fix: check and clippy

Signed-off-by: shuiyisong <xixing.sys@gmail.com>

.github/CODEOWNERS

@@ -5,23 +5,23 @@
 * @GreptimeTeam/db-approver
 
 ## [Module] Database Engine
-/src/index @zhongzc
+/src/index @evenyag @discord9 @WenyXu 
 /src/mito2 @evenyag @v0y4g3r @waynexia
-/src/query @evenyag
+/src/query @evenyag @waynexia @discord9
 
 ## [Module] Distributed
-/src/common/meta @MichaelScofield
-/src/common/procedure @MichaelScofield
-/src/meta-client @MichaelScofield
-/src/meta-srv @MichaelScofield
+/src/common/meta @MichaelScofield @WenyXu
+/src/common/procedure @MichaelScofield @WenyXu
+/src/meta-client @MichaelScofield @WenyXu
+/src/meta-srv @MichaelScofield @WenyXu
 
 ## [Module] Write Ahead Log
-/src/log-store @v0y4g3r
-/src/store-api @v0y4g3r
+/src/log-store @v0y4g3r @WenyXu
+/src/store-api @v0y4g3r @evenyag
 
 ## [Module] Metrics Engine
-/src/metric-engine @waynexia
-/src/promql @waynexia
+/src/metric-engine @waynexia @WenyXu
+/src/promql @waynexia @evenyag @discord9
 
 ## [Module] Flow
-/src/flow @zhongzc @waynexia
+/src/flow @discord9 @waynexia

.github/actions/build-greptime-binary/action.yml

@@ -32,9 +32,23 @@ inputs:
     description: Image Registry
     required: false
     default: 'docker.io'
+  large-page-size:
+    description: Build GreptimeDB with large page size (65536).
+    required: false
+    default: 'false'
+
 runs:
   using: composite
   steps:
+    - name: Set extra build environment variables
+      shell: bash
+      run: |
+        if [[ '${{ inputs.large-page-size }}' == 'true' ]]; then
+          echo 'EXTRA_BUILD_ENVS="JEMALLOC_SYS_WITH_LG_PAGE=16"' >> $GITHUB_ENV
+        else
+          echo 'EXTRA_BUILD_ENVS=' >> $GITHUB_ENV
+        fi
+
     - name: Build greptime binary
       shell: bash
       if: ${{ inputs.build-android-artifacts == 'false' }}
@@ -45,7 +59,8 @@ runs:
           FEATURES=${{ inputs.features }} \
           BASE_IMAGE=${{ inputs.base-image }} \
           IMAGE_NAMESPACE=${{ inputs.image-namespace }} \
-          IMAGE_REGISTRY=${{ inputs.image-registry }}
+          IMAGE_REGISTRY=${{ inputs.image-registry }} \
+          EXTRA_BUILD_ENVS=$EXTRA_BUILD_ENVS
 
     - name: Upload artifacts
       uses: ./.github/actions/upload-artifacts

.github/actions/build-linux-artifacts/action.yml

@@ -27,6 +27,10 @@ inputs:
     description: Working directory to build the artifacts
     required: false
     default: .
+  large-page-size:
+    description: Build GreptimeDB with large page size (65536).
+    required: false
+    default: 'false'
 runs:
   using: composite
   steps:
@@ -59,6 +63,7 @@ runs:
         working-dir: ${{ inputs.working-dir }}
         image-registry: ${{ inputs.image-registry }}
         image-namespace: ${{ inputs.image-namespace }}
+        large-page-size: ${{ inputs.large-page-size }}
 
     - name: Clean up the target directory # Clean up the target directory for the centos7 base image, or it will still use the objects of last build.
       shell: bash
@@ -77,6 +82,7 @@ runs:
         working-dir: ${{ inputs.working-dir }}
         image-registry: ${{ inputs.image-registry }}
         image-namespace: ${{ inputs.image-namespace }}
+        large-page-size: ${{ inputs.large-page-size }}
 
     - name: Build greptime on android base image
       uses: ./.github/actions/build-greptime-binary
@@ -89,3 +95,4 @@ runs:
         build-android-artifacts: true
         image-registry: ${{ inputs.image-registry }}
         image-namespace: ${{ inputs.image-namespace }}
+        large-page-size: ${{ inputs.large-page-size }}

.github/workflows/dev-build.yml

@@ -4,10 +4,11 @@ name: GreptimeDB Development Build
 on:
   workflow_dispatch: # Allows you to run this workflow manually.
     inputs:
-      repository:
-        description: The public repository to build
+      large-page-size:
+        description: Build GreptimeDB with large page size (65536).
+        type: boolean
         required: false
-        default: GreptimeTeam/greptimedb
+        default: false
       commit: # Note: We only pull the source code and use the current workflow to build the artifacts.
         description: The commit to build
         required: true
@@ -181,6 +182,7 @@ jobs:
           working-dir: ${{ env.CHECKOUT_GREPTIMEDB_PATH }}
           image-registry: ${{ vars.ECR_IMAGE_REGISTRY }}
           image-namespace: ${{ vars.ECR_IMAGE_NAMESPACE }}
+          large-page-size: ${{ inputs.large-page-size }}
 
   build-linux-arm64-artifacts:
     name: Build linux-arm64 artifacts
@@ -214,6 +216,7 @@ jobs:
           working-dir: ${{ env.CHECKOUT_GREPTIMEDB_PATH }}
           image-registry: ${{ vars.ECR_IMAGE_REGISTRY }}
           image-namespace: ${{ vars.ECR_IMAGE_NAMESPACE }}
+          large-page-size: ${{ inputs.large-page-size }}
 
   release-images-to-dockerhub:
     name: Build and push images to DockerHub

.github/workflows/multi-lang-tests.yml

@@ -0,0 +1,57 @@
+name: Multi-language Integration Tests
+
+on:
+  push:
+    branches:
+      - main
+  workflow_dispatch:
+
+concurrency:
+  group: ${{ github.workflow }}-${{ github.head_ref || github.run_id }}
+  cancel-in-progress: true
+
+jobs:
+  build-greptimedb:
+    if: ${{ github.repository == 'GreptimeTeam/greptimedb' }}
+    name: Build GreptimeDB binary
+    runs-on: ubuntu-latest
+    timeout-minutes: 60
+    steps:
+      - uses: actions/checkout@v4
+        with:
+          persist-credentials: false
+      - uses: arduino/setup-protoc@v3
+        with:
+          repo-token: ${{ secrets.GITHUB_TOKEN }}
+      - uses: actions-rust-lang/setup-rust-toolchain@v1
+      - uses: Swatinem/rust-cache@v2
+        with:
+          shared-key: "multi-lang-build"
+          cache-all-crates: "true"
+          save-if: ${{ github.ref == 'refs/heads/main' }}
+      - name: Install cargo-gc-bin
+        shell: bash
+        run: cargo install cargo-gc-bin --force
+      - name: Build greptime binary
+        shell: bash
+        run: cargo gc -- --bin greptime --features "pg_kvbackend,mysql_kvbackend"
+      - name: Pack greptime binary
+        shell: bash
+        run: |
+          mkdir bin && \
+          mv ./target/debug/greptime bin
+      - name: Print greptime binary info
+        run: ls -lh bin
+      - name: Upload greptime binary
+        uses: actions/upload-artifact@v4
+        with:
+          name: greptime-bin
+          path: bin/
+          retention-days: 1
+
+  run-multi-lang-tests:
+    name: Run Multi-language SDK Tests
+    needs: build-greptimedb
+    uses: ./.github/workflows/run-multi-lang-tests.yml
+    with:
+      artifact-name: greptime-bin

.github/workflows/nightly-build.yml

@@ -174,6 +174,18 @@ jobs:
           image-registry: ${{ vars.ECR_IMAGE_REGISTRY }}
           image-namespace: ${{ vars.ECR_IMAGE_NAMESPACE }}
 
+  run-multi-lang-tests:
+    name: Run Multi-language SDK Tests
+    if: ${{ inputs.build_linux_amd64_artifacts || github.event_name == 'schedule' }}
+    needs: [
+      allocate-runners,
+      build-linux-amd64-artifacts,
+    ]
+    uses: ./.github/workflows/run-multi-lang-tests.yml
+    with:
+      artifact-name: greptime-linux-amd64-${{ needs.allocate-runners.outputs.version }}
+      artifact-is-tarball: true
+
   release-images-to-dockerhub:
     name: Build and push images to DockerHub
     if: ${{ inputs.release_images || github.event_name == 'schedule' }}
@@ -301,7 +313,8 @@ jobs:
     if: ${{ github.repository == 'GreptimeTeam/greptimedb' && always() }} # Not requiring successful dependent jobs, always run.
     name: Send notification to Greptime team
     needs: [
-      release-images-to-dockerhub
+      release-images-to-dockerhub,
+      run-multi-lang-tests,
     ]
     runs-on: ubuntu-latest
     permissions:
@@ -319,17 +332,17 @@ jobs:
         run: pnpm tsx bin/report-ci-failure.ts
         env:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
-          CI_REPORT_STATUS: ${{ needs.release-images-to-dockerhub.outputs.nightly-build-result == 'success' }}
+          CI_REPORT_STATUS: ${{ needs.release-images-to-dockerhub.outputs.nightly-build-result == 'success' && (needs.run-multi-lang-tests.result == 'success' || needs.run-multi-lang-tests.result == 'skipped') }}
       - name: Notify nightly build successful result
         uses: slackapi/slack-github-action@v1.23.0
-        if: ${{ needs.release-images-to-dockerhub.outputs.nightly-build-result == 'success' }}
+        if: ${{ needs.release-images-to-dockerhub.outputs.nightly-build-result == 'success' && (needs.run-multi-lang-tests.result == 'success' || needs.run-multi-lang-tests.result == 'skipped') }}
         with:
           payload: |
             {"text": "GreptimeDB's ${{ env.NEXT_RELEASE_VERSION }} build has completed successfully."}
 
       - name: Notify nightly build failed result
         uses: slackapi/slack-github-action@v1.23.0
-        if: ${{ needs.release-images-to-dockerhub.outputs.nightly-build-result != 'success' }}
+        if: ${{ needs.release-images-to-dockerhub.outputs.nightly-build-result != 'success' || needs.run-multi-lang-tests.result == 'failure' }}
         with:
           payload: |
             {"text": "GreptimeDB's ${{ env.NEXT_RELEASE_VERSION }} build has failed, please check ${{ steps.report-ci-status.outputs.html_url }}."}

.github/workflows/release.yml

@@ -215,6 +215,18 @@ jobs:
           image-registry: ${{ vars.ECR_IMAGE_REGISTRY }}
           image-namespace: ${{ vars.ECR_IMAGE_NAMESPACE }}
 
+  run-multi-lang-tests:
+    name: Run Multi-language SDK Tests
+    if: ${{ inputs.build_linux_amd64_artifacts || github.event_name == 'push' || github.event_name == 'schedule' }}
+    needs: [
+      allocate-runners,
+      build-linux-amd64-artifacts,
+    ]
+    uses: ./.github/workflows/run-multi-lang-tests.yml
+    with:
+      artifact-name: greptime-linux-amd64-${{ needs.allocate-runners.outputs.version }}
+      artifact-is-tarball: true
+
   build-macos-artifacts:
     name: Build macOS artifacts
     strategy:
@@ -303,6 +315,7 @@ jobs:
       allocate-runners,
       build-linux-amd64-artifacts,
       build-linux-arm64-artifacts,
+      run-multi-lang-tests,
     ]
     runs-on: ubuntu-latest
     outputs:
@@ -381,6 +394,7 @@ jobs:
       build-macos-artifacts,
       build-windows-artifacts,
       release-images-to-dockerhub,
+      run-multi-lang-tests,
     ]
     runs-on: ubuntu-latest
     steps:

.github/workflows/run-multi-lang-tests.yml

@@ -0,0 +1,194 @@
+# Reusable workflow for running multi-language SDK tests against GreptimeDB
+# Used by: multi-lang-tests.yml, release.yml, nightly-build.yml
+# Supports both direct binary artifacts and tarball artifacts
+
+name: Run Multi-language SDK Tests
+
+on:
+  workflow_call:
+    inputs:
+      artifact-name:
+        required: true
+        type: string
+        description: 'Name of the artifact containing greptime binary'
+      http-port:
+        required: false
+        type: string
+        default: '4000'
+        description: 'HTTP server port'
+      mysql-port:
+        required: false
+        type: string
+        default: '4002'
+        description: 'MySQL server port'
+      postgres-port:
+        required: false
+        type: string
+        default: '4003'
+        description: 'PostgreSQL server port'
+      db-name:
+        required: false
+        type: string
+        default: 'test_db'
+        description: 'Test database name'
+      username:
+        required: false
+        type: string
+        default: 'greptime_user'
+        description: 'Authentication username'
+      password:
+        required: false
+        type: string
+        default: 'greptime_pwd'
+        description: 'Authentication password'
+      timeout-minutes:
+        required: false
+        type: number
+        default: 30
+        description: 'Job timeout in minutes'
+      artifact-is-tarball:
+        required: false
+        type: boolean
+        default: false
+        description: 'Whether the artifact is a tarball (tar.gz) that needs to be extracted'
+
+jobs:
+  run-tests:
+    name: Run Multi-language SDK Tests
+    runs-on: ubuntu-latest
+    timeout-minutes: ${{ inputs.timeout-minutes }}
+    steps:
+      - name: Checkout greptimedb-tests repository
+        uses: actions/checkout@v4
+        with:
+          repository: GreptimeTeam/greptimedb-tests
+          persist-credentials: false
+
+      - name: Download pre-built greptime binary
+        uses: actions/download-artifact@v4
+        with:
+          name: ${{ inputs.artifact-name }}
+          path: artifact
+
+      - name: Setup greptime binary
+        run: |
+          mkdir -p bin
+          if [ "${{ inputs.artifact-is-tarball }}" = "true" ]; then
+            # Extract tarball and find greptime binary
+            tar -xzf artifact/*.tar.gz -C artifact
+            find artifact -name "greptime" -type f -exec cp {} bin/greptime \;
+          else
+            # Direct binary format
+            if [ -f artifact/greptime ]; then
+              cp artifact/greptime bin/greptime
+            else
+              cp artifact/* bin/greptime
+            fi
+          fi
+          chmod +x ./bin/greptime
+          ls -lh ./bin/greptime
+          ./bin/greptime --version
+
+      - name: Setup Java 17
+        uses: actions/setup-java@v4
+        with:
+          distribution: 'temurin'
+          java-version: '17'
+          cache: 'maven'
+
+      - name: Setup Python 3.8
+        uses: actions/setup-python@v5
+        with:
+          python-version: '3.8'
+
+      - name: Setup Go 1.24
+        uses: actions/setup-go@v5
+        with:
+          go-version: '1.24'
+          cache: true
+          cache-dependency-path: go-tests/go.sum
+
+      - name: Set up Node.js
+        uses: actions/setup-node@v4
+        with:
+          node-version: '18'
+
+      - name: Install Python dependencies
+        run: |
+          pip install mysql-connector-python psycopg2-binary
+          python3 -c "import mysql.connector; print(f'mysql-connector-python {mysql.connector.__version__}')"
+          python3 -c "import psycopg2; print(f'psycopg2 {psycopg2.__version__}')"
+
+      - name: Install Go dependencies
+        working-directory: go-tests
+        run: |
+          go mod download
+          go mod verify
+          go version
+
+      - name: Kill existing GreptimeDB processes
+        run: |
+          pkill -f greptime || true
+          sleep 2
+
+      - name: Start GreptimeDB standalone
+        run: |
+          ./bin/greptime standalone start \
+            --http-addr 0.0.0.0:${{ inputs.http-port }} \
+            --rpc-addr 0.0.0.0:4001 \
+            --mysql-addr 0.0.0.0:${{ inputs.mysql-port }} \
+            --postgres-addr 0.0.0.0:${{ inputs.postgres-port }} \
+            --user-provider=static_user_provider:cmd:${{ inputs.username }}=${{ inputs.password }} > /tmp/greptimedb.log 2>&1 &
+
+      - name: Wait for GreptimeDB to be ready
+        run: |
+          echo "Waiting for GreptimeDB..."
+          for i in {1..60}; do
+            if curl -sf http://localhost:${{ inputs.http-port }}/health > /dev/null; then
+              echo "✅ GreptimeDB is ready"
+              exit 0
+            fi
+            sleep 2
+          done
+          echo "❌ GreptimeDB failed to start"
+          cat /tmp/greptimedb.log
+          exit 1
+
+      - name: Run multi-language tests
+        env:
+          DB_NAME: ${{ inputs.db-name }}
+          MYSQL_HOST: 127.0.0.1
+          MYSQL_PORT: ${{ inputs.mysql-port }}
+          POSTGRES_HOST: 127.0.0.1
+          POSTGRES_PORT: ${{ inputs.postgres-port }}
+          HTTP_HOST: 127.0.0.1
+          HTTP_PORT: ${{ inputs.http-port }}
+          GREPTIME_USERNAME: ${{ inputs.username }}
+          GREPTIME_PASSWORD: ${{ inputs.password }}
+        run: |
+          chmod +x ./run_tests.sh
+          ./run_tests.sh
+
+      - name: Collect logs on failure
+        if: failure()
+        run: |
+          echo "=== GreptimeDB Logs ==="
+          cat /tmp/greptimedb.log || true
+
+      - name: Upload test logs on failure
+        if: failure()
+        uses: actions/upload-artifact@v4
+        with:
+          name: test-logs
+          path: |
+            /tmp/greptimedb.log
+            java-tests/target/surefire-reports/
+            python-tests/.pytest_cache/
+            go-tests/*.log
+            **/test-output/
+          retention-days: 7
+
+      - name: Cleanup
+        if: always()
+        run: |
+          pkill -f greptime || true

Cargo.lock

@@ -212,7 +212,7 @@ checksum = "d301b3b94cb4b2f23d7917810addbbaff90738e0ca2be692bd027e70d7e0330c"
 
 [[package]]
 name = "api"
-version = "1.0.0-beta.1"
+version = "1.0.0-beta.2"
 dependencies = [
  "arrow-schema",
  "common-base",
@@ -733,7 +733,7 @@ dependencies = [
 
 [[package]]
 name = "auth"
-version = "1.0.0-beta.1"
+version = "1.0.0-beta.2"
 dependencies = [
  "api",
  "async-trait",
@@ -1383,7 +1383,7 @@ dependencies = [
 
 [[package]]
 name = "cache"
-version = "1.0.0-beta.1"
+version = "1.0.0-beta.2"
 dependencies = [
  "catalog",
  "common-error",
@@ -1418,7 +1418,7 @@ checksum = "37b2a672a2cb129a2e41c10b1224bb368f9f37a2b16b612598138befd7b37eb5"
 
 [[package]]
 name = "catalog"
-version = "1.0.0-beta.1"
+version = "1.0.0-beta.2"
 dependencies = [
  "api",
  "arrow",
@@ -1763,7 +1763,7 @@ checksum = "b94f61472cee1439c0b966b47e3aca9ae07e45d070759512cd390ea2bebc6675"
 
 [[package]]
 name = "cli"
-version = "1.0.0-beta.1"
+version = "1.0.0-beta.2"
 dependencies = [
  "async-stream",
  "async-trait",
@@ -1816,7 +1816,7 @@ dependencies = [
 
 [[package]]
 name = "client"
-version = "1.0.0-beta.1"
+version = "1.0.0-beta.2"
 dependencies = [
  "api",
  "arc-swap",
@@ -1849,7 +1849,7 @@ dependencies = [
  "snafu 0.8.6",
  "store-api",
  "substrait 0.37.3",
- "substrait 1.0.0-beta.1",
+ "substrait 1.0.0-beta.2",
  "tokio",
  "tokio-stream",
  "tonic 0.13.1",
@@ -1889,7 +1889,7 @@ dependencies = [
 
 [[package]]
 name = "cmd"
-version = "1.0.0-beta.1"
+version = "1.0.0-beta.2"
 dependencies = [
  "async-trait",
  "auth",
@@ -2012,7 +2012,7 @@ checksum = "55b672471b4e9f9e95499ea597ff64941a309b2cdbffcc46f2cc5e2d971fd335"
 
 [[package]]
 name = "common-base"
-version = "1.0.0-beta.1"
+version = "1.0.0-beta.2"
 dependencies = [
  "anymap2",
  "async-trait",
@@ -2036,14 +2036,14 @@ dependencies = [
 
 [[package]]
 name = "common-catalog"
-version = "1.0.0-beta.1"
+version = "1.0.0-beta.2"
 dependencies = [
  "const_format",
 ]
 
 [[package]]
 name = "common-config"
-version = "1.0.0-beta.1"
+version = "1.0.0-beta.2"
 dependencies = [
  "common-base",
  "common-error",
@@ -2067,7 +2067,7 @@ dependencies = [
 
 [[package]]
 name = "common-datasource"
-version = "1.0.0-beta.1"
+version = "1.0.0-beta.2"
 dependencies = [
  "arrow",
  "arrow-schema",
@@ -2102,7 +2102,7 @@ dependencies = [
 
 [[package]]
 name = "common-decimal"
-version = "1.0.0-beta.1"
+version = "1.0.0-beta.2"
 dependencies = [
  "bigdecimal 0.4.8",
  "common-error",
@@ -2115,7 +2115,7 @@ dependencies = [
 
 [[package]]
 name = "common-error"
-version = "1.0.0-beta.1"
+version = "1.0.0-beta.2"
 dependencies = [
  "common-macro",
  "http 1.3.1",
@@ -2126,7 +2126,7 @@ dependencies = [
 
 [[package]]
 name = "common-event-recorder"
-version = "1.0.0-beta.1"
+version = "1.0.0-beta.2"
 dependencies = [
  "api",
  "async-trait",
@@ -2148,7 +2148,7 @@ dependencies = [
 
 [[package]]
 name = "common-frontend"
-version = "1.0.0-beta.1"
+version = "1.0.0-beta.2"
 dependencies = [
  "api",
  "async-trait",
@@ -2170,7 +2170,7 @@ dependencies = [
 
 [[package]]
 name = "common-function"
-version = "1.0.0-beta.1"
+version = "1.0.0-beta.2"
 dependencies = [
  "ahash 0.8.12",
  "api",
@@ -2208,6 +2208,7 @@ dependencies = [
  "hyperloglogplus",
  "jsonb",
  "memchr",
+ "mito-codec",
  "nalgebra",
  "num",
  "num-traits",
@@ -2229,7 +2230,7 @@ dependencies = [
 
 [[package]]
 name = "common-greptimedb-telemetry"
-version = "1.0.0-beta.1"
+version = "1.0.0-beta.2"
 dependencies = [
  "async-trait",
  "common-runtime",
@@ -2246,7 +2247,7 @@ dependencies = [
 
 [[package]]
 name = "common-grpc"
-version = "1.0.0-beta.1"
+version = "1.0.0-beta.2"
 dependencies = [
  "api",
  "arrow-flight",
@@ -2265,11 +2266,13 @@ dependencies = [
  "hyper 1.6.0",
  "hyper-util",
  "lazy_static",
+ "notify",
  "prost 0.13.5",
  "rand 0.9.1",
  "serde",
  "serde_json",
  "snafu 0.8.6",
+ "tempfile",
  "tokio",
  "tokio-util",
  "tonic 0.13.1",
@@ -2279,7 +2282,7 @@ dependencies = [
 
 [[package]]
 name = "common-grpc-expr"
-version = "1.0.0-beta.1"
+version = "1.0.0-beta.2"
 dependencies = [
  "api",
  "common-base",
@@ -2299,7 +2302,7 @@ dependencies = [
 
 [[package]]
 name = "common-macro"
-version = "1.0.0-beta.1"
+version = "1.0.0-beta.2"
 dependencies = [
  "greptime-proto",
  "once_cell",
@@ -2310,7 +2313,7 @@ dependencies = [
 
 [[package]]
 name = "common-mem-prof"
-version = "1.0.0-beta.1"
+version = "1.0.0-beta.2"
 dependencies = [
  "anyhow",
  "common-error",
@@ -2326,7 +2329,7 @@ dependencies = [
 
 [[package]]
 name = "common-meta"
-version = "1.0.0-beta.1"
+version = "1.0.0-beta.2"
 dependencies = [
  "anymap2",
  "api",
@@ -2398,7 +2401,7 @@ dependencies = [
 
 [[package]]
 name = "common-options"
-version = "1.0.0-beta.1"
+version = "1.0.0-beta.2"
 dependencies = [
  "common-grpc",
  "humantime-serde",
@@ -2407,11 +2410,11 @@ dependencies = [
 
 [[package]]
 name = "common-plugins"
-version = "1.0.0-beta.1"
+version = "1.0.0-beta.2"
 
 [[package]]
 name = "common-pprof"
-version = "1.0.0-beta.1"
+version = "1.0.0-beta.2"
 dependencies = [
  "common-error",
  "common-macro",
@@ -2423,7 +2426,7 @@ dependencies = [
 
 [[package]]
 name = "common-procedure"
-version = "1.0.0-beta.1"
+version = "1.0.0-beta.2"
 dependencies = [
  "api",
  "async-stream",
@@ -2452,7 +2455,7 @@ dependencies = [
 
 [[package]]
 name = "common-procedure-test"
-version = "1.0.0-beta.1"
+version = "1.0.0-beta.2"
 dependencies = [
  "async-trait",
  "common-procedure",
@@ -2462,7 +2465,7 @@ dependencies = [
 
 [[package]]
 name = "common-query"
-version = "1.0.0-beta.1"
+version = "1.0.0-beta.2"
 dependencies = [
  "api",
  "async-trait",
@@ -2488,7 +2491,7 @@ dependencies = [
 
 [[package]]
 name = "common-recordbatch"
-version = "1.0.0-beta.1"
+version = "1.0.0-beta.2"
 dependencies = [
  "arc-swap",
  "common-base",
@@ -2512,7 +2515,7 @@ dependencies = [
 
 [[package]]
 name = "common-runtime"
-version = "1.0.0-beta.1"
+version = "1.0.0-beta.2"
 dependencies = [
  "async-trait",
  "clap 4.5.40",
@@ -2541,7 +2544,7 @@ dependencies = [
 
 [[package]]
 name = "common-session"
-version = "1.0.0-beta.1"
+version = "1.0.0-beta.2"
 dependencies = [
  "serde",
  "strum 0.27.1",
@@ -2549,7 +2552,7 @@ dependencies = [
 
 [[package]]
 name = "common-sql"
-version = "1.0.0-beta.1"
+version = "1.0.0-beta.2"
 dependencies = [
  "common-base",
  "common-decimal",
@@ -2567,7 +2570,7 @@ dependencies = [
 
 [[package]]
 name = "common-stat"
-version = "1.0.0-beta.1"
+version = "1.0.0-beta.2"
 dependencies = [
  "common-base",
  "common-runtime",
@@ -2582,7 +2585,7 @@ dependencies = [
 
 [[package]]
 name = "common-telemetry"
-version = "1.0.0-beta.1"
+version = "1.0.0-beta.2"
 dependencies = [
  "backtrace",
  "common-base",
@@ -2611,7 +2614,7 @@ dependencies = [
 
 [[package]]
 name = "common-test-util"
-version = "1.0.0-beta.1"
+version = "1.0.0-beta.2"
 dependencies = [
  "client",
  "common-grpc",
@@ -2624,7 +2627,7 @@ dependencies = [
 
 [[package]]
 name = "common-time"
-version = "1.0.0-beta.1"
+version = "1.0.0-beta.2"
 dependencies = [
  "arrow",
  "chrono",
@@ -2642,7 +2645,7 @@ dependencies = [
 
 [[package]]
 name = "common-version"
-version = "1.0.0-beta.1"
+version = "1.0.0-beta.2"
 dependencies = [
  "build-data",
  "cargo-manifest",
@@ -2653,7 +2656,7 @@ dependencies = [
 
 [[package]]
 name = "common-wal"
-version = "1.0.0-beta.1"
+version = "1.0.0-beta.2"
 dependencies = [
  "common-base",
  "common-error",
@@ -2676,7 +2679,7 @@ dependencies = [
 
 [[package]]
 name = "common-workload"
-version = "1.0.0-beta.1"
+version = "1.0.0-beta.2"
 dependencies = [
  "common-telemetry",
  "serde",
@@ -3913,7 +3916,7 @@ dependencies = [
 
 [[package]]
 name = "datanode"
-version = "1.0.0-beta.1"
+version = "1.0.0-beta.2"
 dependencies = [
  "api",
  "arrow-flight",
@@ -3977,7 +3980,7 @@ dependencies = [
 
 [[package]]
 name = "datatypes"
-version = "1.0.0-beta.1"
+version = "1.0.0-beta.2"
 dependencies = [
  "arrow",
  "arrow-array",
@@ -4649,7 +4652,7 @@ checksum = "37909eebbb50d72f9059c3b6d82c0463f2ff062c9e95845c43a6c9c0355411be"
 
 [[package]]
 name = "file-engine"
-version = "1.0.0-beta.1"
+version = "1.0.0-beta.2"
 dependencies = [
  "api",
  "async-trait",
@@ -4781,7 +4784,7 @@ checksum = "8bf7cc16383c4b8d58b9905a8509f02926ce3058053c056376248d958c9df1e8"
 
 [[package]]
 name = "flow"
-version = "1.0.0-beta.1"
+version = "1.0.0-beta.2"
 dependencies = [
  "api",
  "arrow",
@@ -4850,7 +4853,7 @@ dependencies = [
  "sql",
  "store-api",
  "strum 0.27.1",
- "substrait 1.0.0-beta.1",
+ "substrait 1.0.0-beta.2",
  "table",
  "tokio",
  "tonic 0.13.1",
@@ -4905,7 +4908,7 @@ checksum = "28dd6caf6059519a65843af8fe2a3ae298b14b80179855aeb4adc2c1934ee619"
 
 [[package]]
 name = "frontend"
-version = "1.0.0-beta.1"
+version = "1.0.0-beta.2"
 dependencies = [
  "api",
  "arc-swap",
@@ -5348,7 +5351,7 @@ dependencies = [
 [[package]]
 name = "greptime-proto"
 version = "0.1.0"
-source = "git+https://github.com/GreptimeTeam/greptime-proto.git?rev=14b9dc40bdc8288742b0cefc7bb024303b7429ef#14b9dc40bdc8288742b0cefc7bb024303b7429ef"
+source = "git+https://github.com/GreptimeTeam/greptime-proto.git?rev=0df99f09f1d6785055b2d9da96fc4ecc2bdf6803#0df99f09f1d6785055b2d9da96fc4ecc2bdf6803"
 dependencies = [
  "prost 0.13.5",
  "prost-types 0.13.5",
@@ -6116,7 +6119,7 @@ dependencies = [
 
 [[package]]
 name = "index"
-version = "1.0.0-beta.1"
+version = "1.0.0-beta.2"
 dependencies = [
  "async-trait",
  "asynchronous-codec",
@@ -7045,7 +7048,7 @@ checksum = "13dc2df351e3202783a1fe0d44375f7295ffb4049267b0f3018346dc122a1d94"
 
 [[package]]
 name = "log-query"
-version = "1.0.0-beta.1"
+version = "1.0.0-beta.2"
 dependencies = [
  "chrono",
  "common-error",
@@ -7057,7 +7060,7 @@ dependencies = [
 
 [[package]]
 name = "log-store"
-version = "1.0.0-beta.1"
+version = "1.0.0-beta.2"
 dependencies = [
  "async-stream",
  "async-trait",
@@ -7364,7 +7367,7 @@ dependencies = [
 
 [[package]]
 name = "meta-client"
-version = "1.0.0-beta.1"
+version = "1.0.0-beta.2"
 dependencies = [
  "api",
  "async-trait",
@@ -7392,7 +7395,7 @@ dependencies = [
 
 [[package]]
 name = "meta-srv"
-version = "1.0.0-beta.1"
+version = "1.0.0-beta.2"
 dependencies = [
  "api",
  "async-trait",
@@ -7440,6 +7443,7 @@ dependencies = [
  "lazy_static",
  "local-ip-address",
  "once_cell",
+ "ordered-float 4.6.0",
  "parking_lot 0.12.4",
  "prometheus",
  "prost 0.13.5",
@@ -7490,7 +7494,7 @@ dependencies = [
 
 [[package]]
 name = "metric-engine"
-version = "1.0.0-beta.1"
+version = "1.0.0-beta.2"
 dependencies = [
  "api",
  "aquamarine",
@@ -7585,7 +7589,7 @@ dependencies = [
 
 [[package]]
 name = "mito-codec"
-version = "1.0.0-beta.1"
+version = "1.0.0-beta.2"
 dependencies = [
  "api",
  "bytes",
@@ -7610,7 +7614,7 @@ dependencies = [
 
 [[package]]
 name = "mito2"
-version = "1.0.0-beta.1"
+version = "1.0.0-beta.2"
 dependencies = [
  "api",
  "aquamarine",
@@ -8348,7 +8352,7 @@ dependencies = [
 
 [[package]]
 name = "object-store"
-version = "1.0.0-beta.1"
+version = "1.0.0-beta.2"
 dependencies = [
  "anyhow",
  "bytes",
@@ -8527,7 +8531,7 @@ dependencies = [
 [[package]]
 name = "opensrv-mysql"
 version = "0.8.0"
-source = "git+https://github.com/datafuselabs/opensrv?rev=a1fb4da215c8693c7e4f62be249a01b7fec52997#a1fb4da215c8693c7e4f62be249a01b7fec52997"
+source = "git+https://github.com/datafuselabs/opensrv?tag=v0.10.0#074bd8fb81da3c9e6d6a098a482f3380478b9c0b"
 dependencies = [
  "async-trait",
  "byteorder",
@@ -8633,7 +8637,7 @@ dependencies = [
 
 [[package]]
 name = "operator"
-version = "1.0.0-beta.1"
+version = "1.0.0-beta.2"
 dependencies = [
  "ahash 0.8.12",
  "api",
@@ -8659,6 +8663,7 @@ dependencies = [
  "common-recordbatch",
  "common-runtime",
  "common-sql",
+ "common-stat",
  "common-telemetry",
  "common-test-util",
  "common-time",
@@ -8670,6 +8675,7 @@ dependencies = [
  "futures",
  "futures-util",
  "humantime",
+ "itertools 0.14.0",
  "jsonb",
  "lazy_static",
  "meta-client",
@@ -8691,7 +8697,7 @@ dependencies = [
  "sql",
  "sqlparser",
  "store-api",
- "substrait 1.0.0-beta.1",
+ "substrait 1.0.0-beta.2",
  "table",
  "tokio",
  "tokio-util",
@@ -8977,7 +8983,7 @@ dependencies = [
 
 [[package]]
 name = "partition"
-version = "1.0.0-beta.1"
+version = "1.0.0-beta.2"
 dependencies = [
  "api",
  "async-trait",
@@ -9181,10 +9187,21 @@ dependencies = [
 ]
 
 [[package]]
-name = "pgwire"
-version = "0.34.2"
+name = "pg_interval"
+version = "0.4.2"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "4f56a81b4fcc69016028f657a68f9b8e8a2a4b7d07684ca3298f2d3e7ff199ce"
+checksum = "fe46640b465e284b048ef065cbed8ef17a622878d310c724578396b4cfd00df2"
+dependencies = [
+ "bytes",
+ "chrono",
+ "postgres-types",
+]
+
+[[package]]
+name = "pgwire"
+version = "0.36.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "d331bb0eef5bc83a221c0a85b1f205bccf094d4f72a26ae1d68a1b1c535123b7"
 dependencies = [
  "async-trait",
  "base64 0.22.1",
@@ -9200,6 +9217,7 @@ dependencies = [
  "ring",
  "rust_decimal",
  "rustls-pki-types",
+ "ryu",
  "serde",
  "serde_json",
  "stringprep",
@@ -9322,7 +9340,7 @@ checksum = "8b870d8c151b6f2fb93e84a13146138f05d02ed11c7e7c54f8826aaaf7c9f184"
 
 [[package]]
 name = "pipeline"
-version = "1.0.0-beta.1"
+version = "1.0.0-beta.2"
 dependencies = [
  "ahash 0.8.12",
  "api",
@@ -9478,7 +9496,7 @@ dependencies = [
 
 [[package]]
 name = "plugins"
-version = "1.0.0-beta.1"
+version = "1.0.0-beta.2"
 dependencies = [
  "auth",
  "clap 4.5.40",
@@ -9778,7 +9796,7 @@ dependencies = [
 
 [[package]]
 name = "promql"
-version = "1.0.0-beta.1"
+version = "1.0.0-beta.2"
 dependencies = [
  "ahash 0.8.12",
  "async-trait",
@@ -10061,7 +10079,7 @@ dependencies = [
 
 [[package]]
 name = "puffin"
-version = "1.0.0-beta.1"
+version = "1.0.0-beta.2"
 dependencies = [
  "async-compression 0.4.19",
  "async-trait",
@@ -10103,7 +10121,7 @@ dependencies = [
 
 [[package]]
 name = "query"
-version = "1.0.0-beta.1"
+version = "1.0.0-beta.2"
 dependencies = [
  "ahash 0.8.12",
  "api",
@@ -10170,7 +10188,7 @@ dependencies = [
  "sql",
  "sqlparser",
  "store-api",
- "substrait 1.0.0-beta.1",
+ "substrait 1.0.0-beta.2",
  "table",
  "tokio",
  "tokio-stream",
@@ -11506,7 +11524,7 @@ dependencies = [
 
 [[package]]
 name = "servers"
-version = "1.0.0-beta.1"
+version = "1.0.0-beta.2"
 dependencies = [
  "ahash 0.8.12",
  "api",
@@ -11583,6 +11601,7 @@ dependencies = [
  "otel-arrow-rust",
  "parking_lot 0.12.4",
  "permutation",
+ "pg_interval",
  "pgwire",
  "pin-project",
  "pipeline",
@@ -11624,6 +11643,7 @@ dependencies = [
  "tower 0.5.2",
  "tower-http 0.6.6",
  "tracing",
+ "tracing-opentelemetry",
  "urlencoding",
  "uuid",
  "vrl",
@@ -11632,7 +11652,7 @@ dependencies = [
 
 [[package]]
 name = "session"
-version = "1.0.0-beta.1"
+version = "1.0.0-beta.2"
 dependencies = [
  "ahash 0.8.12",
  "api",
@@ -11966,7 +11986,7 @@ dependencies = [
 
 [[package]]
 name = "sql"
-version = "1.0.0-beta.1"
+version = "1.0.0-beta.2"
 dependencies = [
  "api",
  "arrow-buffer",
@@ -12026,7 +12046,7 @@ dependencies = [
 
 [[package]]
 name = "sqlness-runner"
-version = "1.0.0-beta.1"
+version = "1.0.0-beta.2"
 dependencies = [
  "async-trait",
  "clap 4.5.40",
@@ -12303,7 +12323,7 @@ dependencies = [
 
 [[package]]
 name = "standalone"
-version = "1.0.0-beta.1"
+version = "1.0.0-beta.2"
 dependencies = [
  "async-trait",
  "catalog",
@@ -12344,7 +12364,7 @@ checksum = "a2eb9349b6444b326872e140eb1cf5e7c522154d69e7a0ffb0fb81c06b37543f"
 
 [[package]]
 name = "store-api"
-version = "1.0.0-beta.1"
+version = "1.0.0-beta.2"
 dependencies = [
  "api",
  "aquamarine",
@@ -12557,7 +12577,7 @@ dependencies = [
 
 [[package]]
 name = "substrait"
-version = "1.0.0-beta.1"
+version = "1.0.0-beta.2"
 dependencies = [
  "async-trait",
  "bytes",
@@ -12680,7 +12700,7 @@ dependencies = [
 
 [[package]]
 name = "table"
-version = "1.0.0-beta.1"
+version = "1.0.0-beta.2"
 dependencies = [
  "api",
  "async-trait",
@@ -12919,7 +12939,7 @@ dependencies = [
  "getrandom 0.3.3",
  "once_cell",
  "rustix 1.0.7",
- "windows-sys 0.59.0",
+ "windows-sys 0.61.2",
 ]
 
 [[package]]
@@ -12949,7 +12969,7 @@ checksum = "8f50febec83f5ee1df3015341d8bd429f2d1cc62bcba7ea2076759d315084683"
 
 [[package]]
 name = "tests-fuzz"
-version = "1.0.0-beta.1"
+version = "1.0.0-beta.2"
 dependencies = [
  "arbitrary",
  "async-trait",
@@ -12993,7 +13013,7 @@ dependencies = [
 
 [[package]]
 name = "tests-integration"
-version = "1.0.0-beta.1"
+version = "1.0.0-beta.2"
 dependencies = [
  "api",
  "arrow-flight",
@@ -13067,7 +13087,7 @@ dependencies = [
  "sqlx",
  "standalone",
  "store-api",
- "substrait 1.0.0-beta.1",
+ "substrait 1.0.0-beta.2",
  "table",
  "tempfile",
  "time",

Cargo.toml

@@ -74,7 +74,7 @@ members = [
 resolver = "2"
 
 [workspace.package]
-version = "1.0.0-beta.1"
+version = "1.0.0-beta.2"
 edition = "2024"
 license = "Apache-2.0"
 
@@ -148,7 +148,7 @@ etcd-client = { git = "https://github.com/GreptimeTeam/etcd-client", rev = "f62d
 fst = "0.4.7"
 futures = "0.3"
 futures-util = "0.3"
-greptime-proto = { git = "https://github.com/GreptimeTeam/greptime-proto.git", rev = "14b9dc40bdc8288742b0cefc7bb024303b7429ef" }
+greptime-proto = { git = "https://github.com/GreptimeTeam/greptime-proto.git", rev = "0df99f09f1d6785055b2d9da96fc4ecc2bdf6803" }
 hex = "0.4"
 http = "1"
 humantime = "2.1"
@@ -234,6 +234,7 @@ tower = "0.5"
 tower-http = "0.6"
 tracing = "0.1"
 tracing-appender = "0.2"
+tracing-opentelemetry = "0.31.0"
 tracing-subscriber = { version = "0.3", features = ["env-filter", "json", "fmt"] }
 typetag = "0.2"
 uuid = { version = "1.17", features = ["serde", "v4", "fast-rng"] }

Makefile

@@ -17,6 +17,8 @@ CARGO_REGISTRY_CACHE ?= ${HOME}/.cargo/registry
 ARCH := $(shell uname -m | sed 's/x86_64/amd64/' | sed 's/aarch64/arm64/')
 OUTPUT_DIR := $(shell if [ "$(RELEASE)" = "true" ]; then echo "release"; elif [ ! -z "$(CARGO_PROFILE)" ]; then echo "$(CARGO_PROFILE)" ; else echo "debug"; fi)
 SQLNESS_OPTS ?=
+EXTRA_BUILD_ENVS ?=
+ASSEMBLED_EXTRA_BUILD_ENV := $(foreach var,$(EXTRA_BUILD_ENVS),-e $(var))
 
