ci: release artifacts after binary and container is ready (#1192 )

ci: release artifacts before binary and container is ready
ci: disable run tests temporarily (#1187 )
2025-12-22 22:20:02 +00:00 · 2023-03-16 09:20:03 +00:00 · 2023-03-16 14:12:19 +08:00 · 2023-03-16 11:56:28 +08:00 · 2023-03-16 11:36:38 +08:00 · 2023-03-15 20:15:34 +08:00
421 changed files with 21967 additions and 7867 deletions
--- a/.env.example
+++ b/.env.example
@@ -2,7 +2,7 @@
 GT_S3_BUCKET=S3 bucket
 GT_S3_ACCESS_KEY_ID=S3 access key id
 GT_S3_ACCESS_KEY=S3 secret access key
-
+GT_S3_ENDPOINT_URL=S3 endpoint url
 # Settings for oss test
 GT_OSS_BUCKET=OSS bucket
 GT_OSS_ACCESS_KEY_ID=OSS access key id
--- a/.github/pr-title-breaking-change-label-config.json
+++ b/.github/pr-title-breaking-change-label-config.json
@@ -0,0 +1,13 @@
+{
+    "LABEL": {
+        "name": "breaking change",
+        "color": "D93F0B"
+    },
+    "CHECKS": {
+        "regexp": "^(?:(?!!:).)*$",
+        "ignoreLabels": [
+            "ignore-title"
+        ],
+        "alwaysPassCI": true
+    }
+}
--- a/.github/pr-title-checker-config.json
+++ b/.github/pr-title-checker-config.json
@@ -1,10 +1,12 @@
 {
-  "LABEL": {
-    "name": "Invalid PR Title",
-    "color": "B60205"
-  },
-  "CHECKS": {
-    "regexp": "^(feat|fix|test|refactor|chore|style|docs|perf|build|ci|revert)(\\(.*\\))?:.*",
-    "ignoreLabels" : ["ignore-title"]
-  }
+    "LABEL": {
+        "name": "Invalid PR Title",
+        "color": "B60205"
+    },
+    "CHECKS": {
+        "regexp": "^(feat|fix|test|refactor|chore|style|docs|perf|build|ci|revert)(\\(.*\\))?\\!?:.*",
+        "ignoreLabels": [
+            "ignore-title"
+        ]
+    }
 }
--- a/.github/workflows/apidoc.yml
+++ b/.github/workflows/apidoc.yml
@@ -0,0 +1,42 @@
+on:
+  push:
+    branches:
+      - develop
+    paths-ignore:
+      - 'docs/**'
+      - 'config/**'
+      - '**.md'
+      - '.dockerignore'
+      - 'docker/**'
+      - '.gitignore'
+
+name: Build API docs
+
+env:
+  RUST_TOOLCHAIN: nightly-2023-02-26
+
+jobs:
+  apidoc:
+    runs-on: ubuntu-latest
+    steps:
+    - uses: actions/checkout@v3
+    - uses: arduino/setup-protoc@v1
+      with:
+        repo-token: ${{ secrets.GITHUB_TOKEN }}
+    - uses: dtolnay/rust-toolchain@master
+      with:
+        toolchain: ${{ env.RUST_TOOLCHAIN }}
+    - run: cargo doc --workspace --no-deps --document-private-items
+    - run: |
+        cat <<EOF > target/doc/index.html
+        <!DOCTYPE html>
+        <html>
+        <head>
+        <meta http-equiv="refresh" content="0; url='greptime/'" />
+        </head>
+        <body></body></html>
+        EOF
+    - name: Publish dist directory
+      uses: JamesIves/github-pages-deploy-action@v4
+      with:
+        folder: target/doc
--- a/.github/workflows/develop.yml
+++ b/.github/workflows/develop.yml
@@ -24,15 +24,15 @@ on:
 name: CI

 env:
-  RUST_TOOLCHAIN: nightly-2023-02-14
+  RUST_TOOLCHAIN: nightly-2023-02-26

 jobs:
  typos:
    name: Spell Check with Typos
    runs-on: ubuntu-latest
    steps:
-      - uses: actions/checkout@v2
-      - uses: crate-ci/typos@v1.0.4
+      - uses: actions/checkout@v3
+      - uses: crate-ci/typos@v1.13.10

  check:
    name: Check
@@ -116,6 +116,7 @@ jobs:
    if: github.event.pull_request.draft == false
    runs-on: ubuntu-latest-8-cores
    timeout-minutes: 60
+    needs: [clippy]
    steps:
      - uses: actions/checkout@v3
      - uses: arduino/setup-protoc@v1
@@ -131,7 +132,7 @@ jobs:
          ETCD_VER=v3.5.7
          DOWNLOAD_URL=https://github.com/etcd-io/etcd/releases/download
          curl -L ${DOWNLOAD_URL}/${ETCD_VER}/etcd-${ETCD_VER}-linux-amd64.tar.gz -o /tmp/etcd-${ETCD_VER}-linux-amd64.tar.gz
-          mkdir -p /tmp/etcd-download 
+          mkdir -p /tmp/etcd-download
          tar xzvf /tmp/etcd-${ETCD_VER}-linux-amd64.tar.gz -C /tmp/etcd-download --strip-components=1
          rm -f /tmp/etcd-${ETCD_VER}-linux-amd64.tar.gz

@@ -188,6 +189,7 @@ jobs:
    if: github.event.pull_request.draft == false
    runs-on: ubuntu-latest-8-cores
    timeout-minutes: 60
+    needs: [clippy]
    steps:
      - uses: actions/checkout@v3
      - uses: arduino/setup-protoc@v1
@@ -205,10 +207,16 @@ jobs:
        uses: Swatinem/rust-cache@v2
      - name: Install latest nextest release
        uses: taiki-e/install-action@nextest
+      - name: Install Python
+        uses: actions/setup-python@v4
+        with:
+          python-version: '3.10'
+      - name: Install PyArrow Package
+        run: pip install pyarrow
      - name: Install cargo-llvm-cov
        uses: taiki-e/install-action@cargo-llvm-cov
      - name: Collect coverage data
-        run: cargo llvm-cov nextest --workspace --lcov --output-path lcov.info
+        run: cargo llvm-cov nextest --workspace --lcov --output-path lcov.info -F pyo3_backend
        env:
          CARGO_BUILD_RUSTFLAGS: "-C link-arg=-fuse-ld=lld"
          RUST_BACKTRACE: 1
@@ -223,5 +231,5 @@ jobs:
          token: ${{ secrets.CODECOV_TOKEN }}
          files: ./lcov.info
          flags: rust
-          fail_ci_if_error: true
+          fail_ci_if_error: false
          verbose: true
--- a/.github/workflows/license.yaml
+++ b/.github/workflows/license.yaml
@@ -13,4 +13,4 @@ jobs:
    steps:
    - uses: actions/checkout@v2
    - name: Check License Header
-      uses: apache/skywalking-eyes/header@main
+      uses: apache/skywalking-eyes/header@df70871af1a8109c9a5b1dc824faaf65246c5236
--- a/.github/workflows/pr-title-checker.yml
+++ b/.github/workflows/pr-title-checker.yml
@@ -18,3 +18,12 @@ jobs:
          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
          pass_on_octokit_error: false
          configuration_path: ".github/pr-title-checker-config.json"
+  breaking:
+    runs-on: ubuntu-latest
+    timeout-minutes: 10
+    steps:
+      - uses: thehanimo/pr-title-checker@v1.3.4
+        with:
+          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+          pass_on_octokit_error: false
+          configuration_path: ".github/pr-title-breaking-change-label-config.json"
--- a/.github/workflows/release.yml
+++ b/.github/workflows/release.yml
@@ -10,15 +10,16 @@ on:
 name: Release

 env:
-  RUST_TOOLCHAIN: nightly-2023-02-14
+  RUST_TOOLCHAIN: nightly-2023-02-26

-  # FIXME(zyy17): Would be better to use `gh release list -L 1 | cut -f 3` to get the latest release version tag, but for a long time, we will stay at 'v0.1.0-alpha-*'.
-  SCHEDULED_BUILD_VERSION_PREFIX: v0.1.0-alpha
+  SCHEDULED_BUILD_VERSION_PREFIX: v0.2.0

-  # In the future, we can change SCHEDULED_PERIOD to nightly.
-  SCHEDULED_PERIOD: weekly
+  SCHEDULED_PERIOD: nightly

-  CARGO_PROFILE: weekly
+  CARGO_PROFILE: nightly
+
+  ## FIXME(zyy17): Enable it after the tests are stabled.
+  DISABLE_RUN_TESTS: true

 jobs:
  build:
@@ -30,16 +31,25 @@ jobs:
          - arch: x86_64-unknown-linux-gnu
            os: ubuntu-2004-16-cores
            file: greptime-linux-amd64
+            continue-on-error: false
+            opts: "-F pyo3_backend"
          - arch: aarch64-unknown-linux-gnu
            os: ubuntu-2004-16-cores
            file: greptime-linux-arm64
+            continue-on-error: false
+            opts: "-F pyo3_backend"
          - arch: aarch64-apple-darwin
            os: macos-latest
            file: greptime-darwin-arm64
+            continue-on-error: false
+            opts: "-F pyo3_backend"
          - arch: x86_64-apple-darwin
            os: macos-latest
            file: greptime-darwin-amd64
+            continue-on-error: false
+            opts: "-F pyo3_backend"
    runs-on: ${{ matrix.os }}
+    continue-on-error: ${{ matrix.continue-on-error }}
    if: github.repository == 'GreptimeTeam/greptimedb'
    steps:
      - name: Checkout sources
@@ -76,10 +86,10 @@ jobs:
          ETCD_VER=v3.5.7
          DOWNLOAD_URL=https://github.com/etcd-io/etcd/releases/download
          curl -L ${DOWNLOAD_URL}/${ETCD_VER}/etcd-${ETCD_VER}-linux-amd64.tar.gz -o /tmp/etcd-${ETCD_VER}-linux-amd64.tar.gz
-          mkdir -p /tmp/etcd-download 
-          tar xzvf /tmp/etcd-${ETCD_VER}-linux-amd64.tar.gz -C /tmp/etcd-download --strip-components=1 
+          mkdir -p /tmp/etcd-download
+          tar xzvf /tmp/etcd-${ETCD_VER}-linux-amd64.tar.gz -C /tmp/etcd-download --strip-components=1
          rm -f /tmp/etcd-${ETCD_VER}-linux-amd64.tar.gz
-          
+
          sudo cp -a /tmp/etcd-download/etcd* /usr/local/bin/
          nohup etcd >/tmp/etcd.log 2>&1 &

@@ -93,7 +103,13 @@ jobs:
        if: contains(matrix.arch, 'linux') && endsWith(matrix.arch, '-gnu')
        run: |
          sudo apt-get -y update
-          sudo apt-get -y install libssl-dev pkg-config g++-aarch64-linux-gnu gcc-aarch64-linux-gnu
+          sudo apt-get -y install libssl-dev pkg-config g++-aarch64-linux-gnu gcc-aarch64-linux-gnu binutils-aarch64-linux-gnu wget
+
+      - name: Compile Python 3.10.10 from source for Aarch64
+        if: contains(matrix.arch, 'aarch64-unknown-linux-gnu')
+        run: |
+          sudo chmod +x ./docker/aarch64/compile-python.sh
+          sudo ./docker/aarch64/compile-python.sh

      - name: Install rust toolchain
        uses: dtolnay/rust-toolchain@master
@@ -105,10 +121,21 @@ jobs:
        run: protoc --version ; cargo version ; rustc --version ; gcc --version ; g++ --version

      - name: Run tests
+        if: env.DISABLE_RUN_TESTS == 'false'
        run: make unit-test integration-test sqlness-test

+      - name: Run cargo build for aarch64-linux
+        if: contains(matrix.arch, 'aarch64-unknown-linux-gnu')
+        run: |
+          # TODO(zyy17): We should make PYO3_CROSS_LIB_DIR configurable.
+          export PYO3_CROSS_LIB_DIR=$(pwd)/python_arm64_build/lib
+          echo "PYO3_CROSS_LIB_DIR: $PYO3_CROSS_LIB_DIR"
+          alias python=python3
+          cargo build --profile ${{ env.CARGO_PROFILE }} --locked --target ${{ matrix.arch }} ${{ matrix.opts }}
+
      - name: Run cargo build
-        run: cargo build ${{ matrix.opts }} --profile ${{ env.CARGO_PROFILE }} --locked --target ${{ matrix.arch }}
+        if: contains(matrix.arch, 'aarch64-unknown-linux-gnu') == false
+        run: cargo build --profile ${{ env.CARGO_PROFILE }} --locked --target ${{ matrix.arch }} ${{ matrix.opts }}

