfg(test)]
-mod tests {
-    use super::*;
-
-    #[test]
-    fn test_sanitize_connection_string() {
-        // Test URL format with username/password
-        let conn_str = "mysql://user:password123@localhost:3306/db";
-        assert_eq!(sanitize_connection_string(conn_str), "localhost:3306/db");
-
-        // Test URL format without credentials
-        let conn_str = "mysql://localhost:3306/db";
-        assert_eq!(sanitize_connection_string(conn_str), "localhost:3306/db");
-
-        // Test parameter format with password
-        let conn_str = "host=localhost port=5432 user=postgres password=secret dbname=mydb";
-        assert_eq!(
-            sanitize_connection_string(conn_str),
-            "host=localhost port=5432 user=postgres dbname=mydb"
-        );
-
-        // Test parameter format without password
-        let conn_str = "host=localhost port=5432 user=postgres dbname=mydb";
-        assert_eq!(
-            sanitize_connection_string(conn_str),
-            "host=localhost port=5432 user=postgres dbname=mydb"
-        );
-    }
-}
--- a/src/common/meta/src/lib.rs
+++ b/src/common/meta/src/lib.rs
@@ -15,8 +15,6 @@
 #![feature(assert_matches)]
 #![feature(btree_extract_if)]
 #![feature(let_chains)]
-#![feature(extract_if)]
-#![feature(hash_extract_if)]

 pub mod cache;
 pub mod cache_invalidator;
--- a/src/common/meta/src/rpc/router.rs
+++ b/src/common/meta/src/rpc/router.rs
@@ -176,15 +176,12 @@ impl TableRoute {
                })?
                .into();

-            let leader_peer = peers
-                .get(region_route.leader_peer_index as usize)
-                .cloned()
-                .map(Into::into);
+            let leader_peer = peers.get(region_route.leader_peer_index as usize).cloned();

            let follower_peers = region_route
                .follower_peer_indexes
                .into_iter()
-                .filter_map(|x| peers.get(x as usize).cloned().map(Into::into))
+                .filter_map(|x| peers.get(x as usize).cloned())
                .collect::<Vec<_>>();

            region_routes.push(RegionRoute {
--- a/src/common/meta/src/test_util.rs
+++ b/src/common/meta/src/test_util.rs
@@ -20,8 +20,6 @@ use api::v1::region::{InsertRequests, RegionRequest};
 pub use common_base::AffectedRows;
 use common_query::request::QueryRequest;
 use common_recordbatch::SendableRecordBatchStream;
-use common_wal::config::kafka::common::{KafkaConnectionConfig, KafkaTopicConfig};
-use common_wal::config::kafka::MetasrvKafkaConfig;

 use crate::cache_invalidator::DummyCacheInvalidator;
 use crate::ddl::flow_meta::FlowMetadataAllocator;
@@ -39,8 +37,7 @@ use crate::peer::{Peer, PeerLookupService};
 use crate::region_keeper::MemoryRegionKeeper;
 use crate::region_registry::LeaderRegionRegistry;
 use crate::sequence::SequenceBuilder;
-use crate::wal_options_allocator::topic_pool::KafkaTopicPool;
-use crate::wal_options_allocator::{build_kafka_topic_creator, WalOptionsAllocator};
+use crate::wal_options_allocator::WalOptionsAllocator;
 use crate::{DatanodeId, FlownodeId};

 #[async_trait::async_trait]
@@ -202,34 +199,3 @@ impl PeerLookupService for NoopPeerLookupService {
        Ok(Some(Peer::empty(id)))
    }
 }
-
-/// Create a kafka topic pool for testing.
-pub async fn test_kafka_topic_pool(
-    broker_endpoints: Vec<String>,
-    num_topics: usize,
-    auto_create_topics: bool,
-    topic_name_prefix: Option<&str>,
-) -> KafkaTopicPool {
-    let mut config = MetasrvKafkaConfig {
-        connection: KafkaConnectionConfig {
-            broker_endpoints,
-            ..Default::default()
-        },
-        kafka_topic: KafkaTopicConfig {
-            num_topics,
-
-            ..Default::default()
-        },
-        auto_create_topics,
-        ..Default::default()
-    };
-    if let Some(prefix) = topic_name_prefix {
-        config.kafka_topic.topic_name_prefix = prefix.to_string();
-    }
-    let kv_backend = Arc::new(MemoryKvBackend::new()) as KvBackendRef;
-    let topic_creator = build_kafka_topic_creator(&config.connection, &config.kafka_topic)
-        .await
-        .unwrap();
-
-    KafkaTopicPool::new(&config, kv_backend, topic_creator)
-}
--- a/src/common/meta/src/wal_options_allocator.rs
+++ b/src/common/meta/src/wal_options_allocator.rs
@@ -112,9 +112,7 @@ pub async fn build_wal_options_allocator(
                NAME_PATTERN_REGEX.is_match(prefix),
                InvalidTopicNamePrefixSnafu { prefix }
            );
-            let topic_creator =
-                build_kafka_topic_creator(&kafka_config.connection, &kafka_config.kafka_topic)
-                    .await?;
+            let topic_creator = build_kafka_topic_creator(kafka_config).await?;
            let topic_pool = KafkaTopicPool::new(kafka_config, kv_backend, topic_creator);
            Ok(WalOptionsAllocator::Kafka(topic_pool))
        }
@@ -153,16 +151,13 @@ pub fn prepare_wal_options(
 mod tests {
    use std::assert_matches::assert_matches;

-    use common_wal::config::kafka::common::KafkaTopicConfig;
+    use common_wal::config::kafka::common::{KafkaConnectionConfig, KafkaTopicConfig};
    use common_wal::config::kafka::MetasrvKafkaConfig;
-    use common_wal::maybe_skip_kafka_integration_test;
-    use common_wal::test_util::get_kafka_endpoints;
+    use common_wal::test_util::run_test_with_kafka_wal;

    use super::*;
    use crate::error::Error;
    use crate::kv_backend::memory::MemoryKvBackend;
-    use crate::test_util::test_kafka_topic_pool;
-    use crate::wal_options_allocator::selector::RoundRobinTopicSelector;

    // Tests that the wal options allocator could successfully allocate raft-engine wal options.
    #[tokio::test]
@@ -202,42 +197,55 @@ mod tests {
        assert_matches!(got, Error::InvalidTopicNamePrefix { .. });
    }

+    // Tests that the wal options allocator could successfully allocate Kafka wal options.
    #[tokio::test]
-    async fn test_allocator_with_kafka_allocate_wal_options() {
-        common_telemetry::init_default_ut_logging();
-        maybe_skip_kafka_integration_test!();
-        let num_topics = 5;
-        let mut topic_pool = test_kafka_topic_pool(
-            get_kafka_endpoints(),
-            num_topics,
-            true,
-            Some("test_allocator_with_kafka"),
-        )
-        .await;
-        topic_pool.selector = Arc::new(RoundRobinTopicSelector::default());
-        let topics = topic_pool.topics.clone();
-        // clean up the topics before test
-        let topic_creator = topic_pool.topic_creator();
-        topic_creator.delete_topics(&topics).await.unwrap();
+    async fn test_allocator_with_kafka() {
+        run_test_with_kafka_wal(|broker_endpoints| {
+            Box::pin(async {
+                let topics = (0..256)
+                    .map(|i| format!("test_allocator_with_kafka_{}_{}", i, uuid::Uuid::new_v4()))
+                    .collect::<Vec<_>>();

-        // Creates an options allocator.
-        let allocator = WalOptionsAllocator::Kafka(topic_pool);
-        allocator.start().await.unwrap();
+                // Creates a topic manager.
+                let kafka_topic = KafkaTopicConfig {
+                    replication_factor: broker_endpoints.len() as i16,
+                    ..Default::default()
+                };
+                let config = MetasrvKafkaConfig {
+                    connection: KafkaConnectionConfig {
+                        broker_endpoints,
+                        ..Default::default()
+                    },
+                    kafka_topic,
+                    ..Default::default()
+                };
+                let kv_backend = Arc::new(MemoryKvBackend::new()) as KvBackendRef;
+                let topic_creator = build_kafka_topic_creator(&config).await.unwrap();
+                let mut topic_pool = KafkaTopicPool::new(&config, kv_backend, topic_creator);
+                topic_pool.topics.clone_from(&topics);
+                topic_pool.selector = Arc::new(selector::RoundRobinTopicSelector::default());

-        let num_regions = 3;
-        let regions = (0..num_regions).collect::<Vec<_>>();
-        let got = allocate_region_wal_options(regions.clone(), &allocator, false).unwrap();
+                // Creates an options allocator.
+                let allocator = WalOptionsAllocator::Kafka(topic_pool);
+                allocator.start().await.unwrap();

-        // Check the allocated wal options contain the expected topics.
-        let expected = (0..num_regions)
-            .map(|i| {
-                let options = WalOptions::Kafka(KafkaWalOptions {
-                    topic: topics[i as usize].clone(),
-                });
-                (i, serde_json::to_string(&options).unwrap())
+                let num_regions = 32;
+                let regions = (0..num_regions).collect::<Vec<_>>();
+                let got = allocate_region_wal_options(regions.clone(), &allocator, false).unwrap();
+
+                // Check the allocated wal options contain the expected topics.
+                let expected = (0..num_regions)
+                    .map(|i| {
+                        let options = WalOptions::Kafka(KafkaWalOptions {
+                            topic: topics[i as usize].clone(),
+                        });
+                        (i, serde_json::to_string(&options).unwrap())
+                    })
+                    .collect::<HashMap<_, _>>();
+                assert_eq!(got, expected);
            })
-            .collect::<HashMap<_, _>>();
-        assert_eq!(got, expected);
+        })
+        .await;
    }

    #[tokio::test]
--- a/src/common/meta/src/wal_options_allocator/topic_creator.rs
+++ b/src/common/meta/src/wal_options_allocator/topic_creator.rs
@@ -12,21 +12,20 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.

-use common_telemetry::{debug, error, info};
-use common_wal::config::kafka::common::{
-    KafkaConnectionConfig, KafkaTopicConfig, DEFAULT_BACKOFF_CONFIG,
-};
+use common_telemetry::{error, info};
+use common_wal::config::kafka::common::DEFAULT_BACKOFF_CONFIG;
+use common_wal::config::kafka::MetasrvKafkaConfig;
 use rskafka::client::error::Error as RsKafkaError;
 use rskafka::client::error::ProtocolError::TopicAlreadyExists;
-use rskafka::client::partition::{Compression, OffsetAt, PartitionClient, UnknownTopicHandling};
+use rskafka::client::partition::{Compression, UnknownTopicHandling};
 use rskafka::client::{Client, ClientBuilder};
 use rskafka::record::Record;
 use snafu::ResultExt;

 use crate::error::{
-    BuildKafkaClientSnafu, BuildKafkaCtrlClientSnafu, CreateKafkaWalTopicSnafu,
-    KafkaGetOffsetSnafu, KafkaPartitionClientSnafu, ProduceRecordSnafu, ResolveKafkaEndpointSnafu,
-    Result, TlsConfigSnafu,
+    BuildKafkaClientSnafu, BuildKafkaCtrlClientSnafu, BuildKafkaPartitionClientSnafu,
+    CreateKafkaWalTopicSnafu, ProduceRecordSnafu, ResolveKafkaEndpointSnafu, Result,
+    TlsConfigSnafu,
 };

 // Each topic only has one partition for now.
@@ -71,47 +70,21 @@ impl KafkaTopicCreator {
                    info!("The topic {} already exists", topic);
                    Ok(())
                } else {
-                    error!(e; "Failed to create a topic {}", topic);
+                    error!("Failed to create a topic {}, error {:?}", topic, e);
                    Err(e).context(CreateKafkaWalTopicSnafu)
                }
            }
        }
    }

-    async fn prepare_topic(&self, topic: &String) -> Result<()> {
-        let partition_client = self.partition_client(topic).await?;
-        self.append_noop_record(topic, &partition_client).await?;
-        Ok(())
-    }
-
-    /// Creates a [PartitionClient] for the given topic.
-    async fn partition_client(&self, topic: &str) -> Result<PartitionClient> {
-        self.client
+    async fn append_noop_record(&self, topic: &String, client: &Client) -> Result<()> {
+        let partition_client = client
            .partition_client(topic, DEFAULT_PARTITION, UnknownTopicHandling::Retry)
            .await
-            .context(KafkaPartitionClientSnafu {
+            .context(BuildKafkaPartitionClientSnafu {
                topic,
                partition: DEFAULT_PARTITION,
-            })
-    }
-
-    /// Appends a noop record to the topic.
-    /// It only appends a noop record if the topic is empty.
-    async fn append_noop_record(
-        &self,
-        topic: &String,
-        partition_client: &PartitionClient,
-    ) -> Result<()> {
-        let end_offset = partition_client
-            .get_offset(OffsetAt::Latest)
-            .await
-            .context(KafkaGetOffsetSnafu {
-                topic: topic.to_string(),
-                partition: DEFAULT_PARTITION,
            })?;
-        if end_offset > 0 {
-            return Ok(());
-        }

        partition_client
            .produce(
@@ -125,28 +98,22 @@ impl KafkaTopicCreator {
            )
            .await
            .context(ProduceRecordSnafu { topic })?;
-        debug!("Appended a noop record to topic {}", topic);

        Ok(())
    }

-    /// Creates topics in Kafka.
-    pub async fn create_topics(&self, topics: &[String]) -> Result<()> {
-        let tasks = topics
-            .iter()
-            .map(|topic| async { self.create_topic(topic, &self.client).await })
-            .collect::<Vec<_>>();
-        futures::future::try_join_all(tasks).await.map(|_| ())
-    }
-
    /// Prepares topics in Kafka.
-    ///
-    /// It appends a noop record to each topic if the topic is empty.
-    pub async fn prepare_topics(&self, topics: &[String]) -> Result<()> {
+    /// 1. Creates missing topics.
+    /// 2. Appends a noop record to each topic.
+    pub async fn prepare_topics(&self, topics: &[&String]) -> Result<()> {
        // Try to create missing topics.
        let tasks = topics
            .iter()
-            .map(|topic| async { self.prepare_topic(topic).await })
+            .map(|topic| async {
+                self.create_topic(topic, &self.client).await?;
+                self.append_noop_record(topic, &self.client).await?;
+                Ok(())
+            })
            .collect::<Vec<_>>();
        futures::future::try_join_all(tasks).await.map(|_| ())
    }
@@ -162,244 +129,34 @@ impl KafkaTopicCreator {
    }
 }

-#[cfg(test)]
-impl KafkaTopicCreator {
-    pub async fn delete_topics(&self, topics: &[String]) -> Result<()> {
-        let tasks = topics
-            .iter()
-            .map(|topic| async { self.delete_topic(topic, &self.client).await })
-            .collect::<Vec<_>>();
-        futures::future::try_join_all(tasks).await.map(|_| ())
-    }
-
-    async fn delete_topic(&self, topic: &String, client: &Client) -> Result<()> {
-        let controller = client
-            .controller_client()
-            .context(BuildKafkaCtrlClientSnafu)?;
-        match controller.delete_topic(topic, 10).await {
-            Ok(_) => {
-                info!("Successfully deleted topic {}", topic);
-                Ok(())
-            }
-            Err(e) => {
-                if Self::is_unknown_topic_err(&e) {
-                    info!("The topic {} does not exist", topic);
-                    Ok(())
-                } else {
-                    panic!("Failed to delete a topic {}, error: {}", topic, e);
-                }
-            }
-        }
-    }
-
-    fn is_unknown_topic_err(e: &RsKafkaError) -> bool {
-        matches!(
-            e,
-            &RsKafkaError::ServerError {
-                protocol_error: rskafka::client::error::ProtocolError::UnknownTopicOrPartition,
-                ..
-            }
-        )
-    }
-
-    pub async fn get_partition_client(&self, topic: &str) -> PartitionClient {
-        self.partition_client(topic).await.unwrap()
-    }
-}
 /// Builds a kafka [Client](rskafka::client::Client).
-pub async fn build_kafka_client(connection: &KafkaConnectionConfig) -> Result<Client> {
+pub async fn build_kafka_client(config: &MetasrvKafkaConfig) -> Result<Client> {
    // Builds an kafka controller client for creating topics.
-    let broker_endpoints = common_wal::resolve_to_ipv4(&connection.broker_endpoints)
+    let broker_endpoints = common_wal::resolve_to_ipv4(&config.connection.broker_endpoints)
        .await
        .context(ResolveKafkaEndpointSnafu)?;
    let mut builder = ClientBuilder::new(broker_endpoints).backoff_config(DEFAULT_BACKOFF_CONFIG);
-    if let Some(sasl) = &connection.sasl {
+    if let Some(sasl) = &config.connection.sasl {
        builder = builder.sasl_config(sasl.config.clone().into_sasl_config());
    };
-    if let Some(tls) = &connection.tls {
+    if let Some(tls) = &config.connection.tls {
        builder = builder.tls_config(tls.to_tls_config().await.context(TlsConfigSnafu)?)
    };
    builder
        .build()
        .await
        .with_context(|_| BuildKafkaClientSnafu {
-            broker_endpoints: connection.broker_endpoints.clone(),
+            broker_endpoints: config.connection.broker_endpoints.clone(),
        })
 }

 /// Builds a [KafkaTopicCreator].
-pub async fn build_kafka_topic_creator(
-    connection: &KafkaConnectionConfig,
-    kafka_topic: &KafkaTopicConfig,
-) -> Result<KafkaTopicCreator> {
-    let client = build_kafka_client(connection).await?;
+pub async fn build_kafka_topic_creator(config: &MetasrvKafkaConfig) -> Result<KafkaTopicCreator> {
+    let client = build_kafka_client(config).await?;
    Ok(KafkaTopicCreator {
        client,
-        num_partitions: kafka_topic.num_partitions,
-        replication_factor: kafka_topic.replication_factor,
-        create_topic_timeout: kafka_topic.create_topic_timeout.as_millis() as i32,
+        num_partitions: config.kafka_topic.num_partitions,
+        replication_factor: config.kafka_topic.replication_factor,
+        create_topic_timeout: config.kafka_topic.create_topic_timeout.as_millis() as i32,
    })
 }
-
-#[cfg(test)]
-mod tests {
-    use common_wal::config::kafka::common::{KafkaConnectionConfig, KafkaTopicConfig};
-    use common_wal::maybe_skip_kafka_integration_test;
-    use common_wal::test_util::get_kafka_endpoints;
-
-    use super::*;
-
-    async fn test_topic_creator(broker_endpoints: Vec<String>) -> KafkaTopicCreator {
-        let connection = KafkaConnectionConfig {
-            broker_endpoints,
-            ..Default::default()
-        };
-        let kafka_topic = KafkaTopicConfig::default();
-
-        build_kafka_topic_creator(&connection, &kafka_topic)
-            .await
-            .unwrap()
-    }
-
-    async fn append_records(partition_client: &PartitionClient, num_records: usize) -> Result<()> {
-        for i in 0..num_records {
-            partition_client
-                .produce(
-                    vec![Record {
-                        key: Some(b"test".to_vec()),
-                        value: Some(format!("test {}", i).as_bytes().to_vec()),
-                        timestamp: chrono::Utc::now(),
-                        headers: Default::default(),
-                    }],
-                    Compression::Lz4,
-                )
-                .await
-                .unwrap();
-        }
-        Ok(())
-    }
-
-    #[tokio::test]
-    async fn test_append_noop_record_to_empty_topic() {
-        common_telemetry::init_default_ut_logging();
-        maybe_skip_kafka_integration_test!();
-        let prefix = "append_noop_record_to_empty_topic";
-        let creator = test_topic_creator(get_kafka_endpoints()).await;
-
-        let topic = format!("{}{}", prefix, "0");
-        // Clean up the topics before test
-        creator.delete_topics(&[topic.to_string()]).await.unwrap();
-        creator.create_topics(&[topic.to_string()]).await.unwrap();
-
-        let partition_client = creator.partition_client(&topic).await.unwrap();
-        let end_offset = partition_client.get_offset(OffsetAt::Latest).await.unwrap();
-        assert_eq!(end_offset, 0);
-
-        // The topic is not empty, so no noop record is appended.
-        creator
-            .append_noop_record(&topic, &partition_client)
-            .await
-            .unwrap();
-        let end_offset = partition_client.get_offset(OffsetAt::Latest).await.unwrap();
-        assert_eq!(end_offset, 1);
-    }
-
-    #[tokio::test]
-    async fn test_append_noop_record_to_non_empty_topic() {
-        common_telemetry::init_default_ut_logging();
-        maybe_skip_kafka_integration_test!();
-        let prefix = "append_noop_record_to_non_empty_topic";
-        let creator = test_topic_creator(get_kafka_endpoints()).await;
-
-        let topic = format!("{}{}", prefix, "0");
-        // Clean up the topics before test
-        creator.delete_topics(&[topic.to_string()]).await.unwrap();
-
-        creator.create_topics(&[topic.to_string()]).await.unwrap();
-        let partition_client = creator.partition_client(&topic).await.unwrap();
-        append_records(&partition_client, 2).await.unwrap();
-
-        let end_offset = partition_client.get_offset(OffsetAt::Latest).await.unwrap();
-        assert_eq!(end_offset, 2);
-
-        // The topic is not empty, so no noop record is appended.
-        creator
-            .append_noop_record(&topic, &partition_client)
-            .await
-            .unwrap();
-        let end_offset = partition_client.get_offset(OffsetAt::Latest).await.unwrap();
-        assert_eq!(end_offset, 2);
-    }
-
-    #[tokio::test]
-    async fn test_create_topic() {
-        common_telemetry::init_default_ut_logging();
-        maybe_skip_kafka_integration_test!();
-        let prefix = "create_topic";
-        let creator = test_topic_creator(get_kafka_endpoints()).await;
-
-        let topic = format!("{}{}", prefix, "0");
-        // Clean up the topics before test
-        creator.delete_topics(&[topic.to_string()]).await.unwrap();
-
-        creator.create_topics(&[topic.to_string()]).await.unwrap();
-        // Should be ok
-        creator.create_topics(&[topic.to_string()]).await.unwrap();
-
-        let partition_client = creator.partition_client(&topic).await.unwrap();
-        let end_offset = partition_client.get_offset(OffsetAt::Latest).await.unwrap();
-        assert_eq!(end_offset, 0);
-    }
-
-    #[tokio::test]
-    async fn test_prepare_topic() {
-        common_telemetry::init_default_ut_logging();
-        maybe_skip_kafka_integration_test!();
-        let prefix = "prepare_topic";
-        let creator = test_topic_creator(get_kafka_endpoints()).await;
-
-        let topic = format!("{}{}", prefix, "0");
-        // Clean up the topics before test
-        creator.delete_topics(&[topic.to_string()]).await.unwrap();
-
-        creator.create_topics(&[topic.to_string()]).await.unwrap();
-        creator.prepare_topic(&topic).await.unwrap();
-
-        let partition_client = creator.partition_client(&topic).await.unwrap();
-        let start_offset = partition_client
-            .get_offset(OffsetAt::Earliest)
-            .await
-            .unwrap();
-        assert_eq!(start_offset, 0);
-
-        let end_offset = partition_client.get_offset(OffsetAt::Latest).await.unwrap();
-        assert_eq!(end_offset, 1);
-    }
-
-    #[tokio::test]
-    async fn test_prepare_topic_with_stale_records_without_pruning() {
-        common_telemetry::init_default_ut_logging();
-        maybe_skip_kafka_integration_test!();
-
-        let prefix = "prepare_topic_with_stale_records_without_pruning";
-        let creator = test_topic_creator(get_kafka_endpoints()).await;
-
-        let topic = format!("{}{}", prefix, "0");
-        // Clean up the topics before test
-        creator.delete_topics(&[topic.to_string()]).await.unwrap();
-
-        creator.create_topics(&[topic.to_string()]).await.unwrap();
-        let partition_client = creator.partition_client(&topic).await.unwrap();
-        append_records(&partition_client, 10).await.unwrap();
-
-        creator.prepare_topic(&topic).await.unwrap();
-
-        let end_offset = partition_client.get_offset(OffsetAt::Latest).await.unwrap();
-        assert_eq!(end_offset, 10);
-        let start_offset = partition_client
-            .get_offset(OffsetAt::Earliest)
-            .await
-            .unwrap();
-        assert_eq!(start_offset, 0);
-    }
-}
--- a/src/common/meta/src/wal_options_allocator/topic_manager.rs
+++ b/src/common/meta/src/wal_options_allocator/topic_manager.rs
@@ -40,21 +40,24 @@ impl KafkaTopicManager {
        Ok(topics)
    }