 # The arguments for running integration tests.
 ETCD_VERSION ?= v3.5.9
@@ -83,6 +85,7 @@ build: ## Build debug version greptime.
 .PHONY: build-by-dev-builder
 build-by-dev-builder: ## Build greptime by dev-builder.
 	docker run --network=host \
+	${ASSEMBLED_EXTRA_BUILD_ENV} \
 	-v ${PWD}:/greptimedb -v ${CARGO_REGISTRY_CACHE}:/root/.cargo/registry \
 	-w /greptimedb ${IMAGE_REGISTRY}/${IMAGE_NAMESPACE}/dev-builder-${BASE_IMAGE}:${DEV_BUILDER_IMAGE_TAG} \
 	make build \

config/config.md

@@ -210,14 +210,6 @@
 | `slow_query.record_type` | String | Unset | The record type of slow queries. It can be `system_table` or `log`. |
 | `slow_query.threshold` | String | Unset | The threshold of slow query. |
 | `slow_query.sample_ratio` | Float | Unset | The sampling ratio of slow query log. The value should be in the range of (0, 1]. |
-| `export_metrics` | -- | -- | The standalone can export its metrics and send to Prometheus compatible service (e.g. `greptimedb`) from remote-write API.<br/>This is only used for `greptimedb` to export its own metrics internally. It's different from prometheus scrape. |
-| `export_metrics.enable` | Bool | `false` | whether enable export metrics. |
-| `export_metrics.write_interval` | String | `30s` | The interval of export metrics. |
-| `export_metrics.self_import` | -- | -- | For `standalone` mode, `self_import` is recommended to collect metrics generated by itself<br/>You must create the database before enabling it. |
-| `export_metrics.self_import.db` | String | Unset | -- |
-| `export_metrics.remote_write` | -- | -- | -- |
-| `export_metrics.remote_write.url` | String | `""` | The prometheus remote write endpoint that the metrics send to. The url example can be: `http://127.0.0.1:4000/v1/prometheus/write?db=greptime_metrics`. |
-| `export_metrics.remote_write.headers` | InlineTable | -- | HTTP headers of Prometheus remote-write carry. |
 | `tracing` | -- | -- | The tracing options. Only effect when compiled with `tokio-console` feature. |
 | `tracing.tokio_console_addr` | String | Unset | The tokio console address. |
 | `memory` | -- | -- | The memory options. |
@@ -335,12 +327,6 @@
 | `slow_query.threshold` | String | `30s` | The threshold of slow query. It can be human readable time string, for example: `10s`, `100ms`, `1s`. |
 | `slow_query.sample_ratio` | Float | `1.0` | The sampling ratio of slow query log. The value should be in the range of (0, 1]. For example, `0.1` means 10% of the slow queries will be logged and `1.0` means all slow queries will be logged. |
 | `slow_query.ttl` | String | `90d` | The TTL of the `slow_queries` system table. Default is `90d` when `record_type` is `system_table`. |
-| `export_metrics` | -- | -- | The frontend can export its metrics and send to Prometheus compatible service (e.g. `greptimedb` itself) from remote-write API.<br/>This is only used for `greptimedb` to export its own metrics internally. It's different from prometheus scrape. |
-| `export_metrics.enable` | Bool | `false` | whether enable export metrics. |
-| `export_metrics.write_interval` | String | `30s` | The interval of export metrics. |
-| `export_metrics.remote_write` | -- | -- | -- |
-| `export_metrics.remote_write.url` | String | `""` | The prometheus remote write endpoint that the metrics send to. The url example can be: `http://127.0.0.1:4000/v1/prometheus/write?db=greptime_metrics`. |
-| `export_metrics.remote_write.headers` | InlineTable | -- | HTTP headers of Prometheus remote-write carry. |
 | `tracing` | -- | -- | The tracing options. Only effect when compiled with `tokio-console` feature. |
 | `tracing.tokio_console_addr` | String | Unset | The tokio console address. |
 | `memory` | -- | -- | The memory options. |
@@ -354,7 +340,7 @@
 | Key | Type | Default | Descriptions |
 | --- | -----| ------- | ----------- |
 | `data_home` | String | `./greptimedb_data` | The working home directory. |
-| `store_addrs` | Array | -- | Store server address default to etcd store.<br/>For postgres store, the format is:<br/>"password=password dbname=postgres user=postgres host=localhost port=5432"<br/>For etcd store, the format is:<br/>"127.0.0.1:2379" |
+| `store_addrs` | Array | -- | Store server address(es). The format depends on the selected backend.<br/><br/>For etcd: a list of "host:port" endpoints.<br/>e.g. ["192.168.1.1:2379", "192.168.1.2:2379"]<br/><br/>For PostgreSQL: a connection string in libpq format or URI.<br/>e.g.<br/>- "host=localhost port=5432 user=postgres password=<PASSWORD> dbname=postgres"<br/>- "postgresql://user:password@localhost:5432/mydb?connect_timeout=10"<br/>The detail see: https://docs.rs/tokio-postgres/latest/tokio_postgres/config/struct.Config.html<br/><br/>For mysql store, the format is a MySQL connection URL.<br/>e.g. "mysql://user:password@localhost:3306/greptime_meta?ssl-mode=VERIFY_CA&ssl-ca=/path/to/ca.pem" |
 | `store_key_prefix` | String | `""` | If it's not empty, the metasrv will store all data with this key prefix. |
 | `backend` | String | `etcd_store` | The datastore for meta server.<br/>Available values:<br/>- `etcd_store` (default value)<br/>- `memory_store`<br/>- `postgres_store`<br/>- `mysql_store` |
 | `meta_table_name` | String | `greptime_metakv` | Table name in RDS to store metadata. Effect when using a RDS kvbackend.<br/>**Only used when backend is `postgres_store`.** |
@@ -370,12 +356,11 @@
 | `runtime` | -- | -- | The runtime options. |
 | `runtime.global_rt_size` | Integer | `8` | The number of threads to execute the runtime for global read operations. |
 | `runtime.compact_rt_size` | Integer | `4` | The number of threads to execute the runtime for global write operations. |
-| `backend_tls` | -- | -- | TLS configuration for kv store backend (applicable for etcd, PostgreSQL, and MySQL backends)<br/>When using etcd, PostgreSQL, or MySQL as metadata store, you can configure TLS here |
+| `backend_tls` | -- | -- | TLS configuration for kv store backend (applicable for etcd, PostgreSQL, and MySQL backends)<br/>When using etcd, PostgreSQL, or MySQL as metadata store, you can configure TLS here<br/><br/>Note: if TLS is configured in both this section and the `store_addrs` connection string, the<br/>settings here will override the TLS settings in `store_addrs`. |
 | `backend_tls.mode` | String | `prefer` | TLS mode, refer to https://www.postgresql.org/docs/current/libpq-ssl.html<br/>- "disable" - No TLS<br/>- "prefer" (default) - Try TLS, fallback to plain<br/>- "require" - Require TLS<br/>- "verify_ca" - Require TLS and verify CA<br/>- "verify_full" - Require TLS and verify hostname |
 | `backend_tls.cert_path` | String | `""` | Path to client certificate file (for client authentication)<br/>Like "/path/to/client.crt" |
 | `backend_tls.key_path` | String | `""` | Path to client private key file (for client authentication)<br/>Like "/path/to/client.key" |
 | `backend_tls.ca_cert_path` | String | `""` | Path to CA certificate file (for server certificate verification)<br/>Required when using custom CAs or self-signed certificates<br/>Leave empty to use system root certificates only<br/>Like "/path/to/ca.crt" |
-| `backend_tls.watch` | Bool | `false` | Watch for certificate file changes and auto reload |
 | `grpc` | -- | -- | The gRPC server options. |
 | `grpc.bind_addr` | String | `127.0.0.1:3002` | The address to bind the gRPC server. |
 | `grpc.server_addr` | String | `127.0.0.1:3002` | The communication server address for the frontend and datanode to connect to metasrv.<br/>If left empty or unset, the server will automatically use the IP address of the first network interface<br/>on the host, with the same port number as the one specified in `bind_addr`. |
@@ -430,12 +415,6 @@
 | `logging.otlp_headers` | -- | -- | Additional OTLP headers, only valid when using OTLP http |
 | `logging.tracing_sample_ratio` | -- | Unset | The percentage of tracing will be sampled and exported.<br/>Valid range `[0, 1]`, 1 means all traces are sampled, 0 means all traces are not sampled, the default value is 1.<br/>ratio > 1 are treated as 1. Fractions < 0 are treated as 0 |
 | `logging.tracing_sample_ratio.default_ratio` | Float | `1.0` | -- |
-| `export_metrics` | -- | -- | The metasrv can export its metrics and send to Prometheus compatible service (e.g. `greptimedb` itself) from remote-write API.<br/>This is only used for `greptimedb` to export its own metrics internally. It's different from prometheus scrape. |
-| `export_metrics.enable` | Bool | `false` | whether enable export metrics. |
-| `export_metrics.write_interval` | String | `30s` | The interval of export metrics. |
-| `export_metrics.remote_write` | -- | -- | -- |
-| `export_metrics.remote_write.url` | String | `""` | The prometheus remote write endpoint that the metrics send to. The url example can be: `http://127.0.0.1:4000/v1/prometheus/write?db=greptime_metrics`. |
-| `export_metrics.remote_write.headers` | InlineTable | -- | HTTP headers of Prometheus remote-write carry. |
 | `tracing` | -- | -- | The tracing options. Only effect when compiled with `tokio-console` feature. |
 | `tracing.tokio_console_addr` | String | Unset | The tokio console address. |
 | `memory` | -- | -- | The memory options. |
@@ -608,12 +587,6 @@
 | `logging.otlp_headers` | -- | -- | Additional OTLP headers, only valid when using OTLP http |
 | `logging.tracing_sample_ratio` | -- | Unset | The percentage of tracing will be sampled and exported.<br/>Valid range `[0, 1]`, 1 means all traces are sampled, 0 means all traces are not sampled, the default value is 1.<br/>ratio > 1 are treated as 1. Fractions < 0 are treated as 0 |
 | `logging.tracing_sample_ratio.default_ratio` | Float | `1.0` | -- |
-| `export_metrics` | -- | -- | The datanode can export its metrics and send to Prometheus compatible service (e.g. `greptimedb` itself) from remote-write API.<br/>This is only used for `greptimedb` to export its own metrics internally. It's different from prometheus scrape. |
-| `export_metrics.enable` | Bool | `false` | whether enable export metrics. |
-| `export_metrics.write_interval` | String | `30s` | The interval of export metrics. |
-| `export_metrics.remote_write` | -- | -- | -- |
-| `export_metrics.remote_write.url` | String | `""` | The prometheus remote write endpoint that the metrics send to. The url example can be: `http://127.0.0.1:4000/v1/prometheus/write?db=greptime_metrics`. |
-| `export_metrics.remote_write.headers` | InlineTable | -- | HTTP headers of Prometheus remote-write carry. |
 | `tracing` | -- | -- | The tracing options. Only effect when compiled with `tokio-console` feature. |
 | `tracing.tokio_console_addr` | String | Unset | The tokio console address. |
 | `memory` | -- | -- | The memory options. |

config/datanode.example.toml

@@ -712,21 +712,6 @@ otlp_export_protocol = "http"
 [logging.tracing_sample_ratio]
 default_ratio = 1.0
 
-## The datanode can export its metrics and send to Prometheus compatible service (e.g. `greptimedb` itself) from remote-write API.
-## This is only used for `greptimedb` to export its own metrics internally. It's different from prometheus scrape.
-[export_metrics]
-## whether enable export metrics.
-enable = false
-## The interval of export metrics.
-write_interval = "30s"
-
-[export_metrics.remote_write]
-## The prometheus remote write endpoint that the metrics send to. The url example can be: `http://127.0.0.1:4000/v1/prometheus/write?db=greptime_metrics`.
-url = ""
-
-## HTTP headers of Prometheus remote-write carry.
-headers = { }
-
 ## The tracing options. Only effect when compiled with `tokio-console` feature.
 #+ [tracing]
 ## The tokio console address.

config/frontend.example.toml

@@ -329,21 +329,6 @@ sample_ratio = 1.0
 ## The TTL of the `slow_queries` system table. Default is `90d` when `record_type` is `system_table`.
 ttl = "90d"
 
-## The frontend can export its metrics and send to Prometheus compatible service (e.g. `greptimedb` itself) from remote-write API.
-## This is only used for `greptimedb` to export its own metrics internally. It's different from prometheus scrape.
-[export_metrics]
-## whether enable export metrics.
-enable = false
-## The interval of export metrics.
-write_interval = "30s"
-
-[export_metrics.remote_write]
-## The prometheus remote write endpoint that the metrics send to. The url example can be: `http://127.0.0.1:4000/v1/prometheus/write?db=greptime_metrics`.
-url = ""
-
-## HTTP headers of Prometheus remote-write carry.
-headers = { }
-
 ## The tracing options. Only effect when compiled with `tokio-console` feature.
 #+ [tracing]
 ## The tokio console address.

config/metasrv.example.toml

@@ -1,11 +1,19 @@
 ## The working home directory.
 data_home = "./greptimedb_data"
 
-## Store server address default to etcd store.
-## For postgres store, the format is:
-## "password=password dbname=postgres user=postgres host=localhost port=5432"
-## For etcd store, the format is:
-## "127.0.0.1:2379"
+## Store server address(es). The format depends on the selected backend.
+##
+## For etcd: a list of "host:port" endpoints.
+## e.g. ["192.168.1.1:2379", "192.168.1.2:2379"]
+##
+## For PostgreSQL: a connection string in libpq format or URI.
+## e.g.
+## - "host=localhost port=5432 user=postgres password=<PASSWORD> dbname=postgres"
+## - "postgresql://user:password@localhost:5432/mydb?connect_timeout=10"
+## The detail see: https://docs.rs/tokio-postgres/latest/tokio_postgres/config/struct.Config.html
+##
+## For mysql store, the format is a MySQL connection URL.
+## e.g. "mysql://user:password@localhost:3306/greptime_meta?ssl-mode=VERIFY_CA&ssl-ca=/path/to/ca.pem"
 store_addrs = ["127.0.0.1:2379"]
 
 ## If it's not empty, the metasrv will store all data with this key prefix.
@@ -75,6 +83,9 @@ node_max_idle_time = "24hours"
 
 ## TLS configuration for kv store backend (applicable for etcd, PostgreSQL, and MySQL backends)
 ## When using etcd, PostgreSQL, or MySQL as metadata store, you can configure TLS here
+##
+## Note: if TLS is configured in both this section and the `store_addrs` connection string, the
+## settings here will override the TLS settings in `store_addrs`.
 [backend_tls]
 ## TLS mode, refer to https://www.postgresql.org/docs/current/libpq-ssl.html
 ## - "disable" - No TLS
@@ -98,9 +109,6 @@ key_path = ""
 ## Like "/path/to/ca.crt"
 ca_cert_path = ""
 
-## Watch for certificate file changes and auto reload
-watch = false
-
 ## The gRPC server options.
 [grpc]
 ## The address to bind the gRPC server.
@@ -323,21 +331,6 @@ otlp_export_protocol = "http"
 [logging.tracing_sample_ratio]
 default_ratio = 1.0
 
-## The metasrv can export its metrics and send to Prometheus compatible service (e.g. `greptimedb` itself) from remote-write API.
-## This is only used for `greptimedb` to export its own metrics internally. It's different from prometheus scrape.
-[export_metrics]
-## whether enable export metrics.
-enable = false
-## The interval of export metrics.
-write_interval = "30s"
-
-[export_metrics.remote_write]
-## The prometheus remote write endpoint that the metrics send to. The url example can be: `http://127.0.0.1:4000/v1/prometheus/write?db=greptime_metrics`.
-url = ""
-
-## HTTP headers of Prometheus remote-write carry.
-headers = { }
-
 ## The tracing options. Only effect when compiled with `tokio-console` feature.
 #+ [tracing]
 ## The tokio console address.

config/standalone.example.toml

@@ -820,27 +820,6 @@ default_ratio = 1.0
 ## @toml2docs:none-default
 #+ sample_ratio = 1.0
 
-## The standalone can export its metrics and send to Prometheus compatible service (e.g. `greptimedb`) from remote-write API.
-## This is only used for `greptimedb` to export its own metrics internally. It's different from prometheus scrape.
-[export_metrics]
-## whether enable export metrics.
-enable = false
-## The interval of export metrics.
-write_interval = "30s"
-
-## For `standalone` mode, `self_import` is recommended to collect metrics generated by itself
-## You must create the database before enabling it.
-[export_metrics.self_import]
-## @toml2docs:none-default
-db = "greptime_metrics"
-
-[export_metrics.remote_write]
-## The prometheus remote write endpoint that the metrics send to. The url example can be: `http://127.0.0.1:4000/v1/prometheus/write?db=greptime_metrics`.
-url = ""
-
-## HTTP headers of Prometheus remote-write carry.
-headers = { }
-
 ## The tracing options. Only effect when compiled with `tokio-console` feature.
 #+ [tracing]
 ## The tokio console address.

docker/buildx/centos/Dockerfile

@@ -1,10 +1,10 @@
-FROM centos:7 as builder
+FROM centos:7 AS builder
 
 ARG CARGO_PROFILE
 ARG FEATURES
 ARG OUTPUT_DIR
 
-ENV LANG en_US.utf8
+ENV LANG=en_US.utf8
 WORKDIR /greptimedb
 
 # Install dependencies
@@ -22,7 +22,7 @@ RUN unzip protoc-3.15.8-linux-x86_64.zip -d /usr/local/
 # Install Rust
 SHELL ["/bin/bash", "-c"]
 RUN curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh -s -- --no-modify-path --default-toolchain none -y
-ENV PATH /usr/local/bin:/root/.cargo/bin/:$PATH
+ENV PATH=/usr/local/bin:/root/.cargo/bin/:$PATH
 
 # Build the project in release mode.
 RUN --mount=target=.,rw \
@@ -33,7 +33,7 @@ RUN --mount=target=.,rw \
     TARGET_DIR=/out/target
 
 # Export the binary to the clean image.
-FROM centos:7 as base
+FROM centos:7 AS base
 
 ARG OUTPUT_DIR
 
@@ -45,7 +45,7 @@ RUN yum install -y epel-release \
 
 WORKDIR /greptime
 COPY --from=builder /out/target/${OUTPUT_DIR}/greptime /greptime/bin/
-ENV PATH /greptime/bin/:$PATH
+ENV PATH=/greptime/bin/:$PATH
 
 ENV MALLOC_CONF="prof:true,prof_active:false"
 

docker/buildx/distroless/Dockerfile

@@ -0,0 +1,65 @@
+FROM ubuntu:22.04 AS builder
+
+ARG CARGO_PROFILE
+ARG FEATURES
+ARG OUTPUT_DIR
+
+ENV LANG=en_US.utf8
+WORKDIR /greptimedb
+
+RUN apt-get update && \
+    DEBIAN_FRONTEND=noninteractive apt-get install -y software-properties-common
+
+# Install dependencies.
+RUN --mount=type=cache,target=/var/cache/apt \
+    apt-get update && apt-get install -y \
+    libssl-dev \
+    protobuf-compiler \
+    curl \
+    git \
+    build-essential \
+    pkg-config
+
+# Install Rust.
+SHELL ["/bin/bash", "-c"]
+RUN curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh -s -- --no-modify-path --default-toolchain none -y
+ENV PATH=/root/.cargo/bin/:$PATH
+
+# Build the project in release mode.
+RUN --mount=target=. \
+    --mount=type=cache,target=/root/.cargo/registry \
+    make build \
+    CARGO_PROFILE=${CARGO_PROFILE} \
+    FEATURES=${FEATURES} \
+    TARGET_DIR=/out/target
+
+FROM ubuntu:22.04 AS libs
+
+ARG TARGETARCH
+
+# Copy required library dependencies based on architecture
+RUN if [ "$TARGETARCH" = "amd64" ]; then \
+        cp /lib/x86_64-linux-gnu/libz.so.1.2.11 /lib/x86_64-linux-gnu/libz.so.1; \
+    elif [ "$TARGETARCH" = "arm64" ]; then \
+        cp /lib/aarch64-linux-gnu/libz.so.1.2.11 /lib/aarch64-linux-gnu/libz.so.1; \
+    else \
+        echo "Unsupported architecture: $TARGETARCH" && exit 1; \
+    fi
+
+# Export the binary to the clean distroless image.
+FROM gcr.io/distroless/cc-debian12:latest AS base
+
+ARG OUTPUT_DIR
+ARG TARGETARCH
+
+# Copy required library dependencies
+COPY --from=libs /lib /lib
+COPY --from=busybox:stable /bin/busybox /bin/busybox
+
+WORKDIR /greptime
+COPY --from=builder /out/target/${OUTPUT_DIR}/greptime /greptime/bin/greptime
+ENV PATH=/greptime/bin/:$PATH
+
+ENV MALLOC_CONF="prof:true,prof_active:false"
+
+ENTRYPOINT ["greptime"]

docker/buildx/ubuntu/Dockerfile

@@ -1,10 +1,10 @@
-FROM ubuntu:22.04 as builder
+FROM ubuntu:22.04 AS builder
 
 ARG CARGO_PROFILE
 ARG FEATURES
 ARG OUTPUT_DIR
 
-ENV LANG en_US.utf8
+ENV LANG=en_US.utf8
 WORKDIR /greptimedb
 
 RUN apt-get update && \
@@ -23,7 +23,7 @@ RUN --mount=type=cache,target=/var/cache/apt \
 # Install Rust.
 SHELL ["/bin/bash", "-c"]
 RUN curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh -s -- --no-modify-path --default-toolchain none -y
-ENV PATH /root/.cargo/bin/:$PATH
+ENV PATH=/root/.cargo/bin/:$PATH
 
 # Build the project in release mode.
 RUN --mount=target=. \
@@ -35,7 +35,7 @@ RUN --mount=target=. \
 
 # Export the binary to the clean image.
 # TODO(zyy17): Maybe should use the more secure container image.
-FROM ubuntu:22.04 as base
+FROM ubuntu:22.04 AS base
 
 ARG OUTPUT_DIR
 
@@ -45,7 +45,7 @@ RUN apt-get update && DEBIAN_FRONTEND=noninteractive apt-get \
 
 WORKDIR /greptime
 COPY --from=builder /out/target/${OUTPUT_DIR}/greptime /greptime/bin/
-ENV PATH /greptime/bin/:$PATH
+ENV PATH=/greptime/bin/:$PATH
 
 ENV MALLOC_CONF="prof:true,prof_active:false"
 

docker/ci/centos/Dockerfile

@@ -13,7 +13,7 @@ ARG TARGETARCH
 
 ADD $TARGETARCH/greptime /greptime/bin/
 
-ENV PATH /greptime/bin/:$PATH
+ENV PATH=/greptime/bin/:$PATH
 
 ENV MALLOC_CONF="prof:true,prof_active:false"
 

docker/ci/distroless/Dockerfile

@@ -0,0 +1,40 @@
+FROM ubuntu:22.04 AS libs
+
+ARG TARGETARCH
+
+# Copy required library dependencies based on architecture
+# TARGETARCH values: amd64, arm64
+# Ubuntu library paths: x86_64-linux-gnu, aarch64-linux-gnu
+RUN if [ "$TARGETARCH" = "amd64" ]; then \
+        mkdir -p /output/x86_64-linux-gnu && \
+        cp /lib/x86_64-linux-gnu/libz.so.1.2.11 /output/x86_64-linux-gnu/libz.so.1; \
+    elif [ "$TARGETARCH" = "arm64" ]; then \
+        mkdir -p /output/aarch64-linux-gnu && \
+        cp /lib/aarch64-linux-gnu/libz.so.1.2.11 /output/aarch64-linux-gnu/libz.so.1; \
+    else \
+        echo "Unsupported architecture: $TARGETARCH" && exit 1; \
+    fi
+
+FROM gcr.io/distroless/cc-debian12:latest
+
+# The root path under which contains all the dependencies to build this Dockerfile.
+ARG DOCKER_BUILD_ROOT=.
+# The binary name of GreptimeDB executable.
+# Defaults to "greptime", but sometimes in other projects it might be different.
+ARG TARGET_BIN=greptime
+
+ARG TARGETARCH
+
+# Copy required library dependencies
+COPY --from=libs /output /lib
+COPY --from=busybox:stable /bin/busybox /bin/busybox
+
+ADD $TARGETARCH/$TARGET_BIN /greptime/bin/
+
+ENV PATH=/greptime/bin/:$PATH
+
+ENV TARGET_BIN=$TARGET_BIN
+
+ENV MALLOC_CONF="prof:true,prof_active:false"
+
+ENTRYPOINT ["greptime"]

docker/ci/ubuntu/Dockerfile

@@ -14,7 +14,7 @@ ARG TARGETARCH
 
 ADD $TARGETARCH/$TARGET_BIN /greptime/bin/
 
-ENV PATH /greptime/bin/:$PATH
+ENV PATH=/greptime/bin/:$PATH
 
 ENV TARGET_BIN=$TARGET_BIN
 

docs/how-to/how-to-change-log-level-on-the-fly.md

@@ -13,4 +13,19 @@ Log Level changed from Some("info") to "trace,flow=debug"%
 
 The data is a string in the format of `global_level,module1=level1,module2=level2,...` that follows the same rule of `RUST_LOG`. 
 
-The module is the module name of the log, and the level is the log level. The log level can be one of the following: `trace`, `debug`, `info`, `warn`, `error`, `off`(case insensitive).
+The module is the module name of the log, and the level is the log level. The log level can be one of the following: `trace`, `debug`, `info`, `warn`, `error`, `off`(case insensitive).
+
+# Enable/Disable Trace on the Fly
+
+## HTTP API
+
+example:
+```bash
+curl --data "true" 127.0.0.1:4000/debug/enable_trace
+```
+And database will reply with something like:
+```
+trace enabled%
+```
+
+Possible values are "true" or "false".

docs/rfcs/2025-07-23-global-gc-worker.md

@@ -106,6 +106,37 @@ This mechanism may be too complex to implement at once. We can consider a two-ph
 Also the read replica shouldn't be later in manifest version for more than the lingering time of obsolete files, otherwise it might ref to files that are already deleted by the GC worker.
 - need to upload tmp manifest to object storage, which may introduce additional complexity and potential performance overhead. But since long-running queries are typically not frequent, the performance impact is expected to be minimal.
 
+one potential race condition with region-migration is illustrated below:
+
+```mermaid
+sequenceDiagram
+    participant gc_worker as GC Worker(same dn as region 1)
+    participant region1 as Region 1 (Leader → Follower)
+    participant region2 as Region 2 (Follower → Leader)
+    participant region_dir as Region Directory
+
+    gc_worker->>region1: Start GC, get region manifest
+    activate region1
+    region1-->>gc_worker: Region 1 manifest
+    deactivate region1
+    gc_worker->>region_dir: Scan region directory
+
+    Note over region1,region2: Region Migration Occurs
+    region1-->>region2: Downgrade to Follower
+    region2-->>region1: Becomes Leader
+
+    region2->>region_dir: Add new file
+
+    gc_worker->>region_dir: Continue scanning
+    gc_worker-->>region_dir: Discovers new file
+    Note over gc_worker: New file not in Region 1's manifest
+    gc_worker->>gc_worker: Mark file as orphan(incorrectly)
+```
+which could cause gc worker to incorrectly mark the new file as orphan and delete it, if config the lingering time for orphan files(files not mentioned anywhere(in used or unused)) is not long enough.
+
+A good enough solution could be to use lock to prevent gc worker to happen on the region if region migration is happening on the region, and vise versa.
+
+The race condition between gc worker and repartition also needs to be considered carefully. For now, acquiring lock for both region-migration and repartition during gc worker process could be a simple solution.
 
 ## Conclusion and Rationale
 

src/api/src/helper.rs

@@ -12,7 +12,7 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
-use std::collections::HashSet;
+use std::collections::{BTreeMap, HashSet};
 use std::sync::Arc;
 
 use common_decimal::Decimal128;
@@ -20,6 +20,7 @@ use common_decimal::decimal128::{DECIMAL128_DEFAULT_SCALE, DECIMAL128_MAX_PRECIS
 use common_time::time::Time;
 use common_time::timestamp::TimeUnit;
 use common_time::{Date, IntervalDayTime, IntervalMonthDayNano, IntervalYearMonth, Timestamp};
+use datatypes::json::value::{JsonNumber, JsonValue, JsonValueRef, JsonVariant};
 use datatypes::prelude::{ConcreteDataType, ValueRef};
 use datatypes::types::{
     IntervalType, JsonFormat, StructField, StructType, TimeType, TimestampType,
@@ -34,9 +35,9 @@ use greptime_proto::v1::greptime_request::Request;
 use greptime_proto::v1::query_request::Query;
 use greptime_proto::v1::value::ValueData;
 use greptime_proto::v1::{
-    self, ColumnDataTypeExtension, DdlRequest, DecimalTypeExtension, JsonNativeTypeExtension,
-    JsonTypeExtension, ListTypeExtension, QueryRequest, Row, SemanticType, StructTypeExtension,
-    VectorTypeExtension,
+    self, ColumnDataTypeExtension, DdlRequest, DecimalTypeExtension, DictionaryTypeExtension,
+    JsonList, JsonNativeTypeExtension, JsonObject, JsonTypeExtension, ListTypeExtension,
+    QueryRequest, Row, SemanticType, StructTypeExtension, VectorTypeExtension, json_value,
 };
 use paste::paste;
 use snafu::prelude::*;
@@ -215,6 +216,26 @@ impl From<ColumnDataTypeWrapper> for ConcreteDataType {
                     ConcreteDataType::null_datatype()
                 }
             }
+            ColumnDataType::Dictionary => {
+                if let Some(TypeExt::DictionaryType(d)) = datatype_wrapper
+                    .datatype_ext
+                    .as_ref()
+                    .and_then(|datatype_ext| datatype_ext.type_ext.as_ref())
+                {
+                    let key_type = ColumnDataTypeWrapper {
+                        datatype: d.key_datatype(),
+                        datatype_ext: d.key_datatype_extension.clone().map(|ext| *ext),
+                    };
+                    let value_type = ColumnDataTypeWrapper {
+                        datatype: d.value_datatype(),
+                        datatype_ext: d.value_datatype_extension.clone().map(|ext| *ext),
+                    };
+                    ConcreteDataType::dictionary_datatype(key_type.into(), value_type.into())
+                } else {
+                    // invalid state: type extension not found
+                    ConcreteDataType::null_datatype()
+                }
+            }
         }
     }
 }
@@ -338,13 +359,30 @@ impl ColumnDataTypeWrapper {
             }),
         }
     }
+
+    pub fn dictionary_datatype(
+        key_type: ColumnDataTypeWrapper,
+        value_type: ColumnDataTypeWrapper,
+    ) -> Self {
+        ColumnDataTypeWrapper {
+            datatype: ColumnDataType::Dictionary,
+            datatype_ext: Some(ColumnDataTypeExtension {
+                type_ext: Some(TypeExt::DictionaryType(Box::new(DictionaryTypeExtension {
+                    key_datatype: key_type.datatype().into(),
+                    key_datatype_extension: key_type.datatype_ext.map(Box::new),
+                    value_datatype: value_type.datatype().into(),
+                    value_datatype_extension: value_type.datatype_ext.map(Box::new),
+                }))),
+            }),
+        }
+    }
 }
 
 impl TryFrom<ConcreteDataType> for ColumnDataTypeWrapper {
     type Error = error::Error;
 
     fn try_from(datatype: ConcreteDataType) -> Result<Self> {
-        let column_datatype = match datatype {
+        let column_datatype = match &datatype {
             ConcreteDataType::Boolean(_) => ColumnDataType::Boolean,
             ConcreteDataType::Int8(_) => ColumnDataType::Int8,
             ConcreteDataType::Int16(_) => ColumnDataType::Int16,
@@ -381,9 +419,8 @@ impl TryFrom<ConcreteDataType> for ColumnDataTypeWrapper {
             ConcreteDataType::Vector(_) => ColumnDataType::Vector,
             ConcreteDataType::List(_) => ColumnDataType::List,
             ConcreteDataType::Struct(_) => ColumnDataType::Struct,
-            ConcreteDataType::Null(_)
-            | ConcreteDataType::Dictionary(_)
-            | ConcreteDataType::Duration(_) => {
+            ConcreteDataType::Dictionary(_) => ColumnDataType::Dictionary,
+            ConcreteDataType::Null(_) | ConcreteDataType::Duration(_) => {
                 return error::IntoColumnDataTypeSnafu { from: datatype }.fail();
             }
         };
@@ -405,7 +442,9 @@ impl TryFrom<ConcreteDataType> for ColumnDataTypeWrapper {
                             type_ext: Some(TypeExt::JsonType(JsonTypeExtension::JsonBinary.into())),
                         }),
                         JsonFormat::Native(inner) => {
-                            let inner_type = ColumnDataTypeWrapper::try_from(*inner.clone())?;
+                            let inner_type = ColumnDataTypeWrapper::try_from(
+                                ConcreteDataType::from(inner.as_ref()),
+                            )?;
                             Some(ColumnDataTypeExtension {
                                 type_ext: Some(TypeExt::JsonNativeType(Box::new(
                                     JsonNativeTypeExtension {
@@ -463,6 +502,25 @@ impl TryFrom<ConcreteDataType> for ColumnDataTypeWrapper {
                     None
                 }
             }
+            ColumnDataType::Dictionary => {
+                if let ConcreteDataType::Dictionary(dict_type) = &datatype {
+                    let key_type = ColumnDataTypeWrapper::try_from(dict_type.key_type().clone())?;
+                    let value_type =
+                        ColumnDataTypeWrapper::try_from(dict_type.value_type().clone())?;
+                    Some(ColumnDataTypeExtension {
+                        type_ext: Some(TypeExt::DictionaryType(Box::new(
+                            DictionaryTypeExtension {
+                                key_datatype: key_type.datatype.into(),
+                                key_datatype_extension: key_type.datatype_ext.map(Box::new),
+                                value_datatype: value_type.datatype.into(),
+                                value_datatype_extension: value_type.datatype_ext.map(Box::new),
+                            },
+                        ))),
+                    })
+                } else {
+                    None
+                }
+            }
             _ => None,
         };
         Ok(Self {
@@ -601,6 +659,9 @@ pub fn values_with_capacity(datatype: ColumnDataType, capacity: usize) -> Values
             struct_values: Vec::with_capacity(capacity),
             ..Default::default()
         },
+        ColumnDataType::Dictionary => Values {
+            ..Default::default()
+        },
     }
 }
 
@@ -801,21 +862,8 @@ pub fn pb_value_to_value_ref<'a>(
         }
 
         ValueData::JsonValue(inner_value) => {
-            let json_datatype_ext = datatype_ext
-                .as_ref()
-                .and_then(|ext| {
-                    if let Some(TypeExt::JsonNativeType(l)) = &ext.type_ext {
-                        Some(l)
-                    } else {
-                        None
-                    }
-                })
-                .expect("json value must contain datatype ext");
-
-            ValueRef::Json(Box::new(pb_value_to_value_ref(
-                inner_value,
-                json_datatype_ext.datatype_extension.as_deref(),
-            )))
+            let value = decode_json_value(inner_value);
+            ValueRef::Json(Box::new(value))
         }
     }
 }
@@ -938,12 +986,72 @@ pub fn to_proto_value(value: Value) -> v1::Value {
             })),
         },
         Value::Json(v) => v1::Value {
-            value_data: Some(ValueData::JsonValue(Box::new(to_proto_value(*v)))),
+            value_data: Some(ValueData::JsonValue(encode_json_value(*v))),
         },
         Value::Duration(_) => v1::Value { value_data: None },
     }
 }
 
+fn encode_json_value(value: JsonValue) -> v1::JsonValue {
+    fn helper(json: JsonVariant) -> v1::JsonValue {
+        let value = match json {
+            JsonVariant::Null => None,
+            JsonVariant::Bool(x) => Some(json_value::Value::Boolean(x)),
+            JsonVariant::Number(x) => Some(match x {
+                JsonNumber::PosInt(i) => json_value::Value::Uint(i),
+                JsonNumber::NegInt(i) => json_value::Value::Int(i),
+                JsonNumber::Float(f) => json_value::Value::Float(f.0),
+            }),
+            JsonVariant::String(x) => Some(json_value::Value::Str(x)),
+            JsonVariant::Array(x) => Some(json_value::Value::Array(JsonList {
+                items: x.into_iter().map(helper).collect::<Vec<_>>(),
+            })),
+            JsonVariant::Object(x) => {
+                let entries = x
+                    .into_iter()
+                    .map(|(key, v)| v1::json_object::Entry {
+                        key,
+                        value: Some(helper(v)),
+                    })
+                    .collect::<Vec<_>>();
+                Some(json_value::Value::Object(JsonObject { entries }))
+            }
+        };
+        v1::JsonValue { value }
+    }
+    helper(value.into_variant())
+}
+
+fn decode_json_value(value: &v1::JsonValue) -> JsonValueRef<'_> {
+    let Some(value) = &value.value else {
+        return JsonValueRef::null();
+    };
+    match value {
+        json_value::Value::Boolean(x) => (*x).into(),
+        json_value::Value::Int(x) => (*x).into(),
+        json_value::Value::Uint(x) => (*x).into(),
+        json_value::Value::Float(x) => (*x).into(),
+        json_value::Value::Str(x) => (x.as_str()).into(),
+        json_value::Value::Array(array) => array
+            .items
+            .iter()
+            .map(|x| decode_json_value(x).into_variant())
+            .collect::<Vec<_>>()
+            .into(),
+        json_value::Value::Object(x) => x
+            .entries
+            .iter()
+            .filter_map(|entry| {
+                entry
+                    .value
+                    .as_ref()
+                    .map(|v| (entry.key.as_str(), decode_json_value(v).into_variant()))
+            })
+            .collect::<BTreeMap<_, _>>()
+            .into(),
+    }
+}
+
 fn convert_list_to_pb_values(list_value: ListValue) -> Vec<v1::Value> {
     list_value
         .take_items()
@@ -1065,9 +1173,7 @@ pub fn value_to_grpc_value(value: Value) -> GrpcValue {
                     .collect();
                 Some(ValueData::StructValue(v1::StructValue { items }))
             }
-            Value::Json(inner_value) => Some(ValueData::JsonValue(Box::new(value_to_grpc_value(
-                *inner_value,
-            )))),
+            Value::Json(v) => Some(ValueData::JsonValue(encode_json_value(*v))),
             Value::Duration(_) => unreachable!(),
         },
     }
@@ -1259,6 +1365,9 @@ mod tests {
         let values = values_with_capacity(ColumnDataType::Json, 2);
         assert_eq!(2, values.json_values.capacity());
         assert_eq!(2, values.string_values.capacity());
+
+        let values = values_with_capacity(ColumnDataType::Dictionary, 2);
+        assert!(values.bool_values.is_empty());
     }
 