      - name: Calculate checksum and rename binary
        shell: bash
@@ -129,48 +156,6 @@ jobs:
        with:
          name: ${{ matrix.file }}.sha256sum
          path: target/${{ matrix.arch }}/${{ env.CARGO_PROFILE }}/${{ matrix.file }}.sha256sum
-  release:
-    name: Release artifacts
-    needs: [build]
-    runs-on: ubuntu-latest
-    if: github.repository == 'GreptimeTeam/greptimedb'
-    steps:
-      - name: Checkout sources
-        uses: actions/checkout@v3
-
-      - name: Download artifacts
-        uses: actions/download-artifact@v3
-
-      - name: Configure scheduled build version # the version would be ${SCHEDULED_BUILD_VERSION_PREFIX}-YYYYMMDD-${SCHEDULED_PERIOD}, like v0.1.0-alpha-20221119-weekly.
-        shell: bash
-        if: github.event_name == 'schedule'
-        run: |
-          buildTime=`date "+%Y%m%d"`
-          SCHEDULED_BUILD_VERSION=${{ env.SCHEDULED_BUILD_VERSION_PREFIX }}-$buildTime-${{ env.SCHEDULED_PERIOD }}
-          echo "SCHEDULED_BUILD_VERSION=${SCHEDULED_BUILD_VERSION}" >> $GITHUB_ENV
-
-      - name: Create scheduled build git tag
-        if: github.event_name == 'schedule'
-        run: |
-          git tag ${{ env.SCHEDULED_BUILD_VERSION }}
-
-      - name: Publish scheduled release # configure the different release title and tags.
-        uses: softprops/action-gh-release@v1
-        if: github.event_name == 'schedule'
-        with:
-          name: "Release ${{ env.SCHEDULED_BUILD_VERSION }}"
-          tag_name: ${{ env.SCHEDULED_BUILD_VERSION }}
-          generate_release_notes: true
-          files: |
-            **/greptime-*
-
-      - name: Publish release
-        uses: softprops/action-gh-release@v1
-        if: github.event_name != 'schedule'
-        with:
-          name: "Release ${{ github.ref_name }}"
-          files: |
-            **/greptime-*

  docker:
    name: Build docker image
@@ -181,37 +166,6 @@ jobs:
      - name: Checkout sources
        uses: actions/checkout@v3

-      - name: Download amd64 binary
-        uses: actions/download-artifact@v3
-        with:
-          name: greptime-linux-amd64
-          path: amd64
-
-      - name: Unzip the amd64 artifacts
-        run: |
-          cd amd64
-          tar xvf greptime-linux-amd64.tgz
-          rm greptime-linux-amd64.tgz
-
-      - name: Download arm64 binary
-        uses: actions/download-artifact@v3
-        with:
-          name: greptime-linux-arm64
-          path: arm64
-
-      - name: Unzip the arm64 artifacts
-        run: |
-          cd arm64
-          tar xvf greptime-linux-arm64.tgz
-          rm greptime-linux-arm64.tgz
-
-      - name: Login to UCloud Container Registry
-        uses: docker/login-action@v2
-        with:
-          registry: uhub.service.ucloud.cn
-          username: ${{ secrets.UCLOUD_USERNAME }}
-          password: ${{ secrets.UCLOUD_PASSWORD }}
-
      - name: Login to Dockerhub
        uses: docker/login-action@v2
        with:
@@ -239,8 +193,36 @@ jobs:
      - name: Set up buildx
        uses: docker/setup-buildx-action@v2

-      - name: Build and push
+      - name: Download amd64 binary
+        uses: actions/download-artifact@v3
+        with:
+          name: greptime-linux-amd64
+          path: amd64
+
+      - name: Unzip the amd64 artifacts
+        run: |
+          cd amd64
+          tar xvf greptime-linux-amd64.tgz
+          rm greptime-linux-amd64.tgz
+
+      - name: Download arm64 binary
+        id: download-arm64
+        uses: actions/download-artifact@v3
+        with:
+          name: greptime-linux-arm64
+          path: arm64
+
+      - name: Unzip the arm64 artifacts
+        id: unzip-arm64
+        if: success() || steps.download-arm64.conclusion == 'success'
+        run: |
+          cd arm64
+          tar xvf greptime-linux-arm64.tgz
+          rm greptime-linux-arm64.tgz
+
+      - name: Build and push all
        uses: docker/build-push-action@v3
+        if: success() || steps.unzip-arm64.conclusion == 'success' # Build and push all platform if unzip-arm64 succeeds
        with:
          context: .
          file: ./docker/ci/Dockerfile
@@ -249,5 +231,105 @@ jobs:
          tags: |
            greptime/greptimedb:latest
            greptime/greptimedb:${{ env.IMAGE_TAG }}
-            uhub.service.ucloud.cn/greptime/greptimedb:latest
-            uhub.service.ucloud.cn/greptime/greptimedb:${{ env.IMAGE_TAG }}
+
+      - name: Build and push amd64 only
+        uses: docker/build-push-action@v3
+        if: success() || steps.download-arm64.conclusion == 'failure' # Only build and push amd64 platform if download-arm64 fails
+        with:
+          context: .
+          file: ./docker/ci/Dockerfile
+          push: true
+          platforms: linux/amd64
+          tags: |
+            greptime/greptimedb:latest
+            greptime/greptimedb:${{ env.IMAGE_TAG }}
+
+  release:
+    name: Release artifacts
+    # Release artifacts only when all the artifacts are built successfully.
+    needs: [build,docker]
+    runs-on: ubuntu-latest
+    if: github.repository == 'GreptimeTeam/greptimedb'
+    steps:
+      - name: Checkout sources
+        uses: actions/checkout@v3
+
+      - name: Download artifacts
+        uses: actions/download-artifact@v3
+
+      - name: Configure scheduled build version # the version would be ${SCHEDULED_BUILD_VERSION_PREFIX}-${SCHEDULED_PERIOD}-YYYYMMDD, like v0.2.0-nigthly-20230313.
+        shell: bash
+        if: github.event_name == 'schedule'
+        run: |
+          buildTime=`date "+%Y%m%d"`
+          SCHEDULED_BUILD_VERSION=${{ env.SCHEDULED_BUILD_VERSION_PREFIX }}-${{ env.SCHEDULED_PERIOD }}-$buildTime
+          echo "SCHEDULED_BUILD_VERSION=${SCHEDULED_BUILD_VERSION}" >> $GITHUB_ENV
+
+      - name: Create scheduled build git tag
+        if: github.event_name == 'schedule'
+        run: |
+          git tag ${{ env.SCHEDULED_BUILD_VERSION }}
+
+      - name: Publish scheduled release # configure the different release title and tags.
+        uses: softprops/action-gh-release@v1
+        if: github.event_name == 'schedule'
+        with:
+          name: "Release ${{ env.SCHEDULED_BUILD_VERSION }}"
+          tag_name: ${{ env.SCHEDULED_BUILD_VERSION }}
+          generate_release_notes: true
+          files: |
+            **/greptime-*
+
+      - name: Publish release
+        uses: softprops/action-gh-release@v1
+        if: github.event_name != 'schedule'
+        with:
+          name: "Release ${{ github.ref_name }}"
+          files: |
+            **/greptime-*
+
+  docker-push-uhub:
+    name: Push docker image to UCloud Container Registry
+    needs: [docker]
+    runs-on: ubuntu-latest
+    if: github.repository == 'GreptimeTeam/greptimedb'
+    # Push to uhub may fail(500 error), but we don't want to block the release process. The failed job will be retried manually.
+    continue-on-error: true
+    steps:
+      - name: Checkout sources
+        uses: actions/checkout@v3
+
+      - name: Set up QEMU
+        uses: docker/setup-qemu-action@v2
+
+      - name: Set up Docker Buildx
+        uses: docker/setup-buildx-action@v2
+
+      - name: Login to UCloud Container Registry
+        uses: docker/login-action@v2
+        with:
+          registry: uhub.service.ucloud.cn
+          username: ${{ secrets.UCLOUD_USERNAME }}
+          password: ${{ secrets.UCLOUD_PASSWORD }}
+
+      - name: Configure scheduled build image tag # the tag would be ${SCHEDULED_BUILD_VERSION_PREFIX}-YYYYMMDD-${SCHEDULED_PERIOD}
+        shell: bash
+        if: github.event_name == 'schedule'
+        run: |
+          buildTime=`date "+%Y%m%d"`
+          SCHEDULED_BUILD_VERSION=${{ env.SCHEDULED_BUILD_VERSION_PREFIX }}-$buildTime-${{ env.SCHEDULED_PERIOD }}
+          echo "IMAGE_TAG=${SCHEDULED_BUILD_VERSION:1}" >> $GITHUB_ENV
+
+      - name: Configure tag # If the release tag is v0.1.0, then the image version tag will be 0.1.0.
+        shell: bash
+        if: github.event_name != 'schedule'
+        run: |
+          VERSION=${{ github.ref_name }}
+          echo "IMAGE_TAG=${VERSION:1}" >> $GITHUB_ENV
+
+      - name: Push image to uhub # Use 'docker buildx imagetools create' to create a new image base on source image.
+        run: |
+          docker buildx imagetools create \
+            --tag uhub.service.ucloud.cn/greptime/greptimedb:latest \
+            --tag uhub.service.ucloud.cn/greptime/greptimedb:${{ env.IMAGE_TAG }} \
+            greptime/greptimedb:${{ env.IMAGE_TAG }}
--- a/CONTRIBUTING.md
+++ b/CONTRIBUTING.md
@@ -1,4 +1,4 @@
-# Welcome!
+# Welcome 👋

 Thanks a lot for considering contributing to GreptimeDB. We believe people like you would make GreptimeDB a great product. We intend to build a community where individuals can have open talks, show respect for one another, and speak with true ❤️. Meanwhile, we are to keep transparency and make your effort count here.

@@ -50,34 +50,33 @@ GreptimeDB uses the [Apache 2.0 license](https://github.com/GreptimeTeam/greptim

 - To ensure that community is free and confident in its ability to use your contributions, please sign the Contributor License Agreement (CLA) which will be incorporated in the pull request process.
 - Make sure all your codes are formatted and follow the [coding style](https://pingcap.github.io/style-guide/rust/).
- Make sure all unit tests are passed.
+- Make sure all unit tests are passed (using `cargo test --workspace` or [nextest](https://nexte.st/index.html) `cargo nextest run`).
 - Make sure all clippy warnings are fixed (you can check it locally by running `cargo clippy --workspace --all-targets -- -D warnings -D clippy::print_stdout -D clippy::print_stderr`).

 #### `pre-commit` Hooks
+
 You could setup the [`pre-commit`](https://pre-commit.com/#plugins) hooks to run these checks on every commit automatically.

 1. Install `pre-commit`
-```
-$ pip install pre-commit
-```
-or
-```
-$ brew install pre-commit
-```
+
+        pip install pre-commit
+
+    or
+
+        brew install pre-commit

 2. Install the `pre-commit` hooks
-```
-$ pre-commit install
-pre-commit installed at .git/hooks/pre-commit

-$ pre-commit install --hook-type commit-msg
-pre-commit installed at .git/hooks/commit-msg
+        $ pre-commit install
+        pre-commit installed at .git/hooks/pre-commit

-$ pre-commit install --hook-type pre-push
-pre-commit installed at .git/hooks/pre-pus
-```
+        $ pre-commit install --hook-type commit-msg
+        pre-commit installed at .git/hooks/commit-msg

-now `pre-commit` will run automatically on `git commit`.
+        $ pre-commit install --hook-type pre-push
+        pre-commit installed at .git/hooks/pre-push
+
+Now, `pre-commit` will run automatically on `git commit`.