-    /// Returns the topics that are not prepared.
-    pub async fn unprepare_topics(&self, all_topics: &[String]) -> Result<Vec<String>> {
+    /// Restores topics from the key-value backend. and returns the topics that are not stored in kvbackend.
+    pub async fn get_topics_to_create<'a>(
+        &self,
+        all_topics: &'a [String],
+    ) -> Result<Vec<&'a String>> {
        let existing_topics = self.restore_topics().await?;
        let existing_topic_set = existing_topics.iter().collect::<HashSet<_>>();
        let mut topics_to_create = Vec::with_capacity(all_topics.len());
        for topic in all_topics {
            if !existing_topic_set.contains(topic) {
-                topics_to_create.push(topic.to_string());
+                topics_to_create.push(topic);
            }
        }
        Ok(topics_to_create)
    }

-    /// Persists prepared topics into the key-value backend.
-    pub async fn persist_prepared_topics(&self, topics: &[String]) -> Result<()> {
+    /// Persists topics into the key-value backend.
+    pub async fn persist_topics(&self, topics: &[String]) -> Result<()> {
        self.topic_name_manager
            .batch_put(
                topics
@@ -67,14 +70,6 @@ impl KafkaTopicManager {
    }
 }

-#[cfg(test)]
-impl KafkaTopicManager {
-    /// Lists all topics in the key-value backend.
-    pub async fn list_topics(&self) -> Result<Vec<String>> {
-        self.topic_name_manager.range().await
-    }
-}
-
 #[cfg(test)]
 mod tests {
    use std::sync::Arc;
@@ -95,11 +90,11 @@ mod tests {

        // No legacy topics.
        let mut topics_to_be_created = topic_kvbackend_manager
-            .unprepare_topics(&all_topics)
+            .get_topics_to_create(&all_topics)
            .await
            .unwrap();
        topics_to_be_created.sort();
-        let mut expected = all_topics.clone();
+        let mut expected = all_topics.iter().collect::<Vec<_>>();
        expected.sort();
        assert_eq!(expected, topics_to_be_created);

@@ -114,7 +109,7 @@ mod tests {
        assert!(res.prev_kv.is_none());

        let topics_to_be_created = topic_kvbackend_manager
-            .unprepare_topics(&all_topics)
+            .get_topics_to_create(&all_topics)
            .await
            .unwrap();
        assert!(topics_to_be_created.is_empty());
@@ -149,21 +144,21 @@ mod tests {
        let topic_kvbackend_manager = KafkaTopicManager::new(kv_backend);

        let mut topics_to_be_created = topic_kvbackend_manager
-            .unprepare_topics(&all_topics)
+            .get_topics_to_create(&all_topics)
            .await
            .unwrap();
        topics_to_be_created.sort();
-        let mut expected = all_topics.clone();
+        let mut expected = all_topics.iter().collect::<Vec<_>>();
        expected.sort();
        assert_eq!(expected, topics_to_be_created);

        // Persists topics to kv backend.
        topic_kvbackend_manager
-            .persist_prepared_topics(&all_topics)
+            .persist_topics(&all_topics)
            .await
            .unwrap();
        let topics_to_be_created = topic_kvbackend_manager
-            .unprepare_topics(&all_topics)
+            .get_topics_to_create(&all_topics)
            .await
            .unwrap();
        assert!(topics_to_be_created.is_empty());
--- a/src/common/meta/src/wal_options_allocator/topic_pool.rs
+++ b/src/common/meta/src/wal_options_allocator/topic_pool.rs
@@ -15,7 +15,6 @@
 use std::fmt::{self, Formatter};
 use std::sync::Arc;

-use common_telemetry::info;
 use common_wal::config::kafka::MetasrvKafkaConfig;
 use common_wal::TopicSelectorType;
 use snafu::ensure;
@@ -78,35 +77,27 @@ impl KafkaTopicPool {
    }

    /// Tries to activate the topic manager when metasrv becomes the leader.
-    ///
    /// First tries to restore persisted topics from the kv backend.
-    /// If there are unprepared topics (topics that exist in the configuration but not in the kv backend),
-    /// it will create these topics in Kafka if `auto_create_topics` is enabled.
-    ///
-    /// Then it prepares all unprepared topics by appending a noop record if the topic is empty,
-    /// and persists them in the kv backend for future use.
+    /// If not enough topics retrieved, it will try to contact the Kafka cluster and request creating more topics.
    pub async fn activate(&self) -> Result<()> {
+        if !self.auto_create_topics {
+            return Ok(());
+        }
+
        let num_topics = self.topics.len();
        ensure!(num_topics > 0, InvalidNumTopicsSnafu { num_topics });

-        let unprepared_topics = self.topic_manager.unprepare_topics(&self.topics).await?;
+        let topics_to_be_created = self
+            .topic_manager
+            .get_topics_to_create(&self.topics)
+            .await?;

-        if !unprepared_topics.is_empty() {
-            if self.auto_create_topics {
-                info!("Creating {} topics", unprepared_topics.len());
-                self.topic_creator.create_topics(&unprepared_topics).await?;
-            } else {
-                info!("Auto create topics is disabled, skipping topic creation.");
-            }
+        if !topics_to_be_created.is_empty() {
            self.topic_creator
-                .prepare_topics(&unprepared_topics)
-                .await?;
-            self.topic_manager
-                .persist_prepared_topics(&unprepared_topics)
+                .prepare_topics(&topics_to_be_created)
                .await?;
+            self.topic_manager.persist_topics(&self.topics).await?;
        }
-        info!("Activated topic pool with {} topics", self.topics.len());
-
        Ok(())
    }

@@ -123,147 +114,77 @@ impl KafkaTopicPool {
    }
 }

-#[cfg(test)]
-impl KafkaTopicPool {
-    pub(crate) fn topic_manager(&self) -> &KafkaTopicManager {
-        &self.topic_manager
-    }
-
-    pub(crate) fn topic_creator(&self) -> &KafkaTopicCreator {
-        &self.topic_creator
-    }
-}
-
 #[cfg(test)]
 mod tests {
-    use std::assert_matches::assert_matches;
-
-    use common_wal::maybe_skip_kafka_integration_test;
-    use common_wal::test_util::get_kafka_endpoints;
+    use common_wal::config::kafka::common::{KafkaConnectionConfig, KafkaTopicConfig};
+    use common_wal::test_util::run_test_with_kafka_wal;

    use super::*;
-    use crate::error::Error;
-    use crate::test_util::test_kafka_topic_pool;
-    use crate::wal_options_allocator::selector::RoundRobinTopicSelector;
-
-    #[tokio::test]
-    async fn test_pool_invalid_number_topics_err() {
-        common_telemetry::init_default_ut_logging();
-        maybe_skip_kafka_integration_test!();
-        let endpoints = get_kafka_endpoints();
-
-        let pool = test_kafka_topic_pool(endpoints.clone(), 0, false, None).await;
-        let err = pool.activate().await.unwrap_err();
-        assert_matches!(err, Error::InvalidNumTopics { .. });
-
-        let pool = test_kafka_topic_pool(endpoints, 0, true, None).await;
-        let err = pool.activate().await.unwrap_err();
-        assert_matches!(err, Error::InvalidNumTopics { .. });
-    }
-
-    #[tokio::test]
-    async fn test_pool_activate_unknown_topics_err() {
-        common_telemetry::init_default_ut_logging();
-        maybe_skip_kafka_integration_test!();
-        let pool =
-            test_kafka_topic_pool(get_kafka_endpoints(), 1, false, Some("unknown_topic")).await;
-        let err = pool.activate().await.unwrap_err();
-        assert_matches!(err, Error::KafkaPartitionClient { .. });
-    }
-
-    #[tokio::test]
-    async fn test_pool_activate() {
-        common_telemetry::init_default_ut_logging();
-        maybe_skip_kafka_integration_test!();
-        let pool =
-            test_kafka_topic_pool(get_kafka_endpoints(), 2, true, Some("pool_activate")).await;
-        // clean up the topics before test
-        let topic_creator = pool.topic_creator();
-        topic_creator.delete_topics(&pool.topics).await.unwrap();
-
-        let topic_manager = pool.topic_manager();
-        pool.activate().await.unwrap();
-        let topics = topic_manager.list_topics().await.unwrap();
-        assert_eq!(topics.len(), 2);
-    }
-
-    #[tokio::test]
-    async fn test_pool_activate_with_existing_topics() {
-        common_telemetry::init_default_ut_logging();
-        maybe_skip_kafka_integration_test!();
-        let prefix = "pool_activate_with_existing_topics";
-        let pool = test_kafka_topic_pool(get_kafka_endpoints(), 2, true, Some(prefix)).await;
-        let topic_creator = pool.topic_creator();
-        topic_creator.delete_topics(&pool.topics).await.unwrap();
-
-        let topic_manager = pool.topic_manager();
-        // persists one topic info, then pool.activate() will create new topics that not persisted.
-        topic_manager
-            .persist_prepared_topics(&pool.topics[0..1])
-            .await
-            .unwrap();
-
-        pool.activate().await.unwrap();
-        let topics = topic_manager.list_topics().await.unwrap();
-        assert_eq!(topics.len(), 2);
-
-        let client = pool.topic_creator().client();
-        let topics = client
-            .list_topics()
-            .await
-            .unwrap()
-            .into_iter()
-            .filter(|t| t.name.starts_with(prefix))
-            .collect::<Vec<_>>();
-        assert_eq!(topics.len(), 1);
-    }
+    use crate::kv_backend::memory::MemoryKvBackend;
+    use crate::wal_options_allocator::topic_creator::build_kafka_topic_creator;

    /// Tests that the topic manager could allocate topics correctly.
    #[tokio::test]
    async fn test_alloc_topics() {
-        common_telemetry::init_default_ut_logging();
-        maybe_skip_kafka_integration_test!();
-        let num_topics = 5;
-        let mut topic_pool = test_kafka_topic_pool(
-            get_kafka_endpoints(),
-            num_topics,
-            true,
-            Some("test_allocator_with_kafka"),
-        )
+        run_test_with_kafka_wal(|broker_endpoints| {
+            Box::pin(async {
+                // Constructs topics that should be created.
+                let topics = (0..256)
+                    .map(|i| format!("test_alloc_topics_{}_{}", i, uuid::Uuid::new_v4()))
+                    .collect::<Vec<_>>();
+
+                // Creates a topic manager.
+                let kafka_topic = KafkaTopicConfig {
+                    replication_factor: broker_endpoints.len() as i16,
+                    ..Default::default()
+                };
+                let config = MetasrvKafkaConfig {
+                    connection: KafkaConnectionConfig {
+                        broker_endpoints,
+                        ..Default::default()
+                    },
+                    kafka_topic,
+                    ..Default::default()
+                };
+                let kv_backend = Arc::new(MemoryKvBackend::new()) as KvBackendRef;
+                let topic_creator = build_kafka_topic_creator(&config).await.unwrap();
+                let mut topic_pool = KafkaTopicPool::new(&config, kv_backend, topic_creator);
+                // Replaces the default topic pool with the constructed topics.
+                topic_pool.topics.clone_from(&topics);
+                // Replaces the default selector with a round-robin selector without shuffled.
+                topic_pool.selector = Arc::new(RoundRobinTopicSelector::default());
+                topic_pool.activate().await.unwrap();
+
+                // Selects exactly the number of `num_topics` topics one by one.
+                let got = (0..topics.len())
+                    .map(|_| topic_pool.select().unwrap())
+                    .cloned()
+                    .collect::<Vec<_>>();
+                assert_eq!(got, topics);
+
+                // Selects exactly the number of `num_topics` topics in a batching manner.
+                let got = topic_pool
+                    .select_batch(topics.len())
+                    .unwrap()
+                    .into_iter()
+                    .map(ToString::to_string)
+                    .collect::<Vec<_>>();
+                assert_eq!(got, topics);
+
+                // Selects more than the number of `num_topics` topics.
+                let got = topic_pool
+                    .select_batch(2 * topics.len())
+                    .unwrap()
+                    .into_iter()
+                    .map(ToString::to_string)
+                    .collect::<Vec<_>>();
+                let expected = vec![topics.clone(); 2]
+                    .into_iter()
+                    .flatten()
+                    .collect::<Vec<_>>();
+                assert_eq!(got, expected);
+            })
+        })
        .await;
-        topic_pool.selector = Arc::new(RoundRobinTopicSelector::default());
-        let topics = topic_pool.topics.clone();
-        // clean up the topics before test
-        let topic_creator = topic_pool.topic_creator();
-        topic_creator.delete_topics(&topics).await.unwrap();
-
-        // Selects exactly the number of `num_topics` topics one by one.
-        let got = (0..topics.len())
-            .map(|_| topic_pool.select().unwrap())
-            .cloned()
-            .collect::<Vec<_>>();
-        assert_eq!(got, topics);
-
-        // Selects exactly the number of `num_topics` topics in a batching manner.
-        let got = topic_pool
-            .select_batch(topics.len())
-            .unwrap()
-            .into_iter()
-            .map(ToString::to_string)
-            .collect::<Vec<_>>();
-        assert_eq!(got, topics);
-
-        // Selects more than the number of `num_topics` topics.
-        let got = topic_pool
-            .select_batch(2 * topics.len())
-            .unwrap()
-            .into_iter()
-            .map(ToString::to_string)
-            .collect::<Vec<_>>();
-        let expected = vec![topics.clone(); 2]
-            .into_iter()
-            .flatten()
-            .collect::<Vec<_>>();
-        assert_eq!(got, expected);
    }
 }
--- a/src/common/query/src/logical_plan/accumulator.rs
+++ b/src/common/query/src/logical_plan/accumulator.rs
@@ -24,7 +24,7 @@ use datatypes::prelude::*;
 use datatypes::vectors::{Helper as VectorHelper, VectorRef};
 use snafu::ResultExt;

-use crate::error::{self, Error, FromScalarValueSnafu, IntoVectorSnafu, Result};
+use crate::error::{self, FromScalarValueSnafu, IntoVectorSnafu, Result};
 use crate::prelude::*;

 pub type AggregateFunctionCreatorRef = Arc<dyn AggregateFunctionCreator>;
@@ -166,8 +166,7 @@ impl DfAccumulator for DfAccumulatorAdaptor {
        let output_type = self.creator.output_type()?;
        let scalar_value = value
            .try_to_scalar_value(&output_type)
-            .context(error::ToScalarValueSnafu)
-            .map_err(Error::from)?;
+            .context(error::ToScalarValueSnafu)?;
        Ok(scalar_value)
    }

--- a/src/common/wal/src/config/kafka/common.rs
+++ b/src/common/wal/src/config/kafka/common.rs
@@ -23,16 +23,11 @@ use serde::{Deserialize, Serialize};
 use snafu::{OptionExt, ResultExt};

 /// The default backoff config for kafka client.
-///
-/// If the operation fails, the client will retry 3 times.
-/// The backoff time is 100ms, 300ms, 900ms.
 pub const DEFAULT_BACKOFF_CONFIG: BackoffConfig = BackoffConfig {
    init_backoff: Duration::from_millis(100),
-    max_backoff: Duration::from_secs(1),
-    base: 3.0,
-    // The deadline shouldn't be too long,
-    // otherwise the client will block the worker loop for a long time.
-    deadline: Some(Duration::from_secs(3)),
+    max_backoff: Duration::from_secs(10),
+    base: 2.0,
+    deadline: Some(Duration::from_secs(120)),
 };

 /// Default interval for auto WAL pruning.
--- a/src/common/wal/src/test_util.rs
+++ b/src/common/wal/src/test_util.rs
@@ -31,33 +31,3 @@ where

    test(endpoints).await
 }
-
-/// Get the kafka endpoints from the environment variable `GT_KAFKA_ENDPOINTS`.
-///
-/// The format of the environment variable is:
-/// ```
-/// GT_KAFKA_ENDPOINTS=localhost:9092,localhost:9093
-/// ```
-pub fn get_kafka_endpoints() -> Vec<String> {
-    let endpoints = std::env::var("GT_KAFKA_ENDPOINTS").unwrap();
-    endpoints
-        .split(',')
-        .map(|s| s.trim().to_string())
-        .collect::<Vec<_>>()
-}
-
-#[macro_export]
-/// Skip the test if the environment variable `GT_KAFKA_ENDPOINTS` is not set.
-///
-/// The format of the environment variable is:
-/// ```
-/// GT_KAFKA_ENDPOINTS=localhost:9092,localhost:9093
-/// ```
-macro_rules! maybe_skip_kafka_integration_test {
-    () => {
-        if std::env::var("GT_KAFKA_ENDPOINTS").is_err() {
-            common_telemetry::warn!("The endpoints is empty, skipping the test");
-            return;
-        }
-    };
-}
--- a/src/datanode/src/datanode.rs
+++ b/src/datanode/src/datanode.rs
@@ -398,46 +398,45 @@ impl DatanodeBuilder {
        schema_metadata_manager: SchemaMetadataManagerRef,
        plugins: Plugins,
    ) -> Result<Vec<RegionEngineRef>> {
-        let mut metric_engine_config = metric_engine::config::EngineConfig::default();
-        let mut mito_engine_config = MitoConfig::default();
-        let mut file_engine_config = file_engine::config::EngineConfig::default();
+        let mut engines = vec![];
+        let mut metric_engine_config = opts.region_engine.iter().find_map(|c| match c {
+            RegionEngineConfig::Metric(config) => Some(config.clone()),
+            _ => None,
+        });

        for engine in &opts.region_engine {
            match engine {
                RegionEngineConfig::Mito(config) => {
-                    mito_engine_config = config.clone();
+                    let mito_engine = Self::build_mito_engine(
+                        opts,
+                        object_store_manager.clone(),
+                        config.clone(),
+                        schema_metadata_manager.clone(),
+                        plugins.clone(),
+                    )
+                    .await?;
+
+                    let metric_engine = MetricEngine::try_new(
+                        mito_engine.clone(),
+                        metric_engine_config.take().unwrap_or_default(),
+                    )
+                    .context(BuildMetricEngineSnafu)?;
+                    engines.push(Arc::new(mito_engine) as _);
+                    engines.push(Arc::new(metric_engine) as _);
                }
                RegionEngineConfig::File(config) => {
-                    file_engine_config = config.clone();
+                    let engine = FileRegionEngine::new(
+                        config.clone(),
+                        object_store_manager.default_object_store().clone(), // TODO: implement custom storage for file engine
+                    );
+                    engines.push(Arc::new(engine) as _);
                }
-                RegionEngineConfig::Metric(metric_config) => {
-                    metric_engine_config = metric_config.clone();
+                RegionEngineConfig::Metric(_) => {
+                    // Already handled in `build_mito_engine`.
                }
            }
        }
-
-        let mito_engine = Self::build_mito_engine(
-            opts,
-            object_store_manager.clone(),
-            mito_engine_config,
-            schema_metadata_manager.clone(),
-            plugins.clone(),
-        )
-        .await?;
-
-        let metric_engine = MetricEngine::try_new(mito_engine.clone(), metric_engine_config)
-            .context(BuildMetricEngineSnafu)?;
-
-        let file_engine = FileRegionEngine::new(
-            file_engine_config,
-            object_store_manager.default_object_store().clone(), // TODO: implement custom storage for file engine
-        );
-
-        Ok(vec![
-            Arc::new(mito_engine) as _,
-            Arc::new(metric_engine) as _,
-            Arc::new(file_engine) as _,
-        ])
+        Ok(engines)
    }

    /// Builds [MitoEngine] according to options.
--- a/src/datanode/src/store.rs
+++ b/src/datanode/src/store.rs
@@ -25,7 +25,6 @@ use std::sync::Arc;
 use std::time::Duration;

 use common_telemetry::{info, warn};
-use mito2::access_layer::{ATOMIC_WRITE_DIR, OLD_ATOMIC_WRITE_DIR};
 use object_store::layers::{LruCacheLayer, RetryInterceptor, RetryLayer};
 use object_store::services::Fs;
 use object_store::util::{join_dir, normalize_dir, with_instrument_layers};
@@ -169,13 +168,9 @@ async fn build_cache_layer(
    if let Some(path) = cache_path.as_ref()
        && !path.trim().is_empty()
    {
-        let atomic_temp_dir = join_dir(path, ATOMIC_WRITE_DIR);
+        let atomic_temp_dir = join_dir(path, ".tmp/");
        clean_temp_dir(&atomic_temp_dir)?;

-        // Compatible code. Remove this after a major release.
-        let old_atomic_temp_dir = join_dir(path, OLD_ATOMIC_WRITE_DIR);
-        clean_temp_dir(&old_atomic_temp_dir)?;
-
        let cache_store = Fs::default()
            .root(path)
            .atomic_write_dir(&atomic_temp_dir)
--- a/src/datanode/src/store/fs.rs
+++ b/src/datanode/src/store/fs.rs
@@ -15,7 +15,6 @@
 use std::{fs, path};

 use common_telemetry::info;
-use mito2::access_layer::{ATOMIC_WRITE_DIR, OLD_ATOMIC_WRITE_DIR};
 use object_store::services::Fs;
 use object_store::util::join_dir;
 use object_store::ObjectStore;
@@ -34,13 +33,9 @@ pub async fn new_fs_object_store(
        .context(error::CreateDirSnafu { dir: data_home })?;
    info!("The file storage home is: {}", data_home);

-    let atomic_write_dir = join_dir(data_home, ATOMIC_WRITE_DIR);
+    let atomic_write_dir = join_dir(data_home, ".tmp/");
    store::clean_temp_dir(&atomic_write_dir)?;

-    // Compatible code. Remove this after a major release.
-    let old_atomic_temp_dir = join_dir(data_home, OLD_ATOMIC_WRITE_DIR);
-    store::clean_temp_dir(&old_atomic_temp_dir)?;
-
    let builder = Fs::default()
        .root(data_home)
        .atomic_write_dir(&atomic_write_dir);
--- a/src/datatypes/src/schema/constraint.rs
+++ b/src/datatypes/src/schema/constraint.rs
@@ -253,9 +253,10 @@ fn create_current_timestamp_vector(
    data_type: &ConcreteDataType,
    num_rows: usize,
 ) -> Result<VectorRef> {
-    let current_timestamp_vector = TimestampMillisecondVector::from_values(
-        std::iter::repeat(util::current_time_millis()).take(num_rows),
-    );
+    let current_timestamp_vector = TimestampMillisecondVector::from_values(std::iter::repeat_n(
+        util::current_time_millis(),
+        num_rows,
+    ));
    if data_type.is_timestamp() {
        current_timestamp_vector.cast(data_type)
    } else {
--- a/src/datatypes/src/vectors/constant.rs
+++ b/src/datatypes/src/vectors/constant.rs
@@ -198,8 +198,7 @@ impl fmt::Debug for ConstantVector {

 impl Serializable for ConstantVector {
    fn serialize_to_json(&self) -> Result<Vec<serde_json::Value>> {
-        std::iter::repeat(self.get(0))
-            .take(self.len())
+        std::iter::repeat_n(self.get(0), self.len())
            .map(serde_json::Value::try_from)
            .collect::<serde_json::Result<_>>()
            .context(SerializeSnafu)
--- a/src/datatypes/src/vectors/decimal.rs
+++ b/src/datatypes/src/vectors/decimal.rs
@@ -412,7 +412,7 @@ pub(crate) fn replicate_decimal128(
                    // Safety: std::iter::Repeat and std::iter::Take implement TrustedLen.
                    builder
                        .mutable_array
-                        .append_trusted_len_iter(std::iter::repeat(data).take(repeat_times));
+                        .append_trusted_len_iter(std::iter::repeat_n(data, repeat_times));
                }
            }
            None => {
--- a/src/datatypes/src/vectors/dictionary.rs
+++ b/src/datatypes/src/vectors/dictionary.rs
@@ -16,8 +16,8 @@ use std::any::Any;
 use std::sync::Arc;

 use arrow::array::Array;
-use arrow::datatypes::Int32Type;
-use arrow_array::{ArrayRef, DictionaryArray, Int32Array};
+use arrow::datatypes::Int64Type;
+use arrow_array::{ArrayRef, DictionaryArray, Int64Array};
 use serde_json::Value as JsonValue;
 use snafu::ResultExt;