     #[test]
@@ -1355,6 +1464,17 @@ mod tests {
             ConcreteDataType::list_datatype(Arc::new(ConcreteDataType::string_datatype())),
             ColumnDataTypeWrapper::list_datatype(ColumnDataTypeWrapper::string_datatype()).into()
         );
+        assert_eq!(
+            ConcreteDataType::dictionary_datatype(
+                ConcreteDataType::int32_datatype(),
+                ConcreteDataType::string_datatype()
+            ),
+            ColumnDataTypeWrapper::dictionary_datatype(
+                ColumnDataTypeWrapper::int32_datatype(),
+                ColumnDataTypeWrapper::string_datatype()
+            )
+            .into()
+        );
         let struct_type = StructType::new(Arc::new(vec![
             StructField::new("id".to_string(), ConcreteDataType::int64_datatype(), true),
             StructField::new(
@@ -1525,6 +1645,18 @@ mod tests {
             ColumnDataTypeWrapper::vector_datatype(3),
             ConcreteDataType::vector_datatype(3).try_into().unwrap()
         );
+        assert_eq!(
+            ColumnDataTypeWrapper::dictionary_datatype(
+                ColumnDataTypeWrapper::int32_datatype(),
+                ColumnDataTypeWrapper::string_datatype()
+            ),
+            ConcreteDataType::dictionary_datatype(
+                ConcreteDataType::int32_datatype(),
+                ConcreteDataType::string_datatype()
+            )
+            .try_into()
+            .unwrap()
+        );
 
         let result: Result<ColumnDataTypeWrapper> = ConcreteDataType::null_datatype().try_into();
         assert!(result.is_err());
@@ -1580,6 +1712,20 @@ mod tests {
                         datatype_extension: Some(Box::new(ColumnDataTypeExtension {
                             type_ext: Some(TypeExt::StructType(StructTypeExtension {
                                 fields: vec![
+                                    v1::StructField {
+                                        name: "address".to_string(),
+                                        datatype: ColumnDataTypeWrapper::string_datatype()
+                                            .datatype()
+                                            .into(),
+                                        datatype_extension: None
+                                    },
+                                    v1::StructField {
+                                        name: "age".to_string(),
+                                        datatype: ColumnDataTypeWrapper::int64_datatype()
+                                            .datatype()
+                                            .into(),
+                                        datatype_extension: None
+                                    },
                                     v1::StructField {
                                         name: "id".to_string(),
                                         datatype: ColumnDataTypeWrapper::int64_datatype()
@@ -1594,20 +1740,6 @@ mod tests {
                                             .into(),
                                         datatype_extension: None
                                     },
-                                    v1::StructField {
-                                        name: "age".to_string(),
-                                        datatype: ColumnDataTypeWrapper::int32_datatype()
-                                            .datatype()
-                                            .into(),
-                                        datatype_extension: None
-                                    },
-                                    v1::StructField {
-                                        name: "address".to_string(),
-                                        datatype: ColumnDataTypeWrapper::string_datatype()
-                                            .datatype()
-                                            .into(),
-                                        datatype_extension: None
-                                    }
                                 ]
                             }))
                         }))
@@ -1778,4 +1910,199 @@ mod tests {
             _ => panic!("Unexpected value type"),
         }
     }
+
+    #[test]
+    fn test_encode_decode_json_value() {
+        let json = JsonValue::null();
+        let proto = encode_json_value(json.clone());
+        assert!(proto.value.is_none());
+        let value = decode_json_value(&proto);
+        assert_eq!(json.as_ref(), value);
+
+        let json: JsonValue = true.into();
+        let proto = encode_json_value(json.clone());
+        assert_eq!(proto.value, Some(json_value::Value::Boolean(true)));
+        let value = decode_json_value(&proto);
+        assert_eq!(json.as_ref(), value);
+
+        let json: JsonValue = (-1i64).into();
+        let proto = encode_json_value(json.clone());
+        assert_eq!(proto.value, Some(json_value::Value::Int(-1)));
+        let value = decode_json_value(&proto);
+        assert_eq!(json.as_ref(), value);
+
+        let json: JsonValue = 1u64.into();
+        let proto = encode_json_value(json.clone());
+        assert_eq!(proto.value, Some(json_value::Value::Uint(1)));
+        let value = decode_json_value(&proto);
+        assert_eq!(json.as_ref(), value);
+
+        let json: JsonValue = 1.0f64.into();
+        let proto = encode_json_value(json.clone());
+        assert_eq!(proto.value, Some(json_value::Value::Float(1.0)));
+        let value = decode_json_value(&proto);
+        assert_eq!(json.as_ref(), value);
+
+        let json: JsonValue = "s".into();
+        let proto = encode_json_value(json.clone());
+        assert_eq!(proto.value, Some(json_value::Value::Str("s".to_string())));
+        let value = decode_json_value(&proto);
+        assert_eq!(json.as_ref(), value);
+
+        let json: JsonValue = [1i64, 2, 3].into();
+        let proto = encode_json_value(json.clone());
+        assert_eq!(
+            proto.value,
+            Some(json_value::Value::Array(JsonList {
+                items: vec![
+                    v1::JsonValue {
+                        value: Some(json_value::Value::Int(1))
+                    },
+                    v1::JsonValue {
+                        value: Some(json_value::Value::Int(2))
+                    },
+                    v1::JsonValue {
+                        value: Some(json_value::Value::Int(3))
+                    }
+                ]
+            }))
+        );
+        let value = decode_json_value(&proto);
+        assert_eq!(json.as_ref(), value);
+
+        let json: JsonValue = [(); 0].into();
+        let proto = encode_json_value(json.clone());
+        assert_eq!(
+            proto.value,
+            Some(json_value::Value::Array(JsonList { items: vec![] }))
+        );
+        let value = decode_json_value(&proto);
+        assert_eq!(json.as_ref(), value);
+
+        let json: JsonValue = [("k3", 3i64), ("k2", 2i64), ("k1", 1i64)].into();
+        let proto = encode_json_value(json.clone());
+        assert_eq!(
+            proto.value,
+            Some(json_value::Value::Object(JsonObject {
+                entries: vec![
+                    v1::json_object::Entry {
+                        key: "k1".to_string(),
+                        value: Some(v1::JsonValue {
+                            value: Some(json_value::Value::Int(1))
+                        }),
+                    },
+                    v1::json_object::Entry {
+                        key: "k2".to_string(),
+                        value: Some(v1::JsonValue {
+                            value: Some(json_value::Value::Int(2))
+                        }),
+                    },
+                    v1::json_object::Entry {
+                        key: "k3".to_string(),
+                        value: Some(v1::JsonValue {
+                            value: Some(json_value::Value::Int(3))
+                        }),
+                    },
+                ]
+            }))
+        );
+        let value = decode_json_value(&proto);
+        assert_eq!(json.as_ref(), value);
+
+        let json: JsonValue = [("null", ()); 0].into();
+        let proto = encode_json_value(json.clone());
+        assert_eq!(
+            proto.value,
+            Some(json_value::Value::Object(JsonObject { entries: vec![] }))
+        );
+        let value = decode_json_value(&proto);
+        assert_eq!(json.as_ref(), value);
+
+        let json: JsonValue = [
+            ("null", JsonVariant::from(())),
+            ("bool", false.into()),
+            ("list", ["hello", "world"].into()),
+            (
+                "object",
+                [
+                    ("positive_i", JsonVariant::from(42u64)),
+                    ("negative_i", (-42i64).into()),
+                    ("nested", [("what", "blah")].into()),
+                ]
+                .into(),
+            ),
+        ]
+        .into();
+        let proto = encode_json_value(json.clone());
+        assert_eq!(
+            proto.value,
+            Some(json_value::Value::Object(JsonObject {
+                entries: vec![
+                    v1::json_object::Entry {
+                        key: "bool".to_string(),
+                        value: Some(v1::JsonValue {
+                            value: Some(json_value::Value::Boolean(false))
+                        }),
+                    },
+                    v1::json_object::Entry {
+                        key: "list".to_string(),
+                        value: Some(v1::JsonValue {
+                            value: Some(json_value::Value::Array(JsonList {
+                                items: vec![
+                                    v1::JsonValue {
+                                        value: Some(json_value::Value::Str("hello".to_string()))
+                                    },
+                                    v1::JsonValue {
+                                        value: Some(json_value::Value::Str("world".to_string()))
+                                    },
+                                ]
+                            }))
+                        }),
+                    },
+                    v1::json_object::Entry {
+                        key: "null".to_string(),
+                        value: Some(v1::JsonValue { value: None }),
+                    },
+                    v1::json_object::Entry {
+                        key: "object".to_string(),
+                        value: Some(v1::JsonValue {
+                            value: Some(json_value::Value::Object(JsonObject {
+                                entries: vec![
+                                    v1::json_object::Entry {
+                                        key: "negative_i".to_string(),
+                                        value: Some(v1::JsonValue {
+                                            value: Some(json_value::Value::Int(-42))
+                                        }),
+                                    },
+                                    v1::json_object::Entry {
+                                        key: "nested".to_string(),
+                                        value: Some(v1::JsonValue {
+                                            value: Some(json_value::Value::Object(JsonObject {
+                                                entries: vec![v1::json_object::Entry {
+                                                    key: "what".to_string(),
+                                                    value: Some(v1::JsonValue {
+                                                        value: Some(json_value::Value::Str(
+                                                            "blah".to_string()
+                                                        ))
+                                                    }),
+                                                },]
+                                            }))
+                                        }),
+                                    },
+                                    v1::json_object::Entry {
+                                        key: "positive_i".to_string(),
+                                        value: Some(v1::JsonValue {
+                                            value: Some(json_value::Value::Uint(42))
+                                        }),
+                                    },
+                                ]
+                            }))
+                        }),
+                    },
+                ]
+            }))
+        );
+        let value = decode_json_value(&proto);
+        assert_eq!(json.as_ref(), value);
+    }
 }

src/catalog/src/system_schema/information_schema.rs

@@ -22,7 +22,6 @@ mod procedure_info;
 pub mod process_list;
 pub mod region_peers;
 mod region_statistics;
-mod runtime_metrics;
 pub mod schemata;
 mod ssts;
 mod table_constraints;
@@ -65,7 +64,6 @@ use crate::system_schema::information_schema::information_memory_table::get_sche
 use crate::system_schema::information_schema::key_column_usage::InformationSchemaKeyColumnUsage;
 use crate::system_schema::information_schema::partitions::InformationSchemaPartitions;
 use crate::system_schema::information_schema::region_peers::InformationSchemaRegionPeers;
-use crate::system_schema::information_schema::runtime_metrics::InformationSchemaMetrics;
 use crate::system_schema::information_schema::schemata::InformationSchemaSchemata;
 use crate::system_schema::information_schema::ssts::{
     InformationSchemaSstsIndexMeta, InformationSchemaSstsManifest, InformationSchemaSstsStorage,
@@ -216,7 +214,6 @@ impl SystemSchemaProviderInner for InformationSchemaProvider {
                 self.catalog_name.clone(),
                 self.catalog_manager.clone(),
             )) as _),
-            RUNTIME_METRICS => Some(Arc::new(InformationSchemaMetrics::new())),
             PARTITIONS => Some(Arc::new(InformationSchemaPartitions::new(
                 self.catalog_name.clone(),
                 self.catalog_manager.clone(),
@@ -311,10 +308,6 @@ impl InformationSchemaProvider {
         // authentication details, and other critical information.
         // Only put these tables under `greptime` catalog to prevent info leak.
         if self.catalog_name == DEFAULT_CATALOG_NAME {
-            tables.insert(
-                RUNTIME_METRICS.to_string(),
-                self.build_table(RUNTIME_METRICS).unwrap(),
-            );
             tables.insert(
                 BUILD_INFO.to_string(),
                 self.build_table(BUILD_INFO).unwrap(),

src/catalog/src/system_schema/information_schema/runtime_metrics.rs

@@ -1,265 +0,0 @@
-// Copyright 2023 Greptime Team
-//
-// Licensed under the Apache License, Version 2.0 (the "License");
-// you may not use this file except in compliance with the License.
-// You may obtain a copy of the License at
-//
-//     http://www.apache.org/licenses/LICENSE-2.0
-//
-// Unless required by applicable law or agreed to in writing, software
-// distributed under the License is distributed on an "AS IS" BASIS,
-// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-// See the License for the specific language governing permissions and
-// limitations under the License.
-
-use std::sync::Arc;
-
-use arrow_schema::SchemaRef as ArrowSchemaRef;
-use common_catalog::consts::INFORMATION_SCHEMA_RUNTIME_METRICS_TABLE_ID;
-use common_error::ext::BoxedError;
-use common_recordbatch::adapter::RecordBatchStreamAdapter;
-use common_recordbatch::{RecordBatch, SendableRecordBatchStream};
-use common_time::util::current_time_millis;
-use datafusion::execution::TaskContext;
-use datafusion::physical_plan::SendableRecordBatchStream as DfSendableRecordBatchStream;
-use datafusion::physical_plan::stream::RecordBatchStreamAdapter as DfRecordBatchStreamAdapter;
-use datafusion::physical_plan::streaming::PartitionStream as DfPartitionStream;
-use datatypes::prelude::{ConcreteDataType, MutableVector};
-use datatypes::scalars::ScalarVectorBuilder;
-use datatypes::schema::{ColumnSchema, Schema, SchemaRef};
-use datatypes::vectors::{
-    ConstantVector, Float64VectorBuilder, StringVectorBuilder, TimestampMillisecondVector,
-    VectorRef,
-};
-use itertools::Itertools;
-use snafu::ResultExt;
-use store_api::storage::{ScanRequest, TableId};
-
-use crate::error::{CreateRecordBatchSnafu, InternalSnafu, Result};
-use crate::system_schema::information_schema::{InformationTable, RUNTIME_METRICS};
-
-#[derive(Debug)]
-pub(super) struct InformationSchemaMetrics {
-    schema: SchemaRef,
-}
-
-const METRIC_NAME: &str = "metric_name";
-const METRIC_VALUE: &str = "value";
-const METRIC_LABELS: &str = "labels";
-const PEER_ADDR: &str = "peer_addr";
-const PEER_TYPE: &str = "peer_type";
-const TIMESTAMP: &str = "timestamp";
-
-/// The `information_schema.runtime_metrics` virtual table.
-/// It provides the GreptimeDB runtime metrics for the users by SQL.
-impl InformationSchemaMetrics {
-    pub(super) fn new() -> Self {
-        Self {
-            schema: Self::schema(),
-        }
-    }
-
-    fn schema() -> SchemaRef {
-        Arc::new(Schema::new(vec![
-            ColumnSchema::new(METRIC_NAME, ConcreteDataType::string_datatype(), false),
-            ColumnSchema::new(METRIC_VALUE, ConcreteDataType::float64_datatype(), false),
-            ColumnSchema::new(METRIC_LABELS, ConcreteDataType::string_datatype(), true),
-            ColumnSchema::new(PEER_ADDR, ConcreteDataType::string_datatype(), true),
-            ColumnSchema::new(PEER_TYPE, ConcreteDataType::string_datatype(), false),
-            ColumnSchema::new(
-                TIMESTAMP,
-                ConcreteDataType::timestamp_millisecond_datatype(),
-                false,
-            ),
-        ]))
-    }
-
-    fn builder(&self) -> InformationSchemaMetricsBuilder {
-        InformationSchemaMetricsBuilder::new(self.schema.clone())
-    }
-}
-
-impl InformationTable for InformationSchemaMetrics {
-    fn table_id(&self) -> TableId {
-        INFORMATION_SCHEMA_RUNTIME_METRICS_TABLE_ID
-    }
-
-    fn table_name(&self) -> &'static str {
-        RUNTIME_METRICS
-    }
-
-    fn schema(&self) -> SchemaRef {
-        self.schema.clone()
-    }
-
-    fn to_stream(&self, request: ScanRequest) -> Result<SendableRecordBatchStream> {
-        let schema = self.schema.arrow_schema().clone();
-        let mut builder = self.builder();
-        let stream = Box::pin(DfRecordBatchStreamAdapter::new(
-            schema,
-            futures::stream::once(async move {
-                builder
-                    .make_metrics(Some(request))
-                    .await
-                    .map(|x| x.into_df_record_batch())
-                    .map_err(Into::into)
-            }),
-        ));
-
-        Ok(Box::pin(
-            RecordBatchStreamAdapter::try_new(stream)
-                .map_err(BoxedError::new)
-                .context(InternalSnafu)?,
-        ))
-    }
-}
-
-struct InformationSchemaMetricsBuilder {
-    schema: SchemaRef,
-
-    metric_names: StringVectorBuilder,
-    metric_values: Float64VectorBuilder,
-    metric_labels: StringVectorBuilder,
-    peer_addrs: StringVectorBuilder,
-    peer_types: StringVectorBuilder,
-}
-
-impl InformationSchemaMetricsBuilder {
-    fn new(schema: SchemaRef) -> Self {
-        Self {
-            schema,
-            metric_names: StringVectorBuilder::with_capacity(42),
-            metric_values: Float64VectorBuilder::with_capacity(42),
-            metric_labels: StringVectorBuilder::with_capacity(42),
-            peer_addrs: StringVectorBuilder::with_capacity(42),
-            peer_types: StringVectorBuilder::with_capacity(42),
-        }
-    }
-
-    fn add_metric(
-        &mut self,
-        metric_name: &str,
-        labels: String,
-        metric_value: f64,
-        peer: Option<&str>,
-        peer_type: &str,
-    ) {
-        self.metric_names.push(Some(metric_name));
-        self.metric_values.push(Some(metric_value));
-        self.metric_labels.push(Some(&labels));
-        self.peer_addrs.push(peer);
-        self.peer_types.push(Some(peer_type));
-    }
-
-    async fn make_metrics(&mut self, _request: Option<ScanRequest>) -> Result<RecordBatch> {
-        let metric_families = prometheus::gather();
-
-        let write_request =
-            common_telemetry::metric::convert_metric_to_write_request(metric_families, None, 0);
-
-        for ts in write_request.timeseries {
-            //Safety: always has `__name__` label
-            let metric_name = ts
-                .labels
-                .iter()
-                .find_map(|label| {
-                    if label.name == "__name__" {
-                        Some(label.value.clone())
-                    } else {
-                        None
-                    }
-                })
-                .unwrap();
-
-            self.add_metric(
-                &metric_name,
-                ts.labels
-                    .into_iter()
-                    .filter_map(|label| {
-                        if label.name == "__name__" {
-                            None
-                        } else {
-                            Some(format!("{}={}", label.name, label.value))
-                        }
-                    })
-                    .join(", "),
-                // Safety: always has a sample
-                ts.samples[0].value,
-                // The peer column is always `None` for standalone
-                None,
-                "STANDALONE",
-            );
-        }
-
-        // FIXME(dennis): fetching other peers metrics
-        self.finish()
-    }
-
-    fn finish(&mut self) -> Result<RecordBatch> {
-        let rows_num = self.metric_names.len();
-
-        let timestamps = Arc::new(ConstantVector::new(
-            Arc::new(TimestampMillisecondVector::from_slice([
-                current_time_millis(),
-            ])),
-            rows_num,
-        ));
-
-        let columns: Vec<VectorRef> = vec![
-            Arc::new(self.metric_names.finish()),
-            Arc::new(self.metric_values.finish()),
-            Arc::new(self.metric_labels.finish()),
-            Arc::new(self.peer_addrs.finish()),
-            Arc::new(self.peer_types.finish()),
-            timestamps,
-        ];
-
-        RecordBatch::new(self.schema.clone(), columns).context(CreateRecordBatchSnafu)
-    }
-}
-
-impl DfPartitionStream for InformationSchemaMetrics {
-    fn schema(&self) -> &ArrowSchemaRef {
-        self.schema.arrow_schema()
-    }
-
-    fn execute(&self, _: Arc<TaskContext>) -> DfSendableRecordBatchStream {
-        let schema = self.schema.arrow_schema().clone();
-        let mut builder = self.builder();
-        Box::pin(DfRecordBatchStreamAdapter::new(
-            schema,
-            futures::stream::once(async move {
-                builder
-                    .make_metrics(None)
-                    .await
-                    .map(|x| x.into_df_record_batch())
-                    .map_err(Into::into)
-            }),
-        ))
-    }
-}
-
-#[cfg(test)]
-mod tests {
-    use common_recordbatch::RecordBatches;
-
-    use super::*;
-
-    #[tokio::test]
-    async fn test_make_metrics() {
-        let metrics = InformationSchemaMetrics::new();
-
-        let stream = metrics.to_stream(ScanRequest::default()).unwrap();
-
-        let batches = RecordBatches::try_collect(stream).await.unwrap();
-
-        let result_literal = batches.pretty_print().unwrap();
-
-        assert!(result_literal.contains(METRIC_NAME));
-        assert!(result_literal.contains(METRIC_VALUE));
-        assert!(result_literal.contains(METRIC_LABELS));
-        assert!(result_literal.contains(PEER_ADDR));
-        assert!(result_literal.contains(PEER_TYPE));
-        assert!(result_literal.contains(TIMESTAMP));
-    }
-}

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

@@ -38,7 +38,6 @@ pub const TABLE_PRIVILEGES: &str = "table_privileges";
 pub const TRIGGERS: &str = "triggers";
 pub const GLOBAL_STATUS: &str = "global_status";
 pub const SESSION_STATUS: &str = "session_status";
-pub const RUNTIME_METRICS: &str = "runtime_metrics";
 pub const PARTITIONS: &str = "partitions";
 pub const REGION_PEERS: &str = "region_peers";
 pub const TABLE_CONSTRAINTS: &str = "table_constraints";

src/catalog/src/system_schema/information_schema/tables.rs

@@ -12,7 +12,6 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
-use std::collections::HashSet;
 use std::sync::{Arc, Weak};
 
 use arrow_schema::SchemaRef as ArrowSchemaRef;
@@ -255,14 +254,17 @@ impl InformationSchemaTablesBuilder {
         // TODO(dennis): `region_stats` API is not stable in distributed cluster because of network issue etc.
         // But we don't want the statements such as `show tables` fail,
         // so using `unwrap_or_else` here instead of `?` operator.
-        let region_stats = information_extension
-            .region_stats()
-            .await
-            .map_err(|e| {
-                error!(e; "Failed to call region_stats");
-                e
-            })
-            .unwrap_or_else(|_| vec![]);
+        let region_stats = {
+            let mut x = information_extension
+                .region_stats()
+                .await
+                .unwrap_or_else(|e| {
+                    error!(e; "Failed to find region stats in information_schema, fallback to all empty");
+                    vec![]
+                });
+            x.sort_unstable_by_key(|x| x.id);
+            x
+        };
 
         for schema_name in catalog_manager.schema_names(&catalog_name, None).await? {
             let mut stream = catalog_manager.tables(&catalog_name, &schema_name, None);
@@ -273,16 +275,16 @@ impl InformationSchemaTablesBuilder {
                 // TODO(dennis): make it working for metric engine
                 let table_region_stats =
                     if table_info.meta.engine == MITO_ENGINE || table_info.is_physical_table() {
-                        let region_ids = table_info
+                        table_info
                             .meta
                             .region_numbers
                             .iter()
                             .map(|n| RegionId::new(table_info.ident.table_id, *n))
-                            .collect::<HashSet<_>>();
-
-                        region_stats
-                            .iter()
-                            .filter(|stat| region_ids.contains(&stat.id))
+                            .flat_map(|region_id| {
+                                region_stats
+                                    .binary_search_by_key(&region_id, |x| x.id)
+                                    .map(|i| &region_stats[i])
+                            })
                             .collect::<Vec<_>>()
                     } else {
                         vec![]

src/cli/src/data/export.rs

@@ -67,9 +67,17 @@ pub struct ExportCommand {
     #[clap(long, default_value_t = default_database())]
     database: String,
 
-    /// Parallelism of the export.
-    #[clap(long, short = 'j', default_value = "1")]
-    export_jobs: usize,
+    /// The number of databases exported in parallel.
+    /// For example, if there are 20 databases and `db_parallelism` is 4,
+    /// 4 databases will be exported concurrently.
+    #[clap(long, short = 'j', default_value = "1", alias = "export-jobs")]
+    db_parallelism: usize,
+
+    /// The number of tables exported in parallel within a single database.
+    /// For example, if a database has 30 tables and `parallelism` is 8,
+    /// 8 tables will be exported concurrently.
+    #[clap(long, default_value = "4")]
+    table_parallelism: usize,
 
     /// Max retry times for each job.
     #[clap(long, default_value = "3")]
@@ -210,10 +218,11 @@ impl ExportCommand {
             schema,
             database_client,
             output_dir: self.output_dir.clone(),
-            parallelism: self.export_jobs,
+            export_jobs: self.db_parallelism,
             target: self.target.clone(),
             start_time: self.start_time.clone(),
             end_time: self.end_time.clone(),
+            parallelism: self.table_parallelism,
             s3: self.s3,
             ddl_local_dir: self.ddl_local_dir.clone(),
             s3_bucket: self.s3_bucket.clone(),
@@ -251,10 +260,11 @@ pub struct Export {
     schema: Option<String>,
     database_client: DatabaseClient,
     output_dir: Option<String>,
-    parallelism: usize,
+    export_jobs: usize,
     target: ExportTarget,
     start_time: Option<String>,
     end_time: Option<String>,
+    parallelism: usize,
     s3: bool,
     ddl_local_dir: Option<String>,
     s3_bucket: Option<String>,
@@ -464,7 +474,7 @@ impl Export {
 
     async fn export_create_table(&self) -> Result<()> {
         let timer = Instant::now();
-        let semaphore = Arc::new(Semaphore::new(self.parallelism));
+        let semaphore = Arc::new(Semaphore::new(self.export_jobs));
         let db_names = self.get_db_names().await?;
         let db_count = db_names.len();
         let operator = Arc::new(self.build_prefer_fs_operator().await?);
@@ -625,13 +635,13 @@ impl Export {
 
     async fn export_database_data(&self) -> Result<()> {
         let timer = Instant::now();
-        let semaphore = Arc::new(Semaphore::new(self.parallelism));
+        let semaphore = Arc::new(Semaphore::new(self.export_jobs));
         let db_names = self.get_db_names().await?;
         let db_count = db_names.len();
         let mut tasks = Vec::with_capacity(db_count);
         let operator = Arc::new(self.build_operator().await?);
         let fs_first_operator = Arc::new(self.build_prefer_fs_operator().await?);
-        let with_options = build_with_options(&self.start_time, &self.end_time);
+        let with_options = build_with_options(&self.start_time, &self.end_time, self.parallelism);
 
         for schema in db_names {
             let semaphore_moved = semaphore.clone();
@@ -888,7 +898,11 @@ impl Tool for Export {
 }
 
 /// Builds the WITH options string for SQL commands, assuming consistent syntax across S3 and local exports.
-fn build_with_options(start_time: &Option<String>, end_time: &Option<String>) -> String {
+fn build_with_options(
+    start_time: &Option<String>,
+    end_time: &Option<String>,
+    parallelism: usize,
+) -> String {
     let mut options = vec!["format = 'parquet'".to_string()];
     if let Some(start) = start_time {
         options.push(format!("start_time = '{}'", start));
@@ -896,5 +910,6 @@ fn build_with_options(start_time: &Option<String>, end_time: &Option<String>) ->
     if let Some(end) = end_time {
         options.push(format!("end_time = '{}'", end));
     }
+    options.push(format!("parallelism = {}", parallelism));
     options.join(", ")
 }

src/cli/src/data/import.rs

@@ -56,9 +56,11 @@ pub struct ImportCommand {
     #[clap(long, default_value_t = default_database())]
     database: String,
 
-    /// Parallelism of the import.
-    #[clap(long, short = 'j', default_value = "1")]
-    import_jobs: usize,
+    /// The number of databases imported in parallel.
+    /// For example, if there are 20 databases and `db_parallelism` is 4,
+    /// 4 databases will be imported concurrently.
+    #[clap(long, short = 'j', default_value = "1", alias = "import-jobs")]
+    db_parallelism: usize,
 
     /// Max retry times for each job.
     #[clap(long, default_value = "3")]
@@ -109,7 +111,7 @@ impl ImportCommand {
             schema,
             database_client,
             input_dir: self.input_dir.clone(),
-            parallelism: self.import_jobs,
+            parallelism: self.db_parallelism,
             target: self.target.clone(),
         }))
     }

src/client/src/client.rs

@@ -21,7 +21,7 @@ use api::v1::prometheus_gateway_client::PrometheusGatewayClient;
 use api::v1::region::region_client::RegionClient as PbRegionClient;
 use arrow_flight::flight_service_client::FlightServiceClient;
 use common_grpc::channel_manager::{
-    ChannelConfig, ChannelManager, ClientTlsOption, load_tls_config,
+    ChannelConfig, ChannelManager, ClientTlsOption, load_client_tls_config,
 };
 use parking_lot::RwLock;
 use snafu::{OptionExt, ResultExt};
@@ -95,9 +95,9 @@ impl Client {
         U: AsRef<str>,
         A: AsRef<[U]>,
     {
-        let channel_config = ChannelConfig::default().client_tls_config(client_tls);
-        let tls_config = load_tls_config(channel_config.client_tls.as_ref())
-            .context(error::CreateTlsChannelSnafu)?;
+        let channel_config = ChannelConfig::default().client_tls_config(client_tls.clone());
+        let tls_config =
+            load_client_tls_config(Some(client_tls)).context(error::CreateTlsChannelSnafu)?;
         let channel_manager = ChannelManager::with_config(channel_config, tls_config);
         Ok(Self::with_manager_and_urls(channel_manager, urls))
     }

src/client/src/database.rs

@@ -435,10 +435,10 @@ impl Database {
                             .context(ExternalSnafu)?;
                         match flight_message {
                             FlightMessage::RecordBatch(arrow_batch) => {
-                                yield RecordBatch::try_from_df_record_batch(
+                                yield Ok(RecordBatch::from_df_record_batch(
                                     schema_cloned.clone(),
                                     arrow_batch,
-                                )
+                                ))
                             }
                             FlightMessage::Metrics(_) => {}
                             FlightMessage::AffectedRows(_) | FlightMessage::Schema(_) => {

src/client/src/region.rs

@@ -182,10 +182,8 @@ impl RegionRequester {
 
                 match flight_message {
                     FlightMessage::RecordBatch(record_batch) => {
-                        let result_to_yield = RecordBatch::try_from_df_record_batch(
-                            schema_cloned.clone(),
-                            record_batch,
-                        );
+                        let result_to_yield =
+                            RecordBatch::from_df_record_batch(schema_cloned.clone(), record_batch);
 
                         // get the next message from the stream. normally it should be a metrics message.
                         if let Some(next_flight_message_result) = flight_message_stream.next().await
@@ -219,7 +217,7 @@ impl RegionRequester {
                             stream_ended = true;
                         }
 
-                        yield result_to_yield;
+                        yield Ok(result_to_yield);
                     }
                     FlightMessage::Metrics(s) => {
                         // just a branch in case of some metrics message comes after other things.

src/cmd/src/datanode/objbench.rs

@@ -161,6 +161,7 @@ impl ObjbenchCommand {
             level: 0,
             file_size,
             available_indexes: Default::default(),
+            indexes: Default::default(),
             index_file_size: 0,
             index_file_id: None,
             num_rows,

src/cmd/src/error.rs

@@ -99,13 +99,6 @@ pub enum Error {
         source: flow::Error,
     },
 
-    #[snafu(display("Servers error"))]
-    Servers {
-        #[snafu(implicit)]
-        location: Location,
-        source: servers::error::Error,
-    },
-
     #[snafu(display("Failed to start frontend"))]
     StartFrontend {
         #[snafu(implicit)]
@@ -336,7 +329,6 @@ impl ErrorExt for Error {
             Error::ShutdownFrontend { source, .. } => source.status_code(),
             Error::StartMetaServer { source, .. } => source.status_code(),
             Error::ShutdownMetaServer { source, .. } => source.status_code(),
-            Error::Servers { source, .. } => source.status_code(),
             Error::BuildMetaServer { source, .. } => source.status_code(),
             Error::UnsupportedSelectorType { source, .. } => source.status_code(),
             Error::BuildCli { source, .. } => source.status_code(),

src/cmd/src/frontend.rs

@@ -43,7 +43,6 @@ use frontend::instance::builder::FrontendBuilder;
 use frontend::server::Services;
 use meta_client::{MetaClientOptions, MetaClientType};
 use servers::addrs;
-use servers::export_metrics::ExportMetricsTask;
 use servers::grpc::GrpcOptions;
 use servers::tls::{TlsMode, TlsOption};
 use snafu::{OptionExt, ResultExt};
@@ -458,9 +457,6 @@ impl StartCommand {
         .context(error::StartFrontendSnafu)?;
         let instance = Arc::new(instance);
 
-        let export_metrics_task = ExportMetricsTask::try_new(&opts.export_metrics, Some(&plugins))
-            .context(error::ServersSnafu)?;
-
         let servers = Services::new(opts, instance.clone(), plugins)
             .build()
             .context(error::StartFrontendSnafu)?;
@@ -469,7 +465,6 @@ impl StartCommand {
             instance,
             servers,
             heartbeat_task,
-            export_metrics_task,
         };
 
         Ok(Instance::new(frontend, guard))

src/cmd/src/standalone.rs

@@ -57,7 +57,6 @@ use frontend::instance::StandaloneDatanodeManager;
 use frontend::instance::builder::FrontendBuilder;
 use frontend::server::Services;
 use meta_srv::metasrv::{FLOW_ID_SEQ, TABLE_ID_SEQ};
-use servers::export_metrics::ExportMetricsTask;
 use servers::tls::{TlsMode, TlsOption};
 use snafu::ResultExt;
 use standalone::StandaloneInformationExtension;
@@ -565,9 +564,6 @@ impl StartCommand {
         .context(StartFlownodeSnafu)?;
         flow_streaming_engine.set_frontend_invoker(invoker).await;
 
-        let export_metrics_task = ExportMetricsTask::try_new(&opts.export_metrics, Some(&plugins))
-            .context(error::ServersSnafu)?;
-
         let servers = Services::new(opts, fe_instance.clone(), plugins.clone())
             .build()
             .context(error::StartFrontendSnafu)?;
@@ -576,7 +572,6 @@ impl StartCommand {
             instance: fe_instance,
             servers,
             heartbeat_task: None,
-            export_metrics_task,
         };
 
         #[cfg(feature = "enterprise")]

src/cmd/tests/load_config_test.rs

@@ -31,7 +31,6 @@ use meta_srv::selector::SelectorType;
 use metric_engine::config::EngineConfig as MetricEngineConfig;
 use mito2::config::MitoConfig;
 use query::options::QueryOptions;
-use servers::export_metrics::ExportMetricsOption;
 use servers::grpc::GrpcOptions;
 use servers::http::HttpOptions;
 use servers::tls::{TlsMode, TlsOption};
@@ -95,11 +94,6 @@ fn test_load_datanode_example_config() {
                 tracing_sample_ratio: Some(Default::default()),
                 ..Default::default()
             },
-            export_metrics: ExportMetricsOption {
-                self_import: None,
-                remote_write: Some(Default::default()),
-                ..Default::default()
-            },
             grpc: GrpcOptions::default()
                 .with_bind_addr("127.0.0.1:3001")
                 .with_server_addr("127.0.0.1:3001"),
@@ -146,11 +140,6 @@ fn test_load_frontend_example_config() {
                     ..Default::default()
                 },
             },
-            export_metrics: ExportMetricsOption {
-                self_import: None,
-                remote_write: Some(Default::default()),
-                ..Default::default()
-            },
             grpc: GrpcOptions {
                 bind_addr: "127.0.0.1:4001".to_string(),
                 server_addr: "127.0.0.1:4001".to_string(),
@@ -201,11 +190,6 @@ fn test_load_metasrv_example_config() {
                     tcp_nodelay: true,
                 },
             },
-            export_metrics: ExportMetricsOption {
-                self_import: None,
-                remote_write: Some(Default::default()),
-                ..Default::default()
-            },
             backend_tls: Some(TlsOption {
                 mode: TlsMode::Prefer,
                 cert_path: String::new(),
@@ -317,11 +301,6 @@ fn test_load_standalone_example_config() {
                 tracing_sample_ratio: Some(Default::default()),
                 ..Default::default()
             },
-            export_metrics: ExportMetricsOption {
-                self_import: Some(Default::default()),
-                remote_write: Some(Default::default()),
-                ..Default::default()
-            },
             http: HttpOptions {
                 cors_allowed_origins: vec!["https://example.com".to_string()],
                 ..Default::default()

src/common/catalog/src/consts.rs

@@ -86,8 +86,6 @@ pub const INFORMATION_SCHEMA_TRIGGERS_TABLE_ID: u32 = 24;
 pub const INFORMATION_SCHEMA_GLOBAL_STATUS_TABLE_ID: u32 = 25;
 /// id for information_schema.SESSION_STATUS
 pub const INFORMATION_SCHEMA_SESSION_STATUS_TABLE_ID: u32 = 26;
-/// id for information_schema.RUNTIME_METRICS
-pub const INFORMATION_SCHEMA_RUNTIME_METRICS_TABLE_ID: u32 = 27;
 /// id for information_schema.PARTITIONS
 pub const INFORMATION_SCHEMA_PARTITIONS_TABLE_ID: u32 = 28;
 /// id for information_schema.REGION_PEERS
@@ -112,6 +110,8 @@ pub const INFORMATION_SCHEMA_SSTS_MANIFEST_TABLE_ID: u32 = 37;
 pub const INFORMATION_SCHEMA_SSTS_STORAGE_TABLE_ID: u32 = 38;
 /// id for information_schema.ssts_index_meta
 pub const INFORMATION_SCHEMA_SSTS_INDEX_META_TABLE_ID: u32 = 39;
+/// id for information_schema.alerts
+pub const INFORMATION_SCHEMA_ALERTS_TABLE_ID: u32 = 40;
 
 // ----- End of information_schema tables -----
 

src/common/datasource/src/buffered_writer.rs

@@ -12,28 +12,11 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
-use std::future::Future;
-
 use arrow::record_batch::RecordBatch;
 use async_trait::async_trait;
 use datafusion::parquet::format::FileMetaData;
-use snafu::{OptionExt, ResultExt};
-use tokio::io::{AsyncWrite, AsyncWriteExt};
 
-use crate::error::{self, Result};
-use crate::share_buffer::SharedBuffer;
-
-pub struct LazyBufferedWriter<T, U, F> {
-    path: String,
-    writer_factory: F,
-    writer: Option<T>,
-    /// None stands for [`LazyBufferedWriter`] closed.
-    encoder: Option<U>,
-    buffer: SharedBuffer,
-    rows_written: usize,
-    bytes_written: u64,
-    threshold: usize,
-}
+use crate::error::Result;
 
 pub trait DfRecordBatchEncoder {
     fn write(&mut self, batch: &RecordBatch) -> Result<()>;
@@ -43,126 +26,3 @@ pub trait DfRecordBatchEncoder {
 pub trait ArrowWriterCloser {
     async fn close(mut self) -> Result<FileMetaData>;
 }
-
-impl<
-    T: AsyncWrite + Send + Unpin,
-    U: DfRecordBatchEncoder + ArrowWriterCloser,
-    F: Fn(String) -> Fut,
-    Fut: Future<Output = Result<T>>,
-> LazyBufferedWriter<T, U, F>
-{
-    /// Closes `LazyBufferedWriter` and optionally flushes all data to underlying storage
-    /// if any row's been written.
-    pub async fn close_with_arrow_writer(mut self) -> Result<(FileMetaData, u64)> {
-        let encoder = self
-            .encoder
-            .take()
-            .context(error::BufferedWriterClosedSnafu)?;
-        let metadata = encoder.close().await?;
-
-        // It's important to shut down! flushes all pending writes
-        self.close_inner_writer().await?;
-        Ok((metadata, self.bytes_written))
-    }
-}
-
-impl<
-    T: AsyncWrite + Send + Unpin,
-    U: DfRecordBatchEncoder,
-    F: Fn(String) -> Fut,
-    Fut: Future<Output = Result<T>>,
-> LazyBufferedWriter<T, U, F>
-{
-    /// Closes the writer and flushes the buffer data.
-    pub async fn close_inner_writer(&mut self) -> Result<()> {
-        // Use `rows_written` to keep a track of if any rows have been written.
-        // If no row's been written, then we can simply close the underlying
-        // writer without flush so that no file will be actually created.
-        if self.rows_written != 0 {
-            self.bytes_written += self.try_flush(true).await?;
-        }
-
-        if let Some(writer) = &mut self.writer {
-            writer.shutdown().await.context(error::AsyncWriteSnafu)?;
-        }
-        Ok(())
-    }
-
-    pub fn new(
-        threshold: usize,
-        buffer: SharedBuffer,
-        encoder: U,
-        path: impl AsRef<str>,
-        writer_factory: F,
-    ) -> Self {
-        Self {
-            path: path.as_ref().to_string(),
-            threshold,
-            encoder: Some(encoder),
-            buffer,
-            rows_written: 0,
-            bytes_written: 0,
-            writer_factory,
-            writer: None,
-        }
-    }
-
-    pub async fn write(&mut self, batch: &RecordBatch) -> Result<()> {
-        let encoder = self
-            .encoder
-            .as_mut()
-            .context(error::BufferedWriterClosedSnafu)?;
-        encoder.write(batch)?;
-        self.rows_written += batch.num_rows();
-        self.bytes_written += self.try_flush(false).await?;
-        Ok(())
-    }
-
-    async fn try_flush(&mut self, all: bool) -> Result<u64> {
-        let mut bytes_written: u64 = 0;
-
-        // Once buffered data size reaches threshold, split the data in chunks (typically 4MB)
-        // and write to underlying storage.
-        while self.buffer.buffer.lock().unwrap().len() >= self.threshold {
-            let chunk = {
-                let mut buffer = self.buffer.buffer.lock().unwrap();
-                buffer.split_to(self.threshold)
-            };
-            let size = chunk.len();
-
-            self.maybe_init_writer()
-                .await?
-                .write_all(&chunk)
-                .await
-                .context(error::AsyncWriteSnafu)?;
-
-            bytes_written += size as u64;
-        }
-
-        if all {
-            bytes_written += self.try_flush_all().await?;
-        }
-        Ok(bytes_written)
-    }
-
-    /// Only initiates underlying file writer when rows have been written.
-    async fn maybe_init_writer(&mut self) -> Result<&mut T> {
-        if let Some(ref mut writer) = self.writer {
-            Ok(writer)
-        } else {
-            let writer = (self.writer_factory)(self.path.clone()).await?;
-            Ok(self.writer.insert(writer))
-        }
-    }
-
-    async fn try_flush_all(&mut self) -> Result<u64> {
-        let remain = self.buffer.buffer.lock().unwrap().split();
-        let size = remain.len();
-        self.maybe_init_writer()
-            .await?
-            .write_all(&remain)
-            .await
-            .context(error::AsyncWriteSnafu)?;
-        Ok(size as u64)
-    }
-}

src/common/datasource/src/compressed_writer.rs

@@ -0,0 +1,202 @@
+// 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::io;
+use std::pin::Pin;
+use std::task::{Context, Poll};
+
+use async_compression::tokio::write::{BzEncoder, GzipEncoder, XzEncoder, ZstdEncoder};
+use snafu::ResultExt;
+use tokio::io::{AsyncWrite, AsyncWriteExt};
+
+use crate::compression::CompressionType;
+use crate::error::{self, Result};
+
+/// A compressed writer that wraps an underlying async writer with compression.
+///
+/// This writer supports multiple compression formats including GZIP, BZIP2, XZ, and ZSTD.
+/// It provides transparent compression for any async writer implementation.
+pub struct CompressedWriter {
+    inner: Box<dyn AsyncWrite + Unpin + Send>,
+    compression_type: CompressionType,
+}
+
+impl CompressedWriter {
+    /// Creates a new compressed writer with the specified compression type.
+    ///
+    /// # Arguments
+    ///
+    /// * `writer` - The underlying writer to wrap with compression
+    /// * `compression_type` - The type of compression to apply
+    pub fn new(
+        writer: impl AsyncWrite + Unpin + Send + 'static,
+        compression_type: CompressionType,
+    ) -> Self {
+        let inner: Box<dyn AsyncWrite + Unpin + Send> = match compression_type {
+            CompressionType::Gzip => Box::new(GzipEncoder::new(writer)),
+            CompressionType::Bzip2 => Box::new(BzEncoder::new(writer)),
+            CompressionType::Xz => Box::new(XzEncoder::new(writer)),
+            CompressionType::Zstd => Box::new(ZstdEncoder::new(writer)),
+            CompressionType::Uncompressed => Box::new(writer),
+        };
+
+        Self {
+            inner,
+            compression_type,
+        }
+    }
+
+    /// Returns the compression type used by this writer.
+    pub fn compression_type(&self) -> CompressionType {
+        self.compression_type
+    }
+
+    /// Flush the writer and shutdown compression
+    pub async fn shutdown(mut self) -> Result<()> {
+        self.inner
+            .shutdown()
+            .await
+            .context(error::AsyncWriteSnafu)?;
+        Ok(())
+    }
+}
+
+impl AsyncWrite for CompressedWriter {
+    fn poll_write(
+        mut self: Pin<&mut Self>,
+        cx: &mut Context<'_>,
+        buf: &[u8],
+    ) -> Poll<io::Result<usize>> {
+        Pin::new(&mut self.inner).poll_write(cx, buf)
+    }
+
+    fn poll_flush(mut self: Pin<&mut Self>, cx: &mut Context<'_>) -> Poll<io::Result<()>> {
+        Pin::new(&mut self.inner).poll_flush(cx)
+    }
+
+    fn poll_shutdown(mut self: Pin<&mut Self>, cx: &mut Context<'_>) -> Poll<io::Result<()>> {
+        Pin::new(&mut self.inner).poll_shutdown(cx)
+    }
+}
+
+/// A trait for converting async writers into compressed writers.
+///
+/// This trait is automatically implemented for all types that implement [`AsyncWrite`].
+pub trait IntoCompressedWriter {
+    /// Converts this writer into a [`CompressedWriter`] with the specified compression type.
+    ///
+    /// # Arguments
+    ///
+    /// * `self` - The underlying writer to wrap with compression
+    /// * `compression_type` - The type of compression to apply
+    fn into_compressed_writer(self, compression_type: CompressionType) -> CompressedWriter
+    where
+        Self: AsyncWrite + Unpin + Send + 'static + Sized,
+    {
+        CompressedWriter::new(self, compression_type)
+    }
+}
+
+impl<W: AsyncWrite + Unpin + Send + 'static> IntoCompressedWriter for W {}
+
+#[cfg(test)]
+mod tests {
+    use tokio::io::{AsyncReadExt, AsyncWriteExt, duplex};
+
+    use super::*;
+
+    #[tokio::test]
+    async fn test_compressed_writer_gzip() {
+        let (duplex_writer, mut duplex_reader) = duplex(1024);
+        let mut writer = duplex_writer.into_compressed_writer(CompressionType::Gzip);
+        let original = b"test data for gzip compression";
+
+        writer.write_all(original).await.unwrap();
+        writer.shutdown().await.unwrap();
+
+        let mut buffer = Vec::new();
+        duplex_reader.read_to_end(&mut buffer).await.unwrap();
+
+        // The compressed data should be different from the original
+        assert_ne!(buffer, original);
+        assert!(!buffer.is_empty());
+    }
+
+    #[tokio::test]
+    async fn test_compressed_writer_bzip2() {
+        let (duplex_writer, mut duplex_reader) = duplex(1024);
+        let mut writer = duplex_writer.into_compressed_writer(CompressionType::Bzip2);
+        let original = b"test data for bzip2 compression";
+
+        writer.write_all(original).await.unwrap();
+        writer.shutdown().await.unwrap();
+
+        let mut buffer = Vec::new();
+        duplex_reader.read_to_end(&mut buffer).await.unwrap();
+
+        // The compressed data should be different from the original
+        assert_ne!(buffer, original);
+        assert!(!buffer.is_empty());
+    }
+
+    #[tokio::test]
+    async fn test_compressed_writer_xz() {
+        let (duplex_writer, mut duplex_reader) = duplex(1024);
+        let mut writer = duplex_writer.into_compressed_writer(CompressionType::Xz);
+        let original = b"test data for xz compression";
+
+        writer.write_all(original).await.unwrap();
+        writer.shutdown().await.unwrap();
+
+        let mut buffer = Vec::new();
+        duplex_reader.read_to_end(&mut buffer).await.unwrap();
+
+        // The compressed data should be different from the original
+        assert_ne!(buffer, original);
+        assert!(!buffer.is_empty());
+    }
+
+    #[tokio::test]
+    async fn test_compressed_writer_zstd() {
+        let (duplex_writer, mut duplex_reader) = duplex(1024);
+        let mut writer = duplex_writer.into_compressed_writer(CompressionType::Zstd);
+        let original = b"test data for zstd compression";
+
+        writer.write_all(original).await.unwrap();
+        writer.shutdown().await.unwrap();
+
+        let mut buffer = Vec::new();
+        duplex_reader.read_to_end(&mut buffer).await.unwrap();
+
+        // The compressed data should be different from the original
+        assert_ne!(buffer, original);
+        assert!(!buffer.is_empty());
+    }
+
+    #[tokio::test]
+    async fn test_compressed_writer_uncompressed() {
+        let (duplex_writer, mut duplex_reader) = duplex(1024);
+        let mut writer = duplex_writer.into_compressed_writer(CompressionType::Uncompressed);
+        let original = b"test data for uncompressed";
+
+        writer.write_all(original).await.unwrap();
+        writer.shutdown().await.unwrap();
+
+        let mut buffer = Vec::new();
+        duplex_reader.read_to_end(&mut buffer).await.unwrap();
+
+        // Uncompressed data should be the same as the original
+        assert_eq!(buffer, original);
+    }
+}

src/common/datasource/src/error.rs

@@ -194,12 +194,6 @@ pub enum Error {
         location: Location,
     },
 
-    #[snafu(display("Buffered writer closed"))]
-    BufferedWriterClosed {
-        #[snafu(implicit)]
-        location: Location,
-    },
-
     #[snafu(display("Failed to write parquet file, path: {}", path))]
     WriteParquet {
         path: String,
@@ -208,6 +202,14 @@ pub enum Error {
         #[snafu(source)]
         error: parquet::errors::ParquetError,
     },
+
+    #[snafu(display("Failed to build file stream"))]
+    BuildFileStream {
+        #[snafu(implicit)]
+        location: Location,
+        #[snafu(source)]
+        error: datafusion::error::DataFusionError,
+    },
 }
 
 pub type Result<T> = std::result::Result<T, Error>;
@@ -239,7 +241,7 @@ impl ErrorExt for Error {
             | ReadRecordBatch { .. }
             | WriteRecordBatch { .. }
             | EncodeRecordBatch { .. }
-            | BufferedWriterClosed { .. }
+            | BuildFileStream { .. }
             | OrcReader { .. } => StatusCode::Unexpected,
         }
     }

src/common/datasource/src/file_format.rs

@@ -30,12 +30,22 @@ use arrow::record_batch::RecordBatch;
 use arrow_schema::{ArrowError, Schema as ArrowSchema};
 use async_trait::async_trait;
 use bytes::{Buf, Bytes};
-use datafusion::datasource::physical_plan::FileOpenFuture;
+use common_recordbatch::DfSendableRecordBatchStream;
+use datafusion::datasource::file_format::file_compression_type::FileCompressionType as DfCompressionType;
+use datafusion::datasource::listing::PartitionedFile;
+use datafusion::datasource::object_store::ObjectStoreUrl;
+use datafusion::datasource::physical_plan::{
+    FileGroup, FileOpenFuture, FileScanConfigBuilder, FileSource, FileStream,
+};
 use datafusion::error::{DataFusionError, Result as DataFusionResult};
 use datafusion::physical_plan::SendableRecordBatchStream;
+use datafusion::physical_plan::metrics::ExecutionPlanMetricsSet;
+use datatypes::arrow::datatypes::SchemaRef;
 use futures::{StreamExt, TryStreamExt};
 use object_store::ObjectStore;
+use object_store_opendal::OpendalStore;
 use snafu::ResultExt;
+use tokio::io::AsyncWriteExt;
 use tokio_util::compat::FuturesAsyncWriteCompatExt;
 
 use self::csv::CsvFormat;
@@ -43,7 +53,8 @@ use self::json::JsonFormat;
 use self::orc::OrcFormat;
 use self::parquet::ParquetFormat;
 use crate::DEFAULT_WRITE_BUFFER_SIZE;
-use crate::buffered_writer::{DfRecordBatchEncoder, LazyBufferedWriter};
+use crate::buffered_writer::DfRecordBatchEncoder;
+use crate::compressed_writer::{CompressedWriter, IntoCompressedWriter};
 use crate::compression::CompressionType;
 use crate::error::{self, Result};
 use crate::share_buffer::SharedBuffer;
@@ -195,33 +206,128 @@ pub async fn infer_schemas(
     ArrowSchema::try_merge(schemas).context(error::MergeSchemaSnafu)
 }
 