 ### Title

@@ -102,10 +101,12 @@ of what you were trying to do and what went wrong. You can also reach for help i
 ## Community

 The core team will be thrilled if you participate in any way you like. When you are stuck, try ask for help by filing an issue, with a detailed description of what you were trying to do and what went wrong. If you have any questions or if you would like to get involved in our community, please check out:
+
 - [GreptimeDB Community Slack](https://greptime.com/slack)
 - [GreptimeDB Github Discussions](https://github.com/GreptimeTeam/greptimedb/discussions)

 Also, see some extra GreptimeDB content:
+
 - [GreptimeDB Docs](https://greptime.com/docs)
 - [Learn GreptimeDB](https://greptime.com/products/db)
 - [Greptime Inc. Website](https://greptime.com)
--- a/Cargo.lock
+++ b/Cargo.lock
--- a/Cargo.toml
+++ b/Cargo.toml
@@ -12,12 +12,14 @@ members = [
    "src/common/function-macro",
    "src/common/grpc",
    "src/common/grpc-expr",
+    "src/common/mem-prof",
    "src/common/procedure",
    "src/common/query",
    "src/common/recordbatch",
    "src/common/runtime",
    "src/common/substrait",
    "src/common/telemetry",
+    "src/common/test-util",
    "src/common/time",
    "src/datanode",
    "src/datatypes",
@@ -37,39 +39,40 @@ members = [
    "src/storage",
    "src/store-api",
    "src/table",
+    "src/table-procedure",
    "tests-integration",
    "tests/runner",
 ]

 [workspace.package]
-version = "0.1.0"
+version = "0.1.1"
 edition = "2021"
 license = "Apache-2.0"

 [workspace.dependencies]
-arrow = "29.0"
-arrow-array = "29.0"
-arrow-flight = "29.0"
-arrow-schema = { version = "29.0", features = ["serde"] }
+arrow = { version = "34.0" }
+arrow-array = "34.0"
+arrow-flight = "34.0"
+arrow-schema = { version = "34.0", features = ["serde"] }
 async-stream = "0.3"
 async-trait = "0.1"
 chrono = { version = "0.4", features = ["serde"] }
-# TODO(LFC): Use released Datafusion when it officially dpendent on Arrow 29.0
-datafusion = { git = "https://github.com/apache/arrow-datafusion.git", rev = "4917235a398ae20145c87d20984e6367dc1a0c1e" }
-datafusion-common = { git = "https://github.com/apache/arrow-datafusion.git", rev = "4917235a398ae20145c87d20984e6367dc1a0c1e" }
-datafusion-expr = { git = "https://github.com/apache/arrow-datafusion.git", rev = "4917235a398ae20145c87d20984e6367dc1a0c1e" }
-datafusion-optimizer = { git = "https://github.com/apache/arrow-datafusion.git", rev = "4917235a398ae20145c87d20984e6367dc1a0c1e" }
-datafusion-physical-expr = { git = "https://github.com/apache/arrow-datafusion.git", rev = "4917235a398ae20145c87d20984e6367dc1a0c1e" }
-datafusion-sql = { git = "https://github.com/apache/arrow-datafusion.git", rev = "4917235a398ae20145c87d20984e6367dc1a0c1e" }
+datafusion = { git = "https://github.com/apache/arrow-datafusion.git", rev = "146a949218ec970784974137277cde3b4e547d0a" }
+datafusion-common = { git = "https://github.com/apache/arrow-datafusion.git", rev = "146a949218ec970784974137277cde3b4e547d0a" }
+datafusion-expr = { git = "https://github.com/apache/arrow-datafusion.git", rev = "146a949218ec970784974137277cde3b4e547d0a" }
+datafusion-optimizer = { git = "https://github.com/apache/arrow-datafusion.git", rev = "146a949218ec970784974137277cde3b4e547d0a" }
+datafusion-physical-expr = { git = "https://github.com/apache/arrow-datafusion.git", rev = "146a949218ec970784974137277cde3b4e547d0a" }
+datafusion-sql = { git = "https://github.com/apache/arrow-datafusion.git", rev = "146a949218ec970784974137277cde3b4e547d0a" }
 futures = "0.3"
 futures-util = "0.3"
-parquet = "29.0"
+parquet = "34.0"
 paste = "1.0"
 prost = "0.11"
 serde = { version = "1.0", features = ["derive"] }
 serde_json = "1.0"
 snafu = { version = "0.7", features = ["backtraces"] }
-sqlparser = "0.28"
+sqlparser = "0.32"
+tempfile = "3"
 tokio = { version = "1.24.2", features = ["full"] }
 tokio-util = "0.7"
 tonic = { version = "0.8", features = ["tls"] }
@@ -78,7 +81,7 @@ uuid = { version = "1", features = ["serde", "v4", "fast-rng"] }
 [profile.release]
 debug = true

-[profile.weekly]
+[profile.nightly]
 inherits = "release"
 strip = true
 lto = "thin"
--- a/4
+++ b/4
@@ -19,6 +19,10 @@ clean: ## Clean the project.
 fmt: ## Format all the Rust code.
 	cargo fmt --all

+.PHONY: fmt-toml
+fmt-toml: ## Format all TOML files.
+	taplo format --check --option "indent_string=    "
+
 .PHONY: docker-image
 docker-image: ## Build docker image.
 	docker build --network host -f docker/Dockerfile -t ${IMAGE_REGISTRY}:${IMAGE_TAG} .
--- a/README.md
+++ b/README.md
@@ -61,6 +61,12 @@ To compile GreptimeDB from source, you'll need:
  find an installation instructions [here](https://grpc.io/docs/protoc-installation/).
  **Note that `protoc` version needs to be >= 3.15** because we have used the `optional`
  keyword. You can check it with `protoc --version`.
+- python3-dev or python3-devel(Optional, only needed if you want to run scripts
+  in cpython): this install a Python shared library required for running python
+  scripting engine(In CPython Mode). This is available as `python3-dev` on
+  ubuntu, you can install it with `sudo apt install python3-dev`, or
+  `python3-devel` on RPM based distributions (e.g. Fedora, Red Hat, SuSE). Mac's
+  `Python3` package should have this shared library by default.

 #### Build with Docker

--- a/benchmarks/src/bin/nyc-taxi.rs
+++ b/benchmarks/src/bin/nyc-taxi.rs
@@ -27,7 +27,7 @@ use arrow::record_batch::RecordBatch;
 use clap::Parser;
 use client::api::v1::column::Values;
 use client::api::v1::{Column, ColumnDataType, ColumnDef, CreateTableExpr, InsertRequest, TableId};
-use client::{Client, Database};
+use client::{Client, Database, DEFAULT_CATALOG_NAME, DEFAULT_SCHEMA_NAME};
 use indicatif::{MultiProgress, ProgressBar, ProgressStyle};
 use parquet::arrow::arrow_reader::ParquetRecordBatchReaderBuilder;
 use tokio::task::JoinSet;
@@ -208,6 +208,7 @@ fn build_values(column: &ArrayRef) -> Values {
        | DataType::Dictionary(_, _)
        | DataType::Decimal128(_, _)
        | DataType::Decimal256(_, _)
+        | DataType::RunEndEncoded(_, _)
        | DataType::Map(_, _) => todo!(),
    }
 }
@@ -422,7 +423,7 @@ fn main() {
        .unwrap()
        .block_on(async {
            let client = Client::with_urls(vec![&args.endpoint]);
-            let db = Database::with_client(client);
+            let db = Database::new(DEFAULT_CATALOG_NAME, DEFAULT_SCHEMA_NAME, client);

            if !args.skip_write {
                do_write(&args, &db).await;
--- a/codecov.yml
+++ b/codecov.yml
@@ -8,3 +8,5 @@ coverage:
 ignore:
  - "**/error*.rs" # ignore all error.rs files
  - "tests/runner/*.rs" # ignore integration test runner
+comment:                  # this is a top-level key
+  layout: "diff"
--- a/config/datanode.example.toml
+++ b/config/datanode.example.toml
@@ -1,31 +1,50 @@
-node_id = 42
-mode = 'distributed'
-rpc_addr = '127.0.0.1:3001'
-rpc_hostname = '127.0.0.1'
-rpc_runtime_size = 8
-mysql_addr = '127.0.0.1:4406'
-mysql_runtime_size = 4
+# Node running mode, see `standalone.example.toml`.
+mode = "distributed"
+# Whether to use in-memory catalog, see `standalone.example.toml`.
 enable_memory_catalog = false
+# The datanode identifier, should be unique.
+node_id = 42
+# gRPC server address, "127.0.0.1:3001" by default.
+rpc_addr = "127.0.0.1:3001"
+# Hostname of this node.
+rpc_hostname = "127.0.0.1"
+# The number of gRPC server worker threads, 8 by default.
+rpc_runtime_size = 8

+# Metasrv client options.
+[meta_client_options]
+# Metasrv address list.
+metasrv_addrs = ["127.0.0.1:3002"]
+# Operation timeout in milliseconds, 3000 by default.
+timeout_millis = 3000
+# Connect server timeout in milliseconds, 5000 by default.
+connect_timeout_millis = 5000
+# `TCP_NODELAY` option for accepted connections, true by default.
+tcp_nodelay = true
+
+# WAL options, see `standalone.example.toml`.
 [wal]
 dir = "/tmp/greptimedb/wal"
-file_size = '1GB'
-purge_interval = '10m'
-purge_threshold = '50GB'
+file_size = "1GB"
+purge_threshold = "50GB"
+purge_interval = "10m"
 read_batch_size = 128
 sync_write = false

+# Storage options, see `standalone.example.toml`.
 [storage]
-type = 'File'
-data_dir = '/tmp/greptimedb/data/'
-
-[meta_client_opts]
-metasrv_addrs = ['127.0.0.1:3002']
-timeout_millis = 3000
-connect_timeout_millis = 5000
-tcp_nodelay = false
+type = "File"
+data_dir = "/tmp/greptimedb/data/"

+# Compaction options, see `standalone.example.toml`.
 [compaction]
 max_inflight_tasks = 4
-max_files_in_level0 = 16
+max_files_in_level0 = 8
 max_purge_tasks = 32
+
+# Procedure storage options, see `standalone.example.toml`.
+# [procedure.store]
+# type = "File"
+# data_dir = "/tmp/greptimedb/procedure/"
+# max_retry_times = 3
+# retry_delay = "500ms"
--- a/config/frontend.example.toml
+++ b/config/frontend.example.toml
@@ -1,12 +1,58 @@
-mode = 'distributed'
-datanode_rpc_addr = '127.0.0.1:3001'
+# Node running mode, see `standalone.example.toml`.
+mode = "distributed"

+# HTTP server options, see `standalone.example.toml`.
 [http_options]
-addr = '127.0.0.1:4000'
+addr = "127.0.0.1:4000"
 timeout = "30s"

-[meta_client_opts]
-metasrv_addrs = ['127.0.0.1:3002']
+# gRPC server options, see `standalone.example.toml`.
+[grpc_options]
+addr = "127.0.0.1:4001"
+runtime_size = 8
+
+# MySQL server options, see `standalone.example.toml`.
+[mysql_options]
+addr = "127.0.0.1:4002"
+runtime_size = 2
+
+# MySQL server TLS options, see `standalone.example.toml`.
+[mysql_options.tls]
+mode = "disable"
+cert_path = ""
+key_path = ""
+
+# PostgresSQL server options, see `standalone.example.toml`.
+[postgres_options]
+addr = "127.0.0.1:4003"
+runtime_size = 2
+
+# PostgresSQL server TLS options, see `standalone.example.toml`.
+[postgres_options.tls]
+mode = "disable"
+cert_path = ""
+key_path = ""
+
+# OpenTSDB protocol options, see `standalone.example.toml`.
+[opentsdb_options]
+addr = "127.0.0.1:4242"
+runtime_size = 2
+
+# InfluxDB protocol options, see `standalone.example.toml`.
+[influxdb_options]
+enable = true
+
+# Prometheus protocol options, see `standalone.example.toml`.
+[prometheus_options]
+enable = true
+
+# Prometheus protocol options, see `standalone.example.toml`.
+[prom_options]
+addr = "127.0.0.1:4004"
+
+# Metasrv client options, see `datanode.example.toml`.
+[meta_client_options]
+metasrv_addrs = ["127.0.0.1:3002"]
 timeout_millis = 3000
 connect_timeout_millis = 5000
-tcp_nodelay = false
+tcp_nodelay = true
--- a/config/metasrv.example.toml
+++ b/config/metasrv.example.toml
@@ -1,6 +1,15 @@
-bind_addr = '127.0.0.1:3002'
-server_addr = '127.0.0.1:3002'
-store_addr = '127.0.0.1:2379'
+# The bind address of metasrv, "127.0.0.1:3002" by default.
+bind_addr = "127.0.0.1:3002"
+# The communication server address for frontend and datanode to connect to metasrv,  "127.0.0.1:3002" by default for localhost.
+server_addr = "127.0.0.1:3002"
+# Etcd server address, "127.0.0.1:2379" by default.
+store_addr = "127.0.0.1:2379"
+# Datanode lease in seconds, 15 seconds by default.
 datanode_lease_secs = 15
-# selector: 'LeaseBased', 'LoadBased'
-selector = 'LeaseBased'
+# Datanode selector type.
+# - "LeaseBased" (default value).
+# - "LoadBased"
+# For details, please see "https://docs.greptime.com/developer-guide/meta/selector".
+selector = "LeaseBased"
+# Store data in memory, false by default.
+use_memory_store = false
--- a/config/standalone.example.toml
+++ b/config/standalone.example.toml
@@ -1,44 +1,120 @@
-node_id = 0
-mode = 'standalone'
+# Node running mode, "standalone" or "distributed".
+mode = "standalone"
+# Whether to use in-memory catalog, `false` by default.
 enable_memory_catalog = false