@@ -32,7 +32,7 @@ use crate::vectors::{self, Helper, Validity, Vector, VectorRef};
 /// Vector of dictionaries, basically backed by Arrow's `DictionaryArray`.
 #[derive(Debug, PartialEq)]
 pub struct DictionaryVector {
-    array: DictionaryArray<Int32Type>,
+    array: DictionaryArray<Int64Type>,
    /// The datatype of the items in the dictionary.
    item_type: ConcreteDataType,
    /// The vector of items in the dictionary.
@@ -41,7 +41,7 @@ pub struct DictionaryVector {

 impl DictionaryVector {
    /// Create a new instance of `DictionaryVector` from a dictionary array and item type
-    pub fn new(array: DictionaryArray<Int32Type>, item_type: ConcreteDataType) -> Result<Self> {
+    pub fn new(array: DictionaryArray<Int64Type>, item_type: ConcreteDataType) -> Result<Self> {
        let item_vector = Helper::try_into_vector(array.values())?;

        Ok(Self {
@@ -52,12 +52,12 @@ impl DictionaryVector {
    }

    /// Returns the underlying Arrow dictionary array
-    pub fn array(&self) -> &DictionaryArray<Int32Type> {
+    pub fn array(&self) -> &DictionaryArray<Int64Type> {
        &self.array
    }

    /// Returns the keys array of this dictionary
-    pub fn keys(&self) -> &arrow_array::PrimitiveArray<Int32Type> {
+    pub fn keys(&self) -> &arrow_array::PrimitiveArray<Int64Type> {
        self.array.keys()
    }

@@ -74,7 +74,7 @@ impl DictionaryVector {
 impl Vector for DictionaryVector {
    fn data_type(&self) -> ConcreteDataType {
        ConcreteDataType::Dictionary(DictionaryType::new(
-            ConcreteDataType::int32_datatype(),
+            ConcreteDataType::int64_datatype(),
            self.item_type.clone(),
        ))
    }
@@ -163,10 +163,10 @@ impl Serializable for DictionaryVector {
    }
 }

-impl TryFrom<DictionaryArray<Int32Type>> for DictionaryVector {
+impl TryFrom<DictionaryArray<Int64Type>> for DictionaryVector {
    type Error = crate::error::Error;

-    fn try_from(array: DictionaryArray<Int32Type>) -> Result<Self> {
+    fn try_from(array: DictionaryArray<Int64Type>) -> Result<Self> {
        let item_type = ConcreteDataType::from_arrow_type(array.values().data_type());
        let item_vector = Helper::try_into_vector(array.values())?;

@@ -243,7 +243,7 @@ impl VectorOp for DictionaryVector {
            previous_offset = offset;
        }

-        let new_keys = Int32Array::from(replicated_keys);
+        let new_keys = Int64Array::from(replicated_keys);
        let new_array = DictionaryArray::try_new(new_keys, self.values().clone())
            .expect("Failed to create replicated dictionary array");

@@ -261,7 +261,7 @@ impl VectorOp for DictionaryVector {
        let filtered_key_array = filtered_key_vector.to_arrow_array();
        let filtered_key_array = filtered_key_array
            .as_any()
-            .downcast_ref::<Int32Array>()
+            .downcast_ref::<Int64Array>()
            .unwrap();

        let new_array = DictionaryArray::try_new(filtered_key_array.clone(), self.values().clone())
@@ -291,7 +291,7 @@ impl VectorOp for DictionaryVector {
        let key_vector = Helper::try_into_vector(&key_array)?;
        let new_key_vector = key_vector.take(indices)?;
        let new_key_array = new_key_vector.to_arrow_array();
-        let new_key_array = new_key_array.as_any().downcast_ref::<Int32Array>().unwrap();
+        let new_key_array = new_key_array.as_any().downcast_ref::<Int64Array>().unwrap();

        let new_array = DictionaryArray::try_new(new_key_array.clone(), self.values().clone())
            .expect("Failed to create filtered dictionary array");
@@ -318,7 +318,7 @@ mod tests {
        // Keys: [0, 1, 2, null, 1, 3]
        // Resulting in: ["a", "b", "c", null, "b", "d"]
        let values = StringArray::from(vec!["a", "b", "c", "d"]);
-        let keys = Int32Array::from(vec![Some(0), Some(1), Some(2), None, Some(1), Some(3)]);
+        let keys = Int64Array::from(vec![Some(0), Some(1), Some(2), None, Some(1), Some(3)]);
        let dict_array = DictionaryArray::new(keys, Arc::new(values));
        DictionaryVector::try_from(dict_array).unwrap()
    }
@@ -404,7 +404,7 @@ mod tests {
        assert_eq!(
            casted.data_type(),
            ConcreteDataType::Dictionary(DictionaryType::new(
-                ConcreteDataType::int32_datatype(),
+                ConcreteDataType::int64_datatype(),
                ConcreteDataType::string_datatype(),
            ))
        );
--- a/src/datatypes/src/vectors/helper.rs
+++ b/src/datatypes/src/vectors/helper.rs
@@ -20,7 +20,7 @@ use std::sync::Arc;
 use arrow::array::{Array, ArrayRef, StringArray};
 use arrow::compute;
 use arrow::compute::kernels::comparison;
-use arrow::datatypes::{DataType as ArrowDataType, Int32Type, TimeUnit};
+use arrow::datatypes::{DataType as ArrowDataType, Int64Type, TimeUnit};
 use arrow_array::DictionaryArray;
 use arrow_schema::IntervalUnit;
 use datafusion_common::ScalarValue;
@@ -348,11 +348,11 @@ impl Helper {
            ArrowDataType::Decimal128(_, _) => {
                Arc::new(Decimal128Vector::try_from_arrow_array(array)?)
            }
-            ArrowDataType::Dictionary(key, value) if matches!(&**key, ArrowDataType::Int32) => {
+            ArrowDataType::Dictionary(key, value) if matches!(&**key, ArrowDataType::Int64) => {
                let array = array
                    .as_ref()
                    .as_any()
-                    .downcast_ref::<DictionaryArray<Int32Type>>()
+                    .downcast_ref::<DictionaryArray<Int64Type>>()
                    .unwrap(); // Safety: the type is guarded by match arm condition
                Arc::new(DictionaryVector::new(
                    array.clone(),
--- a/src/datatypes/src/vectors/null.rs
+++ b/src/datatypes/src/vectors/null.rs
@@ -120,9 +120,7 @@ impl fmt::Debug for NullVector {

 impl Serializable for NullVector {
    fn serialize_to_json(&self) -> Result<Vec<serde_json::Value>> {
-        Ok(std::iter::repeat(serde_json::Value::Null)
-            .take(self.len())
-            .collect())
+        Ok(std::iter::repeat_n(serde_json::Value::Null, self.len()).collect())
    }
 }

--- a/src/datatypes/src/vectors/primitive.rs
+++ b/src/datatypes/src/vectors/primitive.rs
@@ -388,7 +388,7 @@ pub(crate) fn replicate_primitive<T: LogicalPrimitiveType>(
                    // Safety: std::iter::Repeat and std::iter::Take implement TrustedLen.
                    builder
                        .mutable_array
-                        .append_trusted_len_iter(std::iter::repeat(data).take(repeat_times));
+                        .append_trusted_len_iter(std::iter::repeat_n(data, repeat_times));
                }
            }
            None => {
--- a/src/flow/Cargo.toml
+++ b/src/flow/Cargo.toml
@@ -60,7 +60,6 @@ partition.workspace = true
 prometheus.workspace = true
 prost.workspace = true
 query.workspace = true
-rand.workspace = true
 serde.workspace = true
 servers.workspace = true
 session.workspace = true
--- a/src/flow/src/adapter/flownode_impl.rs
+++ b/src/flow/src/adapter/flownode_impl.rs
@@ -14,7 +14,6 @@

 //! impl `FlowNode` trait for FlowNodeManager so standalone can call them
 use std::collections::{HashMap, HashSet};
-use std::sync::atomic::AtomicBool;
 use std::sync::Arc;

 use api::v1::flow::{
@@ -38,12 +37,11 @@ use tokio::sync::{Mutex, RwLock};

 use crate::adapter::{CreateFlowArgs, StreamingEngine};
 use crate::batching_mode::engine::BatchingEngine;
-use crate::batching_mode::{FRONTEND_SCAN_TIMEOUT, MIN_REFRESH_DURATION};
 use crate::engine::FlowEngine;
 use crate::error::{
-    CreateFlowSnafu, ExternalSnafu, FlowNotFoundSnafu, FlowNotRecoveredSnafu,
-    IllegalCheckTaskStateSnafu, InsertIntoFlowSnafu, InternalSnafu, JoinTaskSnafu, ListFlowsSnafu,
-    NoAvailableFrontendSnafu, SyncCheckTaskSnafu, UnexpectedSnafu,
+    CreateFlowSnafu, ExternalSnafu, FlowNotFoundSnafu, IllegalCheckTaskStateSnafu,
+    InsertIntoFlowSnafu, InternalSnafu, JoinTaskSnafu, ListFlowsSnafu, SyncCheckTaskSnafu,
+    UnexpectedSnafu,
 };
 use crate::metrics::METRIC_FLOW_TASK_COUNT;
 use crate::repr::{self, DiffRow};
@@ -64,7 +62,6 @@ pub struct FlowDualEngine {
    flow_metadata_manager: Arc<FlowMetadataManager>,
    catalog_manager: Arc<dyn CatalogManager>,
    check_task: tokio::sync::Mutex<Option<ConsistentCheckTask>>,
-    done_recovering: AtomicBool,
 }

 impl FlowDualEngine {
@@ -81,60 +78,9 @@ impl FlowDualEngine {
            flow_metadata_manager,
            catalog_manager,
            check_task: Mutex::new(None),
-            done_recovering: AtomicBool::new(false),
        }
    }

-    /// Set `done_recovering` to true
-    /// indicate that we are ready to handle requests
-    pub fn set_done_recovering(&self) {
-        info!("FlowDualEngine done recovering");
-        self.done_recovering
-            .store(true, std::sync::atomic::Ordering::Release);
-    }
-
-    /// Check if `done_recovering` is true
-    pub fn is_recover_done(&self) -> bool {
-        self.done_recovering
-            .load(std::sync::atomic::Ordering::Acquire)
-    }
-
-    /// wait for recovering to be done, this will only happen when flownode just started
-    async fn wait_for_all_flow_recover(&self, waiting_req_cnt: usize) -> Result<(), Error> {
-        if self.is_recover_done() {
-            return Ok(());
-        }
-
-        warn!(
-            "FlowDualEngine is not done recovering, {} insert request waiting for recovery",
-            waiting_req_cnt
-        );
-        // wait 3 seconds, check every 1 second
-        // TODO(discord9): make this configurable
-        let mut retry = 0;
-        let max_retry = 3;
-        while retry < max_retry && !self.is_recover_done() {
-            warn!(
-                "FlowDualEngine is not done recovering, retry {} in 1s",
-                retry
-            );
-            tokio::time::sleep(std::time::Duration::from_secs(1)).await;
-            retry += 1;
-        }
-        if retry == max_retry {
-            return FlowNotRecoveredSnafu.fail();
-        } else {
-            info!("FlowDualEngine is done recovering");
-        }
-        // TODO(discord9): also put to centralized logging for flow once it implemented
-        Ok(())
-    }
-
-    /// Determine if the engine is in distributed mode
-    pub fn is_distributed(&self) -> bool {
-        self.streaming_engine.node_id.is_some()
-    }
-
    pub fn streaming_engine(&self) -> Arc<StreamingEngine> {
        self.streaming_engine.clone()
    }
@@ -143,39 +89,6 @@ impl FlowDualEngine {
        self.batching_engine.clone()
    }

-    /// In distributed mode, scan periodically(1s) until available frontend is found, or timeout,
-    /// in standalone mode, return immediately
-    /// notice here if any frontend appear in cluster info this function will return immediately
-    async fn wait_for_available_frontend(&self, timeout: std::time::Duration) -> Result<(), Error> {
-        if !self.is_distributed() {
-            return Ok(());
-        }
-        let frontend_client = self.batching_engine().frontend_client.clone();
-        let sleep_duration = std::time::Duration::from_millis(1_000);
-        let now = std::time::Instant::now();
-        loop {
-            let frontend_list = frontend_client.scan_for_frontend().await?;
-            if !frontend_list.is_empty() {
-                let fe_list = frontend_list
-                    .iter()
-                    .map(|(_, info)| &info.peer.addr)
-                    .collect::<Vec<_>>();
-                info!("Available frontend found: {:?}", fe_list);
-                return Ok(());
-            }
-            let elapsed = now.elapsed();
-            tokio::time::sleep(sleep_duration).await;
-            info!("Waiting for available frontend, elapsed={:?}", elapsed);
-            if elapsed >= timeout {
-                return NoAvailableFrontendSnafu {
-                    timeout,
-                    context: "No available frontend found in cluster info",
-                }
-                .fail();
-            }
-        }
-    }
-
    /// Try to sync with check task, this is only used in drop flow&flush flow, so a flow id is required
    ///
    /// the need to sync is to make sure flush flow actually get called
@@ -283,7 +196,7 @@ impl FlowDualEngine {
                    to_be_created
                );
                let mut errors = vec![];
-                for flow_id in to_be_created.clone() {
+                for flow_id in to_be_created {
                    let flow_id = *flow_id;
                    let info = self
                        .flow_metadata_manager
@@ -342,16 +255,12 @@ impl FlowDualEngine {
                        errors.push((flow_id, err));
                    }
                }
-                if errors.is_empty() {
-                    info!("Recover flows successfully, flows: {:?}", to_be_created);
-                }
-
                for (flow_id, err) in errors {
                    warn!("Failed to recreate flow {}, err={:#?}", flow_id, err);
                }
            } else {
                warn!(
-                    "Flows do not exist in flownode for node {:?}, flow_ids={:?}",
+                    "Flownode {:?} found flows not exist in flownode, flow_ids={:?}",
                    nodeid, to_be_created
                );
            }
@@ -371,7 +280,7 @@ impl FlowDualEngine {
                }
            } else {
                warn!(
-                    "Flows do not exist in metadata for node {:?}, flow_ids={:?}",
+                    "Flownode {:?} found flows not exist in flownode, flow_ids={:?}",
                    nodeid, to_be_dropped
                );
            }
@@ -429,38 +338,18 @@ struct ConsistentCheckTask {

 impl ConsistentCheckTask {
    async fn start_check_task(engine: &Arc<FlowDualEngine>) -> Result<Self, Error> {
-        let engine = engine.clone();
+        // first do recover flows
+        engine.check_flow_consistent(true, false).await?;
+
+        let inner = engine.clone();
        let (tx, mut rx) = tokio::sync::mpsc::channel(1);
        let (trigger_tx, mut trigger_rx) =
            tokio::sync::mpsc::channel::<(bool, bool, tokio::sync::oneshot::Sender<()>)>(10);
        let handle = common_runtime::spawn_global(async move {
-            // first check if available frontend is found
-            if let Err(err) = engine
-                .wait_for_available_frontend(FRONTEND_SCAN_TIMEOUT)
-                .await
-            {
-                warn!("No frontend is available yet:\n {err:?}");
-            }
-
-            // then do recover flows, if failed, always retry
-            let mut recover_retry = 0;
-            while let Err(err) = engine.check_flow_consistent(true, false).await {
-                recover_retry += 1;
-                error!(
-                    "Failed to recover flows:\n {err:?}, retry {} in {}s",
-                    recover_retry,
-                    MIN_REFRESH_DURATION.as_secs()
-                );
-                tokio::time::sleep(MIN_REFRESH_DURATION).await;
-            }
-
-            engine.set_done_recovering();
-
-            // then do check flows, with configurable allow_create and allow_drop
            let (mut allow_create, mut allow_drop) = (false, false);
            let mut ret_signal: Option<tokio::sync::oneshot::Sender<()>> = None;
            loop {
-                if let Err(err) = engine.check_flow_consistent(allow_create, allow_drop).await {
+                if let Err(err) = inner.check_flow_consistent(allow_create, allow_drop).await {
                    error!(err; "Failed to check flow consistent");
                }
                if let Some(done) = ret_signal.take() {
@@ -645,12 +534,7 @@ impl FlowEngine for FlowDualEngine {
        match flow_type {
            Some(FlowType::Batching) => self.batching_engine.flush_flow(flow_id).await,
            Some(FlowType::Streaming) => self.streaming_engine.flush_flow(flow_id).await,
-            None => {
-                warn!(
-                    "Currently flow={flow_id} doesn't exist in flownode, ignore flush_flow request"
-                );
-                Ok(0)
-            }
+            None => Ok(0),
        }
    }

@@ -675,14 +559,11 @@ impl FlowEngine for FlowDualEngine {
        &self,
        request: api::v1::region::InsertRequests,
    ) -> Result<(), Error> {
-        self.wait_for_all_flow_recover(request.requests.len())
-            .await?;
        // TODO(discord9): make as little clone as possible
        let mut to_stream_engine = Vec::with_capacity(request.requests.len());
        let mut to_batch_engine = request.requests;

        {
-            // not locking this, or recover flows will be starved when also handling flow inserts
            let src_table2flow = self.src_table2flow.read().await;
            to_batch_engine.retain(|req| {
                let region_id = RegionId::from(req.region_id);
@@ -818,17 +699,9 @@ fn to_meta_err(
    location: snafu::Location,
 ) -> impl FnOnce(crate::error::Error) -> common_meta::error::Error {
    move |err: crate::error::Error| -> common_meta::error::Error {
-        match err {
-            crate::error::Error::FlowNotFound { id, .. } => {
-                common_meta::error::Error::FlowNotFound {
-                    flow_name: format!("flow_id={id}"),
-                    location,
-                }
-            }
-            _ => common_meta::error::Error::External {
-                location,
-                source: BoxedError::new(err),
-            },
+        common_meta::error::Error::External {
+            location,
+            source: BoxedError::new(err),
        }
    }
 }
--- a/src/flow/src/batching_mode.rs
+++ b/src/flow/src/batching_mode.rs
@@ -31,19 +31,10 @@ pub const DEFAULT_BATCHING_ENGINE_QUERY_TIMEOUT: Duration = Duration::from_secs(
 pub const SLOW_QUERY_THRESHOLD: Duration = Duration::from_secs(60);

 /// The minimum duration between two queries execution by batching mode task
-pub const MIN_REFRESH_DURATION: Duration = Duration::new(5, 0);
+const MIN_REFRESH_DURATION: Duration = Duration::new(5, 0);

 /// Grpc connection timeout
 const GRPC_CONN_TIMEOUT: Duration = Duration::from_secs(5);

 /// Grpc max retry number
 const GRPC_MAX_RETRIES: u32 = 3;
-
-/// Flow wait for available frontend timeout,
-/// if failed to find available frontend after FRONTEND_SCAN_TIMEOUT elapsed, return error
-/// which should prevent flownode from starting
-pub const FRONTEND_SCAN_TIMEOUT: Duration = Duration::from_secs(30);
-
-/// Frontend activity timeout
-/// if frontend is down(not sending heartbeat) for more than FRONTEND_ACTIVITY_TIMEOUT, it will be removed from the list that flownode use to connect
-pub const FRONTEND_ACTIVITY_TIMEOUT: Duration = Duration::from_secs(60);
--- a/src/flow/src/batching_mode/engine.rs
+++ b/src/flow/src/batching_mode/engine.rs
@@ -39,8 +39,7 @@ use crate::batching_mode::time_window::{find_time_window_expr, TimeWindowExpr};
 use crate::batching_mode::utils::sql_to_df_plan;
 use crate::engine::FlowEngine;
 use crate::error::{
-    ExternalSnafu, FlowAlreadyExistSnafu, FlowNotFoundSnafu, TableNotFoundMetaSnafu,
-    UnexpectedSnafu, UnsupportedSnafu,
+    ExternalSnafu, FlowAlreadyExistSnafu, TableNotFoundMetaSnafu, UnexpectedSnafu, UnsupportedSnafu,
 };
 use crate::{CreateFlowArgs, Error, FlowId, TableName};

@@ -50,8 +49,7 @@ use crate::{CreateFlowArgs, Error, FlowId, TableName};
 pub struct BatchingEngine {
    tasks: RwLock<BTreeMap<FlowId, BatchingTask>>,
    shutdown_txs: RwLock<BTreeMap<FlowId, oneshot::Sender<()>>>,
-    /// frontend client for insert request
-    pub(crate) frontend_client: Arc<FrontendClient>,
+    frontend_client: Arc<FrontendClient>,
    flow_metadata_manager: FlowMetadataManagerRef,
    table_meta: TableMetadataManagerRef,
    catalog_manager: CatalogManagerRef,
@@ -331,7 +329,7 @@ impl BatchingEngine {
        let frontend = self.frontend_client.clone();

        // check execute once first to detect any error early
-        task.check_or_create_sink_table(&engine, &frontend).await?;
+        task.check_execute(&engine, &frontend).await?;

        // TODO(discord9): use time wheel or what for better
        let handle = common_runtime::spawn_global(async move {
@@ -350,8 +348,7 @@ impl BatchingEngine {

    pub async fn remove_flow_inner(&self, flow_id: FlowId) -> Result<(), Error> {
        if self.tasks.write().await.remove(&flow_id).is_none() {
-            warn!("Flow {flow_id} not found in tasks");
-            FlowNotFoundSnafu { id: flow_id }.fail()?;
+            warn!("Flow {flow_id} not found in tasks")
        }
        let Some(tx) = self.shutdown_txs.write().await.remove(&flow_id) else {
            UnexpectedSnafu {
@@ -368,7 +365,9 @@ impl BatchingEngine {
    pub async fn flush_flow_inner(&self, flow_id: FlowId) -> Result<usize, Error> {
        debug!("Try flush flow {flow_id}");
        let task = self.tasks.read().await.get(&flow_id).cloned();
-        let task = task.with_context(|| FlowNotFoundSnafu { id: flow_id })?;
+        let task = task.with_context(|| UnexpectedSnafu {
+            reason: format!("Can't found task for flow {flow_id}"),
+        })?;

        task.mark_all_windows_as_dirty()?;

--- a/src/flow/src/batching_mode/frontend_client.rs
+++ b/src/flow/src/batching_mode/frontend_client.rs
@@ -15,7 +15,6 @@
 //! Frontend client to run flow as batching task which is time-window-aware normal query triggered every tick set by user

 use std::sync::{Arc, Weak};
-use std::time::SystemTime;

 use api::v1::greptime_request::Request;
 use api::v1::CreateTableExpr;
@@ -28,17 +27,14 @@ use common_meta::rpc::store::RangeRequest;
 use common_query::Output;
 use common_telemetry::warn;
 use meta_client::client::MetaClient;
-use rand::rng;
-use rand::seq::SliceRandom;
 use servers::query_handler::grpc::GrpcQueryHandler;
 use session::context::{QueryContextBuilder, QueryContextRef};
 use snafu::{OptionExt, ResultExt};

 use crate::batching_mode::{
-    DEFAULT_BATCHING_ENGINE_QUERY_TIMEOUT, FRONTEND_ACTIVITY_TIMEOUT, GRPC_CONN_TIMEOUT,
-    GRPC_MAX_RETRIES,
+    DEFAULT_BATCHING_ENGINE_QUERY_TIMEOUT, GRPC_CONN_TIMEOUT, GRPC_MAX_RETRIES,
 };
-use crate::error::{ExternalSnafu, InvalidRequestSnafu, NoAvailableFrontendSnafu, UnexpectedSnafu};
+use crate::error::{ExternalSnafu, InvalidRequestSnafu, UnexpectedSnafu};
 use crate::Error;