-pub async fn stream_to_file<T: DfRecordBatchEncoder, U: Fn(SharedBuffer) -> T>(
+/// Writes data to a compressed writer if the data is not empty.
+///
+/// Does nothing if `data` is empty; otherwise writes all data and returns any error.
+async fn write_to_compressed_writer(
+    compressed_writer: &mut CompressedWriter,
+    data: &[u8],
+) -> Result<()> {
+    if !data.is_empty() {
+        compressed_writer
+            .write_all(data)
+            .await
+            .context(error::AsyncWriteSnafu)?;
+    }
+    Ok(())
+}
+
+/// Streams [SendableRecordBatchStream] to a file with optional compression support.
+/// Data is buffered and flushed according to the given `threshold`.
+/// Ensures that writer resources are cleanly released and that an empty file is not
+/// created if no rows are written.
+///
+/// Returns the total number of rows successfully written.
+pub async fn stream_to_file<E>(
     mut stream: SendableRecordBatchStream,
     store: ObjectStore,
     path: &str,
     threshold: usize,
     concurrency: usize,
-    encoder_factory: U,
-) -> Result<usize> {
+    compression_type: CompressionType,
+    encoder_factory: impl Fn(SharedBuffer) -> E,
+) -> Result<usize>
+where
+    E: DfRecordBatchEncoder,
+{
+    // Create the file writer with OpenDAL's built-in buffering
+    let writer = store
+        .writer_with(path)
+        .concurrent(concurrency)
+        .chunk(DEFAULT_WRITE_BUFFER_SIZE.as_bytes() as usize)
+        .await
+        .with_context(|_| error::WriteObjectSnafu { path })?
+        .into_futures_async_write()
+        .compat_write();
+
+    // Apply compression if needed
+    let mut compressed_writer = writer.into_compressed_writer(compression_type);
+
+    // Create a buffer for the encoder
     let buffer = SharedBuffer::with_capacity(threshold);
-    let encoder = encoder_factory(buffer.clone());
-    let mut writer = LazyBufferedWriter::new(threshold, buffer, encoder, path, |path| async {
-        store
-            .writer_with(&path)
-            .concurrent(concurrency)
-            .chunk(DEFAULT_WRITE_BUFFER_SIZE.as_bytes() as usize)
-            .await
-            .map(|v| v.into_futures_async_write().compat_write())
-            .context(error::WriteObjectSnafu { path })
-    });
+    let mut encoder = encoder_factory(buffer.clone());
 
     let mut rows = 0;
 
+    // Process each record batch
     while let Some(batch) = stream.next().await {
         let batch = batch.context(error::ReadRecordBatchSnafu)?;
-        writer.write(&batch).await?;
+
+        // Write batch using the encoder
+        encoder.write(&batch)?;
         rows += batch.num_rows();
+
+        loop {
+            let chunk = {
+                let mut buffer_guard = buffer.buffer.lock().unwrap();
+                if buffer_guard.len() < threshold {
+                    break;
+                }
+                buffer_guard.split_to(threshold)
+            };
+            write_to_compressed_writer(&mut compressed_writer, &chunk).await?;
+        }
     }
-    writer.close_inner_writer().await?;
+
+    // If no row's been written, just simply close the underlying writer
+    // without flush so that no file will be actually created.
+    if rows != 0 {
+        // Final flush of any remaining data
+        let final_data = {
+            let mut buffer_guard = buffer.buffer.lock().unwrap();
+            buffer_guard.split()
+        };
+        write_to_compressed_writer(&mut compressed_writer, &final_data).await?;
+    }
+
+    // Shutdown compression and close writer
+    compressed_writer.shutdown().await?;
+
     Ok(rows)
 }
+
+/// Creates a [FileStream] for reading data from a file with optional column projection
+/// and compression support.
+///
+/// Returns [SendableRecordBatchStream].
+pub async fn file_to_stream(
+    store: &ObjectStore,
+    filename: &str,
+    file_schema: SchemaRef,
+    file_source: Arc<dyn FileSource>,
+    projection: Option<Vec<usize>>,
+    compression_type: CompressionType,
+) -> Result<DfSendableRecordBatchStream> {
+    let df_compression: DfCompressionType = compression_type.into();
+    let config = FileScanConfigBuilder::new(
+        ObjectStoreUrl::local_filesystem(),
+        file_schema,
+        file_source.clone(),
+    )
+    .with_file_group(FileGroup::new(vec![PartitionedFile::new(
+        filename.to_string(),
+        0,
+    )]))
+    .with_projection(projection)
+    .with_file_compression_type(df_compression)
+    .build();
+
+    let store = Arc::new(OpendalStore::new(store.clone()));
+    let file_opener = file_source
+        .with_projection(&config)
+        .create_file_opener(store, &config, 0);
+    let stream = FileStream::new(&config, 0, file_opener, &ExecutionPlanMetricsSet::new())
+        .context(error::BuildFileStreamSnafu)?;
+
+    Ok(Box::pin(stream))
+}

src/common/datasource/src/file_format/csv.rs

@@ -157,19 +157,27 @@ pub async fn stream_to_csv(
     concurrency: usize,
     format: &CsvFormat,
 ) -> Result<usize> {
-    stream_to_file(stream, store, path, threshold, concurrency, |buffer| {
-        let mut builder = WriterBuilder::new();
-        if let Some(timestamp_format) = &format.timestamp_format {
-            builder = builder.with_timestamp_format(timestamp_format.to_owned())
-        }
-        if let Some(date_format) = &format.date_format {
-            builder = builder.with_date_format(date_format.to_owned())
-        }
-        if let Some(time_format) = &format.time_format {
-            builder = builder.with_time_format(time_format.to_owned())
-        }
-        builder.build(buffer)
-    })
+    stream_to_file(
+        stream,
+        store,
+        path,
+        threshold,
+        concurrency,
+        format.compression_type,
+        |buffer| {
+            let mut builder = WriterBuilder::new();
+            if let Some(timestamp_format) = &format.timestamp_format {
+                builder = builder.with_timestamp_format(timestamp_format.to_owned())
+            }
+            if let Some(date_format) = &format.date_format {
+                builder = builder.with_date_format(date_format.to_owned())
+            }
+            if let Some(time_format) = &format.time_format {
+                builder = builder.with_time_format(time_format.to_owned())
+            }
+            builder.build(buffer)
+        },
+    )
     .await
 }
 
@@ -181,13 +189,21 @@ impl DfRecordBatchEncoder for csv::Writer<SharedBuffer> {
 
 #[cfg(test)]
 mod tests {
+    use std::sync::Arc;
 
+    use common_recordbatch::adapter::DfRecordBatchStreamAdapter;
+    use common_recordbatch::{RecordBatch, RecordBatches};
     use common_test_util::find_workspace_path;
+    use datafusion::datasource::physical_plan::{CsvSource, FileSource};
+    use datatypes::prelude::ConcreteDataType;
+    use datatypes::schema::{ColumnSchema, Schema};
+    use datatypes::vectors::{Float64Vector, StringVector, UInt32Vector, VectorRef};
+    use futures::TryStreamExt;
 
     use super::*;
     use crate::file_format::{
         FORMAT_COMPRESSION_TYPE, FORMAT_DELIMITER, FORMAT_HAS_HEADER,
-        FORMAT_SCHEMA_INFER_MAX_RECORD, FileFormat,
+        FORMAT_SCHEMA_INFER_MAX_RECORD, FileFormat, file_to_stream,
     };
     use crate::test_util::{format_schema, test_store};
 
@@ -297,4 +313,166 @@ mod tests {
             }
         );
     }
+
+    #[tokio::test]
+    async fn test_compressed_csv() {
+        // Create test data
+        let column_schemas = vec![
+            ColumnSchema::new("id", ConcreteDataType::uint32_datatype(), false),
+            ColumnSchema::new("name", ConcreteDataType::string_datatype(), false),
+            ColumnSchema::new("value", ConcreteDataType::float64_datatype(), false),
+        ];
+        let schema = Arc::new(Schema::new(column_schemas));
+
+        // Create multiple record batches with different data
+        let batch1_columns: Vec<VectorRef> = vec![
+            Arc::new(UInt32Vector::from_slice(vec![1, 2, 3])),
+            Arc::new(StringVector::from(vec!["Alice", "Bob", "Charlie"])),
+            Arc::new(Float64Vector::from_slice(vec![10.5, 20.3, 30.7])),
+        ];
+        let batch1 = RecordBatch::new(schema.clone(), batch1_columns).unwrap();
+
+        let batch2_columns: Vec<VectorRef> = vec![
+            Arc::new(UInt32Vector::from_slice(vec![4, 5, 6])),
+            Arc::new(StringVector::from(vec!["David", "Eva", "Frank"])),
+            Arc::new(Float64Vector::from_slice(vec![40.1, 50.2, 60.3])),
+        ];
+        let batch2 = RecordBatch::new(schema.clone(), batch2_columns).unwrap();
+
+        let batch3_columns: Vec<VectorRef> = vec![
+            Arc::new(UInt32Vector::from_slice(vec![7, 8, 9])),
+            Arc::new(StringVector::from(vec!["Grace", "Henry", "Ivy"])),
+            Arc::new(Float64Vector::from_slice(vec![70.4, 80.5, 90.6])),
+        ];
+        let batch3 = RecordBatch::new(schema.clone(), batch3_columns).unwrap();
+
+        // Combine all batches into a RecordBatches collection
+        let recordbatches = RecordBatches::try_new(schema, vec![batch1, batch2, batch3]).unwrap();
+
+        // Test with different compression types
+        let compression_types = vec![
+            CompressionType::Gzip,
+            CompressionType::Bzip2,
+            CompressionType::Xz,
+            CompressionType::Zstd,
+        ];
+
+        // Create a temporary file path
+        let temp_dir = common_test_util::temp_dir::create_temp_dir("test_compressed_csv");
+        for compression_type in compression_types {
+            let format = CsvFormat {
+                compression_type,
+                ..CsvFormat::default()
+            };
+
+            // Use correct format without Debug formatter
+            let compressed_file_name =
+                format!("test_compressed_csv.{}", compression_type.file_extension());
+            let compressed_file_path = temp_dir.path().join(&compressed_file_name);
+            let compressed_file_path_str = compressed_file_path.to_str().unwrap();
+
+            // Create a simple file store for testing
+            let store = test_store("/");
+
+            // Export CSV with compression
+            let rows = stream_to_csv(
+                Box::pin(DfRecordBatchStreamAdapter::new(recordbatches.as_stream())),
+                store,
+                compressed_file_path_str,
+                1024,
+                1,
+                &format,
+            )
+            .await
+            .unwrap();
+
+            assert_eq!(rows, 9);
+
+            // Verify compressed file was created and has content
+            assert!(compressed_file_path.exists());
+            let file_size = std::fs::metadata(&compressed_file_path).unwrap().len();
+            assert!(file_size > 0);
+
+            // Verify the file is actually compressed
+            let file_content = std::fs::read(&compressed_file_path).unwrap();
+            // Compressed files should not start with CSV header
+            // They should have compression magic bytes
+            match compression_type {
+                CompressionType::Gzip => {
+                    // Gzip magic bytes: 0x1f 0x8b
+                    assert_eq!(file_content[0], 0x1f, "Gzip file should start with 0x1f");
+                    assert_eq!(
+                        file_content[1], 0x8b,
+                        "Gzip file should have 0x8b as second byte"
+                    );
+                }
+                CompressionType::Bzip2 => {
+                    // Bzip2 magic bytes: 'BZ'
+                    assert_eq!(file_content[0], b'B', "Bzip2 file should start with 'B'");
+                    assert_eq!(
+                        file_content[1], b'Z',
+                        "Bzip2 file should have 'Z' as second byte"
+                    );
+                }
+                CompressionType::Xz => {
+                    // XZ magic bytes: 0xFD '7zXZ'
+                    assert_eq!(file_content[0], 0xFD, "XZ file should start with 0xFD");
+                }
+                CompressionType::Zstd => {
+                    // Zstd magic bytes: 0x28 0xB5 0x2F 0xFD
+                    assert_eq!(file_content[0], 0x28, "Zstd file should start with 0x28");
+                    assert_eq!(
+                        file_content[1], 0xB5,
+                        "Zstd file should have 0xB5 as second byte"
+                    );
+                }
+                _ => {}
+            }
+
+            // Verify the compressed file can be decompressed and content matches original data
+            let store = test_store("/");
+            let schema = Arc::new(
+                CsvFormat {
+                    compression_type,
+                    ..Default::default()
+                }
+                .infer_schema(&store, compressed_file_path_str)
+                .await
+                .unwrap(),
+            );
+            let csv_source = CsvSource::new(true, b',', b'"')
+                .with_schema(schema.clone())
+                .with_batch_size(8192);
+
+            let stream = file_to_stream(
+                &store,
+                compressed_file_path_str,
+                schema.clone(),
+                csv_source.clone(),
+                None,
+                compression_type,
+            )
+            .await
+            .unwrap();
+
+            let batches = stream.try_collect::<Vec<_>>().await.unwrap();
+            let pretty_print = arrow::util::pretty::pretty_format_batches(&batches)
+                .unwrap()
+                .to_string();
+            let expected = r#"+----+---------+-------+
+| id | name    | value |
++----+---------+-------+
+| 1  | Alice   | 10.5  |
+| 2  | Bob     | 20.3  |
+| 3  | Charlie | 30.7  |
+| 4  | David   | 40.1  |
+| 5  | Eva     | 50.2  |
+| 6  | Frank   | 60.3  |
+| 7  | Grace   | 70.4  |
+| 8  | Henry   | 80.5  |
+| 9  | Ivy     | 90.6  |
++----+---------+-------+"#;
+            assert_eq!(expected, pretty_print);
+        }
+    }
 }

src/common/datasource/src/file_format/json.rs

@@ -115,10 +115,17 @@ pub async fn stream_to_json(
     path: &str,
     threshold: usize,
     concurrency: usize,
+    format: &JsonFormat,
 ) -> Result<usize> {
-    stream_to_file(stream, store, path, threshold, concurrency, |buffer| {
-        json::LineDelimitedWriter::new(buffer)
-    })
+    stream_to_file(
+        stream,
+        store,
+        path,
+        threshold,
+        concurrency,
+        format.compression_type,
+        json::LineDelimitedWriter::new,
+    )
     .await
 }
 
@@ -130,10 +137,21 @@ impl DfRecordBatchEncoder for json::Writer<SharedBuffer, LineDelimited> {
 
 #[cfg(test)]
 mod tests {
+    use std::sync::Arc;
+
+    use common_recordbatch::adapter::DfRecordBatchStreamAdapter;
+    use common_recordbatch::{RecordBatch, RecordBatches};
     use common_test_util::find_workspace_path;
+    use datafusion::datasource::physical_plan::{FileSource, JsonSource};
+    use datatypes::prelude::ConcreteDataType;
+    use datatypes::schema::{ColumnSchema, Schema};
+    use datatypes::vectors::{Float64Vector, StringVector, UInt32Vector, VectorRef};
+    use futures::TryStreamExt;
 
     use super::*;
-    use crate::file_format::{FORMAT_COMPRESSION_TYPE, FORMAT_SCHEMA_INFER_MAX_RECORD, FileFormat};
+    use crate::file_format::{
+        FORMAT_COMPRESSION_TYPE, FORMAT_SCHEMA_INFER_MAX_RECORD, FileFormat, file_to_stream,
+    };
     use crate::test_util::{format_schema, test_store};
 
     fn test_data_root() -> String {
@@ -203,4 +221,165 @@ mod tests {
             }
         );
     }
+
+    #[tokio::test]
+    async fn test_compressed_json() {
+        // Create test data
+        let column_schemas = vec![
+            ColumnSchema::new("id", ConcreteDataType::uint32_datatype(), false),
+            ColumnSchema::new("name", ConcreteDataType::string_datatype(), false),
+            ColumnSchema::new("value", ConcreteDataType::float64_datatype(), false),
+        ];
+        let schema = Arc::new(Schema::new(column_schemas));
+
+        // Create multiple record batches with different data
+        let batch1_columns: Vec<VectorRef> = vec![
+            Arc::new(UInt32Vector::from_slice(vec![1, 2, 3])),
+            Arc::new(StringVector::from(vec!["Alice", "Bob", "Charlie"])),
+            Arc::new(Float64Vector::from_slice(vec![10.5, 20.3, 30.7])),
+        ];
+        let batch1 = RecordBatch::new(schema.clone(), batch1_columns).unwrap();
+
+        let batch2_columns: Vec<VectorRef> = vec![
+            Arc::new(UInt32Vector::from_slice(vec![4, 5, 6])),
+            Arc::new(StringVector::from(vec!["David", "Eva", "Frank"])),
+            Arc::new(Float64Vector::from_slice(vec![40.1, 50.2, 60.3])),
+        ];
+        let batch2 = RecordBatch::new(schema.clone(), batch2_columns).unwrap();
+
+        let batch3_columns: Vec<VectorRef> = vec![
+            Arc::new(UInt32Vector::from_slice(vec![7, 8, 9])),
+            Arc::new(StringVector::from(vec!["Grace", "Henry", "Ivy"])),
+            Arc::new(Float64Vector::from_slice(vec![70.4, 80.5, 90.6])),
+        ];
+        let batch3 = RecordBatch::new(schema.clone(), batch3_columns).unwrap();
+
+        // Combine all batches into a RecordBatches collection
+        let recordbatches = RecordBatches::try_new(schema, vec![batch1, batch2, batch3]).unwrap();
+
+        // Test with different compression types
+        let compression_types = vec![
+            CompressionType::Gzip,
+            CompressionType::Bzip2,
+            CompressionType::Xz,
+            CompressionType::Zstd,
+        ];
+
+        // Create a temporary file path
+        let temp_dir = common_test_util::temp_dir::create_temp_dir("test_compressed_json");
+        for compression_type in compression_types {
+            let format = JsonFormat {
+                compression_type,
+                ..JsonFormat::default()
+            };
+
+            let compressed_file_name =
+                format!("test_compressed_json.{}", compression_type.file_extension());
+            let compressed_file_path = temp_dir.path().join(&compressed_file_name);
+            let compressed_file_path_str = compressed_file_path.to_str().unwrap();
+
+            // Create a simple file store for testing
+            let store = test_store("/");
+
+            // Export JSON with compression
+            let rows = stream_to_json(
+                Box::pin(DfRecordBatchStreamAdapter::new(recordbatches.as_stream())),
+                store,
+                compressed_file_path_str,
+                1024,
+                1,
+                &format,
+            )
+            .await
+            .unwrap();
+
+            assert_eq!(rows, 9);
+
+            // Verify compressed file was created and has content
+            assert!(compressed_file_path.exists());
+            let file_size = std::fs::metadata(&compressed_file_path).unwrap().len();
+            assert!(file_size > 0);
+
+            // Verify the file is actually compressed
+            let file_content = std::fs::read(&compressed_file_path).unwrap();
+            // Compressed files should not start with '{' (JSON character)
+            // They should have compression magic bytes
+            match compression_type {
+                CompressionType::Gzip => {
+                    // Gzip magic bytes: 0x1f 0x8b
+                    assert_eq!(file_content[0], 0x1f, "Gzip file should start with 0x1f");
+                    assert_eq!(
+                        file_content[1], 0x8b,
+                        "Gzip file should have 0x8b as second byte"
+                    );
+                }
+                CompressionType::Bzip2 => {
+                    // Bzip2 magic bytes: 'BZ'
+                    assert_eq!(file_content[0], b'B', "Bzip2 file should start with 'B'");
+                    assert_eq!(
+                        file_content[1], b'Z',
+                        "Bzip2 file should have 'Z' as second byte"
+                    );
+                }
+                CompressionType::Xz => {
+                    // XZ magic bytes: 0xFD '7zXZ'
+                    assert_eq!(file_content[0], 0xFD, "XZ file should start with 0xFD");
+                }
+                CompressionType::Zstd => {
+                    // Zstd magic bytes: 0x28 0xB5 0x2F 0xFD
+                    assert_eq!(file_content[0], 0x28, "Zstd file should start with 0x28");
+                    assert_eq!(
+                        file_content[1], 0xB5,
+                        "Zstd file should have 0xB5 as second byte"
+                    );
+                }
+                _ => {}
+            }
+
+            // Verify the compressed file can be decompressed and content matches original data
+            let store = test_store("/");
+            let schema = Arc::new(
+                JsonFormat {
+                    compression_type,
+                    ..Default::default()
+                }
+                .infer_schema(&store, compressed_file_path_str)
+                .await
+                .unwrap(),
+            );
+            let json_source = JsonSource::new()
+                .with_schema(schema.clone())
+                .with_batch_size(8192);
+
+            let stream = file_to_stream(
+                &store,
+                compressed_file_path_str,
+                schema.clone(),
+                json_source.clone(),
+                None,
+                compression_type,
+            )
+            .await
+            .unwrap();
+
+            let batches = stream.try_collect::<Vec<_>>().await.unwrap();
+            let pretty_print = arrow::util::pretty::pretty_format_batches(&batches)
+                .unwrap()
+                .to_string();
+            let expected = r#"+----+---------+-------+
+| id | name    | value |
++----+---------+-------+
+| 1  | Alice   | 10.5  |
+| 2  | Bob     | 20.3  |
+| 3  | Charlie | 30.7  |
+| 4  | David   | 40.1  |
+| 5  | Eva     | 50.2  |
+| 6  | Frank   | 60.3  |
+| 7  | Grace   | 70.4  |
+| 8  | Henry   | 80.5  |
+| 9  | Ivy     | 90.6  |
++----+---------+-------+"#;
+            assert_eq!(expected, pretty_print);
+        }
+    }
 }

src/common/datasource/src/lib.rs

@@ -16,6 +16,7 @@
 #![feature(type_alias_impl_trait)]
 
 pub mod buffered_writer;
+pub mod compressed_writer;
 pub mod compression;
 pub mod error;
 pub mod file_format;

src/common/datasource/src/test_util.rs

@@ -28,7 +28,7 @@ use object_store::ObjectStore;
 use object_store::services::Fs;
 
 use crate::file_format::csv::{CsvFormat, stream_to_csv};
-use crate::file_format::json::stream_to_json;
+use crate::file_format::json::{JsonFormat, stream_to_json};
 use crate::test_util;
 
 pub const TEST_BATCH_SIZE: usize = 100;
@@ -122,13 +122,16 @@ pub async fn setup_stream_to_json_test(origin_path: &str, threshold: impl Fn(usi
 
     let output_path = format!("{}/{}", dir.path().display(), "output");
 
+    let json_format = JsonFormat::default();
+
     assert!(
         stream_to_json(
             Box::pin(stream),
             tmp_store.clone(),
             &output_path,
             threshold(size),
-            8
+            8,
+            &json_format,
         )
         .await
         .is_ok()

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

@@ -97,9 +97,9 @@ pub trait Event: Send + Sync + Debug {
         vec![]
     }
 
-    /// Add the extra row to the event with the default row.
-    fn extra_row(&self) -> Result<Row> {
-        Ok(Row { values: vec![] })
+    /// Add the extra rows to the event with the default row.
+    fn extra_rows(&self) -> Result<Vec<Row>> {
+        Ok(vec![Row { values: vec![] }])
     }
 
     /// Returns the event as any type.
@@ -159,15 +159,17 @@ pub fn build_row_inserts_request(events: &[&Box<dyn Event>]) -> Result<RowInsert
 
     let mut rows: Vec<Row> = Vec::with_capacity(events.len());
     for event in events {
-        let extra_row = event.extra_row()?;
-        let mut values = Vec::with_capacity(3 + extra_row.values.len());
-        values.extend([
-            ValueData::StringValue(event.event_type().to_string()).into(),
-            ValueData::BinaryValue(event.json_payload()?.into_bytes()).into(),
-            ValueData::TimestampNanosecondValue(event.timestamp().value()).into(),
-        ]);
-        values.extend(extra_row.values);
-        rows.push(Row { values });
+        let extra_rows = event.extra_rows()?;
+        for extra_row in extra_rows {
+            let mut values = Vec::with_capacity(3 + extra_row.values.len());
+            values.extend([
+                ValueData::StringValue(event.event_type().to_string()).into(),
+                ValueData::BinaryValue(event.json_payload()?.into_bytes()).into(),
+                ValueData::TimestampNanosecondValue(event.timestamp().value()).into(),
+            ]);
+            values.extend(extra_row.values);
+            rows.push(Row { values });
+        }
     }
 
     Ok(RowInsertRequests {

src/common/frontend/src/slow_query_event.rs

@@ -107,8 +107,8 @@ impl Event for SlowQueryEvent {
         ]
     }
 
-    fn extra_row(&self) -> Result<Row> {
-        Ok(Row {
+    fn extra_rows(&self) -> Result<Vec<Row>> {
+        Ok(vec![Row {
             values: vec![
                 ValueData::U64Value(self.cost).into(),
                 ValueData::U64Value(self.threshold).into(),
@@ -119,7 +119,7 @@ impl Event for SlowQueryEvent {
                 ValueData::TimestampMillisecondValue(self.promql_start.unwrap_or(0)).into(),
                 ValueData::TimestampMillisecondValue(self.promql_end.unwrap_or(0)).into(),
             ],
-        })
+        }])
     }
 
     fn json_payload(&self) -> Result<String> {

src/common/function/Cargo.toml

@@ -47,6 +47,7 @@ h3o = { version = "0.6", optional = true }
 hyperloglogplus = "0.4"
 jsonb.workspace = true
 memchr = "2.7"
+mito-codec.workspace = true
 nalgebra.workspace = true
 num = "0.4"
 num-traits = "0.2"

src/common/function/src/admin.rs

@@ -12,6 +12,7 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
+mod build_index_table;
 mod flush_compact_region;
 mod flush_compact_table;
 mod migrate_region;
@@ -26,6 +27,7 @@ use reconcile_catalog::ReconcileCatalogFunction;
 use reconcile_database::ReconcileDatabaseFunction;
 use reconcile_table::ReconcileTableFunction;
 
+use crate::admin::build_index_table::BuildIndexFunction;
 use crate::flush_flow::FlushFlowFunction;
 use crate::function_registry::FunctionRegistry;
 
@@ -40,6 +42,7 @@ impl AdminFunction {
         registry.register(CompactRegionFunction::factory());
         registry.register(FlushTableFunction::factory());
         registry.register(CompactTableFunction::factory());
+        registry.register(BuildIndexFunction::factory());
         registry.register(FlushFlowFunction::factory());
         registry.register(ReconcileCatalogFunction::factory());
         registry.register(ReconcileDatabaseFunction::factory());

src/common/function/src/admin/build_index_table.rs

@@ -0,0 +1,80 @@
+// 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 arrow::datatypes::DataType as ArrowDataType;
+use common_error::ext::BoxedError;
+use common_macro::admin_fn;
+use common_query::error::{
+    InvalidFuncArgsSnafu, MissingTableMutationHandlerSnafu, Result, TableMutationSnafu,
+    UnsupportedInputDataTypeSnafu,
+};
+use datafusion_expr::{Signature, Volatility};
+use datatypes::prelude::*;
+use session::context::QueryContextRef;
+use session::table_name::table_name_to_full_name;
+use snafu::{ResultExt, ensure};
+use table::requests::BuildIndexTableRequest;
+
+use crate::handlers::TableMutationHandlerRef;
+
+#[admin_fn(
+    name = BuildIndexFunction,
+    display_name = build_index,
+    sig_fn = build_index_signature,
+    ret = uint64
+)]
+pub(crate) async fn build_index(
+    table_mutation_handler: &TableMutationHandlerRef,
+    query_ctx: &QueryContextRef,
+    params: &[ValueRef<'_>],
+) -> Result<Value> {
+    ensure!(
+        params.len() == 1,
+        InvalidFuncArgsSnafu {
+            err_msg: format!(
+                "The length of the args is not correct, expect 1, have: {}",
+                params.len()
+            ),
+        }
+    );
+
+    let ValueRef::String(table_name) = params[0] else {
+        return UnsupportedInputDataTypeSnafu {
+            function: "build_index",
+            datatypes: params.iter().map(|v| v.data_type()).collect::<Vec<_>>(),
+        }
+        .fail();
+    };
+
+    let (catalog_name, schema_name, table_name) = table_name_to_full_name(table_name, query_ctx)
+        .map_err(BoxedError::new)
+        .context(TableMutationSnafu)?;
+
+    let affected_rows = table_mutation_handler
+        .build_index(
+            BuildIndexTableRequest {
+                catalog_name,
+                schema_name,
+                table_name,
+            },
+            query_ctx.clone(),
+        )
+        .await?;
+
+    Ok(Value::from(affected_rows as u64))
+}
+
+fn build_index_signature() -> Signature {
+    Signature::uniform(1, vec![ArrowDataType::Utf8], Volatility::Immutable)
+}

src/common/function/src/function_registry.rs

@@ -34,6 +34,7 @@ use crate::scalars::json::JsonFunction;
 use crate::scalars::matches::MatchesFunction;
 use crate::scalars::matches_term::MatchesTermFunction;
 use crate::scalars::math::MathFunction;
+use crate::scalars::primary_key::DecodePrimaryKeyFunction;
 use crate::scalars::string::register_string_functions;
 use crate::scalars::timestamp::TimestampFunction;
 use crate::scalars::uddsketch_calc::UddSketchCalcFunction;
@@ -143,6 +144,7 @@ pub static FUNCTION_REGISTRY: LazyLock<Arc<FunctionRegistry>> = LazyLock::new(||
     ExpressionFunction::register(&function_registry);
     UddSketchCalcFunction::register(&function_registry);
     HllCalcFunction::register(&function_registry);
+    DecodePrimaryKeyFunction::register(&function_registry);
 
     // Full text search function
     MatchesFunction::register(&function_registry);

src/common/function/src/handlers.rs

@@ -25,7 +25,9 @@ use common_query::Output;
 use common_query::error::Result;
 use session::context::QueryContextRef;
 use store_api::storage::RegionId;
-use table::requests::{CompactTableRequest, DeleteRequest, FlushTableRequest, InsertRequest};
+use table::requests::{
+    BuildIndexTableRequest, CompactTableRequest, DeleteRequest, FlushTableRequest, InsertRequest,
+};
 
 /// A trait for handling table mutations in `QueryEngine`.
 #[async_trait]
@@ -47,6 +49,13 @@ pub trait TableMutationHandler: Send + Sync {
         ctx: QueryContextRef,
     ) -> Result<AffectedRows>;
 
+    /// Trigger an index build task for the table.
+    async fn build_index(
+        &self,
+        request: BuildIndexTableRequest,
+        ctx: QueryContextRef,
+    ) -> Result<AffectedRows>;
+
     /// Trigger a flush task for a table region.
     async fn flush_region(&self, region_id: RegionId, ctx: QueryContextRef)
     -> Result<AffectedRows>;

src/common/function/src/scalars.rs

@@ -20,6 +20,7 @@ pub mod json;
 pub mod matches;
 pub mod matches_term;
 pub mod math;
+pub mod primary_key;
 pub(crate) mod string;
 pub mod vector;
 

src/common/function/src/scalars/json.rs

@@ -19,7 +19,7 @@ mod json_path_match;
 mod json_to_string;
 mod parse_json;
 
-use json_get::{JsonGetBool, JsonGetFloat, JsonGetInt, JsonGetString};
+use json_get::{JsonGetBool, JsonGetFloat, JsonGetInt, JsonGetObject, JsonGetString};
 use json_is::{
     JsonIsArray, JsonIsBool, JsonIsFloat, JsonIsInt, JsonIsNull, JsonIsObject, JsonIsString,
 };
@@ -39,6 +39,7 @@ impl JsonFunction {
         registry.register_scalar(JsonGetFloat::default());
         registry.register_scalar(JsonGetString::default());
         registry.register_scalar(JsonGetBool::default());
+        registry.register_scalar(JsonGetObject::default());
 
         registry.register_scalar(JsonIsNull::default());
         registry.register_scalar(JsonIsInt::default());

src/common/function/src/scalars/json/json_get.rs

@@ -16,10 +16,13 @@ use std::fmt::{self, Display};
 use std::sync::Arc;
 
 use arrow::compute;
+use datafusion_common::DataFusionError;
 use datafusion_common::arrow::array::{
-    Array, AsArray, BooleanBuilder, Float64Builder, Int64Builder, StringViewBuilder,
+    Array, AsArray, BinaryViewBuilder, BooleanBuilder, Float64Builder, Int64Builder,
+    StringViewBuilder,
 };
 use datafusion_common::arrow::datatypes::DataType;
+use datafusion_expr::type_coercion::aggregates::STRINGS;
 use datafusion_expr::{ColumnarValue, ScalarFunctionArgs, Signature};
 
 use crate::function::{Function, extract_args};
@@ -212,13 +215,92 @@ impl Display for JsonGetString {
     }
 }
 
+/// Get the object from JSON value by path.
+pub(super) struct JsonGetObject {
+    signature: Signature,
+}
+
+impl JsonGetObject {
+    const NAME: &'static str = "json_get_object";
+}
+
+impl Default for JsonGetObject {
+    fn default() -> Self {
+        Self {
+            signature: helper::one_of_sigs2(
+                vec![
+                    DataType::Binary,
+                    DataType::LargeBinary,
+                    DataType::BinaryView,
+                ],
+                STRINGS.to_vec(),
+            ),
+        }
+    }
+}
+
+impl Function for JsonGetObject {
+    fn name(&self) -> &str {
+        Self::NAME
+    }
+
+    fn return_type(&self, _: &[DataType]) -> datafusion_common::Result<DataType> {
+        Ok(DataType::BinaryView)
+    }
+
+    fn signature(&self) -> &Signature {
+        &self.signature
+    }
+
+    fn invoke_with_args(
+        &self,
+        args: ScalarFunctionArgs,
+    ) -> datafusion_common::Result<ColumnarValue> {
+        let [arg0, arg1] = extract_args(self.name(), &args)?;
+        let arg0 = compute::cast(&arg0, &DataType::BinaryView)?;
+        let jsons = arg0.as_binary_view();
+        let arg1 = compute::cast(&arg1, &DataType::Utf8View)?;
+        let paths = arg1.as_string_view();
+
+        let len = jsons.len();
+        let mut builder = BinaryViewBuilder::with_capacity(len);
+
+        for i in 0..len {
+            let json = jsons.is_valid(i).then(|| jsons.value(i));
+            let path = paths.is_valid(i).then(|| paths.value(i));
+            let result = if let (Some(json), Some(path)) = (json, path) {
+                let result = jsonb::jsonpath::parse_json_path(path.as_bytes()).and_then(|path| {
+                    let mut data = Vec::new();
+                    let mut offset = Vec::new();
+                    jsonb::get_by_path(json, path, &mut data, &mut offset)
+                        .map(|()| jsonb::is_object(&data).then_some(data))
+                });
+                result.map_err(|e| DataFusionError::Execution(e.to_string()))?
+            } else {
+                None
+            };
+            builder.append_option(result);
+        }
+
+        Ok(ColumnarValue::Array(Arc::new(builder.finish())))
+    }
+}
+
+impl Display for JsonGetObject {
+    fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
+        write!(f, "{}", Self::NAME.to_ascii_uppercase())
+    }
+}
+
 #[cfg(test)]
 mod tests {
     use std::sync::Arc;
 
     use arrow_schema::Field;
-    use datafusion_common::arrow::array::{BinaryArray, StringArray};
+    use datafusion_common::ScalarValue;
+    use datafusion_common::arrow::array::{BinaryArray, BinaryViewArray, StringArray};
     use datafusion_common::arrow::datatypes::{Float64Type, Int64Type};
+    use datatypes::types::parse_string_to_jsonb;
 
     use super::*;
 