+# HTTP server options.
 [http_options]
-addr = '127.0.0.1:4000'
+# Server address, "127.0.0.1:4000" by default.
+addr = "127.0.0.1:4000"
+# HTTP request timeout, 30s by default.
 timeout = "30s"

-[wal]
-dir = "/tmp/greptimedb/wal"
-file_size = '1GB'
-purge_interval = '10m'
-purge_threshold = '50GB'
-read_batch_size = 128
-sync_write = false
-
-
-[storage]
-type = 'File'
-data_dir = '/tmp/greptimedb/data/'
-
+# gRPC server options.
 [grpc_options]
-addr = '127.0.0.1:4001'
+# Server address, "127.0.0.1:4001" by default.
+addr = "127.0.0.1:4001"
+# The number of server worker threads, 8 by default.
 runtime_size = 8

+# MySQL server options.
 [mysql_options]
-addr = '127.0.0.1:4002'
+# Server address, "127.0.0.1:4002" by default.
+addr = "127.0.0.1:4002"
+# The number of server worker threads, 2 by default.
 runtime_size = 2

-[influxdb_options]
-enable = true
-
-[opentsdb_options]
-addr = '127.0.0.1:4242'
-enable = true
-runtime_size = 2
-
-[prometheus_options]
-enable = true
+# MySQL server TLS options.
+[mysql_options.tls]
+# TLS mode, refer to https://www.postgresql.org/docs/current/libpq-ssl.html
+# - "disable" (default value)
+# - "prefer"
+# - "require"
+# - "verify-ca"
+# - "verify-full"
+mode = "disable"
+# Certificate file path.
+cert_path = ""
+# Private key file path.
+key_path = ""

+# PostgresSQL server options.
 [postgres_options]
-addr = '127.0.0.1:4003'
+# Server address, "127.0.0.1:4003" by default.
+addr = "127.0.0.1:4003"
+# The number of server worker threads, 2 by default.
 runtime_size = 2
-check_pwd = false
+
+# PostgresSQL server TLS options, see `[mysql_options.tls]` section.
+[postgres_options.tls]
+# TLS mode.
+mode = "disable"
+# certificate file path.
+cert_path = ""
+# private key file path.
+key_path = ""
+
+# OpenTSDB protocol options.
+[opentsdb_options]
+# OpenTSDB telnet API server address, "127.0.0.1:4242" by default.
+addr = "127.0.0.1:4242"
+# The number of server worker threads, 2 by default.
+runtime_size = 2
+
+# InfluxDB protocol options.
+[influxdb_options]
+# Whether to enable InfluxDB protocol in HTTP API, true by default.
+enable = true
+
+# Prometheus protocol options.
+[prometheus_options]
+# Whether to enable Prometheus remote write and read in HTTP API, true by default.
+enable = true
+
+# Prom protocol options.
+[prom_options]
+# Prometheus API server address, "127.0.0.1:4004" by default.
+addr = "127.0.0.1:4004"
+
+# WAL options.
+[wal]
+# WAL data directory.
+dir = "/tmp/greptimedb/wal"
+# WAL file size in bytes.
+file_size = "1GB"
+# WAL purge threshold in bytes.
+purge_threshold = "50GB"
+# WAL purge interval in seconds.
+purge_interval = "10m"
+# WAL read batch size.
+read_batch_size = 128
+# Whether to sync log file after every write.
+sync_write = false
+
+# Storage options.
+[storage]
+# Storage type.
+type = "File"
+# Data directory, "/tmp/greptimedb/data" by default.
+data_dir = "/tmp/greptimedb/data/"
+
+# Compaction options.
+[compaction]
+# Max task number that can concurrently run.
+max_inflight_tasks = 4
+# Max files in level 0 to trigger compaction.
+max_files_in_level0 = 8
+# Max task number for SST purge task after compaction.
+max_purge_tasks = 32
+
+# Procedure storage options.
+# Uncomment to enable.
+# [procedure.store]
+# # Storage type.
+# type = "File"
+# # Procedure data path.
+# data_dir = "/tmp/greptimedb/procedure/"
+# # Procedure max retry time.
+# max_retry_times = 3
+# # Initial retry delay of procedures, increases exponentially
+# retry_delay = "500ms"
--- a/docker/Dockerfile
+++ b/docker/Dockerfile
@@ -9,7 +9,10 @@ RUN apt-get update && apt-get install -y \
    protobuf-compiler \
    curl \
    build-essential \
-    pkg-config
+    pkg-config \
+    python3 \
+    python3-dev \
+    && pip install pyarrow

 # Install Rust.
 SHELL ["/bin/bash", "-c"]
--- a/docker/aarch64/Dockerfile
+++ b/docker/aarch64/Dockerfile
@@ -0,0 +1,57 @@
+FROM ubuntu:22.04 as builder
+
+ENV LANG en_US.utf8
+WORKDIR /greptimedb
+
+# Install dependencies.
+RUN apt-get update && apt-get install -y \
+    libssl-dev \
+    protobuf-compiler \
+    curl \
+    build-essential \
+    pkg-config \
+    wget
+
+# 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/:$PATH
+
+# Install cross platform toolchain
+RUN apt-get -y update && \
+    apt-get -y install g++-aarch64-linux-gnu gcc-aarch64-linux-gnu && \
+    apt-get install binutils-aarch64-linux-gnu
+
+COPY ./docker/aarch64/compile-python.sh ./docker/aarch64/
+RUN chmod +x ./docker/aarch64/compile-python.sh && \
+    ./docker/aarch64/compile-python.sh
+
+COPY ./rust-toolchain.toml .
+# Install rustup target for cross compiling.
+RUN rustup target add aarch64-unknown-linux-gnu
+COPY . .
+# Update dependency, using separate `RUN` to separate cache
+RUN cargo fetch
+
+# This three env var is set in script, so I set it manually in dockerfile.
+ENV LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/local/lib/
+ENV LIBRARY_PATH=$LIBRARY_PATH:/usr/local/lib/
+ENV PY_INSTALL_PATH=/greptimedb/python_arm64_build
+
+# Set the environment variable for cross compiling and compile it
+# cross compiled python is `python3` in path, but pyo3 need `python` in path so alias it
+# Build the project in release mode.
+RUN export PYO3_CROSS_LIB_DIR=$PY_INSTALL_PATH/lib && \ 
+    alias python=python3 && \
+    cargo build --target aarch64-unknown-linux-gnu --release -F pyo3_backend
+
+# Exporting the binary to the clean image
+FROM ubuntu:22.04 as base
+
+RUN apt-get update && DEBIAN_FRONTEND=noninteractive apt-get -y install ca-certificates
+
+WORKDIR /greptime
+COPY --from=builder /greptimedb/target/aarch64-unknown-linux-gnu/release/greptime /greptime/bin/
+ENV PATH /greptime/bin/:$PATH
+
+ENTRYPOINT ["greptime"]
--- a/docker/aarch64/compile-python.sh
+++ b/docker/aarch64/compile-python.sh
@@ -0,0 +1,46 @@
+# this script will download Python source code, compile it, and install it to /usr/local/lib
+# then use this python to compile cross-compiled python for aarch64
+
+wget https://www.python.org/ftp/python/3.10.10/Python-3.10.10.tgz
+tar -xvf Python-3.10.10.tgz
+cd Python-3.10.10
+# explain Python compile options here a bit:s
+# --enable-shared: enable building a shared Python library (default is no) but we do need it for calling from rust
+# CC, CXX, AR, LD, RANLIB: set the compiler, archiver, linker, and ranlib programs to use
+# build: the machine you are building on, host: the machine you will run the compiled program on
+# --with-system-ffi: build _ctypes module using an installed ffi library, see Doc/library/ctypes.rst, not used in here TODO: could remove
+# ac_cv_pthread_is_default=no ac_cv_pthread=yes ac_cv_cxx_thread=yes:
+# allow cross-compiled python to have -pthread set for CXX, see https://github.com/python/cpython/pull/22525
+# ac_cv_have_long_long_format=yes: target platform supports long long type
+# disable-ipv6: disable ipv6 support, we don't need it in here
+# ac_cv_file__dev_ptmx=no ac_cv_file__dev_ptc=no: disable pty support, we don't need it in here
+
+# Build local python first, then build cross-compiled python.
+./configure \
+--enable-shared \
+ac_cv_pthread_is_default=no ac_cv_pthread=yes ac_cv_cxx_thread=yes \
+ac_cv_have_long_long_format=yes \
+--disable-ipv6 ac_cv_file__dev_ptmx=no ac_cv_file__dev_ptc=no && \
+make
+make install
+cd ..
+export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/local/lib/
+export LIBRARY_PATH=$LIBRARY_PATH:/usr/local/lib/
+export PY_INSTALL_PATH=$(pwd)/python_arm64_build
+cd Python-3.10.10 && \
+make clean && \
+make distclean && \
+alias python=python3 && \
+./configure --build=x86_64-linux-gnu --host=aarch64-linux-gnu \
+--prefix=$PY_INSTALL_PATH --enable-optimizations \
+CC=aarch64-linux-gnu-gcc \
+CXX=aarch64-linux-gnu-g++ \
+AR=aarch64-linux-gnu-ar \
+LD=aarch64-linux-gnu-ld \
+RANLIB=aarch64-linux-gnu-ranlib \
+--enable-shared \
+ac_cv_pthread_is_default=no ac_cv_pthread=yes ac_cv_cxx_thread=yes \
+ac_cv_have_long_long_format=yes \
+--disable-ipv6 ac_cv_file__dev_ptmx=no ac_cv_file__dev_ptc=no && \
+make && make altinstall && \
+cd ..
--- a/docs/rfcs/2022-12-20-promql-in-rust/rfc.md
+++ b/docs/rfcs/2022-12-20-promql-in-rust/rfc.md
@@ -149,10 +149,10 @@ inputs:
  - title: 'Series Normalize: \noffset = 0'
    operator: prom
    inputs:
-    - title: 'Filter: \ntimetamp > 2022-12-20T10:00:00 && timestamp < 2022-12-21T10:00:00'
+    - title: 'Filter: \ntimestamp > 2022-12-20T10:00:00 && timestamp < 2022-12-21T10:00:00'
      operator: filter
      inputs:
-      - title: 'Table Scan: \ntable = request_duration, timetamp > 2022-12-20T10:00:00 && timestamp < 2022-12-21T10:00:00'
+      - title: 'Table Scan: \ntable = request_duration, timestamp > 2022-12-20T10:00:00 && timestamp < 2022-12-21T10:00:00'
        operator: scan -->