 /// Just like [`GrpcQueryHandler`] but use BoxedError
@@ -131,24 +127,10 @@ impl DatabaseWithPeer {
    fn new(database: Database, peer: Peer) -> Self {
        Self { database, peer }
    }
-
-    /// Try sending a "SELECT 1" to the database
-    async fn try_select_one(&self) -> Result<(), Error> {
-        // notice here use `sql` for `SELECT 1` return 1 row
-        let _ = self
-            .database
-            .sql("SELECT 1")
-            .await
-            .with_context(|_| InvalidRequestSnafu {
-                context: format!("Failed to handle `SELECT 1` request at {:?}", self.peer),
-            })?;
-        Ok(())
-    }
 }

 impl FrontendClient {
-    /// scan for available frontend from metadata
-    pub(crate) async fn scan_for_frontend(&self) -> Result<Vec<(NodeInfoKey, NodeInfo)>, Error> {
+    async fn scan_for_frontend(&self) -> Result<Vec<(NodeInfoKey, NodeInfo)>, Error> {
        let Self::Distributed { meta_client, .. } = self else {
            return Ok(vec![]);
        };
@@ -178,9 +160,8 @@ impl FrontendClient {
        Ok(res)
    }

-    /// Get the frontend with recent enough(less than 1 minute from now) `last_activity_ts`
-    /// and is able to process query
-    async fn get_random_active_frontend(
+    /// Get the database with max `last_activity_ts`
+    async fn get_last_active_frontend(
        &self,
        catalog: &str,
        schema: &str,
@@ -196,50 +177,22 @@ impl FrontendClient {
            .fail();
        };

-        let mut interval = tokio::time::interval(GRPC_CONN_TIMEOUT);
-        interval.tick().await;
-        for retry in 0..GRPC_MAX_RETRIES {
-            let mut frontends = self.scan_for_frontend().await?;
-            let now_in_ms = SystemTime::now()
-                .duration_since(SystemTime::UNIX_EPOCH)
-                .unwrap()
-                .as_millis() as i64;
-            // shuffle the frontends to avoid always pick the same one
-            frontends.shuffle(&mut rng());
+        let frontends = self.scan_for_frontend().await?;
+        let mut peer = None;

-            // found node with maximum last_activity_ts
-            for (_, node_info) in frontends
-                .iter()
-                // filter out frontend that have been down for more than 1 min
-                .filter(|(_, node_info)| {
-                    node_info.last_activity_ts + FRONTEND_ACTIVITY_TIMEOUT.as_millis() as i64
-                        > now_in_ms
-                })
-            {
-                let addr = &node_info.peer.addr;
-                let client = Client::with_manager_and_urls(chnl_mgr.clone(), vec![addr.clone()]);
-                let database = Database::new(catalog, schema, client);
-                let db = DatabaseWithPeer::new(database, node_info.peer.clone());
-                match db.try_select_one().await {
-                    Ok(_) => return Ok(db),
-                    Err(e) => {
-                        warn!(
-                            "Failed to connect to frontend {} on retry={}: \n{e:?}",
-                            addr, retry
-                        );
-                    }
-                }
+        if let Some((_, val)) = frontends.iter().max_by_key(|(_, val)| val.last_activity_ts) {
+            peer = Some(val.peer.clone());
+        }
+
+        let Some(peer) = peer else {
+            UnexpectedSnafu {
+                reason: format!("No frontend available: {:?}", frontends),
            }
-            // no available frontend
-            // sleep and retry
-            interval.tick().await;
-        }
-
-        NoAvailableFrontendSnafu {
-            timeout: GRPC_CONN_TIMEOUT,
-            context: "No available frontend found that is able to process query",
-        }
-        .fail()
+            .fail()?
+        };
+        let client = Client::with_manager_and_urls(chnl_mgr.clone(), vec![peer.addr.clone()]);
+        let database = Database::new(catalog, schema, client);
+        Ok(DatabaseWithPeer::new(database, peer))
    }

    pub async fn create(
@@ -269,18 +222,38 @@ impl FrontendClient {
    ) -> Result<u32, Error> {
        match self {
            FrontendClient::Distributed { .. } => {
-                let db = self.get_random_active_frontend(catalog, schema).await?;
+                let db = self.get_last_active_frontend(catalog, schema).await?;

                *peer_desc = Some(PeerDesc::Dist {
                    peer: db.peer.clone(),
                });

-                db.database
-                    .handle_with_retry(req.clone(), GRPC_MAX_RETRIES)
-                    .await
-                    .with_context(|_| InvalidRequestSnafu {
-                        context: format!("Failed to handle request at {:?}: {:?}", db.peer, req),
-                    })
+                let mut retry = 0;
+
+                loop {
+                    let ret = db.database.handle(req.clone()).await.with_context(|_| {
+                        InvalidRequestSnafu {
+                            context: format!("Failed to handle request: {:?}", req),
+                        }
+                    });
+                    if let Err(err) = ret {
+                        if retry < GRPC_MAX_RETRIES {
+                            retry += 1;
+                            warn!(
+                                "Failed to send request to grpc handle at Peer={:?}, retry = {}, error = {:?}",
+                                db.peer, retry, err
+                            );
+                            continue;
+                        } else {
+                            common_telemetry::error!(
+                                "Failed to send request to grpc handle at Peer={:?} after {} retries, error = {:?}",
+                                db.peer, retry, err
+                            );
+                            return Err(err);
+                        }
+                    }
+                    return ret;
+                }
            }
            FrontendClient::Standalone { database_client } => {
                let ctx = QueryContextBuilder::default()
--- a/src/flow/src/batching_mode/state.rs
+++ b/src/flow/src/batching_mode/state.rs
@@ -71,33 +71,18 @@ impl TaskState {
        self.last_update_time = Instant::now();
    }

-    /// Compute the next query delay based on the time window size or the last query duration.
-    /// Aiming to avoid too frequent queries. But also not too long delay.
-    /// The delay is computed as follows:
-    /// - If `time_window_size` is set, the delay is half the time window size, constrained to be
-    ///   at least `last_query_duration` and at most `max_timeout`.
-    /// - If `time_window_size` is not set, the delay defaults to `last_query_duration`, constrained
-    ///   to be at least `MIN_REFRESH_DURATION` and at most `max_timeout`.
+    /// wait for at least `last_query_duration`, at most `max_timeout` to start next query
    ///
-    /// If there are dirty time windows, the function returns an immediate execution time to clean them.
-    /// TODO: Make this behavior configurable.
+    /// if have more dirty time window, exec next query immediately
    pub fn get_next_start_query_time(
        &self,
        flow_id: FlowId,
-        time_window_size: &Option<Duration>,
        max_timeout: Option<Duration>,
    ) -> Instant {
-        let last_duration = max_timeout
+        let next_duration = max_timeout
            .unwrap_or(self.last_query_duration)
-            .min(self.last_query_duration)
-            .max(MIN_REFRESH_DURATION);
-
-        let next_duration = time_window_size
-            .map(|t| {
-                let half = t / 2;
-                half.max(last_duration)
-            })
-            .unwrap_or(last_duration);
+            .min(self.last_query_duration);
+        let next_duration = next_duration.max(MIN_REFRESH_DURATION);

        // if have dirty time window, execute immediately to clean dirty time window
        if self.dirty_time_windows.windows.is_empty() {
--- a/src/flow/src/batching_mode/task.rs
+++ b/src/flow/src/batching_mode/task.rs
@@ -142,12 +142,26 @@ impl BatchingTask {
        Ok(())
    }

-    /// Create sink table if not exists
-    pub async fn check_or_create_sink_table(
+    /// Test execute, for check syntax or such
+    pub async fn check_execute(
        &self,
        engine: &QueryEngineRef,
        frontend_client: &Arc<FrontendClient>,
    ) -> Result<Option<(u32, Duration)>, Error> {
+        // use current time to test get a dirty time window, which should be safe
+        let start = SystemTime::now();
+        let ts = Timestamp::new_second(
+            start
+                .duration_since(UNIX_EPOCH)
+                .expect("Time went backwards")
+                .as_secs() as _,
+        );
+        self.state
+            .write()
+            .unwrap()
+            .dirty_time_windows
+            .add_lower_bounds(vec![ts].into_iter());
+
        if !self.is_table_exist(&self.config.sink_table_name).await? {
            let create_table = self.gen_create_table_expr(engine.clone()).await?;
            info!(
@@ -160,8 +174,7 @@ impl BatchingTask {
                self.config.sink_table_name.join(".")
            );
        }
-
-        Ok(None)
+        self.gen_exec_once(engine, frontend_client).await
    }

    async fn is_table_exist(&self, table_name: &[String; 3]) -> Result<bool, Error> {
@@ -380,23 +393,6 @@ impl BatchingTask {
        frontend_client: Arc<FrontendClient>,
    ) {
        loop {
-            // first check if shutdown signal is received
-            // if so, break the loop
-            {
-                let mut state = self.state.write().unwrap();
-                match state.shutdown_rx.try_recv() {
-                    Ok(()) => break,
-                    Err(TryRecvError::Closed) => {
-                        warn!(
-                            "Unexpected shutdown flow {}, shutdown anyway",
-                            self.config.flow_id
-                        );
-                        break;
-                    }
-                    Err(TryRecvError::Empty) => (),
-                }
-            }
-
            let mut new_query = None;
            let mut gen_and_exec = async || {
                new_query = self.gen_insert_plan(&engine).await?;
@@ -410,15 +406,20 @@ impl BatchingTask {
                // normal execute, sleep for some time before doing next query
                Ok(Some(_)) => {
                    let sleep_until = {
-                        let state = self.state.write().unwrap();
-
+                        let mut state = self.state.write().unwrap();
+                        match state.shutdown_rx.try_recv() {
+                            Ok(()) => break,
+                            Err(TryRecvError::Closed) => {
+                                warn!(
+                                    "Unexpected shutdown flow {}, shutdown anyway",
+                                    self.config.flow_id
+                                );
+                                break;
+                            }
+                            Err(TryRecvError::Empty) => (),
+                        }
                        state.get_next_start_query_time(
                            self.config.flow_id,
-                            &self
-                                .config
-                                .time_window_expr
-                                .as_ref()
-                                .and_then(|t| *t.time_window_size()),
                            Some(DEFAULT_BATCHING_ENGINE_QUERY_TIMEOUT),
                        )
                    };
--- a/src/flow/src/batching_mode/time_window.rs
+++ b/src/flow/src/batching_mode/time_window.rs
@@ -55,9 +55,6 @@ use crate::error::{
 use crate::expr::error::DataTypeSnafu;
 use crate::Error;

-/// Represents a test timestamp in seconds since the Unix epoch.
-const DEFAULT_TEST_TIMESTAMP: Timestamp = Timestamp::new_second(17_0000_0000);
-
 /// Time window expr like `date_bin(INTERVAL '1' MINUTE, ts)`, this type help with
 /// evaluating the expr using given timestamp
 ///
@@ -73,7 +70,6 @@ pub struct TimeWindowExpr {
    pub column_name: String,
    logical_expr: Expr,
    df_schema: DFSchema,
-    eval_time_window_size: Option<std::time::Duration>,
 }

 impl std::fmt::Display for TimeWindowExpr {
@@ -88,11 +84,6 @@ impl std::fmt::Display for TimeWindowExpr {
 }

 impl TimeWindowExpr {
-    /// The time window size of the expr, get from calling `eval` with a test timestamp
-    pub fn time_window_size(&self) -> &Option<std::time::Duration> {
-        &self.eval_time_window_size
-    }
-
    pub fn from_expr(
        expr: &Expr,
        column_name: &str,
@@ -100,28 +91,12 @@ impl TimeWindowExpr {
        session: &SessionState,
    ) -> Result<Self, Error> {
        let phy_expr: PhysicalExprRef = to_phy_expr(expr, df_schema, session)?;
-        let mut zelf = Self {
+        Ok(Self {
            phy_expr,
            column_name: column_name.to_string(),
            logical_expr: expr.clone(),
            df_schema: df_schema.clone(),
-            eval_time_window_size: None,
-        };
-        let test_ts = DEFAULT_TEST_TIMESTAMP;
-        let (l, u) = zelf.eval(test_ts)?;
-        let time_window_size = match (l, u) {
-            (Some(l), Some(u)) => u.sub(&l).map(|r| r.to_std()).transpose().map_err(|_| {
-                UnexpectedSnafu {
-                    reason: format!(
-                        "Expect upper bound older than lower bound, found upper={u:?} and lower={l:?}"
-                    ),
-                }
-                .build()
-            })?,
-            _ => None,
-        };
-        zelf.eval_time_window_size = time_window_size;
-        Ok(zelf)
+        })
    }

    pub fn eval(
--- a/src/flow/src/error.rs
+++ b/src/flow/src/error.rs
@@ -46,12 +46,6 @@ pub enum Error {
        location: Location,
    },

-    #[snafu(display("Flow engine is still recovering"))]
-    FlowNotRecovered {
-        #[snafu(implicit)]
-        location: Location,
-    },
-
    #[snafu(display("Error encountered while creating flow: {sql}"))]
    CreateFlow {
        sql: String,
@@ -67,16 +61,6 @@ pub enum Error {
        location: Location,
    },

-    #[snafu(display(
-        "No available frontend found after timeout: {timeout:?}, context: {context}"
-    ))]
-    NoAvailableFrontend {
-        timeout: std::time::Duration,
-        context: String,
-        #[snafu(implicit)]
-        location: Location,
-    },
-
    #[snafu(display("External error"))]
    External {
        source: BoxedError,
@@ -312,14 +296,12 @@ impl ErrorExt for Error {
            Self::Eval { .. }
            | Self::JoinTask { .. }
            | Self::Datafusion { .. }
-            | Self::InsertIntoFlow { .. }
-            | Self::NoAvailableFrontend { .. }
-            | Self::FlowNotRecovered { .. } => StatusCode::Internal,
+            | Self::InsertIntoFlow { .. } => StatusCode::Internal,
            Self::FlowAlreadyExist { .. } => StatusCode::TableAlreadyExists,
            Self::TableNotFound { .. }
            | Self::TableNotFoundMeta { .. }
+            | Self::FlowNotFound { .. }
            | Self::ListFlows { .. } => StatusCode::TableNotFound,
-            Self::FlowNotFound { .. } => StatusCode::FlowNotFound,
            Self::Plan { .. } | Self::Datatypes { .. } => StatusCode::PlanQuery,
            Self::CreateFlow { .. } | Self::Arrow { .. } | Self::Time { .. } => {
                StatusCode::EngineExecuteQuery
--- a/src/flow/src/expr/signature.rs
+++ b/src/flow/src/expr/signature.rs
@@ -60,7 +60,7 @@ pub enum GenericFn {
    Mul,
    Div,
    Mod,
-    // variadic func
+    // varadic func
    And,
    Or,
    // unmaterized func
--- a/src/flow/src/server.rs
+++ b/src/flow/src/server.rs
@@ -43,7 +43,7 @@ use servers::error::{StartGrpcSnafu, TcpBindSnafu, TcpIncomingSnafu};
 use servers::http::HttpServerBuilder;
 use servers::metrics_handler::MetricsHandler;
 use servers::server::{ServerHandler, ServerHandlers};
-use session::context::QueryContextRef;
+use session::context::{QueryContextBuilder, QueryContextRef};
 use snafu::{OptionExt, ResultExt};
 use tokio::net::TcpListener;
 use tokio::sync::{broadcast, oneshot, Mutex};
@@ -54,18 +54,18 @@ use tonic::{Request, Response, Status};
 use crate::adapter::flownode_impl::{FlowDualEngine, FlowDualEngineRef};
 use crate::adapter::{create_worker, FlowStreamingEngineRef};
 use crate::batching_mode::engine::BatchingEngine;
+use crate::engine::FlowEngine;
 use crate::error::{
-    to_status_with_last_err, CacheRequiredSnafu, ExternalSnafu, ListFlowsSnafu, ParseAddrSnafu,
-    ShutdownServerSnafu, StartServerSnafu, UnexpectedSnafu,
+    to_status_with_last_err, CacheRequiredSnafu, CreateFlowSnafu, ExternalSnafu, FlowNotFoundSnafu,
+    ListFlowsSnafu, ParseAddrSnafu, ShutdownServerSnafu, StartServerSnafu, UnexpectedSnafu,
 };
 use crate::heartbeat::HeartbeatTask;
 use crate::metrics::{METRIC_FLOW_PROCESSING_TIME, METRIC_FLOW_ROWS};
 use crate::transform::register_function_to_query_engine;
 use crate::utils::{SizeReportSender, StateReportHandler};
-use crate::{Error, FlownodeOptions, FrontendClient, StreamingEngine};
+use crate::{CreateFlowArgs, Error, FlownodeOptions, FrontendClient, StreamingEngine};

 pub const FLOW_NODE_SERVER_NAME: &str = "FLOW_NODE_SERVER";
-
 /// wrapping flow node manager to avoid orphan rule with Arc<...>
 #[derive(Clone)]
 pub struct FlowService {
@@ -172,8 +172,6 @@ impl FlownodeServer {
    }

    /// Start the background task for streaming computation.
-    ///
-    /// Should be called only after heartbeat is establish, hence can get cluster info
    async fn start_workers(&self) -> Result<(), Error> {
        let manager_ref = self.inner.flow_service.dual_engine.clone();
        let handle = manager_ref
@@ -397,6 +395,109 @@ impl FlownodeBuilder {
        Ok(instance)
    }

+    /// recover all flow tasks in this flownode in distributed mode(nodeid is Some(<num>))
+    ///
+    /// or recover all existing flow tasks if in standalone mode(nodeid is None)
+    ///
+    /// TODO(discord9): persistent flow tasks with internal state
+    async fn recover_flows(&self, manager: &FlowDualEngine) -> Result<usize, Error> {
+        let nodeid = self.opts.node_id;
+        let to_be_recovered: Vec<_> = if let Some(nodeid) = nodeid {
+            let to_be_recover = self
+                .flow_metadata_manager
+                .flownode_flow_manager()
+                .flows(nodeid)
+                .try_collect::<Vec<_>>()
+                .await
+                .context(ListFlowsSnafu { id: Some(nodeid) })?;
+            to_be_recover.into_iter().map(|(id, _)| id).collect()
+        } else {
+            let all_catalogs = self
+                .catalog_manager
+                .catalog_names()
+                .await
+                .map_err(BoxedError::new)
+                .context(ExternalSnafu)?;
+            let mut all_flow_ids = vec![];
+            for catalog in all_catalogs {
+                let flows = self
+                    .flow_metadata_manager
+                    .flow_name_manager()
+                    .flow_names(&catalog)
+                    .await
+                    .try_collect::<Vec<_>>()
+                    .await
+                    .map_err(BoxedError::new)
+                    .context(ExternalSnafu)?;
+
+                all_flow_ids.extend(flows.into_iter().map(|(_, id)| id.flow_id()));
+            }
+            all_flow_ids
+        };
+        let cnt = to_be_recovered.len();
+
+        // TODO(discord9): recover in parallel
+        info!("Recovering {} flows: {:?}", cnt, to_be_recovered);
+        for flow_id in to_be_recovered {
+            let info = self
+                .flow_metadata_manager
+                .flow_info_manager()
+                .get(flow_id)
+                .await
+                .map_err(BoxedError::new)
+                .context(ExternalSnafu)?
+                .context(FlowNotFoundSnafu { id: flow_id })?;
+
+            let sink_table_name = [
+                info.sink_table_name().catalog_name.clone(),
+                info.sink_table_name().schema_name.clone(),
+                info.sink_table_name().table_name.clone(),
+            ];
+
+            let args = CreateFlowArgs {
+                flow_id: flow_id as _,
+                sink_table_name,
+                source_table_ids: info.source_table_ids().to_vec(),
+                // because recover should only happen on restart the `create_if_not_exists` and `or_replace` can be arbitrary value(since flow doesn't exist)
+                // but for the sake of consistency and to make sure recover of flow actually happen, we set both to true
+                // (which is also fine since checks for not allow both to be true is on metasrv and we already pass that)
+                create_if_not_exists: true,
+                or_replace: true,
+                expire_after: info.expire_after(),
+                comment: Some(info.comment().clone()),
+                sql: info.raw_sql().clone(),
+                flow_options: info.options().clone(),
+                query_ctx: info
+                    .query_context()
+                    .clone()
+                    .map(|ctx| {
+                        ctx.try_into()
+                            .map_err(BoxedError::new)
+                            .context(ExternalSnafu)
+                    })
+                    .transpose()?
+                    // or use default QueryContext with catalog_name from info
+                    // to keep compatibility with old version
+                    .or_else(|| {
+                        Some(
+                            QueryContextBuilder::default()
+                                .current_catalog(info.catalog_name().to_string())
+                                .build(),
+                        )
+                    }),
+            };
+            manager
+                .create_flow(args)
+                .await
+                .map_err(BoxedError::new)
+                .with_context(|_| CreateFlowSnafu {
+                    sql: info.raw_sql().clone(),
+                })?;
+        }
+
+        Ok(cnt)
+    }
+
    /// build [`FlowWorkerManager`], note this doesn't take ownership of `self`,
    /// nor does it actually start running the worker.
    async fn build_manager(
@@ -581,7 +682,7 @@ impl FrontendInvoker {
            .start_timer();

        self.inserter
-            .handle_row_inserts(requests, ctx, &self.statement_executor, false)
+            .handle_row_inserts(requests, ctx, &self.statement_executor)
            .await
            .map_err(BoxedError::new)
            .context(common_frontend::error::ExternalSnafu)
--- a/src/frontend/src/instance/grpc.rs
+++ b/src/frontend/src/instance/grpc.rs
@@ -72,10 +72,7 @@ impl GrpcQueryHandler for Instance {

        let output = match request {
            Request::Inserts(requests) => self.handle_inserts(requests, ctx.clone()).await?,
-            Request::RowInserts(requests) => {
-                self.handle_row_inserts(requests, ctx.clone(), false)
-                    .await?
-            }
+            Request::RowInserts(requests) => self.handle_row_inserts(requests, ctx.clone()).await?,
            Request::Deletes(requests) => self.handle_deletes(requests, ctx.clone()).await?,
            Request::RowDeletes(requests) => self.handle_row_deletes(requests, ctx.clone()).await?,
            Request::Query(query_request) => {
@@ -410,15 +407,9 @@ impl Instance {
        &self,
        requests: RowInsertRequests,
        ctx: QueryContextRef,
-        accommodate_existing_schema: bool,
    ) -> Result<Output> {
        self.inserter
-            .handle_row_inserts(
-                requests,
-                ctx,
-                self.statement_executor.as_ref(),
-                accommodate_existing_schema,
-            )
+            .handle_row_inserts(requests, ctx, self.statement_executor.as_ref())
            .await
            .context(TableOperationSnafu)
    }
@@ -430,7 +421,7 @@ impl Instance {
        ctx: QueryContextRef,
    ) -> Result<Output> {
        self.inserter
-            .handle_last_non_null_inserts(requests, ctx, self.statement_executor.as_ref(), true)
+            .handle_last_non_null_inserts(requests, ctx, self.statement_executor.as_ref())
            .await
            .context(TableOperationSnafu)
    }
--- a/src/frontend/src/instance/opentsdb.rs
+++ b/src/frontend/src/instance/opentsdb.rs
@@ -53,7 +53,7 @@ impl OpentsdbProtocolHandler for Instance {
        };

        let output = self
-            .handle_row_inserts(requests, ctx, true)
+            .handle_row_inserts(requests, ctx)
            .await
            .map_err(BoxedError::new)
            .context(servers::error::ExecuteGrpcQuerySnafu)?;
--- a/src/frontend/src/instance/otlp.rs
+++ b/src/frontend/src/instance/otlp.rs
@@ -63,7 +63,7 @@ impl OpenTelemetryProtocolHandler for Instance {
            None
        };