@@ -425,4 +507,49 @@ mod tests {
             assert_eq!(*gt, result);
         }
     }
+
+    #[test]
+    fn test_json_get_object() -> datafusion_common::Result<()> {
+        let udf = JsonGetObject::default();
+        assert_eq!("json_get_object", udf.name());
+        assert_eq!(
+            DataType::BinaryView,
+            udf.return_type(&[DataType::BinaryView, DataType::Utf8View])?
+        );
+
+        let json_value = parse_string_to_jsonb(r#"{"a": {"b": {"c": {"d": 1}}}}"#).unwrap();
+        let paths = vec!["$", "$.a", "$.a.b", "$.a.b.c", "$.a.b.c.d", "$.e", "$.a.e"];
+        let number_rows = paths.len();
+
+        let args = ScalarFunctionArgs {
+            args: vec![
+                ColumnarValue::Scalar(ScalarValue::Binary(Some(json_value))),
+                ColumnarValue::Array(Arc::new(StringArray::from_iter_values(paths))),
+            ],
+            arg_fields: vec![],
+            number_rows,
+            return_field: Arc::new(Field::new("x", DataType::Binary, false)),
+            config_options: Arc::new(Default::default()),
+        };
+        let result = udf
+            .invoke_with_args(args)
+            .and_then(|x| x.to_array(number_rows))?;
+        let result = result.as_binary_view();
+
+        let expected = &BinaryViewArray::from_iter(
+            vec![
+                Some(r#"{"a": {"b": {"c": {"d": 1}}}}"#),
+                Some(r#"{"b": {"c": {"d": 1}}}"#),
+                Some(r#"{"c": {"d": 1}}"#),
+                Some(r#"{"d": 1}"#),
+                None,
+                None,
+                None,
+            ]
+            .into_iter()
+            .map(|x| x.and_then(|s| parse_string_to_jsonb(s).ok())),
+        );
+        assert_eq!(result, expected);
+        Ok(())
+    }
 }

src/common/function/src/scalars/json/json_to_string.rs

@@ -32,7 +32,15 @@ impl Default for JsonToStringFunction {
     fn default() -> Self {
         Self {
             // TODO(LFC): Use a more clear type here instead of "Binary" for Json input, once we have a "Json" type.
-            signature: Signature::exact(vec![DataType::Binary], Volatility::Immutable),
+            signature: Signature::uniform(
+                1,
+                vec![
+                    DataType::Binary,
+                    DataType::LargeBinary,
+                    DataType::BinaryView,
+                ],
+                Volatility::Immutable,
+            ),
         }
     }
 }
@@ -57,7 +65,8 @@ impl Function for JsonToStringFunction {
         args: ScalarFunctionArgs,
     ) -> datafusion_common::Result<ColumnarValue> {
         let [arg0] = extract_args(self.name(), &args)?;
-        let jsons = arg0.as_binary::<i32>();
+        let arg0 = arrow::compute::cast(&arg0, &DataType::BinaryView)?;
+        let jsons = arg0.as_binary_view();
 
         let size = jsons.len();
         let mut builder = StringViewBuilder::with_capacity(size);

src/common/function/src/scalars/primary_key.rs

@@ -0,0 +1,521 @@
+// Copyright 2023 Greptime Team
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+use std::collections::HashMap;
+use std::fmt::{self, Display};
+use std::sync::Arc;
+
+use datafusion_common::arrow::array::{
+    Array, ArrayRef, BinaryArray, BinaryViewArray, DictionaryArray, ListBuilder, StringBuilder,
+};
+use datafusion_common::arrow::datatypes::{DataType, Field};
+use datafusion_common::{DataFusionError, ScalarValue};
+use datafusion_expr::{ColumnarValue, ScalarFunctionArgs, Signature, Volatility};
+use datatypes::arrow::datatypes::UInt32Type;
+use datatypes::value::Value;
+use mito_codec::row_converter::{
+    CompositeValues, PrimaryKeyCodec, SortField, build_primary_key_codec_with_fields,
+};
+use store_api::codec::PrimaryKeyEncoding;
+use store_api::metadata::RegionMetadata;
+use store_api::storage::ColumnId;
+use store_api::storage::consts::{PRIMARY_KEY_COLUMN_NAME, ReservedColumnId};
+
+use crate::function::{Function, extract_args};
+use crate::function_registry::FunctionRegistry;
+
+type NameValuePair = (String, Option<String>);
+
+#[derive(Clone, Debug)]
+pub(crate) struct DecodePrimaryKeyFunction {
+    signature: Signature,
+}
+
+const NAME: &str = "decode_primary_key";
+const NULL_VALUE_LITERAL: &str = "null";
+
+impl Default for DecodePrimaryKeyFunction {
+    fn default() -> Self {
+        Self {
+            signature: Signature::any(3, Volatility::Immutable),
+        }
+    }
+}
+
+impl DecodePrimaryKeyFunction {
+    pub fn register(registry: &FunctionRegistry) {
+        registry.register_scalar(Self::default());
+    }
+
+    fn return_data_type() -> DataType {
+        DataType::List(Arc::new(Field::new("item", DataType::Utf8, true)))
+    }
+}
+
+impl Function for DecodePrimaryKeyFunction {
+    fn name(&self) -> &str {
+        NAME
+    }
+
+    fn return_type(&self, _: &[DataType]) -> datafusion_common::Result<DataType> {
+        Ok(Self::return_data_type())
+    }
+
+    fn signature(&self) -> &Signature {
+        &self.signature
+    }
+
+    fn invoke_with_args(
+        &self,
+        args: ScalarFunctionArgs,
+    ) -> datafusion_common::Result<ColumnarValue> {
+        let [encoded, _, _] = extract_args(self.name(), &args)?;
+        let number_rows = args.number_rows;
+
+        let encoding = parse_encoding(&args.args[1])?;
+        let metadata = parse_region_metadata(&args.args[2])?;
+        let codec = build_codec(&metadata, encoding);
+        let name_lookup: HashMap<_, _> = metadata
+            .column_metadatas
+            .iter()
+            .map(|c| (c.column_id, c.column_schema.name.clone()))
+            .collect();
+
+        let decoded_rows = decode_primary_keys(encoded, number_rows, codec.as_ref(), &name_lookup)?;
+        let array = build_list_array(&decoded_rows)?;
+
+        Ok(ColumnarValue::Array(array))
+    }
+}
+
+impl Display for DecodePrimaryKeyFunction {
+    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
+        write!(f, "DECODE_PRIMARY_KEY")
+    }
+}
+
+fn parse_encoding(arg: &ColumnarValue) -> datafusion_common::Result<PrimaryKeyEncoding> {
+    let encoding = match arg {
+        ColumnarValue::Scalar(ScalarValue::Utf8(Some(v)))
+        | ColumnarValue::Scalar(ScalarValue::LargeUtf8(Some(v))) => v.as_str(),
+        ColumnarValue::Scalar(value) => {
+            return Err(DataFusionError::Execution(format!(
+                "encoding must be a string literal, got {value:?}"
+            )));
+        }
+        ColumnarValue::Array(_) => {
+            return Err(DataFusionError::Execution(
+                "encoding must be a scalar string".to_string(),
+            ));
+        }
+    };
+
+    match encoding.to_ascii_lowercase().as_str() {
+        "dense" => Ok(PrimaryKeyEncoding::Dense),
+        "sparse" => Ok(PrimaryKeyEncoding::Sparse),
+        _ => Err(DataFusionError::Execution(format!(
+            "unsupported primary key encoding: {encoding}"
+        ))),
+    }
+}
+
+fn build_codec(
+    metadata: &RegionMetadata,
+    encoding: PrimaryKeyEncoding,
+) -> Arc<dyn PrimaryKeyCodec> {
+    let fields = metadata.primary_key_columns().map(|c| {
+        (
+            c.column_id,
+            SortField::new(c.column_schema.data_type.clone()),
+        )
+    });
+    build_primary_key_codec_with_fields(encoding, fields)
+}
+
+fn parse_region_metadata(arg: &ColumnarValue) -> datafusion_common::Result<RegionMetadata> {
+    let json = match arg {
+        ColumnarValue::Scalar(ScalarValue::Utf8(Some(v)))
+        | ColumnarValue::Scalar(ScalarValue::LargeUtf8(Some(v))) => v.as_str(),
+        ColumnarValue::Scalar(value) => {
+            return Err(DataFusionError::Execution(format!(
+                "region metadata must be a string literal, got {value:?}"
+            )));
+        }
+        ColumnarValue::Array(_) => {
+            return Err(DataFusionError::Execution(
+                "region metadata must be a scalar string".to_string(),
+            ));
+        }
+    };
+
+    RegionMetadata::from_json(json)
+        .map_err(|e| DataFusionError::Execution(format!("failed to parse region metadata: {e:?}")))
+}
+
+fn decode_primary_keys(
+    encoded: ArrayRef,
+    number_rows: usize,
+    codec: &dyn PrimaryKeyCodec,
+    name_lookup: &HashMap<ColumnId, String>,
+) -> datafusion_common::Result<Vec<Vec<NameValuePair>>> {
+    if let Some(dict) = encoded
+        .as_any()
+        .downcast_ref::<DictionaryArray<UInt32Type>>()
+    {
+        decode_dictionary(dict, number_rows, codec, name_lookup)
+    } else if let Some(array) = encoded.as_any().downcast_ref::<BinaryArray>() {
+        decode_binary_array(array, codec, name_lookup)
+    } else if let Some(array) = encoded.as_any().downcast_ref::<BinaryViewArray>() {
+        decode_binary_view_array(array, codec, name_lookup)
+    } else {
+        Err(DataFusionError::Execution(format!(
+            "column {PRIMARY_KEY_COLUMN_NAME} must be binary or dictionary(binary) array"
+        )))
+    }
+}
+
+fn decode_dictionary(
+    dict: &DictionaryArray<UInt32Type>,
+    number_rows: usize,
+    codec: &dyn PrimaryKeyCodec,
+    name_lookup: &HashMap<ColumnId, String>,
+) -> datafusion_common::Result<Vec<Vec<NameValuePair>>> {
+    let values = dict
+        .values()
+        .as_any()
+        .downcast_ref::<BinaryArray>()
+        .ok_or_else(|| {
+            DataFusionError::Execution("primary key dictionary values are not binary".to_string())
+        })?;
+
+    let mut decoded_values = Vec::with_capacity(values.len());
+    for i in 0..values.len() {
+        let pk = values.value(i);
+        let pairs = decode_one(pk, codec, name_lookup)?;
+        decoded_values.push(pairs);
+    }
+
+    let mut rows = Vec::with_capacity(number_rows);
+    let keys = dict.keys();
+    for i in 0..number_rows {
+        let dict_index = keys.value(i) as usize;
+        rows.push(decoded_values[dict_index].clone());
+    }
+
+    Ok(rows)
+}
+
+fn decode_binary_array(
+    array: &BinaryArray,
+    codec: &dyn PrimaryKeyCodec,
+    name_lookup: &HashMap<ColumnId, String>,
+) -> datafusion_common::Result<Vec<Vec<NameValuePair>>> {
+    (0..array.len())
+        .map(|i| decode_one(array.value(i), codec, name_lookup))
+        .collect()
+}
+
+fn decode_binary_view_array(
+    array: &BinaryViewArray,
+    codec: &dyn PrimaryKeyCodec,
+    name_lookup: &HashMap<ColumnId, String>,
+) -> datafusion_common::Result<Vec<Vec<NameValuePair>>> {
+    (0..array.len())
+        .map(|i| decode_one(array.value(i), codec, name_lookup))
+        .collect()
+}
+
+fn decode_one(
+    pk: &[u8],
+    codec: &dyn PrimaryKeyCodec,
+    name_lookup: &HashMap<ColumnId, String>,
+) -> datafusion_common::Result<Vec<NameValuePair>> {
+    let decoded = codec
+        .decode(pk)
+        .map_err(|e| DataFusionError::Execution(format!("failed to decode primary key: {e}")))?;
+
+    Ok(match decoded {
+        CompositeValues::Dense(values) => values
+            .into_iter()
+            .map(|(column_id, value)| (column_name(column_id, name_lookup), value_to_string(value)))
+            .collect(),
+        CompositeValues::Sparse(values) => {
+            let mut values: Vec<_> = values
+                .iter()
+                .map(|(column_id, value)| {
+                    (
+                        *column_id,
+                        column_name(*column_id, name_lookup),
+                        value_to_string(value.clone()),
+                    )
+                })
+                .collect();
+            values.sort_by_key(|(column_id, _, _)| {
+                (ReservedColumnId::is_reserved(*column_id), *column_id)
+            });
+            values
+                .into_iter()
+                .map(|(_, name, value)| (name, value))
+                .collect()
+        }
+    })
+}
+
+fn column_name(column_id: ColumnId, name_lookup: &HashMap<ColumnId, String>) -> String {
+    if let Some(name) = name_lookup.get(&column_id) {
+        return name.clone();
+    }
+
+    if column_id == ReservedColumnId::table_id() {
+        return "__table_id".to_string();
+    }
+    if column_id == ReservedColumnId::tsid() {
+        return "__tsid".to_string();
+    }
+
+    column_id.to_string()
+}
+
+fn value_to_string(value: Value) -> Option<String> {
+    match value {
+        Value::Null => None,
+        _ => Some(value.to_string()),
+    }
+}
+
+fn build_list_array(rows: &[Vec<NameValuePair>]) -> datafusion_common::Result<ArrayRef> {
+    let mut builder = ListBuilder::new(StringBuilder::new());
+
+    for row in rows {
+        for (key, value) in row {
+            let value = value.as_deref().unwrap_or(NULL_VALUE_LITERAL);
+            builder.values().append_value(format!("{key} : {value}"));
+        }
+        builder.append(true);
+    }
+
+    Ok(Arc::new(builder.finish()))
+}
+
+#[cfg(test)]
+mod tests {
+    use api::v1::SemanticType;
+    use datafusion_common::ScalarValue;
+    use datatypes::arrow::array::builder::BinaryDictionaryBuilder;
+    use datatypes::arrow::array::{BinaryArray, ListArray, StringArray};
+    use datatypes::arrow::datatypes::UInt32Type;
+    use datatypes::prelude::ConcreteDataType;
+    use datatypes::schema::ColumnSchema;
+    use datatypes::value::Value;
+    use mito_codec::row_converter::{
+        DensePrimaryKeyCodec, PrimaryKeyCodecExt, SortField, SparsePrimaryKeyCodec,
+    };
+    use store_api::codec::PrimaryKeyEncoding;
+    use store_api::metadata::{ColumnMetadata, RegionMetadataBuilder};
+    use store_api::storage::consts::ReservedColumnId;
+    use store_api::storage::{ColumnId, RegionId};
+
+    use super::*;
+
+    fn pk_field() -> Arc<Field> {
+        Arc::new(Field::new_dictionary(
+            PRIMARY_KEY_COLUMN_NAME,
+            DataType::UInt32,
+            DataType::Binary,
+            false,
+        ))
+    }
+
+    fn region_metadata_json(
+        columns: &[(ColumnId, &str, ConcreteDataType)],
+        encoding: PrimaryKeyEncoding,
+    ) -> String {
+        let mut builder = RegionMetadataBuilder::new(RegionId::new(1, 1));
+        builder.push_column_metadata(ColumnMetadata {
+            column_schema: ColumnSchema::new(
+                "ts",
+                ConcreteDataType::timestamp_millisecond_datatype(),
+                false,
+            ),
+            semantic_type: SemanticType::Timestamp,
+            column_id: 100,
+        });
+        builder.primary_key_encoding(encoding);
+        for (id, name, ty) in columns {
+            builder.push_column_metadata(ColumnMetadata {
+                column_schema: ColumnSchema::new((*name).to_string(), ty.clone(), true),
+                semantic_type: SemanticType::Tag,
+                column_id: *id,
+            });
+        }
+        builder.primary_key(columns.iter().map(|(id, _, _)| *id).collect());
+
+        builder.build().unwrap().to_json().unwrap()
+    }
+
+    fn list_row(list: &ListArray, row_idx: usize) -> Vec<String> {
+        let values = list.value(row_idx);
+        let values = values.as_any().downcast_ref::<StringArray>().unwrap();
+        (0..values.len())
+            .map(|i| values.value(i).to_string())
+            .collect()
+    }
+
+    #[test]
+    fn test_decode_dense_primary_key() {
+        let columns = vec![
+            (0, "host", ConcreteDataType::string_datatype()),
+            (1, "core", ConcreteDataType::int64_datatype()),
+        ];
+        let metadata_json = region_metadata_json(&columns, PrimaryKeyEncoding::Dense);
+        let codec = DensePrimaryKeyCodec::with_fields(
+            columns
+                .iter()
+                .map(|(id, _, ty)| (*id, SortField::new(ty.clone())))
+                .collect(),
+        );
+
+        let rows = vec![
+            vec![Value::from("a"), Value::from(1_i64)],
+            vec![Value::from("b"), Value::from(2_i64)],
+            vec![Value::from("a"), Value::from(1_i64)],
+        ];
+
+        let mut builder = BinaryDictionaryBuilder::<UInt32Type>::new();
+        for row in &rows {
+            let encoded = codec.encode(row.iter().map(|v| v.as_value_ref())).unwrap();
+            builder.append(encoded.as_slice()).unwrap();
+        }
+        let dict_array: ArrayRef = Arc::new(builder.finish());
+
+        let args = ScalarFunctionArgs {
+            args: vec![
+                ColumnarValue::Array(dict_array),
+                ColumnarValue::Scalar(ScalarValue::Utf8(Some("dense".to_string()))),
+                ColumnarValue::Scalar(ScalarValue::Utf8(Some(metadata_json))),
+            ],
+            arg_fields: vec![
+                pk_field(),
+                Arc::new(Field::new("encoding", DataType::Utf8, false)),
+                Arc::new(Field::new("region_metadata", DataType::Utf8, false)),
+            ],
+            number_rows: 3,
+            return_field: Arc::new(Field::new(
+                "decoded",
+                DecodePrimaryKeyFunction::return_data_type(),
+                false,
+            )),
+            config_options: Default::default(),
+        };
+
+        let func = DecodePrimaryKeyFunction::default();
+        let result = func
+            .invoke_with_args(args)
+            .and_then(|v| v.to_array(3))
+            .unwrap();
+        let list = result.as_any().downcast_ref::<ListArray>().unwrap();
+
+        let expected = [
+            vec!["host : a".to_string(), "core : 1".to_string()],
+            vec!["host : b".to_string(), "core : 2".to_string()],
+            vec!["host : a".to_string(), "core : 1".to_string()],
+        ];
+
+        for (row_idx, expected_row) in expected.iter().enumerate() {
+            assert_eq!(*expected_row, list_row(list, row_idx));
+        }
+    }
+
+    #[test]
+    fn test_decode_sparse_primary_key() {
+        let columns = vec![
+            (10, "k0", ConcreteDataType::string_datatype()),
+            (11, "k1", ConcreteDataType::string_datatype()),
+        ];
+        let metadata_json = region_metadata_json(&columns, PrimaryKeyEncoding::Sparse);
+        let codec = SparsePrimaryKeyCodec::schemaless();
+
+        let rows = vec![
+            vec![
+                (ReservedColumnId::table_id(), Value::UInt32(1)),
+                (ReservedColumnId::tsid(), Value::UInt64(100)),
+                (10, Value::from("a")),
+                (11, Value::from("b")),
+            ],
+            vec![
+                (ReservedColumnId::table_id(), Value::UInt32(1)),
+                (ReservedColumnId::tsid(), Value::UInt64(200)),
+                (10, Value::from("c")),
+                (11, Value::from("d")),
+            ],
+        ];
+
+        let mut encoded_values = Vec::with_capacity(rows.len());
+        for row in &rows {
+            let mut buf = Vec::new();
+            codec.encode_values(row, &mut buf).unwrap();
+            encoded_values.push(buf);
+        }
+
+        let pk_array: ArrayRef = Arc::new(BinaryArray::from_iter_values(
+            encoded_values.iter().cloned(),
+        ));
+
+        let args = ScalarFunctionArgs {
+            args: vec![
+                ColumnarValue::Array(pk_array),
+                ColumnarValue::Scalar(ScalarValue::Utf8(Some("sparse".to_string()))),
+                ColumnarValue::Scalar(ScalarValue::Utf8(Some(metadata_json))),
+            ],
+            arg_fields: vec![
+                pk_field(),
+                Arc::new(Field::new("encoding", DataType::Utf8, false)),
+                Arc::new(Field::new("region_metadata", DataType::Utf8, false)),
+            ],
+            number_rows: rows.len(),
+            return_field: Arc::new(Field::new(
+                "decoded",
+                DecodePrimaryKeyFunction::return_data_type(),
+                false,
+            )),
+            config_options: Default::default(),
+        };
+
+        let func = DecodePrimaryKeyFunction::default();
+        let result = func
+            .invoke_with_args(args)
+            .and_then(|v| v.to_array(rows.len()))
+            .unwrap();
+        let list = result.as_any().downcast_ref::<ListArray>().unwrap();
+
+        let expected = [
+            vec![
+                "k0 : a".to_string(),
+                "k1 : b".to_string(),
+                "__tsid : 100".to_string(),
+                "__table_id : 1".to_string(),
+            ],
+            vec![
+                "k0 : c".to_string(),
+                "k1 : d".to_string(),
+                "__tsid : 200".to_string(),
+                "__table_id : 1".to_string(),
+            ],
+        ];
+
+        for (row_idx, expected_row) in expected.iter().enumerate() {
+            assert_eq!(*expected_row, list_row(list, row_idx));
+        }
+    }
+}

src/common/function/src/state.rs

@@ -44,7 +44,8 @@ impl FunctionState {
         use session::context::QueryContextRef;
         use store_api::storage::RegionId;
         use table::requests::{
-            CompactTableRequest, DeleteRequest, FlushTableRequest, InsertRequest,
+            BuildIndexTableRequest, CompactTableRequest, DeleteRequest, FlushTableRequest,
+            InsertRequest,
         };
 
         use crate::handlers::{FlowServiceHandler, ProcedureServiceHandler, TableMutationHandler};
@@ -120,6 +121,14 @@ impl FunctionState {
                 Ok(ROWS)
             }
 
+            async fn build_index(
+                &self,
+                _request: BuildIndexTableRequest,
+                _ctx: QueryContextRef,
+            ) -> Result<AffectedRows> {
+                Ok(ROWS)
+            }
+
             async fn flush_region(
                 &self,
                 _region_id: RegionId,

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

@@ -12,8 +12,6 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
-mod version;
-
 use std::sync::Arc;
 
 use common_catalog::consts::{
@@ -27,7 +25,6 @@ use datafusion_expr::{ColumnarValue, ScalarFunctionArgs, Signature, TypeSignatur
 use datafusion_pg_catalog::pg_catalog::{self, PgCatalogStaticTables};
 use datatypes::arrow::datatypes::{DataType, Field};
 use derive_more::derive::Display;
-use version::PGVersionFunction;
 
 use crate::function::{Function, find_function_context};
 use crate::function_registry::FunctionRegistry;
@@ -183,7 +180,6 @@ impl PGCatalogFunction {
         let static_tables =
             Arc::new(PgCatalogStaticTables::try_new().expect("load postgres static tables"));
 
-        registry.register_scalar(PGVersionFunction::default());
         registry.register_scalar(CurrentSchemaFunction::default());
         registry.register_scalar(CurrentSchemasFunction::new());
         registry.register_scalar(SessionUserFunction::default());

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

@@ -1,61 +0,0 @@
-// Copyright 2023 Greptime Team
-//
-// Licensed under the Apache License, Version 2.0 (the "License");
-// you may not use this file except in compliance with the License.
-// You may obtain a copy of the License at
-//
-//     http://www.apache.org/licenses/LICENSE-2.0
-//
-// Unless required by applicable law or agreed to in writing, software
-// distributed under the License is distributed on an "AS IS" BASIS,
-// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-// See the License for the specific language governing permissions and
-// limitations under the License.
-
-use std::fmt;
-
-use datafusion::arrow::datatypes::DataType;
-use datafusion_common::ScalarValue;
-use datafusion_expr::{ColumnarValue, ScalarFunctionArgs, Signature, Volatility};
-
-use crate::function::Function;
-
-#[derive(Clone, Debug)]
-pub(crate) struct PGVersionFunction {
-    signature: Signature,
-}
-
-impl Default for PGVersionFunction {
-    fn default() -> Self {
-        Self {
-            signature: Signature::exact(vec![], Volatility::Immutable),
-        }
-    }
-}
-
-impl fmt::Display for PGVersionFunction {
-    fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
-        write!(f, "pg_catalog.VERSION")
-    }
-}
-
-impl Function for PGVersionFunction {
-    fn name(&self) -> &str {
-        "pg_catalog.version"
-    }
-
-    fn return_type(&self, _: &[DataType]) -> datafusion_common::Result<DataType> {
-        Ok(DataType::Utf8View)
-    }
-
-    fn signature(&self) -> &Signature {
-        &self.signature
-    }
-
-    fn invoke_with_args(&self, _: ScalarFunctionArgs) -> datafusion_common::Result<ColumnarValue> {
-        Ok(ColumnarValue::Scalar(ScalarValue::Utf8View(Some(format!(
-            "PostgreSQL 16.3 GreptimeDB {}",
-            common_version::version()
-        )))))
-    }
-}

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

@@ -50,7 +50,7 @@ impl Function for VersionFunction {
                 )
             }
             Channel::Postgres => {
-                format!("16.3-greptimedb-{}", common_version::version())
+                format!("PostgreSQL 16.3 GreptimeDB {}", common_version::version())
             }
             _ => common_version::version().to_string(),
         };

src/common/grpc/Cargo.toml

@@ -23,6 +23,7 @@ datatypes.workspace = true
 flatbuffers = "25.2"
 hyper.workspace = true
 lazy_static.workspace = true
+notify.workspace = true
 prost.workspace = true
 serde.workspace = true
 serde_json.workspace = true
@@ -37,6 +38,7 @@ vec1 = "1.12"
 criterion = "0.4"
 hyper-util = { workspace = true, features = ["tokio"] }
 rand.workspace = true
+tempfile.workspace = true
 
 [[bench]]
 name = "bench_main"

src/common/grpc/src/channel_manager.rs

@@ -12,6 +12,7 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
+use std::path::Path;
 use std::sync::Arc;
 use std::sync::atomic::{AtomicBool, AtomicU64, AtomicUsize, Ordering};
 use std::time::Duration;
@@ -30,6 +31,7 @@ use tonic::transport::{
 use tower::Service;
 
 use crate::error::{CreateChannelSnafu, InvalidConfigFilePathSnafu, Result};
+use crate::reloadable_tls::{ReloadableTlsConfig, TlsConfigLoader, maybe_watch_tls_config};
 
 const RECYCLE_CHANNEL_INTERVAL_SECS: u64 = 60;
 pub const DEFAULT_GRPC_REQUEST_TIMEOUT_SECS: u64 = 10;
@@ -50,7 +52,7 @@ pub struct ChannelManager {
 struct Inner {
     id: u64,
     config: ChannelConfig,
-    client_tls_config: Option<ClientTlsConfig>,
+    reloadable_client_tls_config: Option<Arc<ReloadableClientTlsConfig>>,
     pool: Arc<Pool>,
     channel_recycle_started: AtomicBool,
     cancel: CancellationToken,
@@ -78,7 +80,7 @@ impl Inner {
         Self {
             id,
             config,
-            client_tls_config: None,
+            reloadable_client_tls_config: None,
             pool,
             channel_recycle_started: AtomicBool::new(false),
             cancel,
@@ -91,13 +93,17 @@ impl ChannelManager {
         Default::default()
     }
 
-    /// unified with config function that support tls config
-    /// use [`load_tls_config`] to load tls config from file system
-    pub fn with_config(config: ChannelConfig, tls_config: Option<ClientTlsConfig>) -> Self {
+    /// Create a ChannelManager with configuration and optional TLS config
+    ///
+    /// Use [`load_client_tls_config`] to create TLS configuration from `ClientTlsOption`.
+    /// The TLS config supports both static (watch disabled) and dynamic reloading (watch enabled).
+    /// If you want to use dynamic reloading, please **manually** invoke [`maybe_watch_client_tls_config`] after this method.
+    pub fn with_config(
+        config: ChannelConfig,
+        reloadable_tls_config: Option<Arc<ReloadableClientTlsConfig>>,
+    ) -> Self {
         let mut inner = Inner::with_config(config.clone());
-        if let Some(tls_config) = tls_config {
-            inner.client_tls_config = Some(tls_config);
-        }
+        inner.reloadable_client_tls_config = reloadable_tls_config;
         Self {
             inner: Arc::new(inner),
         }
@@ -172,8 +178,21 @@ impl ChannelManager {
         self.pool().retain_channel(f);
     }
 
+    /// Clear all channels to force reconnection.
+    /// This should be called when TLS configuration changes to ensure new connections use updated certificates.
+    pub fn clear_all_channels(&self) {
+        self.pool().retain_channel(|_, _| false);
+    }
+
     fn build_endpoint(&self, addr: &str) -> Result<Endpoint> {
-        let http_prefix = if self.inner.client_tls_config.is_some() {
+        // Get the latest TLS config from reloadable config (which handles both static and dynamic cases)
+        let tls_config = self
+            .inner
+            .reloadable_client_tls_config
+            .as_ref()
+            .and_then(|c| c.get_config());
+
+        let http_prefix = if tls_config.is_some() {
             "https"
         } else {
             "http"
@@ -212,9 +231,9 @@ impl ChannelManager {
         if let Some(enabled) = self.config().http2_adaptive_window {
             endpoint = endpoint.http2_adaptive_window(enabled);
         }
-        if let Some(tls_config) = &self.inner.client_tls_config {
+        if let Some(tls_config) = tls_config {
             endpoint = endpoint
-                .tls_config(tls_config.clone())
+                .tls_config(tls_config)
                 .context(CreateChannelSnafu { addr })?;
         }
 
@@ -248,7 +267,7 @@ impl ChannelManager {
     }
 }
 
-pub fn load_tls_config(tls_option: Option<&ClientTlsOption>) -> Result<Option<ClientTlsConfig>> {
+fn load_tls_config(tls_option: Option<&ClientTlsOption>) -> Result<Option<ClientTlsConfig>> {
     let path_config = match tls_option {
         Some(path_config) if path_config.enabled => path_config,
         _ => return Ok(None),
@@ -276,13 +295,69 @@ pub fn load_tls_config(tls_option: Option<&ClientTlsOption>) -> Result<Option<Cl
     Ok(Some(tls_config))
 }
 
-#[derive(Clone, Debug, PartialEq, Eq, Serialize, Deserialize)]
+impl TlsConfigLoader<ClientTlsConfig> for ClientTlsOption {
+    type Error = crate::error::Error;
+
+    fn load(&self) -> Result<Option<ClientTlsConfig>> {
+        load_tls_config(Some(self))
+    }
+
+    fn watch_paths(&self) -> Vec<&Path> {
+        let mut paths = Vec::new();
+        if let Some(cert_path) = &self.client_cert_path {
+            paths.push(Path::new(cert_path.as_str()));
+        }
+        if let Some(key_path) = &self.client_key_path {
+            paths.push(Path::new(key_path.as_str()));
+        }
+        if let Some(ca_path) = &self.server_ca_cert_path {
+            paths.push(Path::new(ca_path.as_str()));
+        }
+        paths
+    }
+
+    fn watch_enabled(&self) -> bool {
+        self.enabled && self.watch
+    }
+}
+
+/// Type alias for client-side reloadable TLS config
+pub type ReloadableClientTlsConfig = ReloadableTlsConfig<ClientTlsConfig, ClientTlsOption>;
+
+/// Load client TLS configuration from `ClientTlsOption` and return a `ReloadableClientTlsConfig`.
+/// This is the primary way to create TLS configuration for the ChannelManager.
+pub fn load_client_tls_config(
+    tls_option: Option<ClientTlsOption>,
+) -> Result<Option<Arc<ReloadableClientTlsConfig>>> {
+    match tls_option {
+        Some(option) if option.enabled => {
+            let reloadable = ReloadableClientTlsConfig::try_new(option)?;
+            Ok(Some(Arc::new(reloadable)))
+        }
+        _ => Ok(None),
+    }
+}
+
+pub fn maybe_watch_client_tls_config(
+    client_tls_config: Arc<ReloadableClientTlsConfig>,
+    channel_manager: ChannelManager,
+) -> Result<()> {
+    maybe_watch_tls_config(client_tls_config, move || {
+        // Clear all existing channels to force reconnection with new certificates
+        channel_manager.clear_all_channels();
+        info!("Cleared all existing channels to use new TLS certificates.");
+    })
+}
+
+#[derive(Clone, Debug, PartialEq, Eq, Serialize, Deserialize, Default)]
 pub struct ClientTlsOption {
     /// Whether to enable TLS for client.
     pub enabled: bool,
     pub server_ca_cert_path: Option<String>,
     pub client_cert_path: Option<String>,
     pub client_key_path: Option<String>,
+    #[serde(default)]
+    pub watch: bool,
 }
 
 #[derive(Clone, Debug, PartialEq, Eq)]
@@ -602,6 +677,7 @@ mod tests {
                 server_ca_cert_path: Some("some_server_path".to_string()),
                 client_cert_path: Some("some_cert_path".to_string()),
                 client_key_path: Some("some_key_path".to_string()),
+                watch: false,
             });
 
         assert_eq!(
@@ -623,6 +699,7 @@ mod tests {
                     server_ca_cert_path: Some("some_server_path".to_string()),
                     client_cert_path: Some("some_cert_path".to_string()),
                     client_key_path: Some("some_key_path".to_string()),
+                    watch: false,
                 }),
                 max_recv_message_size: DEFAULT_MAX_GRPC_RECV_MESSAGE_SIZE,
                 max_send_message_size: DEFAULT_MAX_GRPC_SEND_MESSAGE_SIZE,

src/common/grpc/src/error.rs

@@ -38,6 +38,15 @@ pub enum Error {
         location: Location,
     },
 
+    #[snafu(display("Failed to watch config file path: {}", path))]
+    FileWatch {
+        path: String,
+        #[snafu(source)]
+        error: notify::Error,
+        #[snafu(implicit)]
+        location: Location,
+    },
+
     #[snafu(display(
         "Write type mismatch, column name: {}, expected: {}, actual: {}",
         column_name,
@@ -108,6 +117,7 @@ impl ErrorExt for Error {
         match self {
             Error::InvalidTlsConfig { .. }
             | Error::InvalidConfigFilePath { .. }
+            | Error::FileWatch { .. }
             | Error::TypeMismatch { .. }
             | Error::InvalidFlightData { .. }
             | Error::NotSupported { .. } => StatusCode::InvalidArguments,

src/common/grpc/src/lib.rs

@@ -16,6 +16,7 @@ pub mod channel_manager;
 pub mod error;
 pub mod flight;
 pub mod precision;
+pub mod reloadable_tls;
 pub mod select;
 
 pub use arrow_flight::FlightData;

src/common/grpc/src/reloadable_tls.rs

@@ -0,0 +1,163 @@
+// 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::path::Path;
+use std::result::Result as StdResult;
+use std::sync::atomic::{AtomicUsize, Ordering};
+use std::sync::mpsc::channel;
+use std::sync::{Arc, RwLock};
+
+use common_telemetry::{error, info};
+use notify::{EventKind, RecursiveMode, Watcher};
+use snafu::ResultExt;
+
+use crate::error::{FileWatchSnafu, Result};
+
+/// A trait for loading TLS configuration from an option type
+pub trait TlsConfigLoader<T> {
+    type Error;
+
+    /// Load the TLS configuration
+    fn load(&self) -> StdResult<Option<T>, Self::Error>;
+
+    /// Get paths to certificate files for watching
+    fn watch_paths(&self) -> Vec<&Path>;
+
+    /// Check if watching is enabled
+    fn watch_enabled(&self) -> bool;
+}
+
+/// A mutable container for TLS config
+///
+/// This struct allows dynamic reloading of certificates and keys.
+/// It's generic over the config type (e.g., ServerConfig, ClientTlsConfig)
+/// and the option type (e.g., TlsOption, ClientTlsOption).
+#[derive(Debug)]
+pub struct ReloadableTlsConfig<T, O>
+where
+    O: TlsConfigLoader<T>,
+{
+    tls_option: O,
+    config: RwLock<Option<T>>,
+    version: AtomicUsize,
+}
+
+impl<T, O> ReloadableTlsConfig<T, O>
+where
+    O: TlsConfigLoader<T>,
+{
+    /// Create config by loading configuration from the option type
+    pub fn try_new(tls_option: O) -> StdResult<Self, O::Error> {
+        let config = tls_option.load()?;
+        Ok(Self {
+            tls_option,
+            config: RwLock::new(config),
+            version: AtomicUsize::new(0),
+        })
+    }
+
+    /// Reread certificates and keys from file system.
+    pub fn reload(&self) -> StdResult<(), O::Error> {
+        let config = self.tls_option.load()?;
+        *self.config.write().unwrap() = config;
+        self.version.fetch_add(1, Ordering::Relaxed);
+        Ok(())
+    }
+
+    /// Get the config held by this container
+    pub fn get_config(&self) -> Option<T>
+    where
+        T: Clone,
+    {
+        self.config.read().unwrap().clone()
+    }
+
+    /// Get associated option
+    pub fn get_tls_option(&self) -> &O {
+        &self.tls_option
+    }
+
+    /// Get version of current config
+    ///
+    /// this version will auto increase when config get reloaded.
+    pub fn get_version(&self) -> usize {
+        self.version.load(Ordering::Relaxed)
+    }
+}
+
+/// Watch TLS configuration files for changes and reload automatically
+///
+/// This is a generic function that works with any ReloadableTlsConfig.
+/// When changes are detected, it calls the provided callback after reloading.
+///
+/// T: the original TLS config
+/// O: the compiled TLS option
+/// F: the hook function to be called after reloading
+/// E: the error type for the loading operation
+pub fn maybe_watch_tls_config<T, O, F, E>(
+    tls_config: Arc<ReloadableTlsConfig<T, O>>,
+    on_reload: F,
+) -> Result<()>
+where
+    T: Send + Sync + 'static,
+    O: TlsConfigLoader<T, Error = E> + Send + Sync + 'static,
+    E: std::error::Error + Send + Sync + 'static,
+    F: Fn() + Send + 'static,
+{
+    if !tls_config.get_tls_option().watch_enabled() {
+        return Ok(());
+    }
+
+    let tls_config_for_watcher = tls_config.clone();
+
+    let (tx, rx) = channel::<notify::Result<notify::Event>>();
+    let mut watcher = notify::recommended_watcher(tx).context(FileWatchSnafu { path: "<none>" })?;
+
+    // Watch all paths returned by the TlsConfigLoader
+    for path in tls_config.get_tls_option().watch_paths() {
+        watcher
+            .watch(path, RecursiveMode::NonRecursive)
+            .with_context(|_| FileWatchSnafu {
+                path: path.display().to_string(),
+            })?;
+    }
+
+    info!("Spawning background task for watching TLS cert/key file changes");
+    std::thread::spawn(move || {
+        let _watcher = watcher;
+        loop {
+            match rx.recv() {
+                Ok(Ok(event)) => {
+                    if let EventKind::Modify(_) | EventKind::Create(_) = event.kind {
+                        info!("Detected TLS cert/key file change: {:?}", event);
+                        if let Err(err) = tls_config_for_watcher.reload() {
+                            error!("Failed to reload TLS config: {}", err);
+                        } else {
+                            info!("Reloaded TLS cert/key file successfully.");
+                            on_reload();
+                        }
+                    }
+                }
+                Ok(Err(err)) => {
+                    error!("Failed to watch TLS cert/key file: {}", err);
+                }
+                Err(err) => {
+                    error!("TLS cert/key file watcher channel closed: {}", err);
+                }
+            }
+        }
+    });
+
+    Ok(())
+}

src/common/grpc/tests/mod.rs

@@ -13,14 +13,15 @@
 // limitations under the License.
 
 use common_grpc::channel_manager::{
-    ChannelConfig, ChannelManager, ClientTlsOption, load_tls_config,
+    ChannelConfig, ChannelManager, ClientTlsOption, load_client_tls_config,
+    maybe_watch_client_tls_config,
 };
 
 #[tokio::test]
 async fn test_mtls_config() {
     // test no config
     let config = ChannelConfig::new();
-    let re = load_tls_config(config.client_tls.as_ref());
+    let re = load_client_tls_config(config.client_tls.clone());
     assert!(re.is_ok());
     assert!(re.unwrap().is_none());
 
@@ -30,9 +31,10 @@ async fn test_mtls_config() {
         server_ca_cert_path: Some("tests/tls/wrong_ca.pem".to_string()),
         client_cert_path: Some("tests/tls/wrong_client.pem".to_string()),
         client_key_path: Some("tests/tls/wrong_client.key".to_string()),
+        watch: false,
     });
 
-    let re = load_tls_config(config.client_tls.as_ref());
+    let re = load_client_tls_config(config.client_tls.clone());
     assert!(re.is_err());
 
     // test corrupted file content
@@ -41,9 +43,10 @@ async fn test_mtls_config() {
         server_ca_cert_path: Some("tests/tls/ca.pem".to_string()),
         client_cert_path: Some("tests/tls/client.pem".to_string()),
         client_key_path: Some("tests/tls/corrupted".to_string()),
+        watch: false,
     });
 
-    let tls_config = load_tls_config(config.client_tls.as_ref()).unwrap();
+    let tls_config = load_client_tls_config(config.client_tls.clone()).unwrap();
     let re = ChannelManager::with_config(config, tls_config);
 
     let re = re.get("127.0.0.1:0");
@@ -55,10 +58,112 @@ async fn test_mtls_config() {
         server_ca_cert_path: Some("tests/tls/ca.pem".to_string()),
         client_cert_path: Some("tests/tls/client.pem".to_string()),
         client_key_path: Some("tests/tls/client.key".to_string()),
+        watch: false,
     });
 