 ![example](example.png)
--- a/docs/rfcs/2023-03-08-region-fault-tolerance.md
+++ b/docs/rfcs/2023-03-08-region-fault-tolerance.md
@@ -0,0 +1,196 @@
+---
+Feature Name: "Fault Tolerance for Region"
+Tracking Issue: https://github.com/GreptimeTeam/greptimedb/issues/1126
+Date: 2023-03-08
+Author: "Luo Fucong <luofucong@greptime.com>"
+---
+
+Fault Tolerance for Region
+----------------------
+
+# Summary
+
+This RFC proposes a method to achieve fault tolerance for regions in GreptimeDB's distributed mode. Or, put it in another way, achieving region high availability("HA") for GreptimeDB cluster. 
+
+In this RFC, we mainly describe two aspects of region HA: how region availability is detected, and what recovery process is need to be taken. We also discuss some alternatives and future work.
+
+When this feature is done, our users could expect a GreptimeDB cluster that can always handle their requests to regions, despite some requests may failed during the region failover. The optimization to reduce the MTTR(Mean Time To Recovery) is not a concern of this RPC, and is left for future work.
+
+# Motivation
+
+Fault tolerance for regions is a critical feature for our clients to use the GreptimeDB cluster confidently. High availability for users to interact with their stored data is a "must have" for any TSDB products, that include our GreptimeDB cluster.
+
+# Details
+
+## Background
+
+Some backgrounds about region in distributed mode:
+
+- A table is logically split into multiple regions. Each region stores a part of non-overlapping table data.
+- Regions are distributed in Datanodes, the mappings are not static, are assigned and governed by Metasrv.
+- In distributed mode, client requests are scoped in regions. To be more specific, when a request that needs to scan multiple regions arrived in Frontend, Frontend splits the request into multiple sub-requests, each of which scans one region only, and submits them to Datanodes that hold corresponding regions. 
+
+In conclusion, as long as regions remain available, and regions could regain availability when failures do occur, the overall region HA could be achieved. With this in mind, let's see how region failures are detected first. 
+
+## Failure Detection
+
+We detect region failures in Metasrv, and do it both passively and actively. Passively means that Metasrv do not fire some "are you healthy" requests to regions. Instead, we carry region healthy information in the heartbeat requests that are submit to Metasrv by Datanodes. 
+
+Datanode already carries its regions stats in the heartbeat request (the non-relevant fields are omitted):
+
+```protobuf
+message HeartbeatRequest {
+  ...
+  // Region stats on this node
+  repeated RegionStat region_stats = 6;
+  ...
+}
+
+message RegionStat {
+  uint64 region_id = 1;
+  TableName table_name = 2;
+  ...
+}
+```
+
+For the sake of simplicity, we don't add another field `bool available = 3` to the `RegionStat` message; instead, if the region were unavailable in the view of the Datanode that contains it, the Datanode just not includes the `RegionStat` of it in the heartbeat request. Or, if the Datanode itself is not unavailable, the heartbeat request is not submitted, effectively the same with not carrying the `RegionStat`.
+
+> The heartbeat interval is now hardcoded to five seconds.
+
+Metasrv gathers the heartbeat requests, extracts the `RegionStat`s, and treat them as region heartbeat. In this way, Metasrv maintains all regions healthy information. If some region's heartbeats were not received in a period of time, Metasrv speculates the region might be unavailable. To make the decision whether a region is failed or not, Metasrv uses a failure detection algorithm called the "[Phi φ Accrual Failure Detection](https://medium.com/@arpitbhayani/phi-%CF%86-accrual-failure-detection-79c21ce53a7a)". Basically, the algorithm calculates a value called "phi" to represent the possibility of a region's unavailability, based on the historical heartbeats' arrived rate. Once the "phi" is above some pre-defined threshold, Metasrv knows the region is failed.
+
+> This algorithm has been widely adopted in some well known products, like Akka and Cassandra.
+
+When Metasrv decides some region is failed from heartbeats, it's not the final decision. Here comes the "actively" detection. Before Metasrv decides to do region failover, it actively invokes the healthy check interface of the Datanode that the failure region resides. Only this healthy check is failed does Metasrv actually start doing failover upon the region.
+
+To conclude, the failure detection pseudo-codes are like this:
+
+```rust
+// in Metasrv:
+fn failure_detection() {
+    loop {
+        // passive detection
+        let failed_regions = all_regions.iter().filter(|r| r.estimated_failure_possibility() > config.phi).collect();
+
+        // find the datanodes that contains the failed regions
+        let datanodes_and_regions = find_region_resides_datanodes(failed_regions);
+
+        // active detection  
+        for (datanode, regions) in datanodes_and_regions {
+            if !datanode.is_healthy(regions) {
+                do_failover(datanode, regions);
+            }
+        }
+
+        sleep(config.detect_interval);
+    }
+}
+```
+
+Some design considerations:
+
+- Why active detecting while we have passively detection? Because it could be happened that the network is singly connectable sometimes (especially in the complex Cloud environment), then the Datanode's heartbeats cannot reach Metasrv, while Metasrv could request Datanode. Active detecting avoid this false positive situation.
+- Why the detection works on region instead of Datanode? Because we might face the possibility that only part of the regions in the Datanode are not available, not ALL regions. Especially the situation that Datanodes are used by multiple tenants. If this is the case, it's better to do failover upon the designated regions instead of the whole regions that reside on the Datanode. All in all, we want a more subtle control over region failover. 
+
+So we detect some regions are not available. How to regain the availability back?
+
+## Region Failover
+
+Region Failover largely relies on remote WAL, aka "[Bunshin](https://github.com/GreptimeTeam/bunshin)". I'm not including any of the details of it in this RFC, let's just assume we already have it.
+
+In general, region failover is fairly simple. Once Metasrv decides to do failover upon some regions, it first chooses one or more Datanodes to hold the failed region. This can be done easily, as the Metasrv already has the whole picture of Datanodes: it knows which Datanode has the minimum regions, what Datanode historically had the lowest CPU usage and IO rate, and how the Datanodes are assigned to tenants, among other information that can all help the Metasrv choose the most suitable Datanodes. Let's call these chosen Datanodes as "candidates".
+
+> The strategy to choose the most suitable candidates required careful design, but it's another RFC.
+
+Then, Metasrv sets the states of these failed regions as "passive". We should add a field to `Region`:
+
+```protobuf
+message Region {
+  uint64 id = 1;
+  string name = 2;
+  Partition partition = 3;
+  
+  message State {
+    Active,
+    Passive,
+  }
+  State state = 4;
+  
+  map<string, string> attrs = 100;
+}
+```
+
+Here `Region` is used in message `RegionRoute`, which indicates how the write request is split among regions. When a region is set as "passive", Frontend knows the write to it should be rejected at the moment (the region read is not blocked, however).
+
+> Making a region "passive" here is effectively blocking the write to it. It's ok in the failover situation, the region is failed anyway. However, when dealing with active maintenance operations, region state requires more refined design. But that's another story. 
+
+Third, Metasrv fires the "close region" requests to the failed Datanodes, and fires the "open region" requests to those candidates. "Close region" requests might be failed due to the unavailability of Datanodes, but that's fine, it's just a best-effort attempt to reduce the chance of any in-flight writes got handled unintentionally after the region is set as "passive". The "open region" requests must have succeeded though. Datanodes open regions from remote WAL.
+
+> Currently the "close region" is undefined in Datanode. It could be a local cache clean up of region data or other resources tidy up.
+
+Finally, when a candidate successfully opens its region, it calls back to Metasrv, indicating it is ready to handle region. "call back" here is backed by its heartbeat to Metasrv. Metasrv updates the region's state to "active", so as to let Frontend lifts the restrictions of region writes (again, the read part of region is untouched).
+
+All the above steps should be managed by remote procedure framework. It's another implementation challenge in the region failover feature. (One is the remote WAL of course.)
+
+A picture is worth a 1000 words:
+
+```text
+                                    +-------------------------+                                        
+                                    | Metasrv detects region  |                                        
+                                    | failure                 |                                        
+                                    +-------------------------+                                        
+                                                 |                                                     
+                                                 v                                                     
+                                    +----------------------------+                                        
+                                    | Metasrv chooses candidates |                                        
+                                    | to hold failed regions     |                                        
+                                    +----------------------------+                                        
+                                                 |                                                     
+                                                 v                                                     
+                                    +-------------------------+       +-------------------------+      
+                                    | Metasrv "passive" the   |------>| Frontend rejects writes |      
+                                    | failed regions          |       | to "passive" regions    |      
+                                    +-------------------------+       +-------------------------+      
+                                                 |                                                     
+                                                 v                                                     
+--------------------------+        +---------------------------+                                        
+| Candidate Datanodes open |<-------| Metasrv fires "close" and |                                        
+| regions from remote WAL  |        | "open" region requests    |                                        
+--------------------------+        +---------------------------+                                        
+             |                                                                                         
+             |                                                                                         
+             |                      +-------------------------+       +-------------------------+      
+             +--------------------->| Metasrv "active" the    |------>| Frontend lifts write    |      
+                                    | failed regions          |       | restriction to regions  |      
+                                    +-------------------------+       +-------------------------+      
+                                                 |                                                     
+                                                 v                                                     
+                                    +-------------------------+                                        
+                                    | Region failover done,   |                                        
+                                    | HA regain               |                                        
+                                    +-------------------------+                                        
+```
+
+# Alternatives
+
+## The "Neon" Way
+
+Remote WAL raises a problem that could harm the write throughput of GreptimeDB cluster: each write request has to do at least two remote call, one is from Frontend to Datanode, and one is from Datanode to remote WAL. What if we do it the "[Neon](https://github.com/neondatabase/neon)" way, making remote WAL sits in between the Frontend and Datanode, couldn't that improve our write throughput? It could, though there're some consistency issues like "read-your-writes" to solve.
+
+However, the main concerns we don't adopt this method are two-fold:
+
+1. Remote WAL is planned to be quorum based, it can be efficiently written;
+2. More importantly, we are planning to make the remote WAL an option that users could choose not to enable it (at the cost of some reliability reduction).
+
+## No WAL, Replication instead
+
+This method replicates region across Datanodes directly, like the common way in shared-nothing database. Were the main region failed, a standby region in the replicate group is elected as new "main" and take the read/write requests. The main concern to this method is the incompatibility to our current architecture and code structure. It requires a major redesign, but gains no significant advantage over the remote WAL method. 
+
+However, the replication does have its own advantage that we can learn from to optimize this failover procedure.
+
+# Future Work
+
+Some optimizations we could take:
+
+- To reduce the MTTR, we could make Metasrv chooses the candidate to each region at normal time. The candidate does some preparation works to reduce the open region time, effectively accelerate the failover procedure.
+- We can adopt the replication method, to the degree that region replicas are used as the fast catch-up candidates. The data difference among replicas is minor, region failover does not need to load or exchange too much data, greatly reduced the region failover time.
--- a/rust-toolchain.toml
+++ b/rust-toolchain.toml
@@ -1,2 +1,2 @@
 [toolchain]
-channel = "nightly-2023-02-14"
+channel = "nightly-2023-02-26"
--- a/scripts/install.sh
+++ b/scripts/install.sh
@@ -59,5 +59,5 @@ if [ -n "${OS_TYPE}" ] && [ -n "${ARCH_TYPE}" ]; then
        wget "https://github.com/${GITHUB_ORG}/${GITHUB_REPO}/releases/download/${VERSION}/${BIN}-${OS_TYPE}-${ARCH_TYPE}.tgz"
    fi

-    tar xvf ${BIN}-${OS_TYPE}-${ARCH_TYPE}.tgz && rm ${BIN}-${OS_TYPE}-${ARCH_TYPE}.tgz && echo "Run '${BIN} --help' to get started"
+    tar xvf ${BIN}-${OS_TYPE}-${ARCH_TYPE}.tgz && rm ${BIN}-${OS_TYPE}-${ARCH_TYPE}.tgz && echo "Run './${BIN} --help' to get started"
 fi
--- a/src/api/Cargo.toml
+++ b/src/api/Cargo.toml
@@ -10,7 +10,7 @@ common-base = { path = "../common/base" }
 common-error = { path = "../common/error" }
 common-time = { path = "../common/time" }
 datatypes = { path = "../datatypes" }
-greptime-proto = { git = "https://github.com/GreptimeTeam/greptime-proto.git", rev = "3e6349be127b65a8b42a38cda9d527ec423ca77d" }
+greptime-proto = { git = "https://github.com/GreptimeTeam/greptime-proto.git", rev = "3a715150563b89d5dfc81a5838eac1f66a5658a1" }
 prost.workspace = true
 snafu = { version = "0.7", features = ["backtraces"] }
 tonic.workspace = true
--- a/src/api/src/helper.rs
+++ b/src/api/src/helper.rs
@@ -97,7 +97,9 @@ impl TryFrom<ConcreteDataType> for ColumnDataTypeWrapper {
                TimestampType::Microsecond(_) => ColumnDataType::TimestampMicrosecond,
                TimestampType::Nanosecond(_) => ColumnDataType::TimestampNanosecond,
            },
-            ConcreteDataType::Null(_) | ConcreteDataType::List(_) => {
+            ConcreteDataType::Null(_)
+            | ConcreteDataType::List(_)
+            | ConcreteDataType::Dictionary(_) => {
                return error::IntoColumnDataTypeSnafu { from: datatype }.fail()
            }
        });
--- a/src/catalog/Cargo.toml
+++ b/src/catalog/Cargo.toml
@@ -18,25 +18,28 @@ common-recordbatch = { path = "../common/recordbatch" }
 common-runtime = { path = "../common/runtime" }
 common-telemetry = { path = "../common/telemetry" }
 common-time = { path = "../common/time" }
+dashmap = "5.4"
 datafusion.workspace = true
 datatypes = { path = "../datatypes" }
 futures = "0.3"
 futures-util.workspace = true
 lazy_static = "1.4"
 meta-client = { path = "../meta-client" }
+parking_lot = "0.12"
 regex = "1.6"
 serde = "1.0"
 serde_json = "1.0"
+session = { path = "../session" }
 snafu = { version = "0.7", features = ["backtraces"] }
 storage = { path = "../storage" }
 table = { path = "../table" }
 tokio.workspace = true