-        self.handle_row_inserts(requests, ctx, false)
+        self.handle_row_inserts(requests, ctx)
            .await
            .map_err(BoxedError::new)
            .context(error::ExecuteGrpcQuerySnafu)
--- a/src/frontend/src/instance/prom_store.rs
+++ b/src/frontend/src/instance/prom_store.rs
@@ -195,7 +195,7 @@ impl PromStoreProtocolHandler for Instance {
                .map_err(BoxedError::new)
                .context(error::ExecuteGrpcQuerySnafu)?
        } else {
-            self.handle_row_inserts(request, ctx.clone(), true)
+            self.handle_row_inserts(request, ctx.clone())
                .await
                .map_err(BoxedError::new)
                .context(error::ExecuteGrpcQuerySnafu)?
--- a/src/index/src/inverted_index/create/sort/external_sort.rs
+++ b/src/index/src/inverted_index/create/sort/external_sort.rs
@@ -481,7 +481,7 @@ mod tests {

        let mock_values = dic_values
            .iter()
-            .flat_map(|(value, size)| iter::repeat(value.clone()).take(*size))
+            .flat_map(|(value, size)| std::iter::repeat_n(value.clone(), *size))
            .collect::<Vec<_>>();

        let sorted_result = sorted_result(&mock_values, segment_row_count);
--- a/src/log-store/src/kafka/client_manager.rs
+++ b/src/log-store/src/kafka/client_manager.rs
@@ -182,14 +182,6 @@ impl ClientManager {
    }
 }

-#[cfg(test)]
-impl ClientManager {
-    /// Returns the controller client.
-    pub(crate) fn controller_client(&self) -> rskafka::client::controller::ControllerClient {
-        self.client.controller_client().unwrap()
-    }
-}
-
 #[cfg(test)]
 mod tests {
    use common_wal::test_util::run_test_with_kafka_wal;
--- a/src/log-store/src/kafka/log_store.rs
+++ b/src/log-store/src/kafka/log_store.rs
@@ -552,14 +552,6 @@ mod tests {
            .collect()
    }

-    async fn prepare_topic(logstore: &KafkaLogStore, topic_name: &str) {
-        let controller_client = logstore.client_manager.controller_client();
-        controller_client
-            .create_topic(topic_name.to_string(), 1, 1, 5000)
-            .await
-            .unwrap();
-    }
-
    #[tokio::test]
    async fn test_append_batch_basic() {
        common_telemetry::init_default_ut_logging();
@@ -581,9 +573,7 @@ mod tests {
        };
        let logstore = KafkaLogStore::try_new(&config, None).await.unwrap();
        let topic_name = uuid::Uuid::new_v4().to_string();
-        prepare_topic(&logstore, &topic_name).await;
        let provider = Provider::kafka_provider(topic_name);
-
        let region_entries = (0..5)
            .map(|i| {
                let region_id = RegionId::new(1, i);
@@ -657,7 +647,6 @@ mod tests {
        };
        let logstore = KafkaLogStore::try_new(&config, None).await.unwrap();
        let topic_name = uuid::Uuid::new_v4().to_string();
-        prepare_topic(&logstore, &topic_name).await;
        let provider = Provider::kafka_provider(topic_name);
        let region_entries = (0..5)
            .map(|i| {
--- a/src/meta-srv/src/lib.rs
+++ b/src/meta-srv/src/lib.rs
@@ -14,7 +14,6 @@

 #![feature(result_flattening)]
 #![feature(assert_matches)]
-#![feature(extract_if)]
 #![feature(hash_set_entry)]

 pub mod bootstrap;
--- a/src/meta-srv/src/metasrv.rs
+++ b/src/meta-srv/src/metasrv.rs
@@ -14,7 +14,7 @@

 pub mod builder;

-use std::fmt::{self, Display};
+use std::fmt::Display;
 use std::sync::atomic::{AtomicBool, Ordering};
 use std::sync::{Arc, Mutex, RwLock};
 use std::time::Duration;
@@ -96,7 +96,7 @@ pub enum BackendImpl {
    MysqlStore,
 }

-#[derive(Clone, PartialEq, Serialize, Deserialize)]
+#[derive(Clone, Debug, PartialEq, Serialize, Deserialize)]
 #[serde(default)]
 pub struct MetasrvOptions {
    /// The address the server listens on.
@@ -166,47 +166,6 @@ pub struct MetasrvOptions {
    pub node_max_idle_time: Duration,
 }

-impl fmt::Debug for MetasrvOptions {
-    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
-        let mut debug_struct = f.debug_struct("MetasrvOptions");
-        debug_struct
-            .field("bind_addr", &self.bind_addr)
-            .field("server_addr", &self.server_addr)
-            .field("store_addrs", &self.sanitize_store_addrs())
-            .field("selector", &self.selector)
-            .field("use_memory_store", &self.use_memory_store)
-            .field("enable_region_failover", &self.enable_region_failover)
-            .field(
-                "allow_region_failover_on_local_wal",
-                &self.allow_region_failover_on_local_wal,
-            )
-            .field("http", &self.http)
-            .field("logging", &self.logging)
-            .field("procedure", &self.procedure)
-            .field("failure_detector", &self.failure_detector)
-            .field("datanode", &self.datanode)
-            .field("enable_telemetry", &self.enable_telemetry)
-            .field("data_home", &self.data_home)
-            .field("wal", &self.wal)
-            .field("export_metrics", &self.export_metrics)
-            .field("store_key_prefix", &self.store_key_prefix)
-            .field("max_txn_ops", &self.max_txn_ops)
-            .field("flush_stats_factor", &self.flush_stats_factor)
-            .field("tracing", &self.tracing)
-            .field("backend", &self.backend);
-
-        #[cfg(any(feature = "pg_kvbackend", feature = "mysql_kvbackend"))]
-        debug_struct.field("meta_table_name", &self.meta_table_name);
-
-        #[cfg(feature = "pg_kvbackend")]
-        debug_struct.field("meta_election_lock_id", &self.meta_election_lock_id);
-
-        debug_struct
-            .field("node_max_idle_time", &self.node_max_idle_time)
-            .finish()
-    }
-}
-
 const DEFAULT_METASRV_ADDR_PORT: &str = "3002";

 impl Default for MetasrvOptions {
@@ -290,13 +249,6 @@ impl MetasrvOptions {
            common_telemetry::debug!("detect local IP is not supported on Android");
        }
    }
-
-    fn sanitize_store_addrs(&self) -> Vec<String> {
-        self.store_addrs
-            .iter()
-            .map(|addr| common_meta::kv_backend::util::sanitize_connection_string(addr))
-            .collect()
-    }
 }

 pub struct MetasrvInfo {
--- a/src/meta-srv/src/metasrv/builder.rs
+++ b/src/meta-srv/src/metasrv/builder.rs
@@ -365,7 +365,7 @@ impl MetasrvBuilder {
            let (tx, rx) = WalPruneManager::channel();
            // Safety: Must be remote WAL.
            let remote_wal_options = options.wal.remote_wal_options().unwrap();
-            let kafka_client = build_kafka_client(&remote_wal_options.connection)
+            let kafka_client = build_kafka_client(remote_wal_options)
                .await
                .context(error::BuildKafkaClientSnafu)?;
            let wal_prune_context = WalPruneContext {
--- a/src/meta-srv/src/mocks.rs
+++ b/src/meta-srv/src/mocks.rs
@@ -141,10 +141,7 @@ pub async fn mock(
                if let Some(client) = client {
                    Ok(TokioIo::new(client))
                } else {
-                    Err(std::io::Error::new(
-                        std::io::ErrorKind::Other,
-                        "Client already taken",
-                    ))
+                    Err(std::io::Error::other("Client already taken"))
                }
            }
        }),
--- a/src/meta-srv/src/procedure/wal_prune.rs
+++ b/src/meta-srv/src/procedure/wal_prune.rs
@@ -52,7 +52,7 @@ use crate::Result;

 pub type KafkaClientRef = Arc<Client>;

-const DELETE_RECORDS_TIMEOUT: Duration = Duration::from_secs(5);
+const DELETE_RECORDS_TIMEOUT: Duration = Duration::from_secs(1);

 /// The state of WAL pruning.
 #[derive(Debug, Serialize, Deserialize)]
@@ -558,7 +558,6 @@ mod tests {
                topic_name = format!("test_procedure_execution-{}", topic_name);
                let mut env = TestEnv::new();
                let context = env.build_wal_prune_context(broker_endpoints).await;
-                TestEnv::prepare_topic(&context.client, &topic_name).await;
                let mut procedure = WalPruneProcedure::new(topic_name.clone(), context, 10, None);

                // Before any data in kvbackend is mocked, should return a retryable error.
--- a/src/meta-srv/src/procedure/wal_prune/test_util.rs
+++ b/src/meta-srv/src/procedure/wal_prune/test_util.rs
@@ -78,7 +78,7 @@ impl TestEnv {
            kafka_topic,
            ..Default::default()
        };
-        Arc::new(build_kafka_client(&config.connection).await.unwrap())
+        Arc::new(build_kafka_client(&config).await.unwrap())
    }

    pub async fn build_wal_prune_context(&self, broker_endpoints: Vec<String>) -> WalPruneContext {
@@ -91,12 +91,4 @@ impl TestEnv {
            mailbox: self.mailbox.mailbox().clone(),
        }
    }
-
-    pub async fn prepare_topic(client: &Arc<Client>, topic_name: &str) {
-        let controller_client = client.controller_client().unwrap();
-        controller_client
-            .create_topic(topic_name.to_string(), 1, 1, 5000)
-            .await
-            .unwrap();
-    }
 }
--- a/src/meta-srv/src/service/store/cached_kv.rs
+++ b/src/meta-srv/src/service/store/cached_kv.rs
@@ -278,7 +278,7 @@ impl KvBackend for LeaderCachedKvBackend {

        let remote_res = self.store.batch_get(remote_req).await?;
        let put_req = BatchPutRequest {
-            kvs: remote_res.kvs.clone().into_iter().map(Into::into).collect(),
+            kvs: remote_res.kvs.clone().into_iter().collect(),
            ..Default::default()
        };
        let _ = self.cache.batch_put(put_req).await?;
--- a/src/mito2/src/access_layer.rs
+++ b/src/mito2/src/access_layer.rs
@@ -16,7 +16,7 @@ use std::sync::Arc;

 use object_store::services::Fs;
 use object_store::util::{join_dir, with_instrument_layers};
-use object_store::{ErrorKind, ObjectStore};
+use object_store::ObjectStore;
 use smallvec::SmallVec;
 use snafu::ResultExt;
 use store_api::metadata::RegionMetadataRef;
@@ -42,10 +42,6 @@ pub type AccessLayerRef = Arc<AccessLayer>;
 /// SST write results.
 pub type SstInfoArray = SmallVec<[SstInfo; 2]>;

-pub const ATOMIC_WRITE_DIR: &str = "tmp/";
-/// For compatibility. Remove this after a major version release.
-pub const OLD_ATOMIC_WRITE_DIR: &str = ".tmp/";
-
 /// A layer to access SST files under the same directory.
 pub struct AccessLayer {
    region_dir: String,
@@ -164,18 +160,13 @@ impl AccessLayer {
                fulltext_index_config: request.fulltext_index_config,
                bloom_filter_index_config: request.bloom_filter_index_config,
            };
-            // We disable write cache on file system but we still use atomic write.
-            // TODO(yingwen): If we support other non-fs stores without the write cache, then
-            // we may have find a way to check whether we need the cleaner.
-            let cleaner = TempFileCleaner::new(region_id, self.object_store.clone());
            let mut writer = ParquetWriter::new_with_object_store(
                self.object_store.clone(),
                request.metadata,
                indexer_builder,
                path_provider,
            )
-            .await
-            .with_file_cleaner(cleaner);
+            .await;
            writer
                .write_all(request.source, request.max_sequence, write_opts)
                .await?
@@ -222,85 +213,10 @@ pub struct SstWriteRequest {
    pub bloom_filter_index_config: BloomFilterConfig,
 }

-/// Cleaner to remove temp files on the atomic write dir.
-pub(crate) struct TempFileCleaner {
-    region_id: RegionId,
-    object_store: ObjectStore,
-}
-
-impl TempFileCleaner {
-    /// Constructs the cleaner for the region and store.
-    pub(crate) fn new(region_id: RegionId, object_store: ObjectStore) -> Self {
-        Self {
-            region_id,
-            object_store,
-        }
-    }
-
-    /// Removes the SST and index file from the local atomic dir by the file id.
-    pub(crate) async fn clean_by_file_id(&self, file_id: FileId) {
-        let sst_key = IndexKey::new(self.region_id, file_id, FileType::Parquet).to_string();
-        let index_key = IndexKey::new(self.region_id, file_id, FileType::Puffin).to_string();
-
-        Self::clean_atomic_dir_files(&self.object_store, &[&sst_key, &index_key]).await;
-    }
-
-    /// Removes the files from the local atomic dir by their names.
-    pub(crate) async fn clean_atomic_dir_files(
-        local_store: &ObjectStore,
-        names_to_remove: &[&str],
-    ) {
-        // We don't know the actual suffix of the file under atomic dir, so we have
-        // to list the dir. The cost should be acceptable as there won't be to many files.
-        let Ok(entries) = local_store.list(ATOMIC_WRITE_DIR).await.inspect_err(|e| {
-            if e.kind() != ErrorKind::NotFound {
-                common_telemetry::error!(e; "Failed to list tmp files for {:?}", names_to_remove)
-            }
-        }) else {
-            return;
-        };
-
-        // In our case, we can ensure the file id is unique so it is safe to remove all files
-        // with the same file id under the atomic write dir.
-        let actual_files: Vec<_> = entries
-            .into_iter()
-            .filter_map(|entry| {
-                if entry.metadata().is_dir() {
-                    return None;
-                }
-
-                // Remove name that matches files_to_remove.
-                let should_remove = names_to_remove
-                    .iter()
-                    .any(|file| entry.name().starts_with(file));
-                if should_remove {
-                    Some(entry.path().to_string())
-                } else {
-                    None
-                }
-            })
-            .collect();
-
-        common_telemetry::warn!(
-            "Clean files {:?} under atomic write dir for {:?}",
-            actual_files,
-            names_to_remove
-        );
-
-        if let Err(e) = local_store.delete_iter(actual_files).await {
-            common_telemetry::error!(e; "Failed to delete tmp file for {:?}", names_to_remove);
-        }
-    }
-}
-
 pub(crate) async fn new_fs_cache_store(root: &str) -> Result<ObjectStore> {
-    let atomic_write_dir = join_dir(root, ATOMIC_WRITE_DIR);
+    let atomic_write_dir = join_dir(root, ".tmp/");
    clean_dir(&atomic_write_dir).await?;

-    // Compatible code. Remove this after a major release.
-    let old_atomic_temp_dir = join_dir(root, OLD_ATOMIC_WRITE_DIR);
-    clean_dir(&old_atomic_temp_dir).await?;
-
    let builder = Fs::default().root(root).atomic_write_dir(&atomic_write_dir);
    let store = ObjectStore::new(builder).context(OpenDalSnafu)?.finish();

--- a/src/mito2/src/cache/file_cache.rs
+++ b/src/mito2/src/cache/file_cache.rs
@@ -14,7 +14,6 @@

 //! A cache for files.

-use std::fmt;
 use std::ops::Range;
 use std::sync::Arc;
 use std::time::{Duration, Instant};
@@ -340,18 +339,6 @@ impl IndexKey {
    }
 }

-impl fmt::Display for IndexKey {
-    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
-        write!(
-            f,
-            "{}.{}.{}",
-            self.region_id.as_u64(),
-            self.file_id,
-            self.file_type.as_str()
-        )
-    }
-}
-
 /// Type of the file.
 #[derive(Debug, Clone, Copy, PartialEq, Eq, Hash)]
 pub enum FileType {
@@ -393,7 +380,15 @@ pub(crate) struct IndexValue {
 ///
 /// The file name format is `{region_id}.{file_id}.{file_type}`
 fn cache_file_path(cache_file_dir: &str, key: IndexKey) -> String {
-    join_path(cache_file_dir, &key.to_string())
+    join_path(
+        cache_file_dir,
+        &format!(
+            "{}.{}.{}",
+            key.region_id.as_u64(),
+            key.file_id,
+            key.file_type.as_str()
+        ),
+    )
 }

 /// Parse index key from the file name.
--- a/src/mito2/src/cache/write_cache.rs
+++ b/src/mito2/src/cache/write_cache.rs
@@ -26,7 +26,7 @@ use store_api::storage::RegionId;

 use crate::access_layer::{
    new_fs_cache_store, FilePathProvider, RegionFilePathFactory, SstInfoArray, SstWriteRequest,
-    TempFileCleaner, WriteCachePathProvider,
+    WriteCachePathProvider,
 };
 use crate::cache::file_cache::{FileCache, FileCacheRef, FileType, IndexKey, IndexValue};
 use crate::error::{self, Result};
@@ -122,7 +122,7 @@ impl WriteCache {
            row_group_size: write_opts.row_group_size,
            puffin_manager: self
                .puffin_manager_factory
-                .build(store.clone(), path_provider.clone()),
+                .build(store, path_provider.clone()),
            intermediate_manager: self.intermediate_manager.clone(),
            index_options: write_request.index_options,
            inverted_index_config: write_request.inverted_index_config,
@@ -130,16 +130,14 @@ impl WriteCache {
            bloom_filter_index_config: write_request.bloom_filter_index_config,
        };

-        let cleaner = TempFileCleaner::new(region_id, store.clone());
        // Write to FileCache.
        let mut writer = ParquetWriter::new_with_object_store(
-            store.clone(),
+            self.file_cache.local_store(),
            write_request.metadata,
            indexer,
-            path_provider.clone(),
+            path_provider,
        )
-        .await
-        .with_file_cleaner(cleaner);
+        .await;

        let sst_info = writer
            .write_all(write_request.source, write_request.max_sequence, write_opts)
@@ -203,26 +201,6 @@ impl WriteCache {
        remote_path: &str,
        remote_store: &ObjectStore,
        file_size: u64,
-    ) -> Result<()> {
-        if let Err(e) = self
-            .download_without_cleaning(index_key, remote_path, remote_store, file_size)
-            .await
-        {
-            let filename = index_key.to_string();
-            TempFileCleaner::clean_atomic_dir_files(&self.file_cache.local_store(), &[&filename])
-                .await;
-
-            return Err(e);
-        }
-        Ok(())
-    }
-
-    async fn download_without_cleaning(
-        &self,
-        index_key: IndexKey,
-        remote_path: &str,
-        remote_store: &ObjectStore,
-        file_size: u64,
    ) -> Result<()> {
        const DOWNLOAD_READER_CONCURRENCY: usize = 8;
        const DOWNLOAD_READER_CHUNK_SIZE: ReadableSize = ReadableSize::mb(8);
@@ -432,11 +410,9 @@ mod tests {
    use common_test_util::temp_dir::create_temp_dir;

    use super::*;
-    use crate::access_layer::{OperationType, ATOMIC_WRITE_DIR};
+    use crate::access_layer::OperationType;
    use crate::cache::test_util::new_fs_store;
    use crate::cache::{CacheManager, CacheStrategy};
-    use crate::error::InvalidBatchSnafu;
-    use crate::read::Source;
    use crate::region::options::IndexOptions;
    use crate::sst::parquet::reader::ParquetReaderBuilder;
    use crate::test_util::sst_util::{
@@ -602,82 +578,4 @@ mod tests {
        // Check parquet metadata
        assert_parquet_metadata_eq(write_parquet_metadata, reader.parquet_metadata());
    }
-
-    #[tokio::test]
-    async fn test_write_cache_clean_tmp_files() {
-        common_telemetry::init_default_ut_logging();
-        let mut env = TestEnv::new();
-        let data_home = env.data_home().display().to_string();
-        let mock_store = env.init_object_store_manager();
-
-        let write_cache_dir = create_temp_dir("");
-        let write_cache_path = write_cache_dir.path().to_str().unwrap();
-        let write_cache = env
-            .create_write_cache_from_path(write_cache_path, ReadableSize::mb(10))
-            .await;
-
-        // Create a cache manager using only write cache
-        let cache_manager = Arc::new(
-            CacheManager::builder()
-                .write_cache(Some(write_cache.clone()))
-                .build(),
-        );
-
-        // Create source
-        let metadata = Arc::new(sst_region_metadata());
-
-        // Creates a source that can return an error to abort the writer.
-        let source = Source::Iter(Box::new(
-            [
-                Ok(new_batch_by_range(&["a", "d"], 0, 60)),
-                InvalidBatchSnafu {
-                    reason: "Abort the writer",
-                }
-                .fail(),
-            ]
-            .into_iter(),
-        ));
-
-        // Write to local cache and upload sst to mock remote store
-        let write_request = SstWriteRequest {
-            op_type: OperationType::Flush,
-            metadata,
-            source,
-            storage: None,
-            max_sequence: None,
-            cache_manager: cache_manager.clone(),
-            index_options: IndexOptions::default(),
-            inverted_index_config: Default::default(),
-            fulltext_index_config: Default::default(),
-            bloom_filter_index_config: Default::default(),
-        };
-        let write_opts = WriteOptions {
-            row_group_size: 512,
-            ..Default::default()
-        };
-        let upload_request = SstUploadRequest {
-            dest_path_provider: RegionFilePathFactory::new(data_home.clone()),
-            remote_store: mock_store.clone(),
-        };
-
-        write_cache
-            .write_and_upload_sst(write_request, upload_request, &write_opts)
-            .await
-            .unwrap_err();
-        let atomic_write_dir = write_cache_dir.path().join(ATOMIC_WRITE_DIR);
-        let mut entries = tokio::fs::read_dir(&atomic_write_dir).await.unwrap();
-        let mut has_files = false;
-        while let Some(entry) = entries.next_entry().await.unwrap() {
-            if entry.file_type().await.unwrap().is_dir() {
-                continue;
-            }
-            has_files = true;
-            common_telemetry::warn!(
-                "Found remaining temporary file in atomic dir: {}",
-                entry.path().display()
-            );
-        }
-
-        assert!(!has_files);
-    }
 }
--- a/src/mito2/src/error.rs
+++ b/src/mito2/src/error.rs
@@ -710,8 +710,8 @@ pub enum Error {
        error: std::io::Error,
    },

-    #[snafu(display("Failed to filter record batch"))]
-    FilterRecordBatch {
+    #[snafu(display("Record batch error"))]
+    RecordBatch {
        source: common_recordbatch::error::Error,
        #[snafu(implicit)]
        location: Location,
@@ -1032,6 +1032,20 @@ pub enum Error {
        #[snafu(implicit)]
        location: Location,
    },
+
+    #[snafu(display("Failed to scan series"))]
+    ScanSeries {
+        #[snafu(implicit)]
+        location: Location,
+        source: Arc<Error>,
+    },
+
+    #[snafu(display("Partition {} scan multiple times", partition))]
+    ScanMultiTimes {
+        partition: usize,
+        #[snafu(implicit)]
+        location: Location,
+    },
 }

 pub type Result<T, E = Error> = std::result::Result<T, E>;
@@ -1154,7 +1168,7 @@ impl ErrorExt for Error {

            External { source, .. } => source.status_code(),

-            FilterRecordBatch { source, .. } => source.status_code(),
+            RecordBatch { source, .. } => source.status_code(),

            Download { .. } | Upload { .. } => StatusCode::StorageUnavailable,
            ChecksumMismatch { .. } => StatusCode::Unexpected,
@@ -1183,7 +1197,12 @@ impl ErrorExt for Error {
            ManualCompactionOverride {} => StatusCode::Cancelled,

            IncompatibleWalProviderChange { .. } => StatusCode::InvalidArguments,
+
            ConvertDataType { .. } => StatusCode::Internal,
+
+            ScanSeries { source, .. } => source.status_code(),
+
+            ScanMultiTimes { .. } => StatusCode::InvalidArguments,
        }
    }

--- a/src/mito2/src/memtable/partition_tree.rs
+++ b/src/mito2/src/memtable/partition_tree.rs
@@ -302,10 +302,7 @@ impl PartitionTreeMemtable {
    fn update_stats(&self, metrics: &WriteMetrics) {
        // Only let the tracker tracks value bytes.
        self.alloc_tracker.on_allocation(metrics.value_bytes);
-        self.max_timestamp
-            .fetch_max(metrics.max_ts, Ordering::SeqCst);
-        self.min_timestamp
-            .fetch_min(metrics.min_ts, Ordering::SeqCst);
+        metrics.update_timestamp_range(&self.max_timestamp, &self.min_timestamp);
    }
 }

--- a/src/mito2/src/memtable/stats.rs
+++ b/src/mito2/src/memtable/stats.rs
@@ -14,6 +14,8 @@

 //! Internal metrics of the memtable.