-    let tls_config = load_tls_config(config.client_tls.as_ref()).unwrap();
+    let tls_config = load_client_tls_config(config.client_tls.clone()).unwrap();
     let re = ChannelManager::with_config(config, tls_config);
     let re = re.get("127.0.0.1:0");
     let _ = re.unwrap();
 }
+
+#[tokio::test]
+async fn test_reloadable_client_tls_config() {
+    common_telemetry::init_default_ut_logging();
+
+    let dir = tempfile::tempdir().unwrap();
+    let cert_path = dir.path().join("client.pem");
+    let key_path = dir.path().join("client.key");
+
+    std::fs::copy("tests/tls/client.pem", &cert_path).expect("failed to copy cert to tmpdir");
+    std::fs::copy("tests/tls/client.key", &key_path).expect("failed to copy key to tmpdir");
+
+    assert!(std::fs::exists(&cert_path).unwrap());
+    assert!(std::fs::exists(&key_path).unwrap());
+
+    let client_tls_option = ClientTlsOption {
+        enabled: true,
+        server_ca_cert_path: Some("tests/tls/ca.pem".to_string()),
+        client_cert_path: Some(
+            cert_path
+                .clone()
+                .into_os_string()
+                .into_string()
+                .expect("failed to convert path to string"),
+        ),
+        client_key_path: Some(
+            key_path
+                .clone()
+                .into_os_string()
+                .into_string()
+                .expect("failed to convert path to string"),
+        ),
+        watch: true,
+    };
+
+    let reloadable_config = load_client_tls_config(Some(client_tls_option))
+        .expect("failed to load tls config")
+        .expect("tls config should be present");
+
+    let config = ChannelConfig::new();
+    let manager = ChannelManager::with_config(config, Some(reloadable_config.clone()));
+
+    maybe_watch_client_tls_config(reloadable_config.clone(), manager.clone())
+        .expect("failed to watch client config");
+
+    assert_eq!(0, reloadable_config.get_version());
+    assert!(reloadable_config.get_config().is_some());
+
+    // Create a channel to verify it gets cleared on reload
+    let _ = manager.get("127.0.0.1:0").expect("failed to get channel");
+
+    // Simulate file change by copying a different key file
+    let tmp_file = key_path.with_extension("tmp");
+    std::fs::copy("tests/tls/server.key", &tmp_file).expect("Failed to copy temp key file");
+    std::fs::rename(&tmp_file, &key_path).expect("Failed to rename temp key file");
+
+    const MAX_RETRIES: usize = 30;
+    let mut retries = 0;
+    let mut version_updated = false;
+
+    while retries < MAX_RETRIES {
+        if reloadable_config.get_version() > 0 {
+            version_updated = true;
+            break;
+        }
+        std::thread::sleep(std::time::Duration::from_millis(100));
+        retries += 1;
+    }
+
+    assert!(version_updated, "TLS config did not reload in time");
+    assert!(reloadable_config.get_version() > 0);
+    assert!(reloadable_config.get_config().is_some());
+}
+
+#[tokio::test]
+async fn test_channel_manager_with_reloadable_tls() {
+    common_telemetry::init_default_ut_logging();
+
+    let client_tls_option = ClientTlsOption {
+        enabled: true,
+        server_ca_cert_path: Some("tests/tls/ca.pem".to_string()),
+        client_cert_path: Some("tests/tls/client.pem".to_string()),
+        client_key_path: Some("tests/tls/client.key".to_string()),
+        watch: false,
+    };
+
+    let reloadable_config = load_client_tls_config(Some(client_tls_option))
+        .expect("failed to load tls config")
+        .expect("tls config should be present");
+
+    let config = ChannelConfig::new();
+    let manager = ChannelManager::with_config(config, Some(reloadable_config.clone()));
+
+    // Test that we can get a channel
+    let channel = manager.get("127.0.0.1:0");
+    assert!(channel.is_ok());
+
+    // Test that config is properly set
+    assert_eq!(0, reloadable_config.get_version());
+    assert!(reloadable_config.get_config().is_some());
+}

src/common/macro/src/row/schema.rs

@@ -12,8 +12,9 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
+use greptime_proto::v1::ColumnDataTypeExtension;
 use greptime_proto::v1::column_data_type_extension::TypeExt;
-use proc_macro2::TokenStream as TokenStream2;
+use proc_macro2::{Span, TokenStream as TokenStream2};
 use quote::quote;
 use syn::spanned::Spanned;
 use syn::{DeriveInput, Result};
@@ -69,57 +70,7 @@ fn impl_schema_method(fields: &[ParsedField<'_>]) -> Result<TokenStream2> {
             let semantic_type_val = convert_semantic_type_to_proto_semantic_type(column_attribute.semantic_type) as i32;
             let semantic_type = syn::LitInt::new(&semantic_type_val.to_string(), ident.span());
             let extension = match extension {
-                Some(ext) => {
-                    match ext.type_ext {
-                        Some(TypeExt::DecimalType(ext)) => {
-                            let precision = syn::LitInt::new(&ext.precision.to_string(), ident.span());
-                            let scale = syn::LitInt::new(&ext.scale.to_string(), ident.span());
-                            quote! {
-                                Some(ColumnDataTypeExtension { type_ext: Some(TypeExt::DecimalType(DecimalTypeExtension { precision: #precision, scale: #scale })) })
-                            }
-                        }
-                        Some(TypeExt::JsonType(ext)) => {
-                            let json_type = syn::LitInt::new(&ext.to_string(), ident.span());
-                            quote! {
-                                Some(ColumnDataTypeExtension { type_ext: Some(TypeExt::JsonType(#json_type)) })
-                            }
-                        }
-                        Some(TypeExt::VectorType(ext)) => {
-                            let dim = syn::LitInt::new(&ext.dim.to_string(), ident.span());
-                            quote! {
-                                Some(ColumnDataTypeExtension { type_ext: Some(TypeExt::VectorType(VectorTypeExtension { dim: #dim })) })
-                            }
-                        }
-                        // TODO(sunng87): revisit all these implementations
-                        Some(TypeExt::ListType(ext)) => {
-                            let item_type = syn::Ident::new(&ext.datatype.to_string(), ident.span());
-                            quote! {
-                                Some(ColumnDataTypeExtension { type_ext: Some(TypeExt::ListType(ListTypeExtension { item_type: #item_type })) })
-                            }
-                        }
-                        Some(TypeExt::StructType(ext)) => {
-                            let fields = ext.fields.iter().map(|field| {
-                                let field_name = syn::Ident::new(&field.name.clone(), ident.span());
-                                let field_type = syn::Ident::new(&field.datatype.to_string(), ident.span());
-                                quote! {
-                                    StructField { name: #field_name, type_: #field_type }
-                                }
-                            }).collect::<Vec<_>>();
-                            quote! {
-                                Some(ColumnDataTypeExtension { type_ext: Some(TypeExt::StructType(StructTypeExtension { fields: [#(#fields),*] })) })
-                            }
-                        }
-                        Some(TypeExt::JsonNativeType(ext)) => {
-                            let inner = syn::Ident::new(&ext.datatype.to_string(), ident.span());
-                            quote! {
-                                Some(ColumnDataTypeExtension { type_ext: Some(TypeExt::JsonNativeType(JsonNativeTypeExtension { datatype: #inner })) })
-                            }
-                        }
-                        None => {
-                            quote! { None }
-                        }
-                    }
-                }
+                Some(ext) => column_data_type_extension_to_tokens(&ext, ident.span()),
                 None => quote! { None },
             };
 
@@ -141,3 +92,125 @@ fn impl_schema_method(fields: &[ParsedField<'_>]) -> Result<TokenStream2> {
         }
     })
 }
+
+fn column_data_type_extension_to_tokens(
+    extension: &ColumnDataTypeExtension,
+    span: Span,
+) -> TokenStream2 {
+    match extension.type_ext.as_ref() {
+        Some(TypeExt::DecimalType(ext)) => {
+            let precision = syn::LitInt::new(&ext.precision.to_string(), span);
+            let scale = syn::LitInt::new(&ext.scale.to_string(), span);
+            quote! {
+                Some(ColumnDataTypeExtension {
+                    type_ext: Some(TypeExt::DecimalType(DecimalTypeExtension {
+                        precision: #precision,
+                        scale: #scale,
+                    })),
+                })
+            }
+        }
+        Some(TypeExt::JsonType(ext)) => {
+            let json_type = syn::LitInt::new(&ext.to_string(), span);
+            quote! {
+                Some(ColumnDataTypeExtension {
+                    type_ext: Some(TypeExt::JsonType(#json_type)),
+                })
+            }
+        }
+        Some(TypeExt::VectorType(ext)) => {
+            let dim = syn::LitInt::new(&ext.dim.to_string(), span);
+            quote! {
+                Some(ColumnDataTypeExtension {
+                    type_ext: Some(TypeExt::VectorType(VectorTypeExtension { dim: #dim })),
+                })
+            }
+        }
+        Some(TypeExt::ListType(ext)) => {
+            let datatype = syn::LitInt::new(&ext.datatype.to_string(), span);
+            let datatype_extension = ext
+                .datatype_extension
+                .as_deref()
+                .map(|ext| column_data_type_extension_to_tokens(ext, span))
+                .unwrap_or_else(|| quote! { None });
+            quote! {
+                Some(ColumnDataTypeExtension {
+                    type_ext: Some(TypeExt::ListType(Box::new(ListTypeExtension {
+                        datatype: #datatype,
+                        datatype_extension: #datatype_extension,
+                    }))),
+                })
+            }
+        }
+        Some(TypeExt::StructType(ext)) => {
+            let fields = ext.fields.iter().map(|field| {
+                let field_name = &field.name;
+                let datatype = syn::LitInt::new(&field.datatype.to_string(), span);
+                let datatype_extension = field
+                    .datatype_extension
+                    .as_ref()
+                    .map(|ext| column_data_type_extension_to_tokens(ext, span))
+                    .unwrap_or_else(|| quote! { None });
+                quote! {
+                    greptime_proto::v1::StructField {
+                        name: #field_name.to_string(),
+                        datatype: #datatype,
+                        datatype_extension: #datatype_extension,
+                    }
+                }
+            });
+            quote! {
+                Some(ColumnDataTypeExtension {
+                    type_ext: Some(TypeExt::StructType(StructTypeExtension {
+                        fields: vec![#(#fields),*],
+                    })),
+                })
+            }
+        }
+        Some(TypeExt::JsonNativeType(ext)) => {
+            let inner = syn::LitInt::new(&ext.datatype.to_string(), span);
+            let datatype_extension = ext
+                .datatype_extension
+                .as_deref()
+                .map(|ext| column_data_type_extension_to_tokens(ext, span))
+                .unwrap_or_else(|| quote! { None });
+            quote! {
+                Some(ColumnDataTypeExtension {
+                    type_ext: Some(TypeExt::JsonNativeType(Box::new(
+                        JsonNativeTypeExtension {
+                            datatype: #inner,
+                            datatype_extension: #datatype_extension,
+                        },
+                    ))),
+                })
+            }
+        }
+        Some(TypeExt::DictionaryType(ext)) => {
+            let key_datatype = syn::LitInt::new(&ext.key_datatype.to_string(), span);
+            let value_datatype = syn::LitInt::new(&ext.value_datatype.to_string(), span);
+            let key_datatype_extension = ext
+                .key_datatype_extension
+                .as_deref()
+                .map(|ext| column_data_type_extension_to_tokens(ext, span))
+                .unwrap_or_else(|| quote! { None });
+            let value_datatype_extension = ext
+                .value_datatype_extension
+                .as_deref()
+                .map(|ext| column_data_type_extension_to_tokens(ext, span))
+                .unwrap_or_else(|| quote! { None });
+            quote! {
+                Some(ColumnDataTypeExtension {
+                    type_ext: Some(TypeExt::DictionaryType(Box::new(
+                        DictionaryTypeExtension {
+                            key_datatype: #key_datatype,
+                            key_datatype_extension: #key_datatype_extension,
+                            value_datatype: #value_datatype,
+                            value_datatype_extension: #value_datatype_extension,
+                        },
+                    ))),
+                })
+            }
+        }
+        None => quote! { None },
+    }
+}

src/common/macro/src/row/utils.rs

@@ -309,5 +309,8 @@ pub(crate) fn convert_column_data_type_to_value_data_ident(
         ColumnDataType::Vector => format_ident!("VectorValue"),
         ColumnDataType::List => format_ident!("ListValue"),
         ColumnDataType::Struct => format_ident!("StructValue"),
+        ColumnDataType::Dictionary => {
+            panic!("Dictionary data type is not supported in row macros yet")
+        }
     }
 }

src/common/meta/src/datanode.rs

@@ -132,6 +132,8 @@ pub enum RegionManifestInfo {
     Mito {
         manifest_version: u64,
         flushed_entry_id: u64,
+        /// Number of files removed in the manifest's `removed_files` field.
+        file_removed_cnt: u64,
     },
     Metric {
         data_manifest_version: u64,
@@ -271,9 +273,11 @@ impl From<store_api::region_engine::RegionManifestInfo> for RegionManifestInfo {
             store_api::region_engine::RegionManifestInfo::Mito {
                 manifest_version,
                 flushed_entry_id,
+                file_removed_cnt,
             } => RegionManifestInfo::Mito {
                 manifest_version,
                 flushed_entry_id,
+                file_removed_cnt,
             },
             store_api::region_engine::RegionManifestInfo::Metric {
                 data_manifest_version,

src/common/meta/src/ddl/alter_database.rs

@@ -47,6 +47,9 @@ fn build_new_schema_value(
                     SetDatabaseOption::Ttl(ttl) => {
                         value.ttl = Some(*ttl);
                     }
+                    SetDatabaseOption::Other(key, val) => {
+                        value.extra_options.insert(key.clone(), val.clone());
+                    }
                 }
             }
         }
@@ -54,6 +57,9 @@ fn build_new_schema_value(
             for key in keys.0.iter() {
                 match key {
                     UnsetDatabaseOption::Ttl => value.ttl = None,
+                    UnsetDatabaseOption::Other(key) => {
+                        value.extra_options.remove(key);
+                    }
                 }
             }
         }
@@ -234,4 +240,41 @@ mod tests {
             build_new_schema_value(current_schema_value, &unset_ttl_alter_kind).unwrap();
         assert_eq!(new_schema_value.ttl, None);
     }
+
+    #[test]
+    fn test_build_new_schema_value_with_compaction_options() {
+        let set_compaction = AlterDatabaseKind::SetDatabaseOptions(SetDatabaseOptions(vec![
+            SetDatabaseOption::Other("compaction.type".to_string(), "twcs".to_string()),
+            SetDatabaseOption::Other("compaction.twcs.time_window".to_string(), "1d".to_string()),
+        ]));
+
+        let current_schema_value = SchemaNameValue::default();
+        let new_schema_value =
+            build_new_schema_value(current_schema_value.clone(), &set_compaction).unwrap();
+
+        assert_eq!(
+            new_schema_value.extra_options.get("compaction.type"),
+            Some(&"twcs".to_string())
+        );
+        assert_eq!(
+            new_schema_value
+                .extra_options
+                .get("compaction.twcs.time_window"),
+            Some(&"1d".to_string())
+        );
+
+        let unset_compaction = AlterDatabaseKind::UnsetDatabaseOptions(UnsetDatabaseOptions(vec![
+            UnsetDatabaseOption::Other("compaction.type".to_string()),
+        ]));
+
+        let new_schema_value = build_new_schema_value(new_schema_value, &unset_compaction).unwrap();
+
+        assert_eq!(new_schema_value.extra_options.get("compaction.type"), None);
+        assert_eq!(
+            new_schema_value
+                .extra_options
+                .get("compaction.twcs.time_window"),
+            Some(&"1d".to_string())
+        );
+    }
 }

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

@@ -182,7 +182,7 @@ fn alter_request_handler(_peer: Peer, request: RegionRequest) -> Result<RegionRe
         let region_id = RegionId::from(req.region_id);
         response.extensions.insert(
             MANIFEST_INFO_EXTENSION_KEY.to_string(),
-            RegionManifestInfo::encode_list(&[(region_id, RegionManifestInfo::mito(1, 1))])
+            RegionManifestInfo::encode_list(&[(region_id, RegionManifestInfo::mito(1, 1, 0))])
                 .unwrap(),
         );
         response.extensions.insert(

src/common/meta/src/instruction.rs

@@ -420,20 +420,25 @@ where
 /// Instruction to get file references for specified regions.
 #[derive(Debug, Clone, Serialize, Deserialize, PartialEq, Eq)]
 pub struct GetFileRefs {
-    /// List of region IDs to get file references for.
-    pub region_ids: Vec<RegionId>,
+    /// List of region IDs to get file references from active FileHandles (in-memory).
+    pub query_regions: Vec<RegionId>,
+    /// Mapping from the source region ID (where to read the manifest) to
+    /// the target region IDs (whose file references to look for).
+    /// Key: The region ID of the manifest.
+    /// Value: The list of region IDs to find references for in that manifest.
+    pub related_regions: HashMap<RegionId, Vec<RegionId>>,
 }
 
 impl Display for GetFileRefs {
     fn fmt(&self, f: &mut Formatter<'_>) -> std::fmt::Result {
-        write!(f, "GetFileRefs(region_ids={:?})", self.region_ids)
+        write!(f, "GetFileRefs(region_ids={:?})", self.query_regions)
     }
 }
 
 /// Instruction to trigger garbage collection for a region.
 #[derive(Debug, Clone, Serialize, Deserialize, PartialEq, Eq)]
 pub struct GcRegions {
-    /// The region ID to perform GC on.
+    /// The region ID to perform GC on, only regions that are currently on the given datanode can be garbage collected, regions not on the datanode will report errors.
     pub regions: Vec<RegionId>,
     /// The file references manifest containing temporary file references.
     pub file_refs_manifest: FileRefsManifest,

src/common/meta/src/region_registry.rs

@@ -67,6 +67,7 @@ impl LeaderRegionManifestInfo {
             RegionManifestInfo::Mito {
                 manifest_version,
                 flushed_entry_id,
+                file_removed_cnt: _,
             } => LeaderRegionManifestInfo::Mito {
                 manifest_version,
                 flushed_entry_id,

src/common/meta/src/rpc/ddl.rs

@@ -47,6 +47,7 @@ use serde_with::{DefaultOnNull, serde_as};
 use session::context::{QueryContextBuilder, QueryContextRef};
 use snafu::{OptionExt, ResultExt};
 use table::metadata::{RawTableInfo, TableId};
+use table::requests::validate_database_option;
 use table::table_name::TableName;
 use table::table_reference::TableReference;
 
@@ -1059,14 +1060,21 @@ impl TryFrom<PbOption> for SetDatabaseOption {
     type Error = error::Error;
 
     fn try_from(PbOption { key, value }: PbOption) -> Result<Self> {
-        match key.to_ascii_lowercase().as_str() {
+        let key_lower = key.to_ascii_lowercase();
+        match key_lower.as_str() {
             TTL_KEY => {
                 let ttl = DatabaseTimeToLive::from_humantime_or_str(&value)
                     .map_err(|_| InvalidSetDatabaseOptionSnafu { key, value }.build())?;
 
                 Ok(SetDatabaseOption::Ttl(ttl))
             }
-            _ => InvalidSetDatabaseOptionSnafu { key, value }.fail(),
+            _ => {
+                if validate_database_option(&key_lower) {
+                    Ok(SetDatabaseOption::Other(key_lower, value))
+                } else {
+                    InvalidSetDatabaseOptionSnafu { key, value }.fail()
+                }
+            }
         }
     }
 }
@@ -1074,20 +1082,29 @@ impl TryFrom<PbOption> for SetDatabaseOption {
 #[derive(Debug, PartialEq, Clone, Serialize, Deserialize)]
 pub enum SetDatabaseOption {
     Ttl(DatabaseTimeToLive),
+    Other(String, String),
 }
 
 #[derive(Debug, PartialEq, Clone, Serialize, Deserialize)]
 pub enum UnsetDatabaseOption {
     Ttl,
+    Other(String),
 }
 
 impl TryFrom<&str> for UnsetDatabaseOption {
     type Error = error::Error;
 
     fn try_from(key: &str) -> Result<Self> {
-        match key.to_ascii_lowercase().as_str() {
+        let key_lower = key.to_ascii_lowercase();
+        match key_lower.as_str() {
             TTL_KEY => Ok(UnsetDatabaseOption::Ttl),
-            _ => InvalidUnsetDatabaseOptionSnafu { key }.fail(),
+            _ => {
+                if validate_database_option(&key_lower) {
+                    Ok(UnsetDatabaseOption::Other(key_lower))
+                } else {
+                    InvalidUnsetDatabaseOptionSnafu { key }.fail()
+                }
+            }
         }
     }
 }

src/common/procedure/src/error.rs

@@ -246,6 +246,14 @@ pub enum Error {
         #[snafu(implicit)]
         location: Location,
     },
+
+    #[snafu(display("Loader for {type_name} is not implemented: {reason}"))]
+    ProcedureLoaderNotImplemented {
+        #[snafu(implicit)]
+        location: Location,
+        type_name: String,
+        reason: String,
+    },
 }
 
 pub type Result<T> = std::result::Result<T, Error>;
@@ -266,7 +274,8 @@ impl ErrorExt for Error {
             Error::ToJson { .. }
             | Error::DeleteState { .. }
             | Error::FromJson { .. }
-            | Error::WaitWatcher { .. } => StatusCode::Internal,
+            | Error::WaitWatcher { .. }
+            | Error::ProcedureLoaderNotImplemented { .. } => StatusCode::Internal,
 
             Error::RetryTimesExceeded { .. }
             | Error::RollbackTimesExceeded { .. }

src/common/procedure/src/event.rs

@@ -92,25 +92,96 @@ impl Event for ProcedureEvent {
         schema
     }
 
-    fn extra_row(&self) -> Result<Row> {
-        let error_str = match &self.state {
-            ProcedureState::Failed { error } => format!("{:?}", error),
-            ProcedureState::PrepareRollback { error } => format!("{:?}", error),
-            ProcedureState::RollingBack { error } => format!("{:?}", error),
-            ProcedureState::Retrying { error } => format!("{:?}", error),
-            ProcedureState::Poisoned { error, .. } => format!("{:?}", error),
-            _ => "".to_string(),
-        };
-        let mut row = vec![
-            ValueData::StringValue(self.procedure_id.to_string()).into(),
-            ValueData::StringValue(self.state.as_str_name().to_string()).into(),
-            ValueData::StringValue(error_str).into(),
-        ];
-        row.append(&mut self.internal_event.extra_row()?.values);
-        Ok(Row { values: row })
+    fn extra_rows(&self) -> Result<Vec<Row>> {
+        let mut internal_event_extra_rows = self.internal_event.extra_rows()?;
+        let mut rows = Vec::with_capacity(internal_event_extra_rows.len());
+        for internal_event_extra_row in internal_event_extra_rows.iter_mut() {
+            let error_str = match &self.state {
+                ProcedureState::Failed { error } => format!("{:?}", error),
+                ProcedureState::PrepareRollback { error } => format!("{:?}", error),
+                ProcedureState::RollingBack { error } => format!("{:?}", error),
+                ProcedureState::Retrying { error } => format!("{:?}", error),
+                ProcedureState::Poisoned { error, .. } => format!("{:?}", error),
+                _ => "".to_string(),
+            };
+            let mut values = Vec::with_capacity(3 + internal_event_extra_row.values.len());
+            values.extend([
+                ValueData::StringValue(self.procedure_id.to_string()).into(),
+                ValueData::StringValue(self.state.as_str_name().to_string()).into(),
+                ValueData::StringValue(error_str).into(),
+            ]);
+            values.append(&mut internal_event_extra_row.values);
+            rows.push(Row { values });
+        }
+
+        Ok(rows)
     }
 
     fn as_any(&self) -> &dyn Any {
         self
     }
 }
+
+#[cfg(test)]
+mod tests {
+    use api::v1::value::ValueData;
+    use api::v1::{ColumnDataType, ColumnSchema, Row, SemanticType};
+    use common_event_recorder::Event;
+
+    use crate::{ProcedureEvent, ProcedureId, ProcedureState};
+
+    #[derive(Debug)]
+    struct TestEvent;
+
+    impl Event for TestEvent {
+        fn event_type(&self) -> &str {
+            "test_event"
+        }
+
+        fn extra_schema(&self) -> Vec<ColumnSchema> {
+            vec![ColumnSchema {
+                column_name: "test_event_column".to_string(),
+                datatype: ColumnDataType::String.into(),
+                semantic_type: SemanticType::Field.into(),
+                ..Default::default()
+            }]
+        }
+
+        fn extra_rows(&self) -> common_event_recorder::error::Result<Vec<Row>> {
+            Ok(vec![
+                Row {
+                    values: vec![ValueData::StringValue("test_event1".to_string()).into()],
+                },
+                Row {
+                    values: vec![ValueData::StringValue("test_event2".to_string()).into()],
+                },
+            ])
+        }
+
+        fn as_any(&self) -> &dyn std::any::Any {
+            self
+        }
+    }
+
+    #[test]
+    fn test_procedure_event_extra_rows() {
+        let procedure_event = ProcedureEvent::new(
+            ProcedureId::random(),
+            Box::new(TestEvent {}),
+            ProcedureState::Running,
+        );
+
+        let procedure_event_extra_rows = procedure_event.extra_rows().unwrap();
+        assert_eq!(procedure_event_extra_rows.len(), 2);
+        assert_eq!(procedure_event_extra_rows[0].values.len(), 4);
+        assert_eq!(
+            procedure_event_extra_rows[0].values[3],
+            ValueData::StringValue("test_event1".to_string()).into()
+        );
+        assert_eq!(procedure_event_extra_rows[1].values.len(), 4);
+        assert_eq!(
+            procedure_event_extra_rows[1].values[3],
+            ValueData::StringValue("test_event2".to_string()).into()
+        );
+    }
+}

src/common/query/src/error.rs

@@ -52,9 +52,6 @@ pub enum Error {
         data_type: ArrowDatatype,
     },
 
-    #[snafu(display("Failed to downcast vector: {}", err_msg))]
-    DowncastVector { err_msg: String },
-
     #[snafu(display("Invalid input type: {}", err_msg))]
     InvalidInputType {
         #[snafu(implicit)]
@@ -209,8 +206,7 @@ pub type Result<T> = std::result::Result<T, Error>;
 impl ErrorExt for Error {
     fn status_code(&self) -> StatusCode {
         match self {
-            Error::DowncastVector { .. }
-            | Error::InvalidInputState { .. }
+            Error::InvalidInputState { .. }
             | Error::ToScalarValue { .. }
             | Error::GetScalarVector { .. }
             | Error::ArrowCompute { .. }

src/common/recordbatch/src/adapter.rs

@@ -314,10 +314,10 @@ impl Stream for RecordBatchStreamAdapter {
                         metric_collector.record_batch_metrics,
                     );
                 }
-                Poll::Ready(Some(RecordBatch::try_from_df_record_batch(
+                Poll::Ready(Some(Ok(RecordBatch::from_df_record_batch(
                     self.schema(),
                     df_record_batch,
-                )))
+                ))))
             }
             Poll::Ready(None) => {
                 if let Metrics::Unresolved(df_plan) | Metrics::PartialResolved(df_plan, _) =

src/common/recordbatch/src/error.rs

@@ -133,18 +133,6 @@ pub enum Error {
         source: datatypes::error::Error,
     },
 
-    #[snafu(display(
-        "Failed to downcast vector of type '{:?}' to type '{:?}'",
-        from_type,
-        to_type
-    ))]
-    DowncastVector {
-        from_type: ConcreteDataType,
-        to_type: ConcreteDataType,
-        #[snafu(implicit)]
-        location: Location,
-    },
-
     #[snafu(display("Error occurs when performing arrow computation"))]
     ArrowCompute {
         #[snafu(source)]
@@ -217,8 +205,6 @@ impl ErrorExt for Error {
             | Error::PhysicalExpr { .. }
             | Error::RecordBatchSliceIndexOverflow { .. } => StatusCode::Internal,
 
-            Error::DowncastVector { .. } => StatusCode::Unexpected,
-
             Error::PollStream { .. } => StatusCode::EngineExecuteQuery,
 
             Error::ArrowCompute { .. } => StatusCode::IllegalState,

src/common/recordbatch/src/lib.rs

@@ -30,19 +30,20 @@ use adapter::RecordBatchMetrics;
 use arc_swap::ArcSwapOption;
 use common_base::readable_size::ReadableSize;
 pub use datafusion::physical_plan::SendableRecordBatchStream as DfSendableRecordBatchStream;
+use datatypes::arrow::array::{ArrayRef, AsArray, StringBuilder};
 use datatypes::arrow::compute::SortOptions;
 pub use datatypes::arrow::record_batch::RecordBatch as DfRecordBatch;
 use datatypes::arrow::util::pretty;
 use datatypes::prelude::{ConcreteDataType, VectorRef};
-use datatypes::scalars::{ScalarVector, ScalarVectorBuilder};
 use datatypes::schema::{ColumnSchema, Schema, SchemaRef};
 use datatypes::types::{JsonFormat, jsonb_to_string};
-use datatypes::vectors::{BinaryVector, StringVectorBuilder};
 use error::Result;
 use futures::task::{Context, Poll};
 use futures::{Stream, TryStreamExt};
 pub use recordbatch::RecordBatch;
-use snafu::{OptionExt, ResultExt, ensure};
+use snafu::{ResultExt, ensure};
+
+use crate::error::NewDfRecordBatchSnafu;
 
 pub trait RecordBatchStream: Stream<Item = Result<RecordBatch>> {
     fn name(&self) -> &str {
@@ -92,20 +93,14 @@ pub fn map_json_type_to_string(
     mapped_schema: &SchemaRef,
 ) -> Result<RecordBatch> {
     let mut vectors = Vec::with_capacity(original_schema.column_schemas().len());
-    for (vector, schema) in batch.columns.iter().zip(original_schema.column_schemas()) {
+    for (vector, schema) in batch.columns().iter().zip(original_schema.column_schemas()) {
         if let ConcreteDataType::Json(j) = &schema.data_type {
             if matches!(&j.format, JsonFormat::Jsonb) {
-                let mut string_vector_builder = StringVectorBuilder::with_capacity(vector.len());
-                let binary_vector = vector
-                    .as_any()
-                    .downcast_ref::<BinaryVector>()
-                    .with_context(|| error::DowncastVectorSnafu {
-                        from_type: schema.data_type.clone(),
-                        to_type: ConcreteDataType::binary_datatype(),
-                    })?;
-                for value in binary_vector.iter_data() {
+                let mut string_vector_builder = StringBuilder::new();
+                let binary_vector = vector.as_binary::<i32>();
+                for value in binary_vector.iter() {
                     let Some(value) = value else {
-                        string_vector_builder.push(None);
+                        string_vector_builder.append_null();
                         continue;
                     };
                     let string_value =
@@ -113,11 +108,11 @@ pub fn map_json_type_to_string(
                             from_type: schema.data_type.clone(),
                             to_type: ConcreteDataType::string_datatype(),
                         })?;
-                    string_vector_builder.push(Some(string_value.as_str()));
+                    string_vector_builder.append_value(string_value);
                 }
 
                 let string_vector = string_vector_builder.finish();
-                vectors.push(Arc::new(string_vector) as VectorRef);
+                vectors.push(Arc::new(string_vector) as ArrayRef);
             } else {
                 vectors.push(vector.clone());
             }
@@ -126,7 +121,15 @@ pub fn map_json_type_to_string(
         }
     }
 
-    RecordBatch::new(mapped_schema.clone(), vectors)
+    let record_batch = datatypes::arrow::record_batch::RecordBatch::try_new(
+        mapped_schema.arrow_schema().clone(),
+        vectors,
+    )
+    .context(NewDfRecordBatchSnafu)?;
+    Ok(RecordBatch::from_df_record_batch(
+        mapped_schema.clone(),
+        record_batch,
+    ))
 }
 
 /// Maps the json type to string in the schema.
@@ -755,11 +758,7 @@ impl Stream for MemoryTrackedStream {
     fn poll_next(mut self: Pin<&mut Self>, cx: &mut Context<'_>) -> Poll<Option<Self::Item>> {
         match Pin::new(&mut self.inner).poll_next(cx) {
             Poll::Ready(Some(Ok(batch))) => {
-                let additional = batch
-                    .columns()
-                    .iter()
-                    .map(|c| c.memory_size())
-                    .sum::<usize>();
+                let additional = batch.buffer_memory_size();
 
                 if let Err(e) = self.permit.track(additional, self.total_tracked) {
                     return Poll::Ready(Some(Err(e)));

src/common/recordbatch/src/recordbatch.rs

@@ -20,7 +20,7 @@ use datafusion::arrow::util::pretty::pretty_format_batches;
 use datafusion_common::arrow::array::ArrayRef;
 use datafusion_common::arrow::compute;
 use datafusion_common::arrow::datatypes::{DataType as ArrowDataType, SchemaRef as ArrowSchemaRef};
-use datatypes::arrow::array::RecordBatchOptions;
+use datatypes::arrow::array::{Array, AsArray, RecordBatchOptions};
 use datatypes::prelude::DataType;
 use datatypes::schema::SchemaRef;
 use datatypes::vectors::{Helper, VectorRef};
@@ -30,15 +30,14 @@ use snafu::{OptionExt, ResultExt, ensure};
 
 use crate::DfRecordBatch;
 use crate::error::{
-    self, ArrowComputeSnafu, CastVectorSnafu, ColumnNotExistsSnafu, DataTypesSnafu,
-    ProjectArrowRecordBatchSnafu, Result,
+    self, ArrowComputeSnafu, ColumnNotExistsSnafu, DataTypesSnafu, ProjectArrowRecordBatchSnafu,
+    Result,
 };
 
 /// A two-dimensional batch of column-oriented data with a defined schema.
 #[derive(Clone, Debug, PartialEq)]
 pub struct RecordBatch {
     pub schema: SchemaRef,
-    pub columns: Vec<VectorRef>,
     df_record_batch: DfRecordBatch,
 }
 
@@ -65,7 +64,6 @@ impl RecordBatch {
 
         Ok(RecordBatch {
             schema,
-            columns,
             df_record_batch,
         })
     }
@@ -91,14 +89,8 @@ impl RecordBatch {
     /// Create an empty [`RecordBatch`] from `schema`.
     pub fn new_empty(schema: SchemaRef) -> RecordBatch {
         let df_record_batch = DfRecordBatch::new_empty(schema.arrow_schema().clone());
-        let columns = schema
-            .column_schemas()
-            .iter()
-            .map(|col| col.data_type.create_mutable_vector(0).to_vector())
-            .collect();
         RecordBatch {
             schema,
-            columns,
             df_record_batch,
         }
     }
@@ -113,17 +105,12 @@ impl RecordBatch {
         .context(error::NewDfRecordBatchSnafu)?;
         Ok(RecordBatch {
             schema,
-            columns: vec![],
             df_record_batch,
         })
     }
 
     pub fn try_project(&self, indices: &[usize]) -> Result<Self> {
         let schema = Arc::new(self.schema.try_project(indices).context(DataTypesSnafu)?);
-        let mut columns = Vec::with_capacity(indices.len());
-        for index in indices {
-            columns.push(self.columns[*index].clone());
-        }
         let df_record_batch = self.df_record_batch.project(indices).with_context(|_| {
             ProjectArrowRecordBatchSnafu {
                 schema: self.schema.clone(),
@@ -133,7 +120,6 @@ impl RecordBatch {
 
         Ok(Self {
             schema,
-            columns,
             df_record_batch,
         })
     }
@@ -141,21 +127,11 @@ impl RecordBatch {
     /// Create a new [`RecordBatch`] from `schema` and `df_record_batch`.
     ///
     /// This method doesn't check the schema.
-    pub fn try_from_df_record_batch(
-        schema: SchemaRef,
-        df_record_batch: DfRecordBatch,
-    ) -> Result<RecordBatch> {
-        let columns = df_record_batch
-            .columns()
-            .iter()
-            .map(|c| Helper::try_into_vector(c.clone()).context(error::DataTypesSnafu))
-            .collect::<Result<Vec<_>>>()?;
-
-        Ok(RecordBatch {
+    pub fn from_df_record_batch(schema: SchemaRef, df_record_batch: DfRecordBatch) -> RecordBatch {
+        RecordBatch {
             schema,
-            columns,
             df_record_batch,
-        })
+        }
     }
 
     #[inline]
@@ -169,23 +145,22 @@ impl RecordBatch {
     }
 
     #[inline]
-    pub fn columns(&self) -> &[VectorRef] {
-        &self.columns
+    pub fn columns(&self) -> &[ArrayRef] {
+        self.df_record_batch.columns()
     }
 
     #[inline]
-    pub fn column(&self, idx: usize) -> &VectorRef {
-        &self.columns[idx]
+    pub fn column(&self, idx: usize) -> &ArrayRef {
+        self.df_record_batch.column(idx)
     }
 
-    pub fn column_by_name(&self, name: &str) -> Option<&VectorRef> {
-        let idx = self.schema.column_index_by_name(name)?;
-        Some(&self.columns[idx])
+    pub fn column_by_name(&self, name: &str) -> Option<&ArrayRef> {
+        self.df_record_batch.column_by_name(name)
     }
 
     #[inline]
     pub fn num_columns(&self) -> usize {
-        self.columns.len()
+        self.df_record_batch.num_columns()
     }
 
     #[inline]
@@ -201,9 +176,14 @@ impl RecordBatch {
         let mut vectors = HashMap::with_capacity(self.num_columns());
 
         // column schemas in recordbatch must match its vectors, otherwise it's corrupted
-        for (vector_schema, vector) in self.schema.column_schemas().iter().zip(self.columns.iter())
+        for (field, array) in self
+            .df_record_batch
+            .schema()
+            .fields()
+            .iter()
+            .zip(self.df_record_batch.columns().iter())
         {
-            let column_name = &vector_schema.name;
+            let column_name = field.name();
             let column_schema =
                 table_schema
                     .column_schema_by_name(column_name)
@@ -211,15 +191,12 @@ impl RecordBatch {
                         table_name,
                         column_name,
                     })?;
-            let vector = if vector_schema.data_type != column_schema.data_type {
-                vector
-                    .cast(&column_schema.data_type)
-                    .with_context(|_| CastVectorSnafu {
-                        from_type: vector.data_type(),
-                        to_type: column_schema.data_type.clone(),
-                    })?
+            let vector = if field.data_type() != &column_schema.data_type.as_arrow_type() {
+                let array = compute::cast(array, &column_schema.data_type.as_arrow_type())
+                    .context(ArrowComputeSnafu)?;
+                Helper::try_into_vector(array).context(DataTypesSnafu)?
             } else {
-                vector.clone()
+                Helper::try_into_vector(array).context(DataTypesSnafu)?
             };
 
             let _ = vectors.insert(column_name.clone(), vector);
@@ -244,8 +221,69 @@ impl RecordBatch {
                 visit_index: offset + len
             }
         );
-        let columns = self.columns.iter().map(|vector| vector.slice(offset, len));
-        RecordBatch::new(self.schema.clone(), columns)
+        let sliced = self.df_record_batch.slice(offset, len);
+        Ok(RecordBatch::from_df_record_batch(
+            self.schema.clone(),
+            sliced,
+        ))
+    }
+
+    /// Returns the total number of bytes of memory pointed to by the arrays in this `RecordBatch`.
+    ///
+    /// The buffers store bytes in the Arrow memory format, and include the data as well as the validity map.
+    /// Note that this does not always correspond to the exact memory usage of an array,
+    /// since multiple arrays can share the same buffers or slices thereof.
+    pub fn buffer_memory_size(&self) -> usize {
+        self.df_record_batch
+            .columns()
+            .iter()
+            .map(|array| array.get_buffer_memory_size())
+            .sum()
+    }
+
+    /// Iterate the values as strings in the column at index `i`.
+    ///
+    /// Note that if the underlying array is not a valid GreptimeDB vector, an empty iterator is
+    /// returned.
+    ///
+    /// # Panics
+    /// if index `i` is out of bound.
+    pub fn iter_column_as_string(&self, i: usize) -> Box<dyn Iterator<Item = Option<String>> + '_> {
+        macro_rules! iter {
+            ($column: ident) => {
+                Box::new(
+                    (0..$column.len())
+                        .map(|i| $column.is_valid(i).then(|| $column.value(i).to_string())),
+                )
+            };
+        }
+
+        let column = self.df_record_batch.column(i);
+        match column.data_type() {
+            ArrowDataType::Utf8 => {
+                let column = column.as_string::<i32>();
+                let iter = iter!(column);
+                iter as _
+            }
+            ArrowDataType::LargeUtf8 => {
+                let column = column.as_string::<i64>();
+                iter!(column)
+            }
+            ArrowDataType::Utf8View => {
+                let column = column.as_string_view();
+                iter!(column)
+            }
+            _ => {
+                if let Ok(column) = Helper::try_into_vector(column) {
+                    Box::new(
+                        (0..column.len())
+                            .map(move |i| (!column.is_null(i)).then(|| column.get(i).to_string())),
+                    )
+                } else {
+                    Box::new(std::iter::empty())
+                }
+            }
+        }
     }
 }
 
@@ -259,8 +297,9 @@ impl Serialize for RecordBatch {
         let mut s = serializer.serialize_struct("record", 2)?;
         s.serialize_field("schema", &**self.schema.arrow_schema())?;
 
-        let vec = self
-            .columns
+        let columns = self.df_record_batch.columns();
+        let columns = Helper::try_into_vectors(columns).map_err(Error::custom)?;
+        let vec = columns
             .iter()
             .map(|c| c.serialize_to_json())
             .collect::<std::result::Result<Vec<_>, _>>()
@@ -278,27 +317,14 @@ pub fn merge_record_batches(schema: SchemaRef, batches: &[RecordBatch]) -> Resul
         return Ok(RecordBatch::new_empty(schema));
     }
 
-    let n_rows = batches.iter().map(|b| b.num_rows()).sum();
-    let n_columns = schema.num_columns();
-    // Collect arrays from each batch
-    let mut merged_columns = Vec::with_capacity(n_columns);
-
-    for col_idx in 0..n_columns {
-        let mut acc = schema.column_schemas()[col_idx]
-            .data_type
-            .create_mutable_vector(n_rows);
-
-        for batch in batches {
-            let column = batch.column(col_idx);
-            acc.extend_slice_of(column.as_ref(), 0, column.len())
-                .context(error::DataTypesSnafu)?;
-        }
-
-        merged_columns.push(acc.to_vector());
-    }
+    let record_batch = compute::concat_batches(
+        schema.arrow_schema(),
+        batches.iter().map(|x| x.df_record_batch()),
+    )
+    .context(ArrowComputeSnafu)?;
 
     // Create a new RecordBatch with merged columns
-    RecordBatch::new(schema, merged_columns)
+    Ok(RecordBatch::from_df_record_batch(schema, record_batch))
 }
 
 #[cfg(test)]
@@ -326,21 +352,21 @@ mod tests {
         let c2 = Arc::new(UInt32Vector::from_slice([4, 5, 6]));
         let columns: Vec<VectorRef> = vec![c1, c2];
 
+        let expected = vec![
+            Arc::new(UInt32Array::from_iter_values([1, 2, 3])) as ArrayRef,
+            Arc::new(UInt32Array::from_iter_values([4, 5, 6])),
+        ];
+
         let batch = RecordBatch::new(schema.clone(), columns.clone()).unwrap();
         assert_eq!(3, batch.num_rows());
-        assert_eq!(&columns, batch.columns());
-        for (i, expect) in columns.iter().enumerate().take(batch.num_columns()) {
-            let column = batch.column(i);
-            assert_eq!(expect, column);
-        }
+        assert_eq!(expected, batch.df_record_batch().columns());
         assert_eq!(schema, batch.schema);
 
-        assert_eq!(columns[0], *batch.column_by_name("c1").unwrap());
-        assert_eq!(columns[1], *batch.column_by_name("c2").unwrap());
+        assert_eq!(&expected[0], batch.column_by_name("c1").unwrap());
+        assert_eq!(&expected[1], batch.column_by_name("c2").unwrap());
         assert!(batch.column_by_name("c3").is_none());
 
-        let converted =
-            RecordBatch::try_from_df_record_batch(schema, batch.df_record_batch().clone()).unwrap();
+        let converted = RecordBatch::from_df_record_batch(schema, batch.df_record_batch().clone());
         assert_eq!(batch, converted);
         assert_eq!(*batch.df_record_batch(), converted.into_df_record_batch());
     }
@@ -385,12 +411,12 @@ mod tests {
         let recordbatch = recordbatch.slice(1, 2).expect("recordbatch slice");
 
         let expected = &UInt32Array::from_iter_values([2u32, 3]);
-        let array = recordbatch.column(0).to_arrow_array();
+        let array = recordbatch.column(0);
         let actual = array.as_primitive::<UInt32Type>();
         assert_eq!(expected, actual);
 
         let expected = &StringArray::from(vec!["hello", "greptime"]);
-        let array = recordbatch.column(1).to_arrow_array();
+        let array = recordbatch.column(1);
         let actual = array.as_string::<i32>();
         assert_eq!(expected, actual);
 

src/common/sql/src/convert.rs

@@ -211,8 +211,7 @@ pub fn sql_value_to_value(
             | Value::Duration(_)
             | Value::IntervalYearMonth(_)
             | Value::IntervalDayTime(_)
-            | Value::IntervalMonthDayNano(_)
-            | Value::Json(_) => match unary_op {
+            | Value::IntervalMonthDayNano(_) => match unary_op {
                 UnaryOperator::Plus => {}
                 UnaryOperator::Minus => {
                     value = value
@@ -222,7 +221,11 @@ pub fn sql_value_to_value(
                 _ => return InvalidUnaryOpSnafu { unary_op, value }.fail(),
             },
 