 [dev-dependencies]
+common-test-util = { path = "../common/test-util" }
 chrono.workspace = true
 log-store = { path = "../log-store" }
 mito = { path = "../mito", features = ["test"] }
 object-store = { path = "../object-store" }
 storage = { path = "../storage" }
-tempdir = "0.3"
 tokio.workspace = true
--- a/src/catalog/src/error.rs
+++ b/src/catalog/src/error.rs
@@ -201,6 +201,24 @@ pub enum Error {
        #[snafu(backtrace)]
        source: common_catalog::error::Error,
    },
+
+    #[snafu(display("Illegal access to catalog: {} and schema: {}", catalog, schema))]
+    QueryAccessDenied { catalog: String, schema: String },
+
+    #[snafu(display(
+        "Failed to get region stats, catalog: {}, schema: {}, table: {}, source: {}",
+        catalog,
+        schema,
+        table,
+        source
+    ))]
+    RegionStats {
+        catalog: String,
+        schema: String,
+        table: String,
+        #[snafu(backtrace)]
+        source: table::error::Error,
+    },
 }

 pub type Result<T> = std::result::Result<T, Error>;
@@ -235,7 +253,8 @@ impl ErrorExt for Error {
            | Error::InsertCatalogRecord { source, .. }
            | Error::OpenTable { source, .. }
            | Error::CreateTable { source, .. }
-            | Error::DeregisterTable { source, .. } => source.status_code(),
+            | Error::DeregisterTable { source, .. }
+            | Error::RegionStats { source, .. } => source.status_code(),

            Error::MetaSrv { source, .. } => source.status_code(),
            Error::SystemCatalogTableScan { source } => source.status_code(),
@@ -246,6 +265,7 @@ impl ErrorExt for Error {
            }

            Error::Unimplemented { .. } => StatusCode::Unsupported,
+            Error::QueryAccessDenied { .. } => StatusCode::AccessDenied,
        }
    }

--- a/src/catalog/src/helper.rs
+++ b/src/catalog/src/helper.rs
@@ -24,10 +24,10 @@ use serde::{Deserialize, Serialize, Serializer};
 use snafu::{ensure, OptionExt, ResultExt};
 use table::metadata::{RawTableInfo, TableId, TableVersion};

-const CATALOG_KEY_PREFIX: &str = "__c";
-const SCHEMA_KEY_PREFIX: &str = "__s";
-const TABLE_GLOBAL_KEY_PREFIX: &str = "__tg";
-const TABLE_REGIONAL_KEY_PREFIX: &str = "__tr";
+pub const CATALOG_KEY_PREFIX: &str = "__c";
+pub const SCHEMA_KEY_PREFIX: &str = "__s";
+pub const TABLE_GLOBAL_KEY_PREFIX: &str = "__tg";
+pub const TABLE_REGIONAL_KEY_PREFIX: &str = "__tr";

 const ALPHANUMERICS_NAME_PATTERN: &str = "[a-zA-Z_][a-zA-Z0-9_]*";

@@ -370,4 +370,10 @@ mod tests {
        let deserialized = TableGlobalValue::parse(serialized).unwrap();
        assert_eq!(value, deserialized);
    }
+
+    #[test]
+    fn test_table_global_value_compatibility() {
+        let s = r#"{"node_id":1,"regions_id_map":{"1":[0]},"table_info":{"ident":{"table_id":1098,"version":1},"name":"container_cpu_limit","desc":"Created on insertion","catalog_name":"greptime","schema_name":"dd","meta":{"schema":{"column_schemas":[{"name":"container_id","data_type":{"String":null},"is_nullable":true,"is_time_index":false,"default_constraint":null,"metadata":{}},{"name":"container_name","data_type":{"String":null},"is_nullable":true,"is_time_index":false,"default_constraint":null,"metadata":{}},{"name":"docker_image","data_type":{"String":null},"is_nullable":true,"is_time_index":false,"default_constraint":null,"metadata":{}},{"name":"host","data_type":{"String":null},"is_nullable":true,"is_time_index":false,"default_constraint":null,"metadata":{}},{"name":"image_name","data_type":{"String":null},"is_nullable":true,"is_time_index":false,"default_constraint":null,"metadata":{}},{"name":"image_tag","data_type":{"String":null},"is_nullable":true,"is_time_index":false,"default_constraint":null,"metadata":{}},{"name":"interval","data_type":{"String":null},"is_nullable":true,"is_time_index":false,"default_constraint":null,"metadata":{}},{"name":"runtime","data_type":{"String":null},"is_nullable":true,"is_time_index":false,"default_constraint":null,"metadata":{}},{"name":"short_image","data_type":{"String":null},"is_nullable":true,"is_time_index":false,"default_constraint":null,"metadata":{}},{"name":"type","data_type":{"String":null},"is_nullable":true,"is_time_index":false,"default_constraint":null,"metadata":{}},{"name":"dd_value","data_type":{"Float64":{}},"is_nullable":true,"is_time_index":false,"default_constraint":null,"metadata":{}},{"name":"ts","data_type":{"Timestamp":{"Millisecond":null}},"is_nullable":false,"is_time_index":true,"default_constraint":null,"metadata":{"greptime:time_index":"true"}},{"name":"git.repository_url","data_type":{"String":null},"is_nullable":true,"is_time_index":false,"default_constraint":null,"metadata":{}}],"timestamp_index":11,"version":1},"primary_key_indices":[0,1,2,3,4,5,6,7,8,9,12],"value_indices":[10,11],"engine":"mito","next_column_id":12,"region_numbers":[],"engine_options":{},"options":{},"created_on":"1970-01-01T00:00:00Z"},"table_type":"Base"}}"#;
+        TableGlobalValue::parse(s).unwrap();
+    }
 }
--- a/src/catalog/src/lib.rs
+++ b/src/catalog/src/lib.rs
@@ -18,6 +18,7 @@ use std::any::Any;
 use std::fmt::{Debug, Formatter};
 use std::sync::Arc;

+use api::v1::meta::{RegionStat, TableName};
 use common_telemetry::info;
 use snafu::{OptionExt, ResultExt};
 use table::engine::{EngineContext, TableEngineRef};
@@ -34,6 +35,7 @@ pub mod local;
 pub mod remote;
 pub mod schema;
 pub mod system;
+pub mod table_source;
 pub mod tables;

 /// Represent a list of named catalogs
@@ -107,7 +109,12 @@ pub trait CatalogManager: CatalogList {
    fn schema(&self, catalog: &str, schema: &str) -> Result<Option<SchemaProviderRef>>;

    /// Returns the table by catalog, schema and table name.
-    fn table(&self, catalog: &str, schema: &str, table_name: &str) -> Result<Option<TableRef>>;
+    async fn table(
+        &self,
+        catalog: &str,
+        schema: &str,
+        table_name: &str,
+    ) -> Result<Option<TableRef>>;
 }

 pub type CatalogManagerRef = Arc<dyn CatalogManager>;
@@ -186,7 +193,8 @@ pub(crate) async fn handle_system_table_request<'a, M: CatalogManager>(
        let table_name = &req.create_table_request.table_name;
        let table_id = req.create_table_request.id;

-        let table = if let Some(table) = manager.table(catalog_name, schema_name, table_name)? {
+        let table = manager.table(catalog_name, schema_name, table_name).await?;
+        let table = if let Some(table) = table {
            table
        } else {
            let table = engine
@@ -218,10 +226,10 @@ pub(crate) async fn handle_system_table_request<'a, M: CatalogManager>(
    Ok(())
 }

-/// The number of regions in the datanode node.
-pub fn region_number(catalog_manager: &CatalogManagerRef) -> Result<u64> {
-    let mut region_number: u64 = 0;
-
+/// The stat of regions in the datanode node.
+/// The number of regions can be got from len of vec.
+pub async fn region_stats(catalog_manager: &CatalogManagerRef) -> Result<Vec<RegionStat>> {
+    let mut region_stats = Vec::new();
    for catalog_name in catalog_manager.catalog_names()? {
        let catalog =
            catalog_manager
@@ -239,16 +247,36 @@ pub fn region_number(catalog_manager: &CatalogManagerRef) -> Result<u64> {
                })?;

            for table_name in schema.table_names()? {
-                let table = schema
-                    .table(&table_name)?
-                    .context(error::TableNotFoundSnafu {
-                        table_info: &table_name,
-                    })?;
+                let table =
+                    schema
+                        .table(&table_name)
+                        .await?
+                        .context(error::TableNotFoundSnafu {
+                            table_info: &table_name,
+                        })?;

-                let region_numbers = &table.table_info().meta.region_numbers;
-                region_number += region_numbers.len() as u64;
+                region_stats.extend(
+                    table
+                        .region_stats()
+                        .context(error::RegionStatsSnafu {
+                            catalog: &catalog_name,
+                            schema: &schema_name,
+                            table: &table_name,
+                        })?
+                        .into_iter()
+                        .map(|stat| RegionStat {
+                            region_id: stat.region_id,
+                            table_name: Some(TableName {
+                                catalog_name: catalog_name.clone(),
+                                schema_name: schema_name.clone(),
+                                table_name: table_name.clone(),
+                            }),
+                            approximate_bytes: stat.disk_usage_bytes as i64,
+                            ..Default::default()
+                        }),
+                );
            }
        }
    }
-    Ok(region_number)
+    Ok(region_stats)
 }
--- a/src/catalog/src/local/manager.rs
+++ b/src/catalog/src/local/manager.rs
@@ -345,7 +345,7 @@ impl CatalogManager for LocalCatalogManager {

        {
            let _lock = self.register_lock.lock().await;
-            if let Some(existing) = schema.table(&request.table_name)? {
+            if let Some(existing) = schema.table(&request.table_name).await? {
                if existing.table_info().ident.table_id != request.table_id {
                    error!(
                        "Unexpected table register request: {:?}, existing: {:?}",
@@ -434,9 +434,10 @@ impl CatalogManager for LocalCatalogManager {
            } = &request;
            let table_id = self
                .catalogs
-                .table(catalog, schema, table_name)?
+                .table(catalog, schema, table_name)
+                .await?
                .with_context(|| error::TableNotExistSnafu {
-                    table: format!("{catalog}.{schema}.{table_name}"),
+                    table: format_full_table_name(catalog, schema, table_name),
                })?
                .table_info()
                .ident
@@ -505,7 +506,7 @@ impl CatalogManager for LocalCatalogManager {
            .schema(schema)
    }

-    fn table(
+    async fn table(
        &self,
        catalog_name: &str,
        schema_name: &str,
@@ -521,7 +522,7 @@ impl CatalogManager for LocalCatalogManager {
                catalog: catalog_name,
                schema: schema_name,
            })?;
-        schema.table(table_name)
+        schema.table(table_name).await
    }
 }

--- a/src/catalog/src/local/memory.rs
+++ b/src/catalog/src/local/memory.rs
@@ -18,6 +18,7 @@ use std::collections::HashMap;
 use std::sync::atomic::{AtomicU32, Ordering};
 use std::sync::{Arc, RwLock};

+use async_trait::async_trait;
 use common_catalog::consts::MIN_USER_TABLE_ID;
 use common_telemetry::error;
 use snafu::{ensure, OptionExt};
@@ -155,16 +156,20 @@ impl CatalogManager for MemoryCatalogManager {
        }
    }

-    fn table(&self, catalog: &str, schema: &str, table_name: &str) -> Result<Option<TableRef>> {
-        let c = self.catalogs.read().unwrap();
-        let catalog = if let Some(c) = c.get(catalog) {
+    async fn table(
+        &self,
+        catalog: &str,
+        schema: &str,
+        table_name: &str,
+    ) -> Result<Option<TableRef>> {
+        let catalog = {
+            let c = self.catalogs.read().unwrap();
+            let Some(c) = c.get(catalog) else { return Ok(None) };
            c.clone()
-        } else {
-            return Ok(None);
        };
        match catalog.schema(schema)? {
            None => Ok(None),
-            Some(s) => s.table(table_name),
+            Some(s) => s.table(table_name).await,
        }
    }
 }
@@ -283,6 +288,7 @@ impl Default for MemorySchemaProvider {
    }
 }