+use std::sync::atomic::{AtomicI64, Ordering};
+
 /// Metrics of writing memtables.
 pub(crate) struct WriteMetrics {
    /// Size allocated by keys.
@@ -26,6 +28,51 @@ pub(crate) struct WriteMetrics {
    pub(crate) max_ts: i64,
 }

+impl WriteMetrics {
+    /// Update the min/max timestamp range according to current write metric.
+    pub(crate) fn update_timestamp_range(&self, prev_max_ts: &AtomicI64, prev_min_ts: &AtomicI64) {
+        loop {
+            let current_min = prev_min_ts.load(Ordering::Relaxed);
+            if self.min_ts >= current_min {
+                break;
+            }
+
+            let Err(updated) = prev_min_ts.compare_exchange(
+                current_min,
+                self.min_ts,
+                Ordering::Relaxed,
+                Ordering::Relaxed,
+            ) else {
+                break;
+            };
+
+            if updated == self.min_ts {
+                break;
+            }
+        }
+
+        loop {
+            let current_max = prev_max_ts.load(Ordering::Relaxed);
+            if self.max_ts <= current_max {
+                break;
+            }
+
+            let Err(updated) = prev_max_ts.compare_exchange(
+                current_max,
+                self.max_ts,
+                Ordering::Relaxed,
+                Ordering::Relaxed,
+            ) else {
+                break;
+            };
+
+            if updated == self.max_ts {
+                break;
+            }
+        }
+    }
+}
+
 impl Default for WriteMetrics {
    fn default() -> Self {
        Self {
--- a/src/mito2/src/memtable/time_series.rs
+++ b/src/mito2/src/memtable/time_series.rs
@@ -147,8 +147,7 @@ impl TimeSeriesMemtable {
    fn update_stats(&self, stats: WriteMetrics) {
        self.alloc_tracker
            .on_allocation(stats.key_bytes + stats.value_bytes);
-        self.max_timestamp.fetch_max(stats.max_ts, Ordering::SeqCst);
-        self.min_timestamp.fetch_min(stats.min_ts, Ordering::SeqCst);
+        stats.update_timestamp_range(&self.max_timestamp, &self.min_timestamp);
    }

    fn write_key_value(&self, kv: KeyValue, stats: &mut WriteMetrics) -> Result<()> {
--- a/src/mito2/src/read.rs
+++ b/src/mito2/src/read.rs
@@ -24,6 +24,7 @@ pub(crate) mod range;
 pub(crate) mod scan_region;
 pub(crate) mod scan_util;
 pub(crate) mod seq_scan;
+pub(crate) mod series_scan;
 pub(crate) mod unordered_scan;

 use std::collections::{HashMap, HashSet};
--- a/src/mito2/src/read/projection.rs
+++ b/src/mito2/src/read/projection.rs
@@ -363,9 +363,9 @@ mod tests {
            builder
                .push_field_array(
                    *column_id,
-                    Arc::new(Int64Array::from_iter_values(
-                        std::iter::repeat(*field).take(num_rows),
-                    )),
+                    Arc::new(Int64Array::from_iter_values(std::iter::repeat_n(
+                        *field, num_rows,
+                    ))),
                )
                .unwrap();
        }
--- a/src/mito2/src/read/prune.rs
+++ b/src/mito2/src/read/prune.rs
@@ -21,7 +21,7 @@ use datatypes::arrow::array::BooleanArray;
 use datatypes::arrow::buffer::BooleanBuffer;
 use snafu::ResultExt;

-use crate::error::{FilterRecordBatchSnafu, Result};
+use crate::error::{RecordBatchSnafu, Result};
 use crate::memtable::BoxedBatchIterator;
 use crate::read::last_row::RowGroupLastRowCachedReader;
 use crate::read::{Batch, BatchReader};
@@ -201,7 +201,7 @@ impl PruneTimeIterator {
            for filter in filters.iter() {
                let result = filter
                    .evaluate_vector(batch.timestamps())
-                    .context(FilterRecordBatchSnafu)?;
+                    .context(RecordBatchSnafu)?;
                mask = mask.bitand(&result);
            }

--- a/src/mito2/src/read/scan_region.rs
+++ b/src/mito2/src/read/scan_region.rs
@@ -46,6 +46,7 @@ use crate::read::compat::{self, CompatBatch};
 use crate::read::projection::ProjectionMapper;
 use crate::read::range::{FileRangeBuilder, MemRangeBuilder, RangeMeta, RowGroupIndex};
 use crate::read::seq_scan::SeqScan;
+use crate::read::series_scan::SeriesScan;
 use crate::read::unordered_scan::UnorderedScan;
 use crate::read::{Batch, Source};
 use crate::region::options::MergeMode;
@@ -66,6 +67,8 @@ pub(crate) enum Scanner {
    Seq(SeqScan),
    /// Unordered scan.
    Unordered(UnorderedScan),
+    /// Per-series scan.
+    Series(SeriesScan),
 }

 impl Scanner {
@@ -75,6 +78,7 @@ impl Scanner {
        match self {
            Scanner::Seq(seq_scan) => seq_scan.build_stream(),
            Scanner::Unordered(unordered_scan) => unordered_scan.build_stream().await,
+            Scanner::Series(series_scan) => series_scan.build_stream().await,
        }
    }
 }
@@ -86,6 +90,7 @@ impl Scanner {
        match self {
            Scanner::Seq(seq_scan) => seq_scan.input().num_files(),
            Scanner::Unordered(unordered_scan) => unordered_scan.input().num_files(),
+            Scanner::Series(series_scan) => series_scan.input().num_files(),
        }
    }

@@ -94,6 +99,7 @@ impl Scanner {
        match self {
            Scanner::Seq(seq_scan) => seq_scan.input().num_memtables(),
            Scanner::Unordered(unordered_scan) => unordered_scan.input().num_memtables(),
+            Scanner::Series(series_scan) => series_scan.input().num_memtables(),
        }
    }

@@ -102,6 +108,7 @@ impl Scanner {
        match self {
            Scanner::Seq(seq_scan) => seq_scan.input().file_ids(),
            Scanner::Unordered(unordered_scan) => unordered_scan.input().file_ids(),
+            Scanner::Series(series_scan) => series_scan.input().file_ids(),
        }
    }

@@ -113,6 +120,7 @@ impl Scanner {
        match self {
            Scanner::Seq(seq_scan) => seq_scan.prepare(request).unwrap(),
            Scanner::Unordered(unordered_scan) => unordered_scan.prepare(request).unwrap(),
+            Scanner::Series(series_scan) => series_scan.prepare(request).unwrap(),
        }
    }
 }
@@ -248,7 +256,9 @@ impl ScanRegion {

    /// Returns a [Scanner] to scan the region.
    pub(crate) fn scanner(self) -> Result<Scanner> {
-        if self.use_unordered_scan() {
+        if self.use_series_scan() {
+            self.series_scan().map(Scanner::Series)
+        } else if self.use_unordered_scan() {
            // If table is append only and there is no series row selector, we use unordered scan in query.
            // We still use seq scan in compaction.
            self.unordered_scan().map(Scanner::Unordered)
@@ -260,7 +270,9 @@ impl ScanRegion {
    /// Returns a [RegionScanner] to scan the region.
    #[tracing::instrument(level = tracing::Level::DEBUG, skip_all)]
    pub(crate) fn region_scanner(self) -> Result<RegionScannerRef> {
-        if self.use_unordered_scan() {
+        if self.use_series_scan() {
+            self.series_scan().map(|scanner| Box::new(scanner) as _)
+        } else if self.use_unordered_scan() {
            self.unordered_scan().map(|scanner| Box::new(scanner) as _)
        } else {
            self.seq_scan().map(|scanner| Box::new(scanner) as _)
@@ -279,6 +291,12 @@ impl ScanRegion {
        Ok(UnorderedScan::new(input))
    }

+    /// Scans by series.
+    pub(crate) fn series_scan(self) -> Result<SeriesScan> {
+        let input = self.scan_input(true)?;
+        Ok(SeriesScan::new(input))
+    }
+
    #[cfg(test)]
    pub(crate) fn scan_without_filter_deleted(self) -> Result<SeqScan> {
        let input = self.scan_input(false)?;
@@ -299,6 +317,11 @@ impl ScanRegion {
                || self.request.distribution == Some(TimeSeriesDistribution::TimeWindowed))
    }

+    /// Returns true if the region can use series scan for current request.
+    fn use_series_scan(&self) -> bool {
+        self.request.distribution == Some(TimeSeriesDistribution::PerSeries)
+    }
+
    /// Creates a scan input.
    fn scan_input(mut self, filter_deleted: bool) -> Result<ScanInput> {
        let time_range = self.build_time_range_predicate();
@@ -322,10 +345,13 @@ impl ScanRegion {
        let memtables: Vec<_> = memtables
            .into_iter()
            .filter(|mem| {
-                // check if memtable is empty by reading stats.
-                let Some((start, end)) = mem.stats().time_range() else {
+                if mem.is_empty() {
                    return false;
-                };
+                }
+                let stats = mem.stats();
+                // Safety: the memtable is not empty.
+                let (start, end) = stats.time_range().unwrap();
+
                // The time range of the memtable is inclusive.
                let memtable_range = TimestampRange::new_inclusive(Some(start), Some(end));
                memtable_range.intersects(&time_range)
--- a/src/mito2/src/read/scan_util.rs
+++ b/src/mito2/src/read/scan_util.rs
@@ -92,6 +92,8 @@ struct ScanMetricsSet {

    /// Elapsed time before the first poll operation.
    first_poll: Duration,
+    /// Number of send timeout in SeriesScan.
+    num_series_send_timeout: usize,
 }

 impl fmt::Debug for ScanMetricsSet {
@@ -122,6 +124,7 @@ impl fmt::Debug for ScanMetricsSet {
            num_sst_batches,
            num_sst_rows,
            first_poll,
+            num_series_send_timeout,
        } = self;

        write!(
@@ -150,7 +153,8 @@ impl fmt::Debug for ScanMetricsSet {
            num_sst_record_batches={num_sst_record_batches}, \
            num_sst_batches={num_sst_batches}, \
            num_sst_rows={num_sst_rows}, \
-            first_poll={first_poll:?}}}"
+            first_poll={first_poll:?}, \
+            num_series_send_timeout={num_series_send_timeout}}}"
        )
    }
 }
@@ -439,6 +443,12 @@ impl PartitionMetrics {
    pub(crate) fn on_finish(&self) {
        self.0.on_finish();
    }
+
+    /// Sets the `num_series_send_timeout`.
+    pub(crate) fn set_num_series_send_timeout(&self, num_timeout: usize) {
+        let mut metrics = self.0.metrics.lock().unwrap();
+        metrics.num_series_send_timeout = num_timeout;
+    }
 }

 impl fmt::Debug for PartitionMetrics {
--- a/src/mito2/src/read/seq_scan.rs
+++ b/src/mito2/src/read/seq_scan.rs
@@ -30,7 +30,7 @@ use datatypes::schema::SchemaRef;
 use snafu::ResultExt;
 use store_api::metadata::RegionMetadataRef;
 use store_api::region_engine::{PartitionRange, PrepareRequest, RegionScanner, ScannerProperties};
-use store_api::storage::{TimeSeriesDistribution, TimeSeriesRowSelector};
+use store_api::storage::TimeSeriesRowSelector;
 use tokio::sync::Semaphore;

 use crate::error::{PartitionOutOfRangeSnafu, Result};
@@ -149,7 +149,7 @@ impl SeqScan {
    /// Builds a reader to read sources. If `semaphore` is provided, reads sources in parallel
    /// if possible.
    #[tracing::instrument(level = tracing::Level::DEBUG, skip_all)]
-    async fn build_reader_from_sources(
+    pub(crate) async fn build_reader_from_sources(
        stream_ctx: &StreamContext,
        mut sources: Vec<Source>,
        semaphore: Option<Arc<Semaphore>>,
@@ -206,9 +206,13 @@ impl SeqScan {
                .build(),
            ));
        }
-
-        if self.stream_ctx.input.distribution == Some(TimeSeriesDistribution::PerSeries) {
-            return self.scan_partition_by_series(metrics_set, partition);
+        if self.properties.partitions[partition].is_empty() {
+            return Ok(Box::pin(RecordBatchStreamWrapper::new(
+                self.stream_ctx.input.mapper.output_schema(),
+                common_recordbatch::EmptyRecordBatchStream::new(
+                    self.stream_ctx.input.mapper.output_schema(),
+                ),
+            )));
        }

        let stream_ctx = self.stream_ctx.clone();
@@ -237,14 +241,14 @@ impl SeqScan {
                    &mut sources,
                );

+                let mut metrics = ScannerMetrics::default();
+                let mut fetch_start = Instant::now();
                let mut reader =
                    Self::build_reader_from_sources(&stream_ctx, sources, semaphore.clone())
                        .await
                        .map_err(BoxedError::new)
                        .context(ExternalSnafu)?;
                let cache = &stream_ctx.input.cache_strategy;
-                let mut metrics = ScannerMetrics::default();
-                let mut fetch_start = Instant::now();
                #[cfg(debug_assertions)]
                let mut checker = crate::read::BatchChecker::default()
                    .with_start(Some(part_range.start))
@@ -307,97 +311,6 @@ impl SeqScan {
        Ok(stream)
    }

-    /// Scans all ranges in the given partition and merge by time series.
-    /// Otherwise the returned stream might not contains any data.
-    fn scan_partition_by_series(
-        &self,
-        metrics_set: &ExecutionPlanMetricsSet,
-        partition: usize,
-    ) -> Result<SendableRecordBatchStream, BoxedError> {
-        let stream_ctx = self.stream_ctx.clone();
-        let semaphore = self.new_semaphore();
-        let partition_ranges = self.properties.partitions[partition].clone();
-        let distinguish_range = self.properties.distinguish_partition_range;
-        let part_metrics = self.new_partition_metrics(metrics_set, partition);
-        debug_assert!(!self.compaction);
-
-        let stream = try_stream! {
-            part_metrics.on_first_poll();
-
-            let range_builder_list = Arc::new(RangeBuilderList::new(
-                stream_ctx.input.num_memtables(),
-                stream_ctx.input.num_files(),
-            ));
-            // Scans all parts.
-            let mut sources = Vec::with_capacity(partition_ranges.len());
-            for part_range in partition_ranges {
-                build_sources(
-                    &stream_ctx,
-                    &part_range,
-                    false,
-                    &part_metrics,
-                    range_builder_list.clone(),
-                    &mut sources,
-                );
-            }
-
-            // Builds a reader that merge sources from all parts.
-            let mut reader =
-                Self::build_reader_from_sources(&stream_ctx, sources, semaphore.clone())
-                    .await
-                    .map_err(BoxedError::new)
-                    .context(ExternalSnafu)?;
-            let cache = &stream_ctx.input.cache_strategy;
-            let mut metrics = ScannerMetrics::default();
-            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();
-
-                debug_assert!(!batch.is_empty());
-                if batch.is_empty() {
-                    continue;
-                }
-
-                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();
-            }
-
-            // Yields an empty part to indicate this range is terminated.
-            // The query engine can use this to optimize some queries.
-            if distinguish_range {
-                let yield_start = Instant::now();
-                yield stream_ctx.input.mapper.empty_record_batch();
-                metrics.yield_cost += yield_start.elapsed();
-            }
-
-            metrics.scan_cost += fetch_start.elapsed();
-            part_metrics.merge_metrics(&metrics);
-
-            part_metrics.on_finish();
-        };
-
-        let stream = Box::pin(RecordBatchStreamWrapper::new(
-            self.stream_ctx.input.mapper.output_schema(),
-            Box::pin(stream),
-        ));
-
-        Ok(stream)
-    }
-
    fn new_semaphore(&self) -> Option<Arc<Semaphore>> {
        if self.properties.target_partitions() > self.properties.num_partitions() {
            // We can use additional tasks to read the data if we have more target partitions than actual partitions.
@@ -498,7 +411,7 @@ impl fmt::Debug for SeqScan {
 }