-            Value::String(_) | Value::Binary(_) | Value::List(_) | Value::Struct(_) => {
+            Value::String(_)
+            | Value::Binary(_)
+            | Value::List(_)
+            | Value::Struct(_)
+            | Value::Json(_) => {
                 return InvalidUnaryOpSnafu { unary_op, value }.fail();
             }
         }

src/common/telemetry/Cargo.toml

@@ -35,5 +35,5 @@ tokio.workspace = true
 tracing = "0.1"
 tracing-appender.workspace = true
 tracing-log = "0.2"
-tracing-opentelemetry = "0.31.0"
+tracing-opentelemetry.workspace = true
 tracing-subscriber.workspace = true

src/common/telemetry/src/lib.rs

@@ -21,7 +21,10 @@ mod panic_hook;
 pub mod tracing_context;
 mod tracing_sampler;
 
-pub use logging::{RELOAD_HANDLE, init_default_ut_logging, init_global_logging};
+pub use logging::{
+    LOG_RELOAD_HANDLE, TRACE_RELOAD_HANDLE, get_or_init_tracer, init_default_ut_logging,
+    init_global_logging,
+};
 pub use metric::dump_metrics;
 pub use panic_hook::set_panic_hook;
 pub use {common_error, tracing, tracing_subscriber};

src/common/telemetry/src/logging.rs

@@ -16,7 +16,7 @@
 use std::collections::HashMap;
 use std::env;
 use std::io::IsTerminal;
-use std::sync::{Arc, Mutex, Once};
+use std::sync::{Arc, Mutex, Once, RwLock};
 use std::time::Duration;
 
 use common_base::serde::empty_string_as_default;
@@ -25,15 +25,17 @@ use opentelemetry::trace::TracerProvider;
 use opentelemetry::{KeyValue, global};
 use opentelemetry_otlp::{Protocol, SpanExporter, WithExportConfig, WithHttpConfig};
 use opentelemetry_sdk::propagation::TraceContextPropagator;
-use opentelemetry_sdk::trace::Sampler;
+use opentelemetry_sdk::trace::{Sampler, Tracer};
 use opentelemetry_semantic_conventions::resource;
 use serde::{Deserialize, Serialize};
+use tracing::callsite;
+use tracing::metadata::LevelFilter;
 use tracing_appender::non_blocking::WorkerGuard;
 use tracing_appender::rolling::{RollingFileAppender, Rotation};
 use tracing_log::LogTracer;
 use tracing_subscriber::filter::{FilterFn, Targets};
 use tracing_subscriber::fmt::Layer;
-use tracing_subscriber::layer::SubscriberExt;
+use tracing_subscriber::layer::{Layered, SubscriberExt};
 use tracing_subscriber::prelude::*;
 use tracing_subscriber::{EnvFilter, Registry, filter};
 
@@ -48,10 +50,192 @@ pub const DEFAULT_OTLP_HTTP_ENDPOINT: &str = "http://localhost:4318/v1/traces";
 /// The default logs directory.
 pub const DEFAULT_LOGGING_DIR: &str = "logs";
 
-// Handle for reloading log level
-pub static RELOAD_HANDLE: OnceCell<tracing_subscriber::reload::Handle<Targets, Registry>> =
+/// Handle for reloading log level
+pub static LOG_RELOAD_HANDLE: OnceCell<tracing_subscriber::reload::Handle<Targets, Registry>> =
     OnceCell::new();
 
+type DynSubscriber = Layered<tracing_subscriber::reload::Layer<Targets, Registry>, Registry>;
+type OtelTraceLayer = tracing_opentelemetry::OpenTelemetryLayer<DynSubscriber, Tracer>;
+
+#[derive(Clone)]
+pub struct TraceReloadHandle {
+    inner: Arc<RwLock<Option<OtelTraceLayer>>>,
+}
+
+impl TraceReloadHandle {
+    fn new(inner: Arc<RwLock<Option<OtelTraceLayer>>>) -> Self {
+        Self { inner }
+    }
+
+    pub fn reload(&self, new_layer: Option<OtelTraceLayer>) {
+        let mut guard = self.inner.write().unwrap();
+        *guard = new_layer;
+        drop(guard);
+
+        callsite::rebuild_interest_cache();
+    }
+}
+
+/// A tracing layer that can be dynamically reloaded.
+///
+/// Mostly copied from [`tracing_subscriber::reload::Layer`].
+struct TraceLayer {
+    inner: Arc<RwLock<Option<OtelTraceLayer>>>,
+}
+
+impl TraceLayer {
+    fn new(initial: Option<OtelTraceLayer>) -> (Self, TraceReloadHandle) {
+        let inner = Arc::new(RwLock::new(initial));
+        (
+            Self {
+                inner: inner.clone(),
+            },
+            TraceReloadHandle::new(inner),
+        )
+    }
+
+    fn with_layer<R>(&self, f: impl FnOnce(&OtelTraceLayer) -> R) -> Option<R> {
+        self.inner
+            .read()
+            .ok()
+            .and_then(|guard| guard.as_ref().map(f))
+    }
+
+    fn with_layer_mut<R>(&self, f: impl FnOnce(&mut OtelTraceLayer) -> R) -> Option<R> {
+        self.inner
+            .write()
+            .ok()
+            .and_then(|mut guard| guard.as_mut().map(f))
+    }
+}
+
+impl tracing_subscriber::Layer<DynSubscriber> for TraceLayer {
+    fn on_register_dispatch(&self, subscriber: &tracing::Dispatch) {
+        let _ = self.with_layer(|layer| layer.on_register_dispatch(subscriber));
+    }
+
+    fn on_layer(&mut self, subscriber: &mut DynSubscriber) {
+        let _ = self.with_layer_mut(|layer| layer.on_layer(subscriber));
+    }
+
+    fn register_callsite(
+        &self,
+        metadata: &'static tracing::Metadata<'static>,
+    ) -> tracing::subscriber::Interest {
+        self.with_layer(|layer| layer.register_callsite(metadata))
+            .unwrap_or_else(tracing::subscriber::Interest::always)
+    }
+
+    fn enabled(
+        &self,
+        metadata: &tracing::Metadata<'_>,
+        ctx: tracing_subscriber::layer::Context<'_, DynSubscriber>,
+    ) -> bool {
+        self.with_layer(|layer| layer.enabled(metadata, ctx))
+            .unwrap_or(true)
+    }
+
+    fn on_new_span(
+        &self,
+        attrs: &tracing::span::Attributes<'_>,
+        id: &tracing::span::Id,
+        ctx: tracing_subscriber::layer::Context<'_, DynSubscriber>,
+    ) {
+        let _ = self.with_layer(|layer| layer.on_new_span(attrs, id, ctx));
+    }
+
+    fn max_level_hint(&self) -> Option<LevelFilter> {
+        self.with_layer(|layer| layer.max_level_hint()).flatten()
+    }
+
+    fn on_record(
+        &self,
+        span: &tracing::span::Id,
+        values: &tracing::span::Record<'_>,
+        ctx: tracing_subscriber::layer::Context<'_, DynSubscriber>,
+    ) {
+        let _ = self.with_layer(|layer| layer.on_record(span, values, ctx));
+    }
+
+    fn on_follows_from(
+        &self,
+        span: &tracing::span::Id,
+        follows: &tracing::span::Id,
+        ctx: tracing_subscriber::layer::Context<'_, DynSubscriber>,
+    ) {
+        let _ = self.with_layer(|layer| layer.on_follows_from(span, follows, ctx));
+    }
+
+    fn event_enabled(
+        &self,
+        event: &tracing::Event<'_>,
+        ctx: tracing_subscriber::layer::Context<'_, DynSubscriber>,
+    ) -> bool {
+        self.with_layer(|layer| layer.event_enabled(event, ctx))
+            .unwrap_or(true)
+    }
+
+    fn on_event(
+        &self,
+        event: &tracing::Event<'_>,
+        ctx: tracing_subscriber::layer::Context<'_, DynSubscriber>,
+    ) {
+        let _ = self.with_layer(|layer| layer.on_event(event, ctx));
+    }
+
+    fn on_enter(
+        &self,
+        id: &tracing::span::Id,
+        ctx: tracing_subscriber::layer::Context<'_, DynSubscriber>,
+    ) {
+        let _ = self.with_layer(|layer| layer.on_enter(id, ctx));
+    }
+
+    fn on_exit(
+        &self,
+        id: &tracing::span::Id,
+        ctx: tracing_subscriber::layer::Context<'_, DynSubscriber>,
+    ) {
+        let _ = self.with_layer(|layer| layer.on_exit(id, ctx));
+    }
+
+    fn on_close(
+        &self,
+        id: tracing::span::Id,
+        ctx: tracing_subscriber::layer::Context<'_, DynSubscriber>,
+    ) {
+        let _ = self.with_layer(|layer| layer.on_close(id, ctx));
+    }
+
+    fn on_id_change(
+        &self,
+        old: &tracing::span::Id,
+        new: &tracing::span::Id,
+        ctx: tracing_subscriber::layer::Context<'_, DynSubscriber>,
+    ) {
+        let _ = self.with_layer(|layer| layer.on_id_change(old, new, ctx));
+    }
+
+    unsafe fn downcast_raw(&self, id: std::any::TypeId) -> Option<*const ()> {
+        self.inner.read().ok().and_then(|guard| {
+            guard
+                .as_ref()
+                .and_then(|layer| unsafe { layer.downcast_raw(id) })
+        })
+    }
+}
+
+/// Handle for reloading trace level
+pub static TRACE_RELOAD_HANDLE: OnceCell<TraceReloadHandle> = OnceCell::new();
+
+static TRACER: OnceCell<Mutex<TraceState>> = OnceCell::new();
+
+#[derive(Debug)]
+enum TraceState {
+    Ready(Tracer),
+    Deferred(TraceContext),
+}
+
 /// The logging options that used to initialize the logger.
 #[derive(Clone, Debug, Serialize, Deserialize)]
 #[serde(default)]
@@ -167,6 +351,13 @@ impl PartialEq for LoggingOptions {
 
 impl Eq for LoggingOptions {}
 
+#[derive(Clone, Debug)]
+struct TraceContext {
+    app_name: String,
+    node_id: String,
+    logging_opts: LoggingOptions,
+}
+
 impl Default for LoggingOptions {
     fn default() -> Self {
         Self {
@@ -242,6 +433,7 @@ pub fn init_global_logging(
 ) -> Vec<WorkerGuard> {
     static START: Once = Once::new();
     let mut guards = vec![];
+    let node_id = node_id.unwrap_or_else(|| "none".to_string());
 
     START.call_once(|| {
         // Enable log compatible layer to convert log record to tracing span.
@@ -357,10 +549,37 @@ pub fn init_global_logging(
 
         let (dyn_filter, reload_handle) = tracing_subscriber::reload::Layer::new(filter.clone());
 
-        RELOAD_HANDLE
+        LOG_RELOAD_HANDLE
             .set(reload_handle)
             .expect("reload handle already set, maybe init_global_logging get called twice?");
 
+        let mut initial_tracer = None;
+        let trace_state = if opts.enable_otlp_tracing {
+            let tracer = create_tracer(app_name, &node_id, opts);
+            initial_tracer = Some(tracer.clone());
+            TraceState::Ready(tracer)
+        } else {
+            TraceState::Deferred(TraceContext {
+                app_name: app_name.to_string(),
+                node_id: node_id.clone(),
+                logging_opts: opts.clone(),
+            })
+        };
+
+        TRACER
+            .set(Mutex::new(trace_state))
+            .expect("trace state already initialized");
+
+        let initial_trace_layer = initial_tracer
+            .as_ref()
+            .map(|tracer| tracing_opentelemetry::layer().with_tracer(tracer.clone()));
+
+        let (dyn_trace_layer, trace_reload_handle) = TraceLayer::new(initial_trace_layer);
+
+        TRACE_RELOAD_HANDLE
+            .set(trace_reload_handle)
+            .unwrap_or_else(|_| panic!("failed to set trace reload handle"));
+
         // Must enable 'tokio_unstable' cfg to use this feature.
         // For example: `RUSTFLAGS="--cfg tokio_unstable" cargo run -F common-telemetry/console -- standalone start`
         #[cfg(feature = "tokio-console")]
@@ -383,6 +602,7 @@ pub fn init_global_logging(
 
             Registry::default()
                 .with(dyn_filter)
+                .with(dyn_trace_layer)
                 .with(tokio_console_layer)
                 .with(stdout_logging_layer)
                 .with(file_logging_layer)
@@ -396,53 +616,61 @@ pub fn init_global_logging(
         #[cfg(not(feature = "tokio-console"))]
         let subscriber = Registry::default()
             .with(dyn_filter)
+            .with(dyn_trace_layer)
             .with(stdout_logging_layer)
             .with(file_logging_layer)
             .with(err_file_logging_layer)
             .with(slow_query_logging_layer);
 
-        if opts.enable_otlp_tracing {
-            global::set_text_map_propagator(TraceContextPropagator::new());
+        global::set_text_map_propagator(TraceContextPropagator::new());
 
-            let sampler = opts
-                .tracing_sample_ratio
-                .as_ref()
-                .map(create_sampler)
-                .map(Sampler::ParentBased)
-                .unwrap_or(Sampler::ParentBased(Box::new(Sampler::AlwaysOn)));
-
-            let provider = opentelemetry_sdk::trace::SdkTracerProvider::builder()
-                .with_batch_exporter(build_otlp_exporter(opts))
-                .with_sampler(sampler)
-                .with_resource(
-                    opentelemetry_sdk::Resource::builder_empty()
-                        .with_attributes([
-                            KeyValue::new(resource::SERVICE_NAME, app_name.to_string()),
-                            KeyValue::new(
-                                resource::SERVICE_INSTANCE_ID,
-                                node_id.unwrap_or("none".to_string()),
-                            ),
-                            KeyValue::new(resource::SERVICE_VERSION, common_version::version()),
-                            KeyValue::new(resource::PROCESS_PID, std::process::id().to_string()),
-                        ])
-                        .build(),
-                )
-                .build();
-            let tracer = provider.tracer("greptimedb");
-
-            tracing::subscriber::set_global_default(
-                subscriber.with(tracing_opentelemetry::layer().with_tracer(tracer)),
-            )
+        tracing::subscriber::set_global_default(subscriber)
             .expect("error setting global tracing subscriber");
-        } else {
-            tracing::subscriber::set_global_default(subscriber)
-                .expect("error setting global tracing subscriber");
-        }
     });
 
     guards
 }
 
+fn create_tracer(app_name: &str, node_id: &str, opts: &LoggingOptions) -> Tracer {
+    let sampler = opts
+        .tracing_sample_ratio
+        .as_ref()
+        .map(create_sampler)
+        .map(Sampler::ParentBased)
+        .unwrap_or(Sampler::ParentBased(Box::new(Sampler::AlwaysOn)));
+
+    let resource = opentelemetry_sdk::Resource::builder_empty()
+        .with_attributes([
+            KeyValue::new(resource::SERVICE_NAME, app_name.to_string()),
+            KeyValue::new(resource::SERVICE_INSTANCE_ID, node_id.to_string()),
+            KeyValue::new(resource::SERVICE_VERSION, common_version::version()),
+            KeyValue::new(resource::PROCESS_PID, std::process::id().to_string()),
+        ])
+        .build();
+
+    opentelemetry_sdk::trace::SdkTracerProvider::builder()
+        .with_batch_exporter(build_otlp_exporter(opts))
+        .with_sampler(sampler)
+        .with_resource(resource)
+        .build()
+        .tracer("greptimedb")
+}
+
+/// Ensure that the OTLP tracer has been constructed, building it lazily if needed.
+pub fn get_or_init_tracer() -> Result<Tracer, &'static str> {
+    let state = TRACER.get().ok_or("trace state is not initialized")?;
+    let mut guard = state.lock().expect("trace state lock poisoned");
+
+    match &mut *guard {
+        TraceState::Ready(tracer) => Ok(tracer.clone()),
+        TraceState::Deferred(context) => {
+            let tracer = create_tracer(&context.app_name, &context.node_id, &context.logging_opts);
+            *guard = TraceState::Ready(tracer.clone());
+            Ok(tracer)
+        }
+    }
+}
+
 fn build_otlp_exporter(opts: &LoggingOptions) -> SpanExporter {
     let protocol = opts
         .otlp_export_protocol

src/datanode/src/config.rs

@@ -28,7 +28,6 @@ use mito2::config::MitoConfig;
 pub(crate) use object_store::config::ObjectStoreConfig;
 use query::options::QueryOptions;
 use serde::{Deserialize, Serialize};
-use servers::export_metrics::ExportMetricsOption;
 use servers::grpc::GrpcOptions;
 use servers::heartbeat_options::HeartbeatOptions;
 use servers::http::HttpOptions;
@@ -82,7 +81,6 @@ pub struct DatanodeOptions {
     pub region_engine: Vec<RegionEngineConfig>,
     pub logging: LoggingOptions,
     pub enable_telemetry: bool,
-    pub export_metrics: ExportMetricsOption,
     pub tracing: TracingOptions,
     pub query: QueryOptions,
     pub memory: MemoryOptions,
@@ -138,7 +136,6 @@ impl Default for DatanodeOptions {
             logging: LoggingOptions::default(),
             heartbeat: HeartbeatOptions::datanode_default(),
             enable_telemetry: true,
-            export_metrics: ExportMetricsOption::default(),
             tracing: TracingOptions::default(),
             query: QueryOptions::default(),
             memory: MemoryOptions::default(),

src/datanode/src/datanode.rs

@@ -48,7 +48,6 @@ use object_store::manager::{ObjectStoreManager, ObjectStoreManagerRef};
 use object_store::util::normalize_dir;
 use query::QueryEngineFactory;
 use query::dummy_catalog::{DummyCatalogManager, TableProviderFactoryRef};
-use servers::export_metrics::ExportMetricsTask;
 use servers::server::ServerHandlers;
 use snafu::{OptionExt, ResultExt, ensure};
 use store_api::path_utils::WAL_DIR;
@@ -84,7 +83,6 @@ pub struct Datanode {
     greptimedb_telemetry_task: Arc<GreptimeDBTelemetryTask>,
     leases_notifier: Option<Arc<Notify>>,
     plugins: Plugins,
-    export_metrics_task: Option<ExportMetricsTask>,
 }
 
 impl Datanode {
@@ -96,10 +94,6 @@ impl Datanode {
 
         self.start_telemetry();
 
-        if let Some(t) = self.export_metrics_task.as_ref() {
-            t.start(None).context(StartServerSnafu)?
-        }
-
         self.services.start_all().await.context(StartServerSnafu)
     }
 
@@ -319,10 +313,6 @@ impl DatanodeBuilder {
             None
         };
 
-        let export_metrics_task =
-            ExportMetricsTask::try_new(&self.opts.export_metrics, Some(&self.plugins))
-                .context(StartServerSnafu)?;
-
         Ok(Datanode {
             services: ServerHandlers::default(),
             heartbeat_task,
@@ -331,7 +321,6 @@ impl DatanodeBuilder {
             region_event_receiver,
             leases_notifier,
             plugins: self.plugins.clone(),
-            export_metrics_task,
         })
     }
 

src/datanode/src/heartbeat/handler/file_ref.rs

@@ -39,9 +39,8 @@ impl InstructionHandler for GetFileRefsHandler {
                 error: Some("MitoEngine not found".to_string()),
             }));
         };
-
         match mito_engine
-            .get_snapshot_of_unmanifested_refs(get_file_refs.region_ids)
+            .get_snapshot_of_file_refs(get_file_refs.query_regions, get_file_refs.related_regions)
             .await
         {
             Ok(all_file_refs) => {

src/datanode/src/heartbeat/handler/gc_worker.rs

@@ -15,7 +15,7 @@
 use common_meta::instruction::{GcRegions, GcRegionsReply, InstructionReply};
 use common_telemetry::{debug, warn};
 use mito2::gc::LocalGcWorker;
-use snafu::{OptionExt, ResultExt};
+use snafu::{OptionExt, ResultExt, ensure};
 use store_api::storage::{FileRefsManifest, RegionId};
 
 use crate::error::{GcMitoEngineSnafu, InvalidGcArgsSnafu, Result, UnexpectedSnafu};
@@ -35,20 +35,6 @@ impl InstructionHandler for GcRegionsHandler {
         let region_ids = gc_regions.regions.clone();
         debug!("Received gc regions instruction: {:?}", region_ids);
 
-        let is_same_table = region_ids.windows(2).all(|w| {
-            let t1 = w[0].table_id();
-            let t2 = w[1].table_id();
-            t1 == t2
-        });
-        if !is_same_table {
-            return Some(InstructionReply::GcRegions(GcRegionsReply {
-                result: Err(format!(
-                    "Regions to GC should belong to the same table, found: {:?}",
-                    region_ids
-                )),
-            }));
-        }
-
         let (region_id, gc_worker) = match self
             .create_gc_worker(
                 ctx,
@@ -103,6 +89,8 @@ impl InstructionHandler for GcRegionsHandler {
 }
 
 impl GcRegionsHandler {
+    /// Create a GC worker for the given region IDs.
+    /// Return the first region ID(after sort by given region id) and the GC worker.
     async fn create_gc_worker(
         &self,
         ctx: &HandlerContext,
@@ -112,22 +100,37 @@ impl GcRegionsHandler {
     ) -> Result<(RegionId, LocalGcWorker)> {
         // always use the smallest region id on datanode as the target region id
         region_ids.sort_by_key(|r| r.region_number());
+
         let mito_engine = ctx
             .region_server
             .mito_engine()
             .with_context(|| UnexpectedSnafu {
                 violated: "MitoEngine not found".to_string(),
             })?;
-        let region_id = *region_ids.first().with_context(|| UnexpectedSnafu {
-            violated: "No region ids provided".to_string(),
+
+        let region_id = *region_ids.first().with_context(|| InvalidGcArgsSnafu {
+            msg: "No region ids provided".to_string(),
         })?;
 
-        let mito_config = mito_engine.mito_config();
+        // also need to ensure all regions are on this datanode
+        ensure!(
+            region_ids
+                .iter()
+                .all(|rid| mito_engine.find_region(*rid).is_some()),
+            InvalidGcArgsSnafu {
+                msg: format!(
+                    "Some regions are not on current datanode:{:?}",
+                    region_ids
+                        .iter()
+                        .filter(|rid| mito_engine.find_region(**rid).is_none())
+                        .collect::<Vec<_>>()
+                ),
+            }
+        );
 
         // Find the access layer from one of the regions that exists on this datanode
-        let access_layer = region_ids
-            .iter()
-            .find_map(|rid| mito_engine.find_region(*rid))
+        let access_layer = mito_engine
+            .find_region(region_id)
             .with_context(|| InvalidGcArgsSnafu {
                 msg: format!(
                     "None of the regions is on current datanode:{:?}",
@@ -136,14 +139,22 @@ impl GcRegionsHandler {
             })?
             .access_layer();
 
+        // if region happen to be dropped before this but after gc scheduler send gc instr,
+        // need to deal with it properly(it is ok for region to be dropped after GC worker started)
+        // region not found here can only be drop table/database case, since region migration is prevented by lock in gc procedure
+        // TODO(discord9): add integration test for this drop case
+        let mito_regions = region_ids
+            .iter()
+            .filter_map(|rid| mito_engine.find_region(*rid).map(|r| (*rid, r)))
+            .collect();
+
         let cache_manager = mito_engine.cache_manager();
 
         let gc_worker = LocalGcWorker::try_new(
             access_layer.clone(),
             Some(cache_manager),
-            region_ids.into_iter().collect(),
-            Default::default(),
-            mito_config.clone().into(),
+            mito_regions,
+            mito_engine.mito_config().gc.clone(),
             file_ref_manifest.clone(),
             &mito_engine.gc_limiter(),
             full_file_listing,

src/datanode/src/region_server.rs

@@ -518,7 +518,7 @@ impl RegionServer {
 
         let manifest_info = match manifest_info {
             ManifestInfo::MitoManifestInfo(info) => {
-                RegionManifestInfo::mito(info.data_manifest_version, 0)
+                RegionManifestInfo::mito(info.data_manifest_version, 0, 0)
             }
             ManifestInfo::MetricManifestInfo(info) => RegionManifestInfo::metric(
                 info.data_manifest_version,

src/datanode/src/tests.rs

@@ -34,7 +34,8 @@ use session::context::QueryContextRef;
 use store_api::metadata::RegionMetadataRef;
 use store_api::region_engine::{
     RegionEngine, RegionManifestInfo, RegionRole, RegionScannerRef, RegionStatistic,
-    SetRegionRoleStateResponse, SettableRegionRoleState, SyncManifestResponse,
+    RemapManifestsRequest, RemapManifestsResponse, SetRegionRoleStateResponse,
+    SettableRegionRoleState, SyncManifestResponse,
 };
 use store_api::region_request::{AffectedRows, RegionRequest};
 use store_api::storage::{RegionId, ScanRequest, SequenceNumber};
@@ -291,6 +292,13 @@ impl RegionEngine for MockRegionEngine {
         unimplemented!()
     }
 
+    async fn remap_manifests(
+        &self,
+        _request: RemapManifestsRequest,
+    ) -> Result<RemapManifestsResponse, BoxedError> {
+        unimplemented!()
+    }
+
     fn as_any(&self) -> &dyn Any {
         self
     }

src/datatypes/src/arrow_array.rs

@@ -12,9 +12,117 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
+use arrow::array::{ArrayRef, AsArray};
+use arrow::datatypes::{
+    DataType, DurationMicrosecondType, DurationMillisecondType, DurationNanosecondType,
+    DurationSecondType, Time32MillisecondType, Time32SecondType, Time64MicrosecondType,
+    Time64NanosecondType, TimeUnit, TimestampMicrosecondType, TimestampMillisecondType,
+    TimestampNanosecondType, TimestampSecondType,
+};
+use common_time::time::Time;
+use common_time::{Duration, Timestamp};
+
 pub type BinaryArray = arrow::array::BinaryArray;
 pub type MutableBinaryArray = arrow::array::BinaryBuilder;
 pub type StringArray = arrow::array::StringArray;
 pub type MutableStringArray = arrow::array::StringBuilder;
 pub type LargeStringArray = arrow::array::LargeStringArray;
 pub type MutableLargeStringArray = arrow::array::LargeStringBuilder;
+
+/// Get the [Timestamp] value at index `i` of the timestamp array.
+///
+/// Note: This method does not check for nulls and the value is arbitrary
+/// if [`is_null`](arrow::array::Array::is_null) returns true for the index.
+///
+/// # Panics
+/// 1. if index `i` is out of bounds;
+/// 2. or the array is not timestamp type.
+pub fn timestamp_array_value(array: &ArrayRef, i: usize) -> Timestamp {
+    let DataType::Timestamp(time_unit, _) = &array.data_type() else {
+        unreachable!()
+    };
+    let v = match time_unit {
+        TimeUnit::Second => {
+            let array = array.as_primitive::<TimestampSecondType>();
+            array.value(i)
+        }
+        TimeUnit::Millisecond => {
+            let array = array.as_primitive::<TimestampMillisecondType>();
+            array.value(i)
+        }
+        TimeUnit::Microsecond => {
+            let array = array.as_primitive::<TimestampMicrosecondType>();
+            array.value(i)
+        }
+        TimeUnit::Nanosecond => {
+            let array = array.as_primitive::<TimestampNanosecondType>();
+            array.value(i)
+        }
+    };
+    Timestamp::new(v, time_unit.into())
+}
+
+/// Get the [Time] value at index `i` of the time array.
+///
+/// Note: This method does not check for nulls and the value is arbitrary
+/// if [`is_null`](arrow::array::Array::is_null) returns true for the index.
+///
+/// # Panics
+/// 1. if index `i` is out of bounds;
+/// 2. or the array is not `Time32` or `Time64` type.
+pub fn time_array_value(array: &ArrayRef, i: usize) -> Time {
+    match array.data_type() {
+        DataType::Time32(time_unit) | DataType::Time64(time_unit) => match time_unit {
+            TimeUnit::Second => {
+                let array = array.as_primitive::<Time32SecondType>();
+                Time::new_second(array.value(i) as i64)
+            }
+            TimeUnit::Millisecond => {
+                let array = array.as_primitive::<Time32MillisecondType>();
+                Time::new_millisecond(array.value(i) as i64)
+            }
+            TimeUnit::Microsecond => {
+                let array = array.as_primitive::<Time64MicrosecondType>();
+                Time::new_microsecond(array.value(i))
+            }
+            TimeUnit::Nanosecond => {
+                let array = array.as_primitive::<Time64NanosecondType>();
+                Time::new_nanosecond(array.value(i))
+            }
+        },
+        _ => unreachable!(),
+    }
+}
+
+/// Get the [Duration] value at index `i` of the duration array.
+///
+/// Note: This method does not check for nulls and the value is arbitrary
+/// if [`is_null`](arrow::array::Array::is_null) returns true for the index.
+///
+/// # Panics
+/// 1. if index `i` is out of bounds;
+/// 2. or the array is not duration type.
+pub fn duration_array_value(array: &ArrayRef, i: usize) -> Duration {
+    let DataType::Duration(time_unit) = array.data_type() else {
+        unreachable!();
+    };
+    let v = match time_unit {
+        TimeUnit::Second => {
+            let array = array.as_primitive::<DurationSecondType>();
+            array.value(i)
+        }
+        TimeUnit::Millisecond => {
+            let array = array.as_primitive::<DurationMillisecondType>();
+            array.value(i)
+        }
+        TimeUnit::Microsecond => {
+            let array = array.as_primitive::<DurationMicrosecondType>();
+            array.value(i)
+        }
+        TimeUnit::Nanosecond => {
+            let array = array.as_primitive::<DurationNanosecondType>();
+            array.value(i)
+        }
+    };
+    Duration::new(v, time_unit.into())
+}

src/datatypes/src/data_type.rs

@@ -33,8 +33,8 @@ use crate::types::{
     BinaryType, BooleanType, DateType, Decimal128Type, DictionaryType, DurationMicrosecondType,
     DurationMillisecondType, DurationNanosecondType, DurationSecondType, DurationType, Float32Type,
     Float64Type, Int8Type, Int16Type, Int32Type, Int64Type, IntervalDayTimeType,
-    IntervalMonthDayNanoType, IntervalType, IntervalYearMonthType, JsonFormat, JsonType, ListType,
-    NullType, StringType, StructType, TimeMillisecondType, TimeType, TimestampMicrosecondType,
+    IntervalMonthDayNanoType, IntervalType, IntervalYearMonthType, JsonType, ListType, NullType,
+    StringType, StructType, TimeMillisecondType, TimeType, TimestampMicrosecondType,
     TimestampMillisecondType, TimestampNanosecondType, TimestampSecondType, TimestampType,
     UInt8Type, UInt16Type, UInt32Type, UInt64Type, VectorType,
 };
@@ -677,7 +677,7 @@ impl ConcreteDataType {
     }
 
     pub fn json_native_datatype(inner_type: ConcreteDataType) -> ConcreteDataType {
-        ConcreteDataType::Json(JsonType::new(JsonFormat::Native(Box::new(inner_type))))
+        ConcreteDataType::Json(JsonType::new_native((&inner_type).into()))
     }
 }
 

src/datatypes/src/error.rs

@@ -189,7 +189,7 @@ pub enum Error {
         location: Location,
     },
 
-    #[snafu(display("Invalid JSON text: {}", value))]
+    #[snafu(display("Invalid JSON: {}", value))]
     InvalidJson {
         value: String,
         #[snafu(implicit)]

src/datatypes/src/json.rs

@@ -19,18 +19,19 @@
 //! The struct will carry all the fields of the Json object. We will not flatten any json object in this implementation.
 //!
 
-use std::collections::HashSet;
+pub mod value;
+
+use std::collections::{BTreeMap, HashSet};
 use std::sync::Arc;
 
-use common_base::bytes::StringBytes;
-use ordered_float::OrderedFloat;
 use serde::{Deserialize, Serialize};
 use serde_json::{Map, Value as Json};
 use snafu::{ResultExt, ensure};
 
-use crate::data_type::{ConcreteDataType, DataType};
 use crate::error::{self, Error};
-use crate::types::{ListType, StructField, StructType};
+use crate::json::value::{JsonValue, JsonVariant};
+use crate::types::json_type::{JsonNativeType, JsonNumberType, JsonObjectType};
+use crate::types::{StructField, StructType};
 use crate::value::{ListValue, StructValue, Value};
 
 /// The configuration of JSON encoding
@@ -102,7 +103,7 @@ impl JsonStructureSettings {
     pub fn encode_with_type(
         &self,
         json: Json,
-        data_type: Option<&ConcreteDataType>,
+        data_type: Option<&JsonNativeType>,
     ) -> Result<Value, Error> {
         let context = JsonContext {
             key_path: String::new(),
@@ -146,70 +147,65 @@ impl<'a> JsonContext<'a> {
 /// Main encoding function with key path tracking
 pub fn encode_json_with_context<'a>(
     json: Json,
-    data_type: Option<&ConcreteDataType>,
+    data_type: Option<&JsonNativeType>,
     context: &JsonContext<'a>,
-) -> Result<Value, Error> {
+) -> Result<JsonValue, Error> {
     // Check if the entire encoding should be unstructured
     if matches!(context.settings, JsonStructureSettings::UnstructuredRaw) {
         let json_string = json.to_string();
-        let struct_value = StructValue::try_new(
-            vec![Value::String(json_string.into())],
-            StructType::new(Arc::new(vec![StructField::new(
-                JsonStructureSettings::RAW_FIELD.to_string(),
-                ConcreteDataType::string_datatype(),
-                true,
-            )])),
-        )?;
-        return Ok(Value::Struct(struct_value));
+        return Ok([(JsonStructureSettings::RAW_FIELD, json_string)].into());
     }
 
     // Check if current key should be treated as unstructured
     if context.is_unstructured_key() {
-        return Ok(Value::String(json.to_string().into()));
+        return Ok(json.to_string().into());
     }
 
     match json {
         Json::Object(json_object) => {
-            ensure!(
-                matches!(data_type, Some(ConcreteDataType::Struct(_)) | None),
-                error::InvalidJsonSnafu {
-                    value: "JSON object can only be encoded to Struct type".to_string(),
+            let object_type = match data_type.as_ref() {
+                Some(JsonNativeType::Object(x)) => Some(x),
+                None => None,
+                _ => {
+                    return error::InvalidJsonSnafu {
+                        value: "JSON object value must be encoded with object type",
+                    }
+                    .fail();
                 }
-            );
-
-            let data_type = data_type.and_then(|x| x.as_struct());
-            let struct_value = encode_json_object_with_context(json_object, data_type, context)?;
-            Ok(Value::Struct(struct_value))
+            };
+            encode_json_object_with_context(json_object, object_type, context)
         }
         Json::Array(json_array) => {
-            let item_type = if let Some(ConcreteDataType::List(list_type)) = data_type {
-                Some(list_type.item_type())
-            } else {
-                None
+            let item_type = match data_type.as_ref() {
+                Some(JsonNativeType::Array(x)) => Some(x.as_ref()),
+                None => None,
+                _ => {
+                    return error::InvalidJsonSnafu {
+                        value: "JSON array value must be encoded with array type",
+                    }
+                    .fail();
+                }
             };
-            let list_value = encode_json_array_with_context(json_array, item_type, context)?;
-            Ok(Value::List(list_value))
+            encode_json_array_with_context(json_array, item_type, context)
         }
         _ => {
             // For non-collection types, verify type compatibility
             if let Some(expected_type) = data_type {
-                let (value, actual_type) =
-                    encode_json_value_with_context(json, Some(expected_type), context)?;
-                if &actual_type == expected_type {
+                let value = encode_json_value_with_context(json, Some(expected_type), context)?;
+                let actual_type = value.json_type().native_type();
+                if actual_type == expected_type {
                     Ok(value)
                 } else {
                     Err(error::InvalidJsonSnafu {
                         value: format!(
                             "JSON value type {} does not match expected type {}",
-                            actual_type.name(),
-                            expected_type.name()
+                            actual_type, expected_type
                         ),
                     }
                     .build())
                 }
             } else {
-                let (value, _) = encode_json_value_with_context(json, None, context)?;
-                Ok(value)
+                encode_json_value_with_context(json, None, context)
             }
         }
     }
@@ -217,31 +213,21 @@ pub fn encode_json_with_context<'a>(
 
 fn encode_json_object_with_context<'a>(
     mut json_object: Map<String, Json>,
-    fields: Option<&StructType>,
+    fields: Option<&JsonObjectType>,
     context: &JsonContext<'a>,
-) -> Result<StructValue, Error> {
-    let total_json_keys = json_object.len();
-    let mut items = Vec::with_capacity(total_json_keys);
-    let mut struct_fields = Vec::with_capacity(total_json_keys);
+) -> Result<JsonValue, Error> {
+    let mut object = BTreeMap::new();
     // First, process fields from the provided schema in their original order
     if let Some(fields) = fields {
-        for field in fields.fields().iter() {
-            let field_name = field.name();
-
+        for (field_name, field_type) in fields {
             if let Some(value) = json_object.remove(field_name) {
                 let field_context = context.with_key(field_name);
-                let (value, data_type) =
-                    encode_json_value_with_context(value, Some(field.data_type()), &field_context)?;
-                items.push(value);
-                struct_fields.push(StructField::new(
-                    field_name.to_string(),
-                    data_type,
-                    true, // JSON fields are always nullable
-                ));
+                let value =
+                    encode_json_value_with_context(value, Some(field_type), &field_context)?;
+                object.insert(field_name.clone(), value.into_variant());
             } else {
                 // Field exists in schema but not in JSON - add null value
-                items.push(Value::Null);
-                struct_fields.push(field.clone());
+                object.insert(field_name.clone(), ().into());
             }
         }
     }
@@ -250,139 +236,111 @@ fn encode_json_object_with_context<'a>(
     for (key, value) in json_object {
         let field_context = context.with_key(&key);
 
-        let (value, data_type) = encode_json_value_with_context(value, None, &field_context)?;
-        items.push(value);
+        let value = encode_json_value_with_context(value, None, &field_context)?;
 
-        struct_fields.push(StructField::new(
-            key.clone(),
-            data_type,
-            true, // JSON fields are always nullable
-        ));
+        object.insert(key, value.into_variant());
     }
 
-    let struct_type = StructType::new(Arc::new(struct_fields));
-    StructValue::try_new(items, struct_type)
+    Ok(JsonValue::new(JsonVariant::Object(object)))
 }
 
 fn encode_json_array_with_context<'a>(
     json_array: Vec<Json>,
-    item_type: Option<&ConcreteDataType>,
+    item_type: Option<&JsonNativeType>,
     context: &JsonContext<'a>,
-) -> Result<ListValue, Error> {
+) -> Result<JsonValue, Error> {
     let json_array_len = json_array.len();
     let mut items = Vec::with_capacity(json_array_len);
-    let mut element_type = None;
+    let mut element_type = item_type.cloned();
 
     for (index, value) in json_array.into_iter().enumerate() {
         let array_context = context.with_key(&index.to_string());
-        let (item_value, item_type) =
-            encode_json_value_with_context(value, item_type, &array_context)?;
-        items.push(item_value);
+        let item_value =
+            encode_json_value_with_context(value, element_type.as_ref(), &array_context)?;
+        let item_type = item_value.json_type().native_type().clone();
+        items.push(item_value.into_variant());
 
         // Determine the common type for the list
         if let Some(current_type) = &element_type {
-            // For now, we'll use the first non-null type we encounter
-            // In a more sophisticated implementation, we might want to find a common supertype
-            if *current_type == ConcreteDataType::null_datatype()
-                && item_type != ConcreteDataType::null_datatype()
-            {
-                element_type = Some(item_type);
-            }
+            // It's valid for json array to have different types of items, for example,
+            // ["a string", 1]. However, the `JsonValue` will be converted to Arrow list array,
+            // which requires all items have exactly same type. So we forbid the different types
+            // case here. Besides, it's not common for items in a json array to differ. So I think
+            // we are good here.
+            ensure!(
+                item_type == *current_type,
+                error::InvalidJsonSnafu {
+                    value: "all items in json array must have the same type"
+                }
+            );
         } else {
             element_type = Some(item_type);
         }
     }
 
-    // Use provided item_type if available, otherwise determine from elements
-    let element_type = if let Some(item_type) = item_type {
-        item_type.clone()
-    } else {
-        element_type.unwrap_or_else(ConcreteDataType::string_datatype)
-    };
-
-    Ok(ListValue::new(items, Arc::new(element_type)))
+    Ok(JsonValue::new(JsonVariant::Array(items)))
 }
 
 /// Helper function to encode a JSON value to a Value and determine its ConcreteDataType with context
 fn encode_json_value_with_context<'a>(
     json: Json,
-    expected_type: Option<&ConcreteDataType>,
+    expected_type: Option<&JsonNativeType>,
     context: &JsonContext<'a>,
-) -> Result<(Value, ConcreteDataType), Error> {
+) -> Result<JsonValue, Error> {
     // Check if current key should be treated as unstructured
     if context.is_unstructured_key() {
-        return Ok((
-            Value::String(json.to_string().into()),
-            ConcreteDataType::string_datatype(),
-        ));
+        return Ok(json.to_string().into());
     }
 
     match json {
-        Json::Null => Ok((Value::Null, ConcreteDataType::null_datatype())),
-        Json::Bool(b) => Ok((Value::Boolean(b), ConcreteDataType::boolean_datatype())),
+        Json::Null => Ok(JsonValue::null()),
+        Json::Bool(b) => Ok(b.into()),
         Json::Number(n) => {
             if let Some(i) = n.as_i64() {
                 // Use int64 for all integer numbers when possible
                 if let Some(expected) = expected_type
                     && let Ok(value) = try_convert_to_expected_type(i, expected)
                 {
-                    return Ok((value, expected.clone()));
+                    return Ok(value);
                 }
-                Ok((Value::Int64(i), ConcreteDataType::int64_datatype()))
+                Ok(i.into())
             } else if let Some(u) = n.as_u64() {
                 // Use int64 for unsigned integers that fit, otherwise use u64
                 if let Some(expected) = expected_type
                     && let Ok(value) = try_convert_to_expected_type(u, expected)
                 {
-                    return Ok((value, expected.clone()));
+                    return Ok(value);
                 }
                 if u <= i64::MAX as u64 {
-                    Ok((Value::Int64(u as i64), ConcreteDataType::int64_datatype()))
+                    Ok((u as i64).into())
                 } else {
-                    Ok((Value::UInt64(u), ConcreteDataType::uint64_datatype()))
+                    Ok(u.into())
                 }
             } else if let Some(f) = n.as_f64() {
                 // Try to use the expected type if provided
                 if let Some(expected) = expected_type
                     && let Ok(value) = try_convert_to_expected_type(f, expected)
                 {
-                    return Ok((value, expected.clone()));
+                    return Ok(value);
                 }
 