+#[async_trait]
 impl SchemaProvider for MemorySchemaProvider {
    fn as_any(&self) -> &dyn Any {
        self
@@ -293,7 +299,7 @@ impl SchemaProvider for MemorySchemaProvider {
        Ok(tables.keys().cloned().collect())
    }

-    fn table(&self, name: &str) -> Result<Option<TableRef>> {
+    async fn table(&self, name: &str) -> Result<Option<TableRef>> {
        let tables = self.tables.read().unwrap();
        Ok(tables.get(name).cloned())
    }
@@ -355,8 +361,8 @@ mod tests {

    use super::*;

-    #[test]
-    fn test_new_memory_catalog_list() {
+    #[tokio::test]
+    async fn test_new_memory_catalog_list() {
        let catalog_list = new_memory_catalog_list().unwrap();
        let default_catalog = catalog_list.catalog(DEFAULT_CATALOG_NAME).unwrap().unwrap();

@@ -369,9 +375,9 @@ mod tests {
            .register_table("numbers".to_string(), Arc::new(NumbersTable::default()))
            .unwrap();

-        let table = default_schema.table("numbers").unwrap();
+        let table = default_schema.table("numbers").await.unwrap();
        assert!(table.is_some());
-        assert!(default_schema.table("not_exists").unwrap().is_none());
+        assert!(default_schema.table("not_exists").await.unwrap().is_none());
    }

    #[tokio::test]
@@ -419,7 +425,7 @@ mod tests {

        // test new table name exists
        assert!(provider.table_exist(new_table_name).unwrap());
-        let registered_table = provider.table(new_table_name).unwrap().unwrap();
+        let registered_table = provider.table(new_table_name).await.unwrap().unwrap();
        assert_eq!(
            registered_table.table_info().ident.table_id,
            test_table.table_info().ident.table_id
@@ -468,6 +474,7 @@ mod tests {

        let registered_table = catalog
            .table(DEFAULT_CATALOG_NAME, DEFAULT_SCHEMA_NAME, new_table_name)
+            .await
            .unwrap()
            .unwrap();
        assert_eq!(registered_table.table_info().ident.table_id, table_id);
--- a/src/catalog/src/remote/manager.rs
+++ b/src/catalog/src/remote/manager.rs
@@ -13,16 +13,19 @@
 // limitations under the License.

 use std::any::Any;
-use std::collections::HashMap;
+use std::collections::{HashMap, HashSet};
 use std::pin::Pin;
 use std::sync::Arc;

 use arc_swap::ArcSwap;
 use async_stream::stream;
+use async_trait::async_trait;
 use common_catalog::consts::{DEFAULT_CATALOG_NAME, DEFAULT_SCHEMA_NAME, MIN_USER_TABLE_ID};
-use common_telemetry::{debug, info};
+use common_telemetry::{debug, error, info};
+use dashmap::DashMap;
 use futures::Stream;
 use futures_util::StreamExt;
+use parking_lot::RwLock;
 use snafu::{OptionExt, ResultExt};
 use table::engine::{EngineContext, TableEngineRef};
 use table::metadata::TableId;
@@ -38,6 +41,7 @@ use crate::error::{
 use crate::helper::{
    build_catalog_prefix, build_schema_prefix, build_table_global_prefix, CatalogKey, CatalogValue,
    SchemaKey, SchemaValue, TableGlobalKey, TableGlobalValue, TableRegionalKey, TableRegionalValue,
+    CATALOG_KEY_PREFIX,
 };
 use crate::remote::{Kv, KvBackendRef};
 use crate::{
@@ -50,10 +54,9 @@ use crate::{
 pub struct RemoteCatalogManager {
    node_id: u64,
    backend: KvBackendRef,
-    catalogs: Arc<ArcSwap<HashMap<String, CatalogProviderRef>>>,
+    catalogs: Arc<RwLock<DashMap<String, CatalogProviderRef>>>,
    engine: TableEngineRef,
    system_table_requests: Mutex<Vec<RegisterSystemTableRequest>>,
-    mutex: Arc<Mutex<()>>,
 }

 impl RemoteCatalogManager {
@@ -64,7 +67,6 @@ impl RemoteCatalogManager {
            backend,
            catalogs: Default::default(),
            system_table_requests: Default::default(),
-            mutex: Default::default(),
        }
    }

@@ -108,9 +110,13 @@ impl RemoteCatalogManager {
                    debug!("Ignoring non-catalog key: {}", String::from_utf8_lossy(&k));
                    continue;
                }
-                let key = CatalogKey::parse(&String::from_utf8_lossy(&k))
-                    .context(InvalidCatalogValueSnafu)?;
-                yield Ok(key)
+
+                let catalog_key = String::from_utf8_lossy(&k);
+                if let Ok(key) = CatalogKey::parse(&catalog_key) {
+                    yield Ok(key)
+                } else {
+                    error!("Invalid catalog key: {:?}", catalog_key);
+                }
            }
        }))
    }
@@ -381,7 +387,14 @@ impl CatalogManager for RemoteCatalogManager {
            "Initialized catalogs: {:?}",
            catalogs.keys().cloned().collect::<Vec<_>>()
        );
-        self.catalogs.store(Arc::new(catalogs));
+
+        {
+            let self_catalogs = self.catalogs.read();
+            catalogs.into_iter().for_each(|(k, v)| {
+                self_catalogs.insert(k, v);
+            });
+        }
+
        info!("Max table id allocated: {}", max_table_id);

        let mut system_table_requests = self.system_table_requests.lock().await;
@@ -468,7 +481,7 @@ impl CatalogManager for RemoteCatalogManager {
            .schema(schema)
    }

-    fn table(
+    async fn table(
        &self,
        catalog_name: &str,
        schema_name: &str,
@@ -483,7 +496,7 @@ impl CatalogManager for RemoteCatalogManager {
                catalog: catalog_name,
                schema: schema_name,
            })?;
-        schema.table(table_name)
+        schema.table(table_name).await
    }
 }

@@ -499,12 +512,10 @@ impl CatalogList for RemoteCatalogManager {
    ) -> Result<Option<CatalogProviderRef>> {
        let key = self.build_catalog_key(&name).to_string();
        let backend = self.backend.clone();
-        let mutex = self.mutex.clone();
        let catalogs = self.catalogs.clone();

        std::thread::spawn(|| {
            common_runtime::block_on_write(async move {
-                let _guard = mutex.lock().await;
                backend
                    .set(
                        key.as_bytes(),
@@ -513,11 +524,10 @@ impl CatalogList for RemoteCatalogManager {
                            .context(InvalidCatalogValueSnafu)?,
                    )
                    .await?;
-                let prev_catalogs = catalogs.load();
-                let mut new_catalogs = HashMap::with_capacity(prev_catalogs.len() + 1);
-                new_catalogs.clone_from(&prev_catalogs);
-                let prev = new_catalogs.insert(name, catalog);
-                catalogs.store(Arc::new(new_catalogs));
+
+                let catalogs = catalogs.read();
+                let prev = catalogs.insert(name, catalog.clone());
+
                Ok(prev)
            })
        })
@@ -527,12 +537,65 @@ impl CatalogList for RemoteCatalogManager {

    /// List all catalogs from metasrv
    fn catalog_names(&self) -> Result<Vec<String>> {
-        Ok(self.catalogs.load().keys().cloned().collect::<Vec<_>>())
+        let catalogs = self.catalogs.read();
+        Ok(catalogs.iter().map(|k| k.key().to_string()).collect())
    }

    /// Read catalog info of given name from metasrv.
    fn catalog(&self, name: &str) -> Result<Option<CatalogProviderRef>> {
-        Ok(self.catalogs.load().get(name).cloned())
+        {
+            let catalogs = self.catalogs.read();
+            let catalog = catalogs.get(name);
+
+            if let Some(catalog) = catalog {
+                return Ok(Some(catalog.clone()));
+            }
+        }
+
+        let catalogs = self.catalogs.write();
+
+        let catalog = catalogs.get(name);
+        if let Some(catalog) = catalog {
+            return Ok(Some(catalog.clone()));
+        }
+
+        // It's for lack of incremental catalog syncing between datanode and meta. Here we fetch catalog
+        // from meta on demand. This can be removed when incremental catalog syncing is done in datanode.
+
+        let backend = self.backend.clone();
+
+        let catalogs_from_meta: HashSet<String> = std::thread::spawn(|| {
+            common_runtime::block_on_read(async move {
+                let mut stream = backend.range(CATALOG_KEY_PREFIX.as_bytes());
+                let mut catalogs = HashSet::new();
+
+                while let Some(catalog) = stream.next().await {
+                    if let Ok(catalog) = catalog {
+                        let catalog_key = String::from_utf8_lossy(&catalog.0);
+
+                        if let Ok(key) = CatalogKey::parse(&catalog_key) {
+                            catalogs.insert(key.catalog_name);
+                        }
+                    }
+                }
+
+                catalogs
+            })
+        })
+        .join()
+        .unwrap();
+
+        catalogs.retain(|catalog_name, _| catalogs_from_meta.get(catalog_name).is_some());
+
+        for catalog in catalogs_from_meta {
+            catalogs
+                .entry(catalog.clone())
+                .or_insert(self.new_catalog_provider(&catalog));
+        }
+
+        let catalog = catalogs.get(name);
+
+        Ok(catalog.as_deref().cloned())
    }
 }

@@ -692,6 +755,7 @@ impl RemoteSchemaProvider {
    }
 }

+#[async_trait]
 impl SchemaProvider for RemoteSchemaProvider {
    fn as_any(&self) -> &dyn Any {
        self
@@ -701,7 +765,7 @@ impl SchemaProvider for RemoteSchemaProvider {
        Ok(self.tables.load().keys().cloned().collect::<Vec<_>>())
    }

-    fn table(&self, name: &str) -> Result<Option<TableRef>> {
+    async fn table(&self, name: &str) -> Result<Option<TableRef>> {
        Ok(self.tables.load().get(name).cloned())
    }

--- a/src/catalog/src/schema.rs
+++ b/src/catalog/src/schema.rs
@@ -15,11 +15,13 @@
 use std::any::Any;
 use std::sync::Arc;

+use async_trait::async_trait;
 use table::TableRef;

 use crate::error::Result;

 /// Represents a schema, comprising a number of named tables.
+#[async_trait]
 pub trait SchemaProvider: Sync + Send {
    /// Returns the schema provider as [`Any`](std::any::Any)
    /// so that it can be downcast to a specific implementation.
@@ -29,7 +31,7 @@ pub trait SchemaProvider: Sync + Send {
    fn table_names(&self) -> Result<Vec<String>>;

    /// Retrieves a specific table from the schema by name, provided it exists.
-    fn table(&self, name: &str) -> Result<Option<TableRef>>;
+    async fn table(&self, name: &str) -> Result<Option<TableRef>>;

    /// If supported by the implementation, adds a new table to this schema.
    /// If a table of the same name existed before, it returns "Table already exists" error.
--- a/src/catalog/src/system.rs
+++ b/src/catalog/src/system.rs
@@ -32,7 +32,9 @@ use serde::{Deserialize, Serialize};
 use snafu::{ensure, OptionExt, ResultExt};
 use table::engine::{EngineContext, TableEngineRef};
 use table::metadata::{TableId, TableInfoRef};
-use table::requests::{CreateTableRequest, DeleteRequest, InsertRequest, OpenTableRequest};
+use table::requests::{
+    CreateTableRequest, DeleteRequest, InsertRequest, OpenTableRequest, TableOptions,
+};
 use table::{Table, TableRef};

 use crate::error::{
@@ -109,7 +111,7 @@ impl SystemCatalogTable {
                region_numbers: vec![0],
                primary_key_indices: vec![ENTRY_TYPE_INDEX, KEY_INDEX],
                create_if_not_exists: true,
-                table_options: HashMap::new(),
+                table_options: TableOptions::default(),
            };

            let table = engine
@@ -217,7 +219,7 @@ fn build_primary_key_columns(entry_type: EntryType, key: &[u8]) -> HashMap<Strin
    let mut m = HashMap::with_capacity(3);
    m.insert(
        "entry_type".to_string(),
-        Arc::new(UInt8Vector::from_slice(&[entry_type as u8])) as _,
+        Arc::new(UInt8Vector::from_slice([entry_type as u8])) as _,
    );
    m.insert(
        "key".to_string(),
@@ -226,7 +228,7 @@ fn build_primary_key_columns(entry_type: EntryType, key: &[u8]) -> HashMap<Strin
    // Timestamp in key part is intentionally left to 0
    m.insert(
        "timestamp".to_string(),
-        Arc::new(TimestampMillisecondVector::from_slice(&[0])) as _,
+        Arc::new(TimestampMillisecondVector::from_slice([0])) as _,
    );
    m
 }
@@ -256,12 +258,12 @@ pub fn build_insert_request(entry_type: EntryType, key: &[u8], value: &[u8]) ->
    let now = util::current_time_millis();
    columns_values.insert(
        "gmt_created".to_string(),
-        Arc::new(TimestampMillisecondVector::from_slice(&[now])) as _,
+        Arc::new(TimestampMillisecondVector::from_slice([now])) as _,
    );

    columns_values.insert(
        "gmt_modified".to_string(),
-        Arc::new(TimestampMillisecondVector::from_slice(&[now])) as _,
+        Arc::new(TimestampMillisecondVector::from_slice([now])) as _,
    );