 /// Builds sources for the partition range and push them to the `sources` vector.
-fn build_sources(
+pub(crate) fn build_sources(
    stream_ctx: &Arc<StreamContext>,
    part_range: &PartitionRange,
    compaction: bool,
@@ -509,8 +422,8 @@ fn build_sources(
    // Gets range meta.
    let range_meta = &stream_ctx.ranges[part_range.identifier];
    #[cfg(debug_assertions)]
-    if compaction || stream_ctx.input.distribution == Some(TimeSeriesDistribution::PerSeries) {
-        // Compaction or per series distribution expects input sources are not been split.
+    if compaction {
+        // Compaction expects input sources are not been split.
        debug_assert_eq!(range_meta.indices.len(), range_meta.row_group_indices.len());
        for (i, row_group_idx) in range_meta.row_group_indices.iter().enumerate() {
            // It should scan all row groups.
--- a/src/mito2/src/read/series_scan.rs
+++ b/src/mito2/src/read/series_scan.rs
@@ -0,0 +1,547 @@
+// 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.
+
+//! Per-series scan implementation.
+
+use std::fmt;
+use std::sync::{Arc, Mutex};
+use std::time::{Duration, Instant};
+
+use async_stream::try_stream;
+use common_error::ext::BoxedError;
+use common_recordbatch::error::ExternalSnafu;
+use common_recordbatch::util::ChainedRecordBatchStream;
+use common_recordbatch::{RecordBatch, RecordBatchStreamWrapper, SendableRecordBatchStream};
+use datafusion::physical_plan::metrics::ExecutionPlanMetricsSet;
+use datafusion::physical_plan::{DisplayAs, DisplayFormatType};
+use datatypes::compute::concat_batches;
+use datatypes::schema::SchemaRef;
+use smallvec::{smallvec, SmallVec};
+use snafu::{ensure, OptionExt, ResultExt};
+use store_api::metadata::RegionMetadataRef;
+use store_api::region_engine::{PartitionRange, PrepareRequest, RegionScanner, ScannerProperties};
+use tokio::sync::mpsc::error::{SendTimeoutError, TrySendError};
+use tokio::sync::mpsc::{self, Receiver, Sender};
+use tokio::sync::Semaphore;
+
+use crate::error::{
+    ComputeArrowSnafu, Error, InvalidSenderSnafu, PartitionOutOfRangeSnafu, Result,
+    ScanMultiTimesSnafu, ScanSeriesSnafu,
+};
+use crate::read::range::RangeBuilderList;
+use crate::read::scan_region::{ScanInput, StreamContext};
+use crate::read::scan_util::{PartitionMetrics, PartitionMetricsList};
+use crate::read::seq_scan::{build_sources, SeqScan};
+use crate::read::{Batch, ScannerMetrics};
+
+/// Timeout to send a batch to a sender.
+const SEND_TIMEOUT: Duration = Duration::from_millis(10);
+
+/// List of receivers.
+type ReceiverList = Vec<Option<Receiver<Result<SeriesBatch>>>>;
+
+/// Scans a region and returns sorted rows of a series in the same partition.
+///
+/// The output order is always order by `(primary key, time index)` inside every
+/// partition.
+/// Always returns the same series (primary key) to the same partition.
+pub struct SeriesScan {
+    /// Properties of the scanner.
+    properties: ScannerProperties,
+    /// Context of streams.
+    stream_ctx: Arc<StreamContext>,
+    /// Receivers of each partition.
+    receivers: Mutex<ReceiverList>,
+    /// Metrics for each partition.
+    /// The scanner only sets in query and keeps it empty during compaction.
+    metrics_list: Arc<PartitionMetricsList>,
+}
+
+impl SeriesScan {
+    /// Creates a new [SeriesScan].
+    pub(crate) fn new(input: ScanInput) -> Self {
+        let mut properties = ScannerProperties::default()
+            .with_append_mode(input.append_mode)
+            .with_total_rows(input.total_rows());
+        let stream_ctx = Arc::new(StreamContext::seq_scan_ctx(input, false));
+        properties.partitions = vec![stream_ctx.partition_ranges()];
+
+        Self {
+            properties,
+            stream_ctx,
+            receivers: Mutex::new(Vec::new()),
+            metrics_list: Arc::new(PartitionMetricsList::default()),
+        }
+    }
+
+    fn scan_partition_impl(
+        &self,
+        metrics_set: &ExecutionPlanMetricsSet,
+        partition: usize,
+    ) -> Result<SendableRecordBatchStream, BoxedError> {
+        if partition >= self.properties.num_partitions() {
+            return Err(BoxedError::new(
+                PartitionOutOfRangeSnafu {
+                    given: partition,
+                    all: self.properties.num_partitions(),
+                }
+                .build(),
+            ));
+        }
+
+        self.maybe_start_distributor(metrics_set, &self.metrics_list);
+
+        let part_metrics =
+            new_partition_metrics(&self.stream_ctx, metrics_set, partition, &self.metrics_list);
+        let mut receiver = self.take_receiver(partition).map_err(BoxedError::new)?;
+        let stream_ctx = self.stream_ctx.clone();
+
+        let stream = try_stream! {
+            part_metrics.on_first_poll();
+
+            let cache = &stream_ctx.input.cache_strategy;
+            let mut df_record_batches = Vec::new();
+            let mut fetch_start = Instant::now();
+            while let Some(result) = receiver.recv().await {
+                let mut metrics = ScannerMetrics::default();
+                let series = result.map_err(BoxedError::new).context(ExternalSnafu)?;
+                metrics.scan_cost += fetch_start.elapsed();
+                fetch_start = Instant::now();
+
+                let convert_start = Instant::now();
+                df_record_batches.reserve(series.batches.len());
+                for batch in series.batches {
+                    metrics.num_batches += 1;
+                    metrics.num_rows += batch.num_rows();
+
+                    let record_batch = stream_ctx.input.mapper.convert(&batch, cache)?;
+                    df_record_batches.push(record_batch.into_df_record_batch());
+                }
+
+                let output_schema = stream_ctx.input.mapper.output_schema();
+                let df_record_batch =
+                    concat_batches(output_schema.arrow_schema(), &df_record_batches)
+                        .context(ComputeArrowSnafu)
+                        .map_err(BoxedError::new)
+                        .context(ExternalSnafu)?;
+                df_record_batches.clear();
+                let record_batch =
+                    RecordBatch::try_from_df_record_batch(output_schema, df_record_batch)?;
+                metrics.convert_cost += convert_start.elapsed();
+
+                let yield_start = Instant::now();
+                yield record_batch;
+                metrics.yield_cost += yield_start.elapsed();
+
+                part_metrics.merge_metrics(&metrics);
+            }
+        };
+
+        let stream = Box::pin(RecordBatchStreamWrapper::new(
+            self.stream_ctx.input.mapper.output_schema(),
+            Box::pin(stream),
+        ));
+
+        Ok(stream)
+    }
+
+    /// Takes the receiver for the partition.
+    fn take_receiver(&self, partition: usize) -> Result<Receiver<Result<SeriesBatch>>> {
+        let mut rx_list = self.receivers.lock().unwrap();
+        rx_list[partition]
+            .take()
+            .context(ScanMultiTimesSnafu { partition })
+    }
+
+    /// Starts the distributor if the receiver list is empty.
+    fn maybe_start_distributor(
+        &self,
+        metrics_set: &ExecutionPlanMetricsSet,
+        metrics_list: &Arc<PartitionMetricsList>,
+    ) {
+        let mut rx_list = self.receivers.lock().unwrap();
+        if !rx_list.is_empty() {
+            return;
+        }
+
+        let (senders, receivers) = new_channel_list(self.properties.num_partitions());
+        let mut distributor = SeriesDistributor {
+            stream_ctx: self.stream_ctx.clone(),
+            semaphore: Some(Arc::new(Semaphore::new(self.properties.num_partitions()))),
+            partitions: self.properties.partitions.clone(),
+            senders,
+            metrics_set: metrics_set.clone(),
+            metrics_list: metrics_list.clone(),
+        };
+        common_runtime::spawn_global(async move {
+            distributor.execute().await;
+        });
+
+        *rx_list = receivers;
+    }
+
+    /// Scans the region and returns a stream.
+    pub(crate) async fn build_stream(&self) -> Result<SendableRecordBatchStream, BoxedError> {
+        let part_num = self.properties.num_partitions();
+        let metrics_set = ExecutionPlanMetricsSet::default();
+        let streams = (0..part_num)
+            .map(|i| self.scan_partition(&metrics_set, i))
+            .collect::<Result<Vec<_>, BoxedError>>()?;
+        let chained_stream = ChainedRecordBatchStream::new(streams).map_err(BoxedError::new)?;
+        Ok(Box::pin(chained_stream))
+    }
+}
+
+fn new_channel_list(num_partitions: usize) -> (SenderList, ReceiverList) {
+    let (senders, receivers): (Vec<_>, Vec<_>) = (0..num_partitions)
+        .map(|_| {
+            let (sender, receiver) = mpsc::channel(1);
+            (Some(sender), Some(receiver))
+        })
+        .unzip();
+    (SenderList::new(senders), receivers)
+}
+
+impl RegionScanner for SeriesScan {
+    fn properties(&self) -> &ScannerProperties {
+        &self.properties
+    }
+
+    fn schema(&self) -> SchemaRef {
+        self.stream_ctx.input.mapper.output_schema()
+    }
+
+    fn metadata(&self) -> RegionMetadataRef {
+        self.stream_ctx.input.mapper.metadata().clone()
+    }
+
+    fn scan_partition(
+        &self,
+        metrics_set: &ExecutionPlanMetricsSet,
+        partition: usize,
+    ) -> Result<SendableRecordBatchStream, BoxedError> {
+        self.scan_partition_impl(metrics_set, partition)
+    }
+
+    fn prepare(&mut self, request: PrepareRequest) -> Result<(), BoxedError> {
+        self.properties.prepare(request);
+        Ok(())
+    }
+
+    fn has_predicate(&self) -> bool {
+        let predicate = self.stream_ctx.input.predicate();
+        predicate.map(|p| !p.exprs().is_empty()).unwrap_or(false)
+    }
+
+    fn set_logical_region(&mut self, logical_region: bool) {
+        self.properties.set_logical_region(logical_region);
+    }
+}
+
+impl DisplayAs for SeriesScan {
+    fn fmt_as(&self, t: DisplayFormatType, f: &mut fmt::Formatter) -> fmt::Result {
+        write!(
+            f,
+            "SeriesScan: region={}, ",
+            self.stream_ctx.input.mapper.metadata().region_id
+        )?;
+        match t {
+            DisplayFormatType::Default => self.stream_ctx.format_for_explain(false, f),
+            DisplayFormatType::Verbose => {
+                self.stream_ctx.format_for_explain(true, f)?;
+                self.metrics_list.format_verbose_metrics(f)
+            }
+        }
+    }
+}
+
+impl fmt::Debug for SeriesScan {
+    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
+        f.debug_struct("SeriesScan")
+            .field("num_ranges", &self.stream_ctx.ranges.len())
+            .finish()
+    }
+}
+
+#[cfg(test)]
+impl SeriesScan {
+    /// Returns the input.
+    pub(crate) fn input(&self) -> &ScanInput {
+        &self.stream_ctx.input
+    }
+}
+
+/// The distributor scans series and distributes them to different partitions.
+struct SeriesDistributor {
+    /// Context for the scan stream.
+    stream_ctx: Arc<StreamContext>,
+    /// Optional semaphore for limiting the number of concurrent scans.
+    semaphore: Option<Arc<Semaphore>>,
+    /// Partition ranges to scan.
+    partitions: Vec<Vec<PartitionRange>>,
+    /// Senders of all partitions.
+    senders: SenderList,
+    /// Metrics set to report.
+    /// The distributor report the metrics as an additional partition.
+    /// This may double the scan cost of the [SeriesScan] metrics. We can
+    /// get per-partition metrics in verbose mode to see the metrics of the
+    /// distributor.
+    metrics_set: ExecutionPlanMetricsSet,
+    metrics_list: Arc<PartitionMetricsList>,
+}
+
+impl SeriesDistributor {
+    /// Executes the distributor.
+    async fn execute(&mut self) {
+        if let Err(e) = self.scan_partitions().await {
+            self.senders.send_error(e).await;
+        }
+    }
+
+    /// Scans all parts.
+    async fn scan_partitions(&mut self) -> Result<()> {
+        let part_metrics = new_partition_metrics(
+            &self.stream_ctx,
+            &self.metrics_set,
+            self.partitions.len(),
+            &self.metrics_list,
+        );
+        part_metrics.on_first_poll();
+
+        let range_builder_list = Arc::new(RangeBuilderList::new(
+            self.stream_ctx.input.num_memtables(),
+            self.stream_ctx.input.num_files(),
+        ));
+        // Scans all parts.
+        let mut sources = Vec::with_capacity(self.partitions.len());
+        for partition in &self.partitions {
+            sources.reserve(partition.len());
+            for part_range in partition {
+                build_sources(
+                    &self.stream_ctx,
+                    part_range,
+                    false,
+                    &part_metrics,
+                    range_builder_list.clone(),
+                    &mut sources,
+                );
+            }
+        }
+
+        // Builds a reader that merge sources from all parts.
+        let mut reader =
+            SeqScan::build_reader_from_sources(&self.stream_ctx, sources, self.semaphore.clone())
+                .await?;
+        let mut metrics = ScannerMetrics::default();
+        let mut fetch_start = Instant::now();
+
+        let mut current_series = SeriesBatch::default();
+        while let Some(batch) = reader.next_batch().await? {
+            metrics.scan_cost += fetch_start.elapsed();
+            fetch_start = Instant::now();
+            metrics.num_batches += 1;
+            metrics.num_rows += batch.num_rows();
+
+            debug_assert!(!batch.is_empty());
+            if batch.is_empty() {
+                continue;
+            }
+
+            let Some(last_key) = current_series.current_key() else {
+                current_series.push(batch);
+                continue;
+            };
+
+            if last_key == batch.primary_key() {
+                current_series.push(batch);
+                continue;
+            }
+
+            // We find a new series, send the current one.
+            let to_send = std::mem::replace(&mut current_series, SeriesBatch::single(batch));
+            let yield_start = Instant::now();
+            self.senders.send_batch(to_send).await?;
+            metrics.yield_cost += yield_start.elapsed();
+        }
+
+        if !current_series.is_empty() {
+            let yield_start = Instant::now();
+            self.senders.send_batch(current_series).await?;
+            metrics.yield_cost += yield_start.elapsed();
+        }
+
+        metrics.scan_cost += fetch_start.elapsed();
+        part_metrics.merge_metrics(&metrics);
+        part_metrics.set_num_series_send_timeout(self.senders.num_timeout);
+
+        part_metrics.on_finish();
+
+        Ok(())
+    }
+}
+
+/// Batches of the same series.
+#[derive(Default)]
+struct SeriesBatch {
+    batches: SmallVec<[Batch; 4]>,
+}
+
+impl SeriesBatch {
+    /// Creates a new [SeriesBatch] from a single [Batch].
+    fn single(batch: Batch) -> Self {
+        Self {
+            batches: smallvec![batch],
+        }
+    }
+
+    fn current_key(&self) -> Option<&[u8]> {
+        self.batches.first().map(|batch| batch.primary_key())
+    }
+
+    fn push(&mut self, batch: Batch) {
+        self.batches.push(batch);
+    }
+
+    /// Returns true if there is no batch.
+    fn is_empty(&self) -> bool {
+        self.batches.is_empty()
+    }
+}
+
+/// List of senders.
+struct SenderList {
+    senders: Vec<Option<Sender<Result<SeriesBatch>>>>,
+    /// Number of None senders.
+    num_nones: usize,
+    /// Index of the current partition to send.
+    sender_idx: usize,
+    /// Number of timeout.
+    num_timeout: usize,
+}
+
+impl SenderList {
+    fn new(senders: Vec<Option<Sender<Result<SeriesBatch>>>>) -> Self {
+        let num_nones = senders.iter().filter(|sender| sender.is_none()).count();
+        Self {
+            senders,
+            num_nones,
+            sender_idx: 0,
+            num_timeout: 0,
+        }
+    }
+
+    /// Finds a partition and tries to send the batch to the partition.
+    /// Returns None if it sends successfully.
+    fn try_send_batch(&mut self, mut batch: SeriesBatch) -> Result<Option<SeriesBatch>> {
+        for _ in 0..self.senders.len() {
+            ensure!(self.num_nones < self.senders.len(), InvalidSenderSnafu);
+
+            let sender_idx = self.fetch_add_sender_idx();
+            let Some(sender) = &self.senders[sender_idx] else {
+                continue;
+            };
+
+            match sender.try_send(Ok(batch)) {
+                Ok(()) => return Ok(None),
+                Err(TrySendError::Full(res)) => {
+                    // Safety: we send Ok.
+                    batch = res.unwrap();
+                }
+                Err(TrySendError::Closed(res)) => {
+                    self.senders[sender_idx] = None;
+                    self.num_nones += 1;
+                    // Safety: we send Ok.
+                    batch = res.unwrap();
+                }
+            }
+        }
+
+        Ok(Some(batch))
+    }
+
+    /// Finds a partition and sends the batch to the partition.
+    async fn send_batch(&mut self, mut batch: SeriesBatch) -> Result<()> {
+        // Sends the batch without blocking first.
+        match self.try_send_batch(batch)? {
+            Some(b) => {
+                // Unable to send batch to partition.
+                batch = b;
+            }
+            None => {
+                return Ok(());
+            }
+        }
+
+        loop {
+            ensure!(self.num_nones < self.senders.len(), InvalidSenderSnafu);
+
+            let sender_idx = self.fetch_add_sender_idx();
+            let Some(sender) = &self.senders[sender_idx] else {
+                continue;
+            };
+            // Adds a timeout to avoid blocking indefinitely and sending
+            // the batch in a round-robin fashion when some partitions
+            // don't poll their inputs. This may happen if we have a
+            // node like sort merging. But it is rare when we are using SeriesScan.
+            match sender.send_timeout(Ok(batch), SEND_TIMEOUT).await {
+                Ok(()) => break,
+                Err(SendTimeoutError::Timeout(res)) => {
+                    self.num_timeout += 1;
+                    // Safety: we send Ok.
+                    batch = res.unwrap();
+                }
+                Err(SendTimeoutError::Closed(res)) => {
+                    self.senders[sender_idx] = None;
+                    self.num_nones += 1;
+                    // Safety: we send Ok.
+                    batch = res.unwrap();
+                }
+            }
+        }
+
+        Ok(())
+    }
+
+    async fn send_error(&self, error: Error) {
+        let error = Arc::new(error);
+        for sender in self.senders.iter().flatten() {
+            let result = Err(error.clone()).context(ScanSeriesSnafu);
+            let _ = sender.send(result).await;
+        }
+    }
+
+    fn fetch_add_sender_idx(&mut self) -> usize {
+        let sender_idx = self.sender_idx;
+        self.sender_idx = (self.sender_idx + 1) % self.senders.len();
+        sender_idx
+    }
+}
+
+fn new_partition_metrics(
+    stream_ctx: &StreamContext,
+    metrics_set: &ExecutionPlanMetricsSet,
+    partition: usize,
+    metrics_list: &PartitionMetricsList,
+) -> PartitionMetrics {
+    let metrics = PartitionMetrics::new(
+        stream_ctx.input.mapper.metadata().region_id,
+        partition,
+        "SeriesScan",
+        stream_ctx.query_start,
+        metrics_set,
+    );
+
+    metrics_list.set(partition, metrics.clone());
+    metrics
+}
--- a/src/mito2/src/sst/index/bloom_filter/creator.rs
+++ b/src/mito2/src/sst/index/bloom_filter/creator.rs
@@ -346,7 +346,6 @@ impl BloomFilterIndexer {

 #[cfg(test)]
 pub(crate) mod tests {
-    use std::iter;

    use api::v1::SemanticType;
    use datatypes::data_type::ConcreteDataType;
@@ -461,15 +460,15 @@ pub(crate) mod tests {

        Batch::new(
            primary_key,
-            Arc::new(UInt64Vector::from_iter_values(
-                iter::repeat(0).take(num_rows),
-            )),
-            Arc::new(UInt64Vector::from_iter_values(
-                iter::repeat(0).take(num_rows),
-            )),
-            Arc::new(UInt8Vector::from_iter_values(
-                iter::repeat(1).take(num_rows),
-            )),
+            Arc::new(UInt64Vector::from_iter_values(std::iter::repeat_n(
+                0, num_rows,
+            ))),
+            Arc::new(UInt64Vector::from_iter_values(std::iter::repeat_n(
+                0, num_rows,
+            ))),
+            Arc::new(UInt8Vector::from_iter_values(std::iter::repeat_n(
+                1, num_rows,
+            ))),
            vec![u64_field],
        )
        .unwrap()
--- a/src/mito2/src/sst/index/fulltext_index/creator.rs
+++ b/src/mito2/src/sst/index/fulltext_index/creator.rs
@@ -489,12 +489,12 @@ mod tests {
            Arc::new(UInt64Vector::from_iter_values(
                (0..num_rows).map(|n| n as u64),
            )),
-            Arc::new(UInt64Vector::from_iter_values(
-                std::iter::repeat(0).take(num_rows),
-            )),
-            Arc::new(UInt8Vector::from_iter_values(
-                std::iter::repeat(1).take(num_rows),
-            )),
+            Arc::new(UInt64Vector::from_iter_values(std::iter::repeat_n(
+                0, num_rows,
+            ))),
+            Arc::new(UInt8Vector::from_iter_values(std::iter::repeat_n(
+                1, num_rows,
+            ))),
            vec![
                BatchColumn {
                    column_id: 1,
--- a/src/mito2/src/sst/index/inverted_index/creator.rs
+++ b/src/mito2/src/sst/index/inverted_index/creator.rs
@@ -326,7 +326,6 @@ impl InvertedIndexer {
 #[cfg(test)]
 mod tests {
    use std::collections::BTreeSet;
-    use std::iter;

    use api::v1::SemanticType;
    use datafusion_expr::{binary_expr, col, lit, Expr as DfExpr, Operator};
@@ -424,15 +423,15 @@ mod tests {

        Batch::new(
            primary_key,
-            Arc::new(UInt64Vector::from_iter_values(
-                iter::repeat(0).take(num_rows),
-            )),
-            Arc::new(UInt64Vector::from_iter_values(
-                iter::repeat(0).take(num_rows),
-            )),
-            Arc::new(UInt8Vector::from_iter_values(
-                iter::repeat(1).take(num_rows),
-            )),
+            Arc::new(UInt64Vector::from_iter_values(std::iter::repeat_n(
+                0, num_rows,
+            ))),
+            Arc::new(UInt64Vector::from_iter_values(std::iter::repeat_n(
+                0, num_rows,
+            ))),
+            Arc::new(UInt8Vector::from_iter_values(std::iter::repeat_n(
+                1, num_rows,
+            ))),
            vec![u64_field],
        )
        .unwrap()
--- a/src/mito2/src/sst/parquet/file_range.rs
+++ b/src/mito2/src/sst/parquet/file_range.rs
@@ -27,7 +27,7 @@ use snafu::{OptionExt, ResultExt};
 use store_api::storage::TimeSeriesRowSelector;

 use crate::error::{
-    DecodeStatsSnafu, FieldTypeMismatchSnafu, FilterRecordBatchSnafu, Result, StatsNotPresentSnafu,
+    DecodeStatsSnafu, FieldTypeMismatchSnafu, RecordBatchSnafu, Result, StatsNotPresentSnafu,
 };
 use crate::read::compat::CompatBatch;
 use crate::read::last_row::RowGroupLastRowCachedReader;
@@ -294,7 +294,7 @@ impl RangeBase {
                    };
                    if filter
                        .evaluate_scalar(&pk_value)
-                        .context(FilterRecordBatchSnafu)?
+                        .context(RecordBatchSnafu)?
                    {
                        continue;
                    } else {
@@ -311,11 +311,11 @@ impl RangeBase {
                    let field_col = &input.fields()[field_index].data;
                    filter
                        .evaluate_vector(field_col)
-                        .context(FilterRecordBatchSnafu)?
+                        .context(RecordBatchSnafu)?
                }
                SemanticType::Timestamp => filter
                    .evaluate_vector(input.timestamps())
-                    .context(FilterRecordBatchSnafu)?,
+                    .context(RecordBatchSnafu)?,
            };

            mask = mask.bitand(&result);
--- a/src/mito2/src/sst/parquet/format.rs
+++ b/src/mito2/src/sst/parquet/format.rs
@@ -134,7 +134,6 @@ impl WriteFormat {

 /// Helper for reading the SST format.
 pub struct ReadFormat {
-    /// The metadata stored in the SST.
    metadata: RegionMetadataRef,
    /// SST file schema.
    arrow_schema: SchemaRef,
@@ -306,23 +305,17 @@ impl ReadFormat {
        &self,
        row_groups: &[impl Borrow<RowGroupMetaData>],
        column_id: ColumnId,
-    ) -> StatValues {
-        let Some(column) = self.metadata.column_by_id(column_id) else {
-            // No such column in the SST.
-            return StatValues::NoColumn;
-        };
+    ) -> Option<ArrayRef> {
+        let column = self.metadata.column_by_id(column_id)?;
        match column.semantic_type {
            SemanticType::Tag => self.tag_values(row_groups, column, true),
            SemanticType::Field => {
-                // Safety: `field_id_to_index` is initialized by the semantic type.
-                let index = self.field_id_to_index.get(&column_id).unwrap();
-                let stats = Self::column_values(row_groups, column, *index, true);
-                StatValues::from_stats_opt(stats)
+                let index = self.field_id_to_index.get(&column_id)?;
+                Self::column_values(row_groups, column, *index, true)
            }
            SemanticType::Timestamp => {
                let index = self.time_index_position();
-                let stats = Self::column_values(row_groups, column, index, true);
-                StatValues::from_stats_opt(stats)
+                Self::column_values(row_groups, column, index, true)
            }
        }
    }
@@ -332,23 +325,17 @@ impl ReadFormat {
        &self,
        row_groups: &[impl Borrow<RowGroupMetaData>],
        column_id: ColumnId,
-    ) -> StatValues {
-        let Some(column) = self.metadata.column_by_id(column_id) else {
-            // No such column in the SST.
-            return StatValues::NoColumn;
-        };
+    ) -> Option<ArrayRef> {
+        let column = self.metadata.column_by_id(column_id)?;
        match column.semantic_type {
            SemanticType::Tag => self.tag_values(row_groups, column, false),
            SemanticType::Field => {
-                // Safety: `field_id_to_index` is initialized by the semantic type.
-                let index = self.field_id_to_index.get(&column_id).unwrap();
-                let stats = Self::column_values(row_groups, column, *index, false);
-                StatValues::from_stats_opt(stats)
+                let index = self.field_id_to_index.get(&column_id)?;
+                Self::column_values(row_groups, column, *index, false)
            }
            SemanticType::Timestamp => {
                let index = self.time_index_position();
-                let stats = Self::column_values(row_groups, column, index, false);
-                StatValues::from_stats_opt(stats)
+                Self::column_values(row_groups, column, index, false)
            }
        }
    }
@@ -358,23 +345,17 @@ impl ReadFormat {
        &self,
        row_groups: &[impl Borrow<RowGroupMetaData>],
        column_id: ColumnId,
-    ) -> StatValues {
-        let Some(column) = self.metadata.column_by_id(column_id) else {
-            // No such column in the SST.
-            return StatValues::NoColumn;
-        };
+    ) -> Option<ArrayRef> {
+        let column = self.metadata.column_by_id(column_id)?;
        match column.semantic_type {
-            SemanticType::Tag => StatValues::NoStats,
+            SemanticType::Tag => None,
            SemanticType::Field => {
-                // Safety: `field_id_to_index` is initialized by the semantic type.
-                let index = self.field_id_to_index.get(&column_id).unwrap();
-                let stats = Self::column_null_counts(row_groups, *index);
-                StatValues::from_stats_opt(stats)
+                let index = self.field_id_to_index.get(&column_id)?;
+                Self::column_null_counts(row_groups, *index)
            }
            SemanticType::Timestamp => {
                let index = self.time_index_position();
-                let stats = Self::column_null_counts(row_groups, index);
-                StatValues::from_stats_opt(stats)
+                Self::column_null_counts(row_groups, index)
            }
        }
    }
@@ -409,7 +390,8 @@ impl ReadFormat {
        row_groups: &[impl Borrow<RowGroupMetaData>],
        column: &ColumnMetadata,
        is_min: bool,
-    ) -> StatValues {
+    ) -> Option<ArrayRef> {
+        let primary_key_encoding = self.metadata.primary_key_encoding;
        let is_first_tag = self
            .metadata
            .primary_key
@@ -418,28 +400,9 @@ impl ReadFormat {
            .unwrap_or(false);
        if !is_first_tag {
            // Only the min-max of the first tag is available in the primary key.
-            return StatValues::NoStats;
+            return None;
        }

-        StatValues::from_stats_opt(self.first_tag_values(row_groups, column, is_min))
-    }
-
-    /// Returns min/max values of the first tag.
-    /// Returns None if the tag does not have statistics.
-    fn first_tag_values(
-        &self,
-        row_groups: &[impl Borrow<RowGroupMetaData>],
-        column: &ColumnMetadata,
-        is_min: bool,
-    ) -> Option<ArrayRef> {
-        debug_assert!(self
-            .metadata
-            .primary_key
-            .first()
-            .map(|id| *id == column.column_id)
-            .unwrap_or(false));
-
-        let primary_key_encoding = self.metadata.primary_key_encoding;
        let converter = build_primary_key_codec_with_fields(
            primary_key_encoding,
            [(
@@ -489,7 +452,6 @@ impl ReadFormat {
    }

    /// Returns min/max values of specific non-tag columns.
-    /// Returns None if the column does not have statistics.
    fn column_values(
        row_groups: &[impl Borrow<RowGroupMetaData>],
        column: &ColumnMetadata,
@@ -582,29 +544,6 @@ impl ReadFormat {
    }
 }

-/// Values of column statistics of the SST.
-///
-/// It also distinguishes the case that a column is not found and
-/// the column exists but has no statistics.
-pub enum StatValues {
-    /// Values of each row group.
-    Values(ArrayRef),
-    /// No such column.
-    NoColumn,
-    /// Column exists but has no statistics.
-    NoStats,
-}
-
-impl StatValues {
-    /// Creates a new `StatValues` instance from optional statistics.
-    pub fn from_stats_opt(stats: Option<ArrayRef>) -> Self {
-        match stats {
-            Some(stats) => StatValues::Values(stats),
-            None => StatValues::NoStats,
-        }
-    }
-}
-
 #[cfg(test)]
 impl ReadFormat {
    /// Creates a helper with existing `metadata` and all columns.
@@ -816,7 +755,7 @@ mod tests {
        ));
        let mut keys = vec![];
        for (index, num_rows) in pk_row_nums.iter().map(|v| v.1).enumerate() {
-            keys.extend(std::iter::repeat(index as u32).take(num_rows));
+            keys.extend(std::iter::repeat_n(index as u32, num_rows));
        }
        let keys = UInt32Array::from(keys);
        Arc::new(DictionaryArray::new(keys, values))
--- a/src/mito2/src/sst/parquet/stats.rs
+++ b/src/mito2/src/sst/parquet/stats.rs
@@ -25,7 +25,7 @@ use parquet::file::metadata::RowGroupMetaData;
 use store_api::metadata::RegionMetadataRef;
 use store_api::storage::ColumnId;

-use crate::sst::parquet::format::{ReadFormat, StatValues};
+use crate::sst::parquet::format::ReadFormat;

 /// Statistics for pruning row groups.
 pub(crate) struct RowGroupPruningStats<'a, T> {
@@ -100,18 +100,16 @@ impl<T: Borrow<RowGroupMetaData>> PruningStatistics for RowGroupPruningStats<'_,
    fn min_values(&self, column: &Column) -> Option<ArrayRef> {
        let column_id = self.column_id_to_prune(&column.name)?;
        match self.read_format.min_values(self.row_groups, column_id) {
-            StatValues::Values(values) => Some(values),
-            StatValues::NoColumn => self.compat_default_value(&column.name),
-            StatValues::NoStats => None,
+            Some(values) => Some(values),
+            None => self.compat_default_value(&column.name),
        }
    }

    fn max_values(&self, column: &Column) -> Option<ArrayRef> {
        let column_id = self.column_id_to_prune(&column.name)?;
        match self.read_format.max_values(self.row_groups, column_id) {
-            StatValues::Values(values) => Some(values),
-            StatValues::NoColumn => self.compat_default_value(&column.name),
-            StatValues::NoStats => None,
+            Some(values) => Some(values),
+            None => self.compat_default_value(&column.name),
        }
    }