                 // Default to f64 for floating point numbers
-                Ok((
-                    Value::Float64(OrderedFloat(f)),
-                    ConcreteDataType::float64_datatype(),
-                ))
+                Ok(f.into())
             } else {
                 // Fallback to string representation
-                Ok((
-                    Value::String(StringBytes::from(n.to_string())),
-                    ConcreteDataType::string_datatype(),
-                ))
+                Ok(n.to_string().into())
             }
         }
         Json::String(s) => {
             if let Some(expected) = expected_type
                 && let Ok(value) = try_convert_to_expected_type(s.as_str(), expected)
             {
-                return Ok((value, expected.clone()));
+                return Ok(value);
             }
-            Ok((
-                Value::String(StringBytes::from(s.clone())),
-                ConcreteDataType::string_datatype(),
-            ))
-        }
-        Json::Array(arr) => {
-            let list_value = encode_json_array_with_context(arr, expected_type, context)?;
-            let datatype = ConcreteDataType::List(ListType::new(list_value.datatype()));
-            Ok((Value::List(list_value), datatype))
-        }
-        Json::Object(obj) => {
-            let struct_value = encode_json_object_with_context(obj, None, context)?;
-            let data_type = ConcreteDataType::Struct(struct_value.struct_type().clone());
-            Ok((Value::Struct(struct_value), data_type))
+            Ok(s.into())
         }
+        Json::Array(arr) => encode_json_array_with_context(arr, expected_type, context),
+        Json::Object(obj) => encode_json_object_with_context(obj, None, context),
     }
 }
 
@@ -402,7 +360,6 @@ pub fn decode_value_with_context<'a>(
     }
 
     match value {
-        Value::Json(inner) => decode_value_with_context(*inner, context),
         Value::Struct(struct_value) => decode_struct_with_context(struct_value, context),
         Value::List(list_value) => decode_list_with_context(list_value, context),
         _ => decode_primitive_value(value),
@@ -569,11 +526,13 @@ fn decode_struct_with_settings<'a>(
                 key_path: field_context.key_path.clone(),
                 settings: &JsonStructureSettings::Structured(None),
             };
-            let (decoded_value, data_type) = encode_json_value_with_context(
+            let decoded_value = encode_json_value_with_context(
                 json_value,
                 None, // Don't force a specific type, let it be inferred from JSON
                 &structured_context,
-            )?;
+            )?
+            .into_value();
+            let data_type = decoded_value.data_type();
 
             items.push(decoded_value);
             struct_fields.push(StructField::new(
@@ -651,8 +610,9 @@ fn decode_unstructured_raw_struct(struct_value: StructValue) -> Result<StructVal
                 key_path: String::new(),
                 settings: &JsonStructureSettings::Structured(None),
             };
-            let (decoded_value, data_type) =
-                encode_json_value_with_context(json_value, None, &context)?;
+            let decoded_value =
+                encode_json_value_with_context(json_value, None, &context)?.into_value();
+            let data_type = decoded_value.data_type();
 
             if let Value::Struct(decoded_struct) = decoded_value {
                 return Ok(decoded_struct);
@@ -678,22 +638,48 @@ fn decode_unstructured_raw_struct(struct_value: StructValue) -> Result<StructVal
 /// Helper function to try converting a value to an expected type
 fn try_convert_to_expected_type<T>(
     value: T,
-    expected_type: &ConcreteDataType,
-) -> Result<Value, Error>
+    expected_type: &JsonNativeType,
+) -> Result<JsonValue, Error>
 where
-    T: Into<Value>,
+    T: Into<JsonValue>,
 {
     let value = value.into();
-    expected_type.try_cast(value.clone()).ok_or_else(|| {
+    let cast_error = || {
         error::CastTypeSnafu {
-            msg: format!(
-                "Cannot cast from {} to {}",
-                value.data_type().name(),
-                expected_type.name()
-            ),
+            msg: format!("Cannot cast value {value} to {expected_type}"),
         }
-        .build()
-    })
+        .fail()
+    };
+    let actual_type = value.json_type().native_type();
+    match (actual_type, expected_type) {
+        (x, y) if x == y => Ok(value),
+        (JsonNativeType::Number(x), JsonNativeType::Number(y)) => match (x, y) {
+            (JsonNumberType::U64, JsonNumberType::I64) => {
+                if let Some(i) = value.as_i64() {
+                    Ok(i.into())
+                } else {
+                    cast_error()
+                }
+            }
+            (JsonNumberType::I64, JsonNumberType::U64) => {
+                if let Some(i) = value.as_u64() {
+                    Ok(i.into())
+                } else {
+                    cast_error()
+                }
+            }
+            (_, JsonNumberType::F64) => {
+                if let Some(f) = value.as_f64() {
+                    Ok(f.into())
+                } else {
+                    cast_error()
+                }
+            }
+            _ => cast_error(),
+        },
+        (_, JsonNativeType::String) => Ok(value.to_string().into()),
+        _ => cast_error(),
+    }
 }
 
 #[cfg(test)]
@@ -702,6 +688,7 @@ mod tests {
     use serde_json::json;
 
     use super::*;
+    use crate::data_type::ConcreteDataType;
     use crate::types::ListType;
 
     #[test]
@@ -898,15 +885,15 @@ mod tests {
         let json = Json::from(42);
         let settings = JsonStructureSettings::Structured(None);
         let result = settings
-            .encode_with_type(json.clone(), Some(&ConcreteDataType::int8_datatype()))
+            .encode_with_type(json.clone(), Some(&JsonNativeType::u64()))
             .unwrap()
             .into_json_inner()
             .unwrap();
-        assert_eq!(result, Value::Int8(42));
+        assert_eq!(result, Value::UInt64(42));
 
         // Test with expected string type
         let result = settings
-            .encode_with_type(json, Some(&ConcreteDataType::string_datatype()))
+            .encode_with_type(json, Some(&JsonNativeType::String))
             .unwrap()
             .into_json_inner()
             .unwrap();
@@ -917,23 +904,11 @@ mod tests {
     fn test_encode_json_array_mixed_types() {
         let json = json!([1, "hello", true, 3.15]);
         let settings = JsonStructureSettings::Structured(None);
-        let result = settings
-            .encode_with_type(json, None)
-            .unwrap()
-            .into_json_inner()
-            .unwrap();
-
-        if let Value::List(list_value) = result {
-            assert_eq!(list_value.items().len(), 4);
-            // The first non-null type should determine the list type
-            // In this case, it should be string since we can't find a common numeric type
-            assert_eq!(
-                list_value.datatype(),
-                Arc::new(ConcreteDataType::int64_datatype())
-            );
-        } else {
-            panic!("Expected List value");
-        }
+        let result = settings.encode_with_type(json, None);
+        assert_eq!(
+            result.unwrap_err().to_string(),
+            "Invalid JSON: all items in json array must have the same type"
+        );
     }
 
     #[test]
@@ -951,7 +926,7 @@ mod tests {
             // Empty arrays default to string type
             assert_eq!(
                 list_value.datatype(),
-                Arc::new(ConcreteDataType::string_datatype())
+                Arc::new(ConcreteDataType::null_datatype())
             );
         } else {
             panic!("Expected List value");
@@ -987,16 +962,10 @@ mod tests {
         });
 
         // Define expected struct type
-        let fields = vec![
-            StructField::new(
-                "name".to_string(),
-                ConcreteDataType::string_datatype(),
-                true,
-            ),
-            StructField::new("age".to_string(), ConcreteDataType::int64_datatype(), true),
-        ];
-        let struct_type = StructType::new(Arc::new(fields));
-        let concrete_type = ConcreteDataType::Struct(struct_type);
+        let concrete_type = JsonNativeType::Object(JsonObjectType::from([
+            ("name".to_string(), JsonNativeType::String),
+            ("age".to_string(), JsonNativeType::i64()),
+        ]));
 
         let settings = JsonStructureSettings::Structured(None);
         let result = settings
@@ -1008,15 +977,15 @@ mod tests {
         if let Value::Struct(struct_value) = result {
             assert_eq!(struct_value.items().len(), 2);
             let struct_fields = struct_value.struct_type().fields();
-            assert_eq!(struct_fields[0].name(), "name");
+            assert_eq!(struct_fields[0].name(), "age");
             assert_eq!(
                 struct_fields[0].data_type(),
-                &ConcreteDataType::string_datatype()
+                &ConcreteDataType::int64_datatype()
             );
-            assert_eq!(struct_fields[1].name(), "age");
+            assert_eq!(struct_fields[1].name(), "name");
             assert_eq!(
                 struct_fields[1].data_type(),
-                &ConcreteDataType::int64_datatype()
+                &ConcreteDataType::string_datatype()
             );
         } else {
             panic!("Expected Struct value");
@@ -1032,34 +1001,24 @@ mod tests {
         });
 
         // Define schema with specific field order
-        let fields = vec![
-            StructField::new(
-                "a_field".to_string(),
-                ConcreteDataType::string_datatype(),
-                true,
-            ),
-            StructField::new(
-                "m_field".to_string(),
-                ConcreteDataType::string_datatype(),
-                true,
-            ),
-            StructField::new(
-                "z_field".to_string(),
-                ConcreteDataType::string_datatype(),
-                true,
-            ),
-        ];
-        let struct_type = StructType::new(Arc::new(fields));
+        let json_type = JsonObjectType::from([
+            ("a_field".to_string(), JsonNativeType::String),
+            ("m_field".to_string(), JsonNativeType::String),
+            ("z_field".to_string(), JsonNativeType::String),
+        ]);
 
-        let result = encode_json_object_with_context(
+        let Value::Struct(result) = encode_json_object_with_context(
             json.as_object().unwrap().clone(),
-            Some(&struct_type),
+            Some(&json_type),
             &JsonContext {
                 key_path: String::new(),
                 settings: &JsonStructureSettings::Structured(None),
             },
         )
-        .unwrap();
+        .map(|x| x.into_value())
+        .unwrap() else {
+            unreachable!()
+        };
 
         // Verify field order is preserved from schema
         let struct_fields = result.struct_type().fields();
@@ -1083,37 +1042,35 @@ mod tests {
         });
 
         // Define schema with only name and age
-        let fields = vec![
-            StructField::new(
-                "name".to_string(),
-                ConcreteDataType::string_datatype(),
-                true,
-            ),
-            StructField::new("age".to_string(), ConcreteDataType::int64_datatype(), true),
-        ];
-        let struct_type = StructType::new(Arc::new(fields));
+        let json_type = JsonObjectType::from([
+            ("name".to_string(), JsonNativeType::String),
+            ("age".to_string(), JsonNativeType::i64()),
+        ]);
 
-        let result = encode_json_object_with_context(
+        let Value::Struct(result) = encode_json_object_with_context(
             json.as_object().unwrap().clone(),
-            Some(&struct_type),
+            Some(&json_type),
             &JsonContext {
                 key_path: String::new(),
                 settings: &JsonStructureSettings::Structured(None),
             },
         )
-        .unwrap();
+        .map(|x| x.into_value())
+        .unwrap() else {
+            unreachable!()
+        };
 
-        // Verify schema fields come first in order
+        // verify fields are sorted in json value
         let struct_fields = result.struct_type().fields();
-        assert_eq!(struct_fields[0].name(), "name");
+        assert_eq!(struct_fields[0].name(), "active");
         assert_eq!(struct_fields[1].name(), "age");
-        assert_eq!(struct_fields[2].name(), "active");
+        assert_eq!(struct_fields[2].name(), "name");
 
         // Verify values are correct
         let items = result.items();
-        assert_eq!(items[0], Value::String("Alice".into()));
+        assert_eq!(items[0], Value::Boolean(true));
         assert_eq!(items[1], Value::Int64(25));
-        assert_eq!(items[2], Value::Boolean(true));
+        assert_eq!(items[2], Value::String("Alice".into()));
     }
 
     #[test]
@@ -1124,35 +1081,33 @@ mod tests {
         });
 
         // Define schema with name and age
-        let fields = vec![
-            StructField::new(
-                "name".to_string(),
-                ConcreteDataType::string_datatype(),
-                true,
-            ),
-            StructField::new("age".to_string(), ConcreteDataType::int64_datatype(), true),
-        ];
-        let struct_type = StructType::new(Arc::new(fields));
+        let json_type = JsonObjectType::from([
+            ("name".to_string(), JsonNativeType::String),
+            ("age".to_string(), JsonNativeType::i64()),
+        ]);
 
-        let result = encode_json_object_with_context(
+        let Value::Struct(result) = encode_json_object_with_context(
             json.as_object().unwrap().clone(),
-            Some(&struct_type),
+            Some(&json_type),
             &JsonContext {
                 key_path: String::new(),
                 settings: &JsonStructureSettings::Structured(None),
             },
         )
-        .unwrap();
+        .map(|x| x.into_value())
+        .unwrap() else {
+            unreachable!()
+        };
 
         // Verify both schema fields are present
         let struct_fields = result.struct_type().fields();
-        assert_eq!(struct_fields[0].name(), "name");
-        assert_eq!(struct_fields[1].name(), "age");
+        assert_eq!(struct_fields[0].name(), "age");
+        assert_eq!(struct_fields[1].name(), "name");
 
         // Verify values - name has value, age is null
         let items = result.items();
-        assert_eq!(items[0], Value::String("Bob".into()));
-        assert_eq!(items[1], Value::Null);
+        assert_eq!(items[0], Value::Null);
+        assert_eq!(items[1], Value::String("Bob".into()));
     }
 
     #[test]
@@ -1175,21 +1130,22 @@ mod tests {
     #[test]
     fn test_encode_json_array_with_item_type() {
         let json = json!([1, 2, 3]);
-        let item_type = Arc::new(ConcreteDataType::int8_datatype());
-        let list_type = ListType::new(item_type.clone());
-        let concrete_type = ConcreteDataType::List(list_type);
+        let item_type = Arc::new(ConcreteDataType::uint64_datatype());
         let settings = JsonStructureSettings::Structured(None);
         let result = settings
-            .encode_with_type(json, Some(&concrete_type))
+            .encode_with_type(
+                json,
+                Some(&JsonNativeType::Array(Box::new(JsonNativeType::u64()))),
+            )
             .unwrap()
             .into_json_inner()
             .unwrap();
 
         if let Value::List(list_value) = result {
             assert_eq!(list_value.items().len(), 3);
-            assert_eq!(list_value.items()[0], Value::Int8(1));
-            assert_eq!(list_value.items()[1], Value::Int8(2));
-            assert_eq!(list_value.items()[2], Value::Int8(3));
+            assert_eq!(list_value.items()[0], Value::UInt64(1));
+            assert_eq!(list_value.items()[1], Value::UInt64(2));
+            assert_eq!(list_value.items()[2], Value::UInt64(3));
             assert_eq!(list_value.datatype(), item_type);
         } else {
             panic!("Expected List value");
@@ -1199,12 +1155,13 @@ mod tests {
     #[test]
     fn test_encode_json_array_empty_with_item_type() {
         let json = json!([]);
-        let item_type = Arc::new(ConcreteDataType::string_datatype());
-        let list_type = ListType::new(item_type.clone());
-        let concrete_type = ConcreteDataType::List(list_type);
+        let item_type = Arc::new(ConcreteDataType::null_datatype());
         let settings = JsonStructureSettings::Structured(None);
         let result = settings
-            .encode_with_type(json, Some(&concrete_type))
+            .encode_with_type(
+                json,
+                Some(&JsonNativeType::Array(Box::new(JsonNativeType::Null))),
+            )
             .unwrap()
             .into_json_inner()
             .unwrap();
@@ -1219,6 +1176,7 @@ mod tests {
 
     #[cfg(test)]
     mod decode_tests {
+        use ordered_float::OrderedFloat;
         use serde_json::json;
 
         use super::*;
@@ -1473,7 +1431,7 @@ mod tests {
         // Test encoding JSON number with expected int64 type
         let json = Json::from(42);
         let result = settings
-            .encode_with_type(json, Some(&ConcreteDataType::int64_datatype()))
+            .encode_with_type(json, Some(&JsonNativeType::i64()))
             .unwrap()
             .into_json_inner()
             .unwrap();
@@ -1482,7 +1440,7 @@ mod tests {
         // Test encoding JSON string with expected string type
         let json = Json::String("hello".to_string());
         let result = settings
-            .encode_with_type(json, Some(&ConcreteDataType::string_datatype()))
+            .encode_with_type(json, Some(&JsonNativeType::String))
             .unwrap()
             .into_json_inner()
             .unwrap();
@@ -1491,7 +1449,7 @@ mod tests {
         // Test encoding JSON boolean with expected boolean type
         let json = Json::Bool(true);
         let result = settings
-            .encode_with_type(json, Some(&ConcreteDataType::boolean_datatype()))
+            .encode_with_type(json, Some(&JsonNativeType::Bool))
             .unwrap()
             .into_json_inner()
             .unwrap();
@@ -1503,12 +1461,12 @@ mod tests {
         // Test encoding JSON number with mismatched string type
         let json = Json::from(42);
         let settings = JsonStructureSettings::Structured(None);
-        let result = settings.encode_with_type(json, Some(&ConcreteDataType::string_datatype()));
+        let result = settings.encode_with_type(json, Some(&JsonNativeType::String));
         assert!(result.is_ok()); // Should succeed due to type conversion
 
         // Test encoding JSON object with mismatched non-struct type
         let json = json!({"name": "test"});
-        let result = settings.encode_with_type(json, Some(&ConcreteDataType::int64_datatype()));
+        let result = settings.encode_with_type(json, Some(&JsonNativeType::i64()));
         assert!(result.is_err()); // Should fail - object can't be converted to int64
     }
 
@@ -1516,12 +1474,13 @@ mod tests {
     fn test_encode_json_array_with_list_type() {
         let json = json!([1, 2, 3]);
         let item_type = Arc::new(ConcreteDataType::int64_datatype());
-        let list_type = ListType::new(item_type.clone());
-        let concrete_type = ConcreteDataType::List(list_type);
 
         let settings = JsonStructureSettings::Structured(None);
         let result = settings
-            .encode_with_type(json, Some(&concrete_type))
+            .encode_with_type(
+                json,
+                Some(&JsonNativeType::Array(Box::new(JsonNativeType::i64()))),
+            )
             .unwrap()
             .into_json_inner()
             .unwrap();
@@ -1543,7 +1502,7 @@ mod tests {
         let json = Json::Null;
         let settings = JsonStructureSettings::Structured(None);
         let result = settings
-            .encode_with_type(json.clone(), Some(&ConcreteDataType::null_datatype()))
+            .encode_with_type(json.clone(), Some(&JsonNativeType::Null))
             .unwrap()
             .into_json_inner()
             .unwrap();
@@ -1552,7 +1511,7 @@ mod tests {
         // Test float with float64 type
         let json = Json::from(3.15);
         let result = settings
-            .encode_with_type(json, Some(&ConcreteDataType::float64_datatype()))
+            .encode_with_type(json, Some(&JsonNativeType::f64()))
             .unwrap()
             .into_json_inner()
             .unwrap();
@@ -1644,20 +1603,11 @@ mod tests {
         }
 
         // Test with encode_with_type (with type)
-        let struct_type = StructType::new(Arc::new(vec![
-            StructField::new(
-                "name".to_string(),
-                ConcreteDataType::string_datatype(),
-                true,
-            ),
-            StructField::new("age".to_string(), ConcreteDataType::int64_datatype(), true),
-            StructField::new(
-                "active".to_string(),
-                ConcreteDataType::boolean_datatype(),
-                true,
-            ),
+        let concrete_type = JsonNativeType::Object(JsonObjectType::from([
+            ("name".to_string(), JsonNativeType::String),
+            ("age".to_string(), JsonNativeType::i64()),
+            ("active".to_string(), JsonNativeType::Bool),
         ]));
-        let concrete_type = ConcreteDataType::Struct(struct_type);
 
         let result2 = settings
             .encode_with_type(json, Some(&concrete_type))
@@ -2153,20 +2103,11 @@ mod tests {
             )])),
         );
 
-        let decoded_struct = settings.decode_struct(array_struct).unwrap();
-        let fields = decoded_struct.struct_type().fields();
-        let decoded_fields: Vec<&str> = fields.iter().map(|f| f.name()).collect();
-        assert!(decoded_fields.contains(&"value"));
-
-        if let Value::List(list_value) = &decoded_struct.items()[0] {
-            assert_eq!(list_value.items().len(), 4);
-            assert_eq!(list_value.items()[0], Value::Int64(1));
-            assert_eq!(list_value.items()[1], Value::String("hello".into()));
-            assert_eq!(list_value.items()[2], Value::Boolean(true));
-            assert_eq!(list_value.items()[3], Value::Float64(OrderedFloat(3.15)));
-        } else {
-            panic!("Expected array to be decoded as ListValue");
-        }
+        let decoded_struct = settings.decode_struct(array_struct);
+        assert_eq!(
+            decoded_struct.unwrap_err().to_string(),
+            "Invalid JSON: all items in json array must have the same type"
+        );
     }
 
     #[test]

src/datatypes/src/json/value.rs

@@ -0,0 +1,683 @@
+// Copyright 2023 Greptime Team
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+use std::collections::BTreeMap;
+use std::fmt::{Display, Formatter};
+use std::hash::{Hash, Hasher};
+use std::sync::{Arc, OnceLock};
+
+use num_traits::ToPrimitive;
+use ordered_float::OrderedFloat;
+use serde::{Deserialize, Serialize};
+use serde_json::Number;
+
+use crate::data_type::ConcreteDataType;
+use crate::types::json_type::JsonNativeType;
+use crate::types::{JsonType, StructField, StructType};
+use crate::value::{ListValue, ListValueRef, StructValue, StructValueRef, Value, ValueRef};
+
+/// Number in json, can be a positive integer, a negative integer, or a floating number.
+/// Each of which is represented as `u64`, `i64` and `f64`.
+///
+/// This follows how `serde_json` designs number.
+#[derive(Debug, Clone, Copy, PartialEq, Eq, Hash, Serialize, Deserialize)]
+pub enum JsonNumber {
+    PosInt(u64),
+    NegInt(i64),
+    Float(OrderedFloat<f64>),
+}
+
+impl JsonNumber {
+    fn as_u64(&self) -> Option<u64> {
+        match self {
+            JsonNumber::PosInt(n) => Some(*n),
+            JsonNumber::NegInt(n) => (*n >= 0).then_some(*n as u64),
+            _ => None,
+        }
+    }
+
+    fn as_i64(&self) -> Option<i64> {
+        match self {
+            JsonNumber::PosInt(n) => (*n <= i64::MAX as u64).then_some(*n as i64),
+            JsonNumber::NegInt(n) => Some(*n),
+            _ => None,
+        }
+    }
+
+    fn as_f64(&self) -> f64 {
+        match self {
+            JsonNumber::PosInt(n) => *n as f64,
+            JsonNumber::NegInt(n) => *n as f64,
+            JsonNumber::Float(n) => n.0,
+        }
+    }
+}
+
+impl From<u64> for JsonNumber {
+    fn from(i: u64) -> Self {
+        Self::PosInt(i)
+    }
+}
+
+impl From<i64> for JsonNumber {
+    fn from(n: i64) -> Self {
+        Self::NegInt(n)
+    }
+}
+
+impl From<f64> for JsonNumber {
+    fn from(i: f64) -> Self {
+        Self::Float(i.into())
+    }
+}
+
+impl Display for JsonNumber {
+    fn fmt(&self, f: &mut Formatter<'_>) -> std::fmt::Result {
+        match self {
+            Self::PosInt(x) => write!(f, "{x}"),
+            Self::NegInt(x) => write!(f, "{x}"),
+            Self::Float(x) => write!(f, "{x}"),
+        }
+    }
+}
+
+/// Variants of json.
+///
+/// This follows how [serde_json::Value] designs except that we only choose to use [BTreeMap] to
+/// preserve the fields order by their names in the json object. (By default `serde_json` uses
+/// [BTreeMap], too. But it additionally supports "IndexMap" which preserves the order by insertion
+/// times of fields.)
+#[derive(Debug, Clone, PartialEq, Eq, Hash, Serialize, Deserialize)]
+pub enum JsonVariant {
+    Null,
+    Bool(bool),
+    Number(JsonNumber),
+    String(String),
+    Array(Vec<JsonVariant>),
+    Object(BTreeMap<String, JsonVariant>),
+}
+
+impl JsonVariant {
+    fn native_type(&self) -> JsonNativeType {
+        match self {
+            JsonVariant::Null => JsonNativeType::Null,
+            JsonVariant::Bool(_) => JsonNativeType::Bool,
+            JsonVariant::Number(n) => match n {
+                JsonNumber::PosInt(_) => JsonNativeType::u64(),
+                JsonNumber::NegInt(_) => JsonNativeType::i64(),
+                JsonNumber::Float(_) => JsonNativeType::f64(),
+            },
+            JsonVariant::String(_) => JsonNativeType::String,
+            JsonVariant::Array(array) => {
+                let item_type = if let Some(first) = array.first() {
+                    first.native_type()
+                } else {
+                    JsonNativeType::Null
+                };
+                JsonNativeType::Array(Box::new(item_type))
+            }
+            JsonVariant::Object(object) => JsonNativeType::Object(
+                object
+                    .iter()
+                    .map(|(k, v)| (k.clone(), v.native_type()))
+                    .collect(),
+            ),
+        }
+    }
+
+    fn json_type(&self) -> JsonType {
+        JsonType::new_native(self.native_type())
+    }
+
+    fn as_ref(&self) -> JsonVariantRef<'_> {
+        match self {
+            JsonVariant::Null => JsonVariantRef::Null,
+            JsonVariant::Bool(x) => (*x).into(),
+            JsonVariant::Number(x) => match x {
+                JsonNumber::PosInt(i) => (*i).into(),
+                JsonNumber::NegInt(i) => (*i).into(),
+                JsonNumber::Float(f) => (f.0).into(),
+            },
+            JsonVariant::String(x) => x.as_str().into(),
+            JsonVariant::Array(array) => {
+                JsonVariantRef::Array(array.iter().map(|x| x.as_ref()).collect())
+            }
+            JsonVariant::Object(object) => JsonVariantRef::Object(
+                object
+                    .iter()
+                    .map(|(k, v)| (k.as_str(), v.as_ref()))
+                    .collect(),
+            ),
+        }
+    }
+}
+
+impl From<()> for JsonVariant {
+    fn from(_: ()) -> Self {
+        Self::Null
+    }
+}
+
+impl From<bool> for JsonVariant {
+    fn from(v: bool) -> Self {
+        Self::Bool(v)
+    }
+}
+
+impl<T: Into<JsonNumber>> From<T> for JsonVariant {
+    fn from(v: T) -> Self {
+        Self::Number(v.into())
+    }
+}
+
+impl From<&str> for JsonVariant {
+    fn from(v: &str) -> Self {
+        Self::String(v.to_string())
+    }
+}
+
+impl From<String> for JsonVariant {
+    fn from(v: String) -> Self {
+        Self::String(v)
+    }
+}
+
+impl<const N: usize, T: Into<JsonVariant>> From<[T; N]> for JsonVariant {
+    fn from(vs: [T; N]) -> Self {
+        Self::Array(vs.into_iter().map(|x| x.into()).collect())
+    }
+}
+
+impl<K: Into<String>, V: Into<JsonVariant>, const N: usize> From<[(K, V); N]> for JsonVariant {
+    fn from(vs: [(K, V); N]) -> Self {
+        Self::Object(vs.into_iter().map(|(k, v)| (k.into(), v.into())).collect())
+    }
+}
+
+impl Display for JsonVariant {
+    fn fmt(&self, f: &mut Formatter<'_>) -> std::fmt::Result {
+        match self {
+            Self::Null => write!(f, "null"),
+            Self::Bool(x) => write!(f, "{x}"),
+            Self::Number(x) => write!(f, "{x}"),
+            Self::String(x) => write!(f, "{x}"),
+            Self::Array(array) => write!(
+                f,
+                "[{}]",
+                array
+                    .iter()
+                    .map(|x| x.to_string())
+                    .collect::<Vec<_>>()
+                    .join(", ")
+            ),
+            Self::Object(object) => {
+                write!(
+                    f,
+                    "{{ {} }}",
+                    object
+                        .iter()
+                        .map(|(k, v)| format!("{k}: {v}"))
+                        .collect::<Vec<_>>()
+                        .join(", ")
+                )
+            }
+        }
+    }
+}
+
+/// Represents any valid JSON value.
+#[derive(Debug, Eq, Serialize, Deserialize)]
+pub struct JsonValue {
+    #[serde(skip)]
+    json_type: OnceLock<JsonType>,
+    json_variant: JsonVariant,
+}
+
+impl JsonValue {
+    pub fn null() -> Self {
+        ().into()
+    }
+
+    pub(crate) fn new(json_variant: JsonVariant) -> Self {
+        Self {
+            json_type: OnceLock::new(),
+            json_variant,
+        }
+    }
+
+    pub(crate) fn data_type(&self) -> ConcreteDataType {
+        ConcreteDataType::Json(self.json_type().clone())
+    }
+
+    pub(crate) fn json_type(&self) -> &JsonType {
+        self.json_type.get_or_init(|| self.json_variant.json_type())
+    }
+
+    pub(crate) fn is_null(&self) -> bool {
+        matches!(self.json_variant, JsonVariant::Null)
+    }
+
+    pub(crate) fn as_i64(&self) -> Option<i64> {
+        match self.json_variant {
+            JsonVariant::Number(n) => n.as_i64(),
+            _ => None,
+        }
+    }
+
+    pub(crate) fn as_u64(&self) -> Option<u64> {
+        match self.json_variant {
+            JsonVariant::Number(n) => n.as_u64(),
+            _ => None,
+        }
+    }
+
+    pub(crate) fn as_f64(&self) -> Option<f64> {
+        match self.json_variant {
+            JsonVariant::Number(n) => Some(n.as_f64()),
+            _ => None,
+        }
+    }
+
+    pub(crate) fn as_f64_lossy(&self) -> Option<f64> {
+        match self.json_variant {
+            JsonVariant::Number(n) => Some(match n {
+                JsonNumber::PosInt(i) => i as f64,
+                JsonNumber::NegInt(i) => i as f64,
+                JsonNumber::Float(f) => f.0,
+            }),
+            _ => None,
+        }
+    }
+
+    pub(crate) fn as_bool(&self) -> Option<bool> {
+        match self.json_variant {
+            JsonVariant::Bool(b) => Some(b),
+            _ => None,
+        }
+    }
+
+    pub fn as_ref(&self) -> JsonValueRef<'_> {
+        JsonValueRef {
+            json_type: OnceLock::new(),
+            json_variant: self.json_variant.as_ref(),
+        }
+    }
+
+    pub fn into_variant(self) -> JsonVariant {
+        self.json_variant
+    }
+
+    pub(crate) fn into_value(self) -> Value {
+        fn helper(v: JsonVariant) -> Value {
+            match v {
+                JsonVariant::Null => Value::Null,
+                JsonVariant::Bool(x) => Value::Boolean(x),
+                JsonVariant::Number(x) => match x {
+                    JsonNumber::PosInt(i) => Value::UInt64(i),
+                    JsonNumber::NegInt(i) => Value::Int64(i),
+                    JsonNumber::Float(f) => Value::Float64(f),
+                },
+                JsonVariant::String(x) => Value::String(x.into()),
+                JsonVariant::Array(array) => {
+                    let item_type = if let Some(first) = array.first() {
+                        first.native_type()
+                    } else {
+                        JsonNativeType::Null
+                    };
+                    Value::List(ListValue::new(
+                        array.into_iter().map(helper).collect(),
+                        Arc::new((&item_type).into()),
+                    ))
+                }
+                JsonVariant::Object(object) => {
+                    let mut fields = Vec::with_capacity(object.len());
+                    let mut items = Vec::with_capacity(object.len());
+                    for (k, v) in object {
+                        fields.push(StructField::new(k, (&v.native_type()).into(), true));
+                        items.push(helper(v));
+                    }
+                    Value::Struct(StructValue::new(items, StructType::new(Arc::new(fields))))
+                }
+            }
+        }
+        helper(self.json_variant)
+    }
+}
+
+impl<T: Into<JsonVariant>> From<T> for JsonValue {
+    fn from(v: T) -> Self {
+        Self {
+            json_type: OnceLock::new(),
+            json_variant: v.into(),
+        }
+    }
+}
+
+impl From<JsonValue> for serde_json::Value {
+    fn from(v: JsonValue) -> Self {
+        fn helper(v: JsonVariant) -> serde_json::Value {
+            match v {
+                JsonVariant::Null => serde_json::Value::Null,
+                JsonVariant::Bool(x) => serde_json::Value::Bool(x),
+                JsonVariant::Number(x) => match x {
+                    JsonNumber::PosInt(i) => serde_json::Value::Number(i.into()),
+                    JsonNumber::NegInt(i) => serde_json::Value::Number(i.into()),
+                    JsonNumber::Float(f) => {
+                        if let Some(x) = Number::from_f64(f.0) {
+                            serde_json::Value::Number(x)
+                        } else {
+                            serde_json::Value::String("NaN".into())
+                        }
+                    }
+                },
+                JsonVariant::String(x) => serde_json::Value::String(x),
+                JsonVariant::Array(array) => {
+                    serde_json::Value::Array(array.into_iter().map(helper).collect())
+                }
+                JsonVariant::Object(object) => serde_json::Value::Object(
+                    object.into_iter().map(|(k, v)| (k, helper(v))).collect(),
+                ),
+            }
+        }
+        helper(v.json_variant)
+    }
+}
+
+impl Clone for JsonValue {
+    fn clone(&self) -> Self {
+        let Self {
+            json_type: _,
+            json_variant,
+        } = self;
+        Self {
+            json_type: OnceLock::new(),
+            json_variant: json_variant.clone(),
+        }
+    }
+}
+
+impl PartialEq<JsonValue> for JsonValue {
+    fn eq(&self, other: &JsonValue) -> bool {
+        let Self {
+            json_type: _,
+            json_variant,
+        } = self;
+        json_variant.eq(&other.json_variant)
+    }
+}
+
+impl Hash for JsonValue {
+    fn hash<H: Hasher>(&self, state: &mut H) {
+        let Self {
+            json_type: _,
+            json_variant,
+        } = self;
+        json_variant.hash(state);
+    }
+}
+
+impl Display for JsonValue {
+    fn fmt(&self, f: &mut Formatter<'_>) -> std::fmt::Result {
+        write!(f, "{}", self.json_variant)
+    }
+}
+
+/// References of variants of json.
+#[derive(Debug, Clone, PartialEq, Eq, Serialize)]
+pub enum JsonVariantRef<'a> {
+    Null,
+    Bool(bool),
+    Number(JsonNumber),
+    String(&'a str),
+    Array(Vec<JsonVariantRef<'a>>),
+    Object(BTreeMap<&'a str, JsonVariantRef<'a>>),
+}
+
+impl JsonVariantRef<'_> {
+    fn json_type(&self) -> JsonType {
+        fn native_type(v: &JsonVariantRef<'_>) -> JsonNativeType {
+            match v {
+                JsonVariantRef::Null => JsonNativeType::Null,
+                JsonVariantRef::Bool(_) => JsonNativeType::Bool,
+                JsonVariantRef::Number(n) => match n {
+                    JsonNumber::PosInt(_) => JsonNativeType::u64(),
+                    JsonNumber::NegInt(_) => JsonNativeType::i64(),
+                    JsonNumber::Float(_) => JsonNativeType::f64(),
+                },
+                JsonVariantRef::String(_) => JsonNativeType::String,
+                JsonVariantRef::Array(array) => {
+                    let item_type = if let Some(first) = array.first() {
+                        native_type(first)
+                    } else {
+                        JsonNativeType::Null
+                    };
+                    JsonNativeType::Array(Box::new(item_type))
+                }
+                JsonVariantRef::Object(object) => JsonNativeType::Object(
+                    object
+                        .iter()
+                        .map(|(k, v)| (k.to_string(), native_type(v)))
+                        .collect(),
+                ),
+            }
+        }
+        JsonType::new_native(native_type(self))
+    }
+}
+
+impl From<()> for JsonVariantRef<'_> {
+    fn from(_: ()) -> Self {
+        Self::Null
+    }
+}
+
+impl From<bool> for JsonVariantRef<'_> {
+    fn from(v: bool) -> Self {
+        Self::Bool(v)
+    }
+}
+
+impl<T: Into<JsonNumber>> From<T> for JsonVariantRef<'_> {
+    fn from(v: T) -> Self {
+        Self::Number(v.into())
+    }
+}
+
+impl<'a> From<&'a str> for JsonVariantRef<'a> {
+    fn from(v: &'a str) -> Self {
+        Self::String(v)
+    }
+}
+
+impl<'a, const N: usize, T: Into<JsonVariantRef<'a>>> From<[T; N]> for JsonVariantRef<'a> {
+    fn from(vs: [T; N]) -> Self {
+        Self::Array(vs.into_iter().map(|x| x.into()).collect())
+    }
+}
+
+impl<'a, V: Into<JsonVariantRef<'a>>, const N: usize> From<[(&'a str, V); N]>
+    for JsonVariantRef<'a>
+{
+    fn from(vs: [(&'a str, V); N]) -> Self {
+        Self::Object(vs.into_iter().map(|(k, v)| (k, v.into())).collect())
+    }
+}
+
+impl<'a> From<Vec<JsonVariantRef<'a>>> for JsonVariantRef<'a> {
+    fn from(v: Vec<JsonVariantRef<'a>>) -> Self {
+        Self::Array(v)
+    }
+}
+
+impl<'a> From<BTreeMap<&'a str, JsonVariantRef<'a>>> for JsonVariantRef<'a> {
+    fn from(v: BTreeMap<&'a str, JsonVariantRef<'a>>) -> Self {
+        Self::Object(v)
+    }
+}
+
+impl From<JsonVariantRef<'_>> for JsonVariant {
+    fn from(v: JsonVariantRef) -> Self {
+        match v {
+            JsonVariantRef::Null => Self::Null,
+            JsonVariantRef::Bool(x) => Self::Bool(x),
+            JsonVariantRef::Number(x) => Self::Number(x),
+            JsonVariantRef::String(x) => Self::String(x.to_string()),
+            JsonVariantRef::Array(array) => {
+                Self::Array(array.into_iter().map(Into::into).collect())
+            }
+            JsonVariantRef::Object(object) => Self::Object(
+                object
+                    .into_iter()
+                    .map(|(k, v)| (k.to_string(), v.into()))
+                    .collect(),
+            ),
+        }
+    }
+}
+
+/// Reference to representation of any valid JSON value.
+#[derive(Debug, Serialize)]
+pub struct JsonValueRef<'a> {
+    #[serde(skip)]
+    json_type: OnceLock<JsonType>,
+    json_variant: JsonVariantRef<'a>,
+}
+
+impl<'a> JsonValueRef<'a> {
+    pub fn null() -> Self {
+        ().into()
+    }
+
+    pub(crate) fn data_type(&self) -> ConcreteDataType {
+        ConcreteDataType::Json(self.json_type().clone())
+    }
+
+    pub(crate) fn json_type(&self) -> &JsonType {
+        self.json_type.get_or_init(|| self.json_variant.json_type())
+    }
+
+    pub fn into_variant(self) -> JsonVariantRef<'a> {
+        self.json_variant
+    }
+
+    pub(crate) fn is_null(&self) -> bool {
+        matches!(self.json_variant, JsonVariantRef::Null)
+    }
+
+    pub fn is_object(&self) -> bool {
+        matches!(self.json_variant, JsonVariantRef::Object(_))
+    }
+
+    pub(crate) fn as_f32(&self) -> Option<f32> {
+        match self.json_variant {
+            JsonVariantRef::Number(JsonNumber::Float(f)) => f.to_f32(),
+            _ => None,
+        }
+    }
+
+    pub(crate) fn as_f64(&self) -> Option<f64> {
+        match self.json_variant {
+            JsonVariantRef::Number(JsonNumber::Float(f)) => Some(f.0),
+            _ => None,
+        }
+    }
+
+    pub fn as_value_ref(&self) -> ValueRef<'_> {
+        fn helper<'a>(v: &'a JsonVariantRef) -> ValueRef<'a> {
+            match v {
+                JsonVariantRef::Null => ValueRef::Null,
+                JsonVariantRef::Bool(x) => ValueRef::Boolean(*x),
+                JsonVariantRef::Number(x) => match x {
+                    JsonNumber::PosInt(i) => ValueRef::UInt64(*i),
+                    JsonNumber::NegInt(i) => ValueRef::Int64(*i),
+                    JsonNumber::Float(f) => ValueRef::Float64(*f),
+                },
+                JsonVariantRef::String(x) => ValueRef::String(x),
+                JsonVariantRef::Array(array) => {
+                    let val = array.iter().map(helper).collect::<Vec<_>>();
+                    let item_datatype = if let Some(first) = val.first() {
+                        first.data_type()
+                    } else {
+                        ConcreteDataType::null_datatype()
+                    };
+                    ValueRef::List(ListValueRef::RefList {
+                        val,
+                        item_datatype: Arc::new(item_datatype),
+                    })
+                }
+                JsonVariantRef::Object(object) => {
+                    let mut fields = Vec::with_capacity(object.len());
+                    let mut val = Vec::with_capacity(object.len());
+                    for (k, v) in object.iter() {
+                        let v = helper(v);
+                        fields.push(StructField::new(k.to_string(), v.data_type(), true));
+                        val.push(v);
+                    }
+                    ValueRef::Struct(StructValueRef::RefList {
+                        val,
+                        fields: StructType::new(Arc::new(fields)),
+                    })
+                }
+            }
+        }
+        helper(&self.json_variant)
+    }
+
+    pub(crate) fn data_size(&self) -> usize {
+        size_of_val(self)
+    }
+}
+
+impl<'a, T: Into<JsonVariantRef<'a>>> From<T> for JsonValueRef<'a> {
+    fn from(v: T) -> Self {
+        Self {
+            json_type: OnceLock::new(),
+            json_variant: v.into(),
+        }
+    }
+}
+
+impl From<JsonValueRef<'_>> for JsonValue {
+    fn from(v: JsonValueRef<'_>) -> Self {
+        Self {
+            json_type: OnceLock::new(),
+            json_variant: v.json_variant.into(),
+        }
+    }
+}
+
+impl PartialEq for JsonValueRef<'_> {
+    fn eq(&self, other: &Self) -> bool {
+        let Self {
+            json_type: _,
+            json_variant,
+        } = self;
+        json_variant == &other.json_variant
+    }
+}
+
+impl Eq for JsonValueRef<'_> {}
+
+impl Clone for JsonValueRef<'_> {
+    fn clone(&self) -> Self {
+        let Self {
+            json_type: _,
+            json_variant,
+        } = self;
+        Self {
+            json_type: OnceLock::new(),
+            json_variant: json_variant.clone(),
+        }
+    }
+}