    InsertRequest {
@@ -393,17 +395,17 @@ pub struct TableEntryValue {
 #[cfg(test)]
 mod tests {
    use common_recordbatch::RecordBatches;
+    use common_test_util::temp_dir::{create_temp_dir, TempDir};
    use datatypes::value::Value;
    use log_store::NoopLogStore;
    use mito::config::EngineConfig;
    use mito::engine::MitoEngine;
-    use object_store::ObjectStore;
+    use object_store::{ObjectStore, ObjectStoreBuilder};
    use storage::compaction::noop::NoopCompactionScheduler;
    use storage::config::EngineConfig as StorageEngineConfig;
    use storage::EngineImpl;
    use table::metadata::TableType;
    use table::metadata::TableType::Base;
-    use tempdir::TempDir;

    use super::*;

@@ -478,13 +480,13 @@ mod tests {
    }

    pub async fn prepare_table_engine() -> (TempDir, TableEngineRef) {
-        let dir = TempDir::new("system-table-test").unwrap();
+        let dir = create_temp_dir("system-table-test");
        let store_dir = dir.path().to_string_lossy();
-        let accessor = object_store::backend::fs::Builder::default()
+        let accessor = object_store::services::Fs::default()
            .root(&store_dir)
            .build()
            .unwrap();
-        let object_store = ObjectStore::new(accessor);
+        let object_store = ObjectStore::new(accessor).finish();
        let noop_compaction_scheduler = Arc::new(NoopCompactionScheduler::default());
        let table_engine = Arc::new(MitoEngine::new(
            EngineConfig::default(),
--- a/src/catalog/src/table_source.rs
+++ b/src/catalog/src/table_source.rs
@@ -0,0 +1,178 @@
+// 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::sync::Arc;
+
+use common_catalog::format_full_table_name;
+use datafusion::common::{OwnedTableReference, ResolvedTableReference, TableReference};
+use datafusion::datasource::provider_as_source;
+use datafusion::logical_expr::TableSource;
+use session::context::QueryContext;
+use snafu::{ensure, OptionExt};
+use table::table::adapter::DfTableProviderAdapter;
+
+use crate::error::{
+    CatalogNotFoundSnafu, QueryAccessDeniedSnafu, Result, SchemaNotFoundSnafu, TableNotExistSnafu,
+};
+use crate::CatalogListRef;
+
+pub struct DfTableSourceProvider {
+    catalog_list: CatalogListRef,
+    resolved_tables: HashMap<String, Arc<dyn TableSource>>,
+    disallow_cross_schema_query: bool,
+    default_catalog: String,
+    default_schema: String,
+}
+
+impl DfTableSourceProvider {
+    pub fn new(
+        catalog_list: CatalogListRef,
+        disallow_cross_schema_query: bool,
+        query_ctx: &QueryContext,
+    ) -> Self {
+        Self {
+            catalog_list,
+            disallow_cross_schema_query,
+            resolved_tables: HashMap::new(),
+            default_catalog: query_ctx.current_catalog(),
+            default_schema: query_ctx.current_schema(),
+        }
+    }
+
+    pub fn resolve_table_ref<'a>(
+        &'a self,
+        table_ref: TableReference<'a>,
+    ) -> Result<ResolvedTableReference<'a>> {
+        if self.disallow_cross_schema_query {
+            match &table_ref {
+                TableReference::Bare { .. } => (),
+                TableReference::Partial { schema, .. } => {
+                    ensure!(
+                        schema.as_ref() == self.default_schema,
+                        QueryAccessDeniedSnafu {
+                            catalog: &self.default_catalog,
+                            schema: schema.as_ref(),
+                        }
+                    );
+                }
+                TableReference::Full {
+                    catalog, schema, ..
+                } => {
+                    ensure!(
+                        catalog.as_ref() == self.default_catalog
+                            && schema.as_ref() == self.default_schema,
+                        QueryAccessDeniedSnafu {
+                            catalog: catalog.as_ref(),
+                            schema: schema.as_ref()
+                        }
+                    );
+                }
+            };
+        }
+
+        Ok(table_ref.resolve(&self.default_catalog, &self.default_schema))
+    }
+
+    pub async fn resolve_table(
+        &mut self,
+        table_ref: OwnedTableReference,
+    ) -> Result<Arc<dyn TableSource>> {
+        let table_ref = table_ref.as_table_reference();
+        let table_ref = self.resolve_table_ref(table_ref)?;
+
+        let resolved_name = table_ref.to_string();
+        if let Some(table) = self.resolved_tables.get(&resolved_name) {
+            return Ok(table.clone());
+        }
+
+        let catalog_name = table_ref.catalog.as_ref();
+        let schema_name = table_ref.schema.as_ref();
+        let table_name = table_ref.table.as_ref();
+
+        let catalog = self
+            .catalog_list
+            .catalog(catalog_name)?
+            .context(CatalogNotFoundSnafu { catalog_name })?;
+        let schema = catalog.schema(schema_name)?.context(SchemaNotFoundSnafu {
+            catalog: catalog_name,
+            schema: schema_name,
+        })?;
+        let table = schema
+            .table(table_name)
+            .await?
+            .with_context(|| TableNotExistSnafu {
+                table: format_full_table_name(catalog_name, schema_name, table_name),
+            })?;
+
+        let table = DfTableProviderAdapter::new(table);
+        let table = provider_as_source(Arc::new(table));
+        self.resolved_tables.insert(resolved_name, table.clone());
+        Ok(table)
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use std::borrow::Cow;
+
+    use session::context::QueryContext;
+
+    use super::*;
+    use crate::local::MemoryCatalogManager;
+
+    #[test]
+    fn test_validate_table_ref() {
+        let query_ctx = &QueryContext::with("greptime", "public");
+
+        let table_provider =
+            DfTableSourceProvider::new(Arc::new(MemoryCatalogManager::default()), true, query_ctx);
+
+        let table_ref = TableReference::Bare {
+            table: Cow::Borrowed("table_name"),
+        };
+        let result = table_provider.resolve_table_ref(table_ref);
+        assert!(result.is_ok());
+
+        let table_ref = TableReference::Partial {
+            schema: Cow::Borrowed("public"),
+            table: Cow::Borrowed("table_name"),
+        };
+        let result = table_provider.resolve_table_ref(table_ref);
+        assert!(result.is_ok());
+
+        let table_ref = TableReference::Partial {
+            schema: Cow::Borrowed("wrong_schema"),
+            table: Cow::Borrowed("table_name"),
+        };
+        let result = table_provider.resolve_table_ref(table_ref);
+        assert!(result.is_err());
+
+        let table_ref = TableReference::Full {
+            catalog: Cow::Borrowed("greptime"),
+            schema: Cow::Borrowed("public"),
+            table: Cow::Borrowed("table_name"),
+        };
+        let result = table_provider.resolve_table_ref(table_ref);
+        assert!(result.is_ok());
+
+        let table_ref = TableReference::Full {
+            catalog: Cow::Borrowed("wrong_catalog"),
+            schema: Cow::Borrowed("public"),
+            table: Cow::Borrowed("table_name"),
+        };
+        let result = table_provider.resolve_table_ref(table_ref);
+        assert!(result.is_err());
+    }
+}
--- a/src/catalog/src/tables.rs
+++ b/src/catalog/src/tables.rs
@@ -20,6 +20,7 @@ use std::sync::Arc;
 use std::task::{Context, Poll};

 use async_stream::stream;
+use async_trait::async_trait;
 use common_catalog::consts::{INFORMATION_SCHEMA_NAME, SYSTEM_CATALOG_TABLE_NAME};
 use common_error::ext::BoxedError;
 use common_query::logical_plan::Expr;
@@ -200,6 +201,7 @@ pub struct InformationSchema {
    pub system: Arc<SystemCatalogTable>,
 }

+#[async_trait]
 impl SchemaProvider for InformationSchema {
    fn as_any(&self) -> &dyn Any {
        self
@@ -212,7 +214,7 @@ impl SchemaProvider for InformationSchema {
        ])
    }

-    fn table(&self, name: &str) -> Result<Option<TableRef>, Error> {
+    async fn table(&self, name: &str) -> Result<Option<TableRef>, Error> {
        if name.eq_ignore_ascii_case("tables") {
            Ok(Some(self.tables.clone()))
        } else if name.eq_ignore_ascii_case(SYSTEM_CATALOG_TABLE_NAME) {
--- a/src/catalog/tests/local_catalog_tests.rs
+++ b/src/catalog/tests/local_catalog_tests.rs
@@ -71,6 +71,7 @@ mod tests {

        let registered_table = catalog_manager
            .table(DEFAULT_CATALOG_NAME, DEFAULT_SCHEMA_NAME, new_table_name)
+            .await
            .unwrap()
            .unwrap();
        assert_eq!(registered_table.table_info().ident.table_id, table_id);
@@ -158,6 +159,7 @@ mod tests {
            let table = guard.as_ref().unwrap();
            let table_registered = catalog_manager
                .table(DEFAULT_CATALOG_NAME, DEFAULT_SCHEMA_NAME, "test_table")
+                .await
                .unwrap()
                .unwrap();
            assert_eq!(
--- a/src/catalog/tests/mock.rs
+++ b/src/catalog/tests/mock.rs
@@ -147,6 +147,7 @@ impl TableEngine for MockTableEngine {
        let table_id = TableId::from_str(
            request
                .table_options
+                .extra_options
                .get("table_id")
                .unwrap_or(&default_table_id),
        )
@@ -220,4 +221,8 @@ impl TableEngine for MockTableEngine {
    ) -> table::Result<bool> {
        unimplemented!()
    }
+
+    async fn close(&self) -> table::Result<()> {
+        Ok(())
+    }
 }
--- a/src/client/Cargo.toml
+++ b/src/client/Cargo.toml
@@ -16,6 +16,7 @@ common-grpc-expr = { path = "../common/grpc-expr" }
 common-query = { path = "../common/query" }
 common-recordbatch = { path = "../common/recordbatch" }
 common-time = { path = "../common/time" }
+common-telemetry = { path = "../common/telemetry" }
 datafusion.workspace = true
 datatypes = { path = "../datatypes" }
 enum_dispatch = "0.3"
@@ -32,12 +33,8 @@ substrait = { path = "../common/substrait" }
 tokio.workspace = true
 tracing = "0.1"
 tracing-subscriber = { version = "0.3", features = ["env-filter"] }
-
-# TODO(ruihang): upgrade to 0.11 once substrait-rs supports it.
-[dev-dependencies.prost_09]
-package = "prost"
-version = "0.9"
+prost.workspace = true

 [dev-dependencies.substrait_proto]
 package = "substrait"
-version = "0.2"
+version = "0.4"
--- a/src/client/examples/logical.rs
+++ b/src/client/examples/logical.rs
@@ -14,11 +14,12 @@

 use api::v1::{ColumnDataType, ColumnDef, CreateTableExpr, TableId};
 use client::{Client, Database};
-use prost_09::Message;
-use substrait_proto::protobuf::plan_rel::RelType as PlanRelType;
-use substrait_proto::protobuf::read_rel::{NamedTable, ReadType};
-use substrait_proto::protobuf::rel::RelType;
-use substrait_proto::protobuf::{PlanRel, ReadRel, Rel};
+use common_catalog::consts::{DEFAULT_CATALOG