@@ -120,12 +118,10 @@ impl<T: Borrow<RowGroupMetaData>> PruningStatistics for RowGroupPruningStats<'_,
    }

    fn null_counts(&self, column: &Column) -> Option<ArrayRef> {
-        let column_id = self.column_id_to_prune(&column.name)?;
-        match self.read_format.null_counts(self.row_groups, column_id) {
-            StatValues::Values(values) => Some(values),
-            StatValues::NoColumn => self.compat_null_count(&column.name),
-            StatValues::NoStats => None,
-        }
+        let Some(column_id) = self.column_id_to_prune(&column.name) else {
+            return self.compat_null_count(&column.name);
+        };
+        self.read_format.null_counts(self.row_groups, column_id)
    }

    fn row_counts(&self, _column: &Column) -> Option<ArrayRef> {
--- a/src/mito2/src/sst/parquet/writer.rs
+++ b/src/mito2/src/sst/parquet/writer.rs
@@ -36,7 +36,7 @@ use store_api::storage::SequenceNumber;
 use tokio::io::AsyncWrite;
 use tokio_util::compat::{Compat, FuturesAsyncWriteCompatExt};

-use crate::access_layer::{FilePathProvider, SstInfoArray, TempFileCleaner};
+use crate::access_layer::{FilePathProvider, SstInfoArray};
 use crate::error::{InvalidMetadataSnafu, OpenDalSnafu, Result, WriteParquetSnafu};
 use crate::read::{Batch, Source};
 use crate::sst::file::FileId;
@@ -61,8 +61,6 @@ pub struct ParquetWriter<F: WriterFactory, I: IndexerBuilder, P: FilePathProvide
    /// Current active indexer.
    current_indexer: Option<Indexer>,
    bytes_written: Arc<AtomicUsize>,
-    /// Cleaner to remove temp files on failure.
-    file_cleaner: Option<TempFileCleaner>,
 }

 pub trait WriterFactory {
@@ -107,11 +105,6 @@ where
        )
        .await
    }
-
-    pub(crate) fn with_file_cleaner(mut self, cleaner: TempFileCleaner) -> Self {
-        self.file_cleaner = Some(cleaner);
-        self
-    }
 }

 impl<F, I, P> ParquetWriter<F, I, P>
@@ -139,7 +132,6 @@ where
            indexer_builder,
            current_indexer: Some(indexer),
            bytes_written: Arc::new(AtomicUsize::new(0)),
-            file_cleaner: None,
        }
    }

@@ -160,25 +152,6 @@ where
    ///
    /// Returns the [SstInfo] if the SST is written.
    pub async fn write_all(
-        &mut self,
-        source: Source,
-        override_sequence: Option<SequenceNumber>, // override the `sequence` field from `Source`
-        opts: &WriteOptions,
-    ) -> Result<SstInfoArray> {
-        let res = self
-            .write_all_without_cleaning(source, override_sequence, opts)
-            .await;
-        if res.is_err() {
-            // Clean tmp files explicitly on failure.
-            let file_id = self.current_file;
-            if let Some(cleaner) = &self.file_cleaner {
-                cleaner.clean_by_file_id(file_id).await;
-            }
-        }
-        res
-    }
-
-    async fn write_all_without_cleaning(
        &mut self,
        mut source: Source,
        override_sequence: Option<SequenceNumber>, // override the `sequence` field from `Source`
--- a/src/mito2/src/test_util.rs
+++ b/src/mito2/src/test_util.rs
@@ -145,12 +145,6 @@ pub(crate) async fn prepare_test_for_kafka_log_store(factory: &LogStoreFactory)
 }

 pub(crate) async fn append_noop_record(client: &Client, topic: &str) {
-    let controller_client = client.controller_client().unwrap();
-    controller_client
-        .create_topic(topic, 1, 1, 5000)
-        .await
-        .unwrap();
-
    let partition_client = client
        .partition_client(topic, 0, UnknownTopicHandling::Retry)
        .await
@@ -665,27 +659,6 @@ impl TestEnv {
        Arc::new(write_cache)
    }

-    /// Creates a write cache from a path.
-    pub async fn create_write_cache_from_path(
-        &self,
-        path: &str,
-        capacity: ReadableSize,
-    ) -> WriteCacheRef {
-        let index_aux_path = self.data_home.path().join("index_aux");
-        let puffin_mgr = PuffinManagerFactory::new(&index_aux_path, 4096, None, None)
-            .await
-            .unwrap();
-        let intm_mgr = IntermediateManager::init_fs(index_aux_path.to_str().unwrap())
-            .await
-            .unwrap();
-
-        let write_cache = WriteCache::new_fs(path, capacity, None, puffin_mgr, intm_mgr)
-            .await
-            .unwrap();
-
-        Arc::new(write_cache)
-    }
-
    pub fn get_schema_metadata_manager(&self) -> SchemaMetadataManagerRef {
        self.schema_metadata_manager.clone()
    }
--- a/src/operator/Cargo.toml
+++ b/src/operator/Cargo.toml
@@ -63,6 +63,5 @@ tokio-util.workspace = true
 tonic.workspace = true

 [dev-dependencies]
-common-meta = { workspace = true, features = ["testing"] }
 common-test-util.workspace = true
 path-slash = "0.2"
--- a/src/operator/src/insert.rs
+++ b/src/operator/src/insert.rs
@@ -147,7 +147,7 @@ impl Inserter {
        statement_executor: &StatementExecutor,
    ) -> Result<Output> {
        let row_inserts = ColumnToRow::convert(requests)?;
-        self.handle_row_inserts(row_inserts, ctx, statement_executor, false)
+        self.handle_row_inserts(row_inserts, ctx, statement_executor)
            .await
    }

@@ -157,7 +157,6 @@ impl Inserter {
        mut requests: RowInsertRequests,
        ctx: QueryContextRef,
        statement_executor: &StatementExecutor,
-        accommodate_existing_schema: bool,
    ) -> Result<Output> {
        preprocess_row_insert_requests(&mut requests.inserts)?;
        self.handle_row_inserts_with_create_type(
@@ -165,7 +164,6 @@ impl Inserter {
            ctx,
            statement_executor,
            AutoCreateTableType::Physical,
-            accommodate_existing_schema,
        )
        .await
    }
@@ -182,7 +180,6 @@ impl Inserter {
            ctx,
            statement_executor,
            AutoCreateTableType::Log,
-            false,
        )
        .await
    }
@@ -198,7 +195,6 @@ impl Inserter {
            ctx,
            statement_executor,
            AutoCreateTableType::Trace,
-            false,
        )
        .await
    }
@@ -209,14 +205,12 @@ impl Inserter {
        requests: RowInsertRequests,
        ctx: QueryContextRef,
        statement_executor: &StatementExecutor,
-        accommodate_existing_schema: bool,
    ) -> Result<Output> {
        self.handle_row_inserts_with_create_type(
            requests,
            ctx,
            statement_executor,
            AutoCreateTableType::LastNonNull,
-            accommodate_existing_schema,
        )
        .await
    }
@@ -228,7 +222,6 @@ impl Inserter {
        ctx: QueryContextRef,
        statement_executor: &StatementExecutor,
        create_type: AutoCreateTableType,
-        accommodate_existing_schema: bool,
    ) -> Result<Output> {
        // remove empty requests
        requests.inserts.retain(|req| {
@@ -243,13 +236,7 @@ impl Inserter {
            instant_table_ids,
            table_infos,
        } = self
-            .create_or_alter_tables_on_demand(
-                &mut requests,
-                &ctx,
-                create_type,
-                statement_executor,
-                accommodate_existing_schema,
-            )
+            .create_or_alter_tables_on_demand(&requests, &ctx, create_type, statement_executor)
            .await?;

        let name_to_info = table_infos
@@ -294,11 +281,10 @@ impl Inserter {
            table_infos,
        } = self
            .create_or_alter_tables_on_demand(
-                &mut requests,
+                &requests,
                &ctx,
                AutoCreateTableType::Logical(physical_table.to_string()),
                statement_executor,
-                true,
            )
            .await?;
        let name_to_info = table_infos
@@ -462,18 +448,12 @@ impl Inserter {
    ///
    /// Returns a mapping from table name to table id, where table name is the table name involved in the requests.
    /// This mapping is used in the conversion of RowToRegion.
-    ///
-    /// `accommodate_existing_schema` is used to determine if the existing schema should override the new schema.
-    /// It only works for TIME_INDEX and VALUE columns. This is for the case where the user creates a table with
-    /// custom schema, and then inserts data with endpoints that have default schema setting, like prometheus
-    /// remote write. This will modify the `RowInsertRequests` in place.
    async fn create_or_alter_tables_on_demand(
        &self,
-        requests: &mut RowInsertRequests,
+        requests: &RowInsertRequests,
        ctx: &QueryContextRef,
        auto_create_table_type: AutoCreateTableType,
        statement_executor: &StatementExecutor,
-        accommodate_existing_schema: bool,
    ) -> Result<CreateAlterTableResult> {
        let _timer = crate::metrics::CREATE_ALTER_ON_DEMAND
            .with_label_values(&[auto_create_table_type.as_str()])
@@ -524,7 +504,7 @@ impl Inserter {
        let mut alter_tables = vec![];
        let mut instant_table_ids = HashSet::new();

-        for req in &mut requests.inserts {
+        for req in &requests.inserts {
            match self.get_table(catalog, &schema, &req.table_name).await? {
                Some(table) => {
                    let table_info = table.table_info();
@@ -532,12 +512,9 @@ impl Inserter {
                        instant_table_ids.insert(table_info.table_id());
                    }
                    table_infos.insert(table_info.table_id(), table.table_info());
-                    if let Some(alter_expr) = self.get_alter_table_expr_on_demand(
-                        req,
-                        &table,
-                        ctx,
-                        accommodate_existing_schema,
-                    )? {
+                    if let Some(alter_expr) =
+                        self.get_alter_table_expr_on_demand(req, &table, ctx)?
+                    {
                        alter_tables.push(alter_expr);
                    }
                }
@@ -807,16 +784,12 @@ impl Inserter {
    }

    /// Returns an alter table expression if it finds new columns in the request.
-    /// When `accommodate_existing_schema` is false, it always adds columns if not exist.
-    /// When `accommodate_existing_schema` is true, it may modify the input `req` to
-    /// accommodate it with existing schema. See [`create_or_alter_tables_on_demand`](Self::create_or_alter_tables_on_demand)
-    /// for more details.
+    /// It always adds columns if not exist.
    fn get_alter_table_expr_on_demand(
        &self,
-        req: &mut RowInsertRequest,
+        req: &RowInsertRequest,
        table: &TableRef,
        ctx: &QueryContextRef,
-        accommodate_existing_schema: bool,
    ) -> Result<Option<AlterTableExpr>> {
        let catalog_name = ctx.current_catalog();
        let schema_name = ctx.current_schema();
@@ -825,64 +798,10 @@ impl Inserter {
        let request_schema = req.rows.as_ref().unwrap().schema.as_slice();
        let column_exprs = ColumnExpr::from_column_schemas(request_schema);
        let add_columns = expr_helper::extract_add_columns_expr(&table.schema(), column_exprs)?;
-        let Some(mut add_columns) = add_columns else {
+        let Some(add_columns) = add_columns else {
            return Ok(None);
        };

-        // If accommodate_existing_schema is true, update request schema for Timestamp/Field columns
-        if accommodate_existing_schema {
-            let table_schema = table.schema();
-            // Find timestamp column name
-            let ts_col_name = table_schema.timestamp_column().map(|c| c.name.clone());
-            // Find field column name if there is only one
-            let mut field_col_name = None;
-            let mut multiple_field_cols = false;
-            table.field_columns().for_each(|col| {
-                if field_col_name.is_none() {
-                    field_col_name = Some(col.name.clone());
-                } else {
-                    multiple_field_cols = true;
-                }
-            });
-            if multiple_field_cols {
-                field_col_name = None;
-            }
-
-            // Update column name in request schema for Timestamp/Field columns
-            if let Some(rows) = req.rows.as_mut() {
-                for col in &mut rows.schema {
-                    match col.semantic_type {
-                        x if x == SemanticType::Timestamp as i32 => {
-                            if let Some(ref ts_name) = ts_col_name {
-                                if col.column_name != *ts_name {
-                                    col.column_name = ts_name.clone();
-                                }
-                            }
-                        }
-                        x if x == SemanticType::Field as i32 => {
-                            if let Some(ref field_name) = field_col_name {
-                                if col.column_name != *field_name {
-                                    col.column_name = field_name.clone();
-                                }
-                            }
-                        }
-                        _ => {}
-                    }
-                }
-            }
-
-            // Remove from add_columns any column that is timestamp or field (if there is only one field column)
-            add_columns.add_columns.retain(|col| {
-                let def = col.column_def.as_ref().unwrap();
-                def.semantic_type != SemanticType::Timestamp as i32
-                    && (def.semantic_type != SemanticType::Field as i32 && field_col_name.is_some())
-            });
-
-            if add_columns.add_columns.is_empty() {
-                return Ok(None);
-            }
-        }
-
        Ok(Some(AlterTableExpr {
            catalog_name: catalog_name.to_string(),
            schema_name: schema_name.to_string(),
@@ -1027,7 +946,6 @@ impl FlowMirrorTask {
                // already know this is not source table
                Some(None) => continue,
                _ => {
-                    // dedup peers
                    let peers = cache
                        .get(table_id)
                        .await
@@ -1035,8 +953,6 @@ impl FlowMirrorTask {
                        .unwrap_or_default()
                        .values()
                        .cloned()
-                        .collect::<HashSet<_>>()
-                        .into_iter()
                        .collect::<Vec<_>>();

                    if !peers.is_empty() {
@@ -1116,124 +1032,3 @@ impl FlowMirrorTask {
        Ok(())
    }
 }
-
-#[cfg(test)]
-mod tests {
-    use std::sync::Arc;
-
-    use api::v1::{ColumnSchema as GrpcColumnSchema, RowInsertRequest, Rows, SemanticType, Value};
-    use common_catalog::consts::{DEFAULT_CATALOG_NAME, DEFAULT_SCHEMA_NAME};
-    use common_meta::cache::new_table_flownode_set_cache;
-    use common_meta::ddl::test_util::datanode_handler::NaiveDatanodeHandler;
-    use common_meta::test_util::MockDatanodeManager;
-    use datatypes::data_type::ConcreteDataType;
-    use datatypes::schema::ColumnSchema;
-    use moka::future::Cache;
-    use session::context::QueryContext;
-    use table::dist_table::DummyDataSource;
-    use table::metadata::{TableInfoBuilder, TableMetaBuilder, TableType};
-    use table::TableRef;
-
-    use super::*;
-    use crate::tests::{create_partition_rule_manager, prepare_mocked_backend};
-
-    fn make_table_ref_with_schema(ts_name: &str, field_name: &str) -> TableRef {
-        let schema = datatypes::schema::SchemaBuilder::try_from_columns(vec![
-            ColumnSchema::new(
-                ts_name,
-                ConcreteDataType::timestamp_millisecond_datatype(),
-                false,
-            )
-            .with_time_index(true),
-            ColumnSchema::new(field_name, ConcreteDataType::float64_datatype(), true),
-        ])
-        .unwrap()
-        .build()
-        .unwrap();
-        let meta = TableMetaBuilder::empty()
-            .schema(Arc::new(schema))
-            .primary_key_indices(vec![])
-            .value_indices(vec![1])
-            .engine("mito")
-            .next_column_id(0)
-            .options(Default::default())
-            .created_on(Default::default())
-            .region_numbers(vec![0])
-            .build()
-            .unwrap();
-        let info = Arc::new(
-            TableInfoBuilder::default()
-                .table_id(1)
-                .table_version(0)
-                .name("test_table")
-                .schema_name(DEFAULT_SCHEMA_NAME)
-                .catalog_name(DEFAULT_CATALOG_NAME)
-                .desc(None)
-                .table_type(TableType::Base)
-                .meta(meta)
-                .build()
-                .unwrap(),
-        );
-        Arc::new(table::Table::new(
-            info,
-            table::metadata::FilterPushDownType::Unsupported,
-            Arc::new(DummyDataSource),
-        ))
-    }
-
-    #[tokio::test]
-    async fn test_accommodate_existing_schema_logic() {
-        let ts_name = "my_ts";
-        let field_name = "my_field";
-        let table = make_table_ref_with_schema(ts_name, field_name);
-
-        // The request uses different names for timestamp and field columns
-        let mut req = RowInsertRequest {
-            table_name: "test_table".to_string(),
-            rows: Some(Rows {
-                schema: vec![
-                    GrpcColumnSchema {
-                        column_name: "ts_wrong".to_string(),
-                        datatype: api::v1::ColumnDataType::TimestampMillisecond as i32,
-                        semantic_type: SemanticType::Timestamp as i32,
-                        ..Default::default()
-                    },
-                    GrpcColumnSchema {
-                        column_name: "field_wrong".to_string(),
-                        datatype: api::v1::ColumnDataType::Float64 as i32,
-                        semantic_type: SemanticType::Field as i32,
-                        ..Default::default()
-                    },
-                ],
-                rows: vec![api::v1::Row {
-                    values: vec![Value::default(), Value::default()],
-                }],
-            }),
-        };
-        let ctx = Arc::new(QueryContext::with(
-            DEFAULT_CATALOG_NAME,
-            DEFAULT_SCHEMA_NAME,
-        ));
-
-        let kv_backend = prepare_mocked_backend().await;
-        let inserter = Inserter::new(
-            catalog::memory::MemoryCatalogManager::new(),
-            create_partition_rule_manager(kv_backend.clone()).await,
-            Arc::new(MockDatanodeManager::new(NaiveDatanodeHandler)),
-            Arc::new(new_table_flownode_set_cache(
-                String::new(),
-                Cache::new(100),
-                kv_backend.clone(),
-            )),
-        );
-        let alter_expr = inserter
-            .get_alter_table_expr_on_demand(&mut req, &table, &ctx, true)
-            .unwrap();
-        assert!(alter_expr.is_none());
-
-        // The request's schema should have updated names for timestamp and field columns
-        let req_schema = req.rows.as_ref().unwrap().schema.clone();
-        assert_eq!(req_schema[0].column_name, ts_name);
-        assert_eq!(req_schema[1].column_name, field_name);
-    }
-}
--- a/Show More
+++ b/Show More
Author	SHA1	Message	Date
Ruihang Xia	edd8cb6710	add rate, increase and delta Signed-off-by: Ruihang Xia <waynestxia@gmail.com>	2025-04-29 10:54:32 +08:00
Ruihang Xia	7ee61e5d28	disable recursion limit Signed-off-by: Ruihang Xia <waynestxia@gmail.com>	2025-04-28 16:58:30 +08:00
Ruihang Xia	1b30aca5a5	Merge branch 'prom-plan-commutativity' into jkt	2025-04-27 19:23:42 +08:00
Ruihang Xia	99b352cea1	Merge branch 'main' into prom-plan-commutativity	2025-04-27 17:40:24 +08:00
Ruihang Xia	0f521956bf	update tests Signed-off-by: Ruihang Xia <waynestxia@gmail.com>	2025-04-27 17:35:50 +08:00
Ruihang Xia	aee72ab363	fix clippy Signed-off-by: Ruihang Xia <waynestxia@gmail.com>	2025-04-27 17:35:29 +08:00
shuiyisong	3c943be189	chore: update rust toolchain (#5818 ) * chore: update nightly version * chore: sort lint lines * chore: minor fix * chore: update nix * chore: update toolchain to 2024-04-14 * chore: update toolchain to 2024-04-15 * chore: remove unnecessory test * chore: do not assert oid in sqlness test * chore: fix margin issue * chore: fix cr issues * chore: fix cr issues --------- Co-authored-by: Ning Sun <sunning@greptime.com>	2025-04-27 09:02:36 +00:00
Ruihang Xia	5b78d76fc5	handle partition and ordering Signed-off-by: Ruihang Xia <waynestxia@gmail.com>	2025-04-27 15:38:57 +08:00
Ning Sun	eeba466717	ci: read next release version from toml by default (#5986 ) * ci: read next release version from toml by default * ci: send error message to stderr * ci: take the first version only	2025-04-27 04:43:44 +00:00
Zhenchi	2ff54486d3	chore: bump main branch version to 0.15 (#5984 ) Signed-off-by: Zhenchi <zhongzc_arch@outlook.com>	2025-04-27 01:39:44 +00:00
evenyag	a166430650	chore: fix typos	2025-04-24 06:08:42 +08:00
evenyag	007a2b3dfe	refactor: use ChainedRecordBatchStream to simplify codes	2025-04-24 06:03:07 +08:00
evenyag	f35e957ddd	style: fix clippy	2025-04-24 06:03:07 +08:00
evenyag	68414bf593	feat: metrics for send series timeout	2025-04-24 06:03:07 +08:00
evenyag	5e836a0d1b	chore: display more info in explain	2025-04-24 06:03:07 +08:00
evenyag	f5e0da2fc8	fix: scanner metrics init	2025-04-24 06:03:07 +08:00
evenyag	fb96d26ebf	chore: reset instant	2025-04-24 06:03:07 +08:00
evenyag	0046d3f65b	fix: correct scan cost metrics	2025-04-24 06:03:06 +08:00
evenyag	d7b97fc877	fix: add metrics to partition metrics list	2025-04-24 06:03:06 +08:00
evenyag	bfdaa28b25	chore: add comments	2025-04-24 06:03:06 +08:00
evenyag	6293bb1f5b	feat: try send before send reduce the send timeout to 10ms	2025-04-24 06:03:06 +08:00
Ruihang Xia	8fa1ebcc3e	assign partition_ranges Signed-off-by: Ruihang Xia <waynestxia@gmail.com>	2025-04-24 06:03:06 +08:00
Ruihang Xia	c18c3f5839	Revert "feat: keep parallelize_scan unchanged" This reverts commit `96ba00d175`.	2025-04-24 06:03:06 +08:00
evenyag	629e72d8c0	chore: update comment	2025-04-24 06:03:06 +08:00
evenyag	e4065505ab	feat: use smallvec	2025-04-24 06:03:05 +08:00
evenyag	aafd164483	fix: include build merge reader cost to scan cost	2025-04-24 06:03:05 +08:00
evenyag	1386e903d6	fix: address compiler errors	2025-04-24 06:03:05 +08:00
evenyag	12692a940c	feat: keep parallelize_scan unchanged	2025-04-24 06:03:05 +08:00
evenyag	4d44cbb8b2	fix: use series scan in PerSeries distribution	2025-04-24 06:03:05 +08:00
evenyag	f4911aa3bb	refactor: remove per series scan from SeqScan	2025-04-24 06:03:05 +08:00
evenyag	5ac61f17bc	feat: parallelize PerSeries	2025-04-24 06:03:05 +08:00
evenyag	e0d34c6d95	feat: use series scan when distribution is PerSeries	2025-04-24 06:03:04 +08:00
evenyag	8a98b9c433	feat: implement scan logic of each partition	2025-04-24 06:03:03 +08:00
evenyag	1f5d36a203	chore: add to scanner enum	2025-04-24 05:57:37 +08:00
evenyag	6fc7168893	chore: basic methods for SeriesScan	2025-04-24 05:57:37 +08:00
Ruihang Xia	2799d67212	change dictionary type Signed-off-by: Ruihang Xia <waynestxia@gmail.com>	2025-04-11 17:07:55 +08:00
Ruihang Xia	d97a76c312	blocklist in commutativity rule Signed-off-by: Ruihang Xia <waynestxia@gmail.com>	2025-04-11 17:07:16 +08:00
Ruihang Xia	15caca244e	fix range manipulate deserializer Signed-off-by: Ruihang Xia <waynestxia@gmail.com>	2025-04-10 19:38:03 +08:00
Ruihang Xia	8638075cdd	feat: implement commutativity rule for prom-related plans Signed-off-by: Ruihang Xia <waynestxia@gmail.com>	2025-04-10 16:26:31 +08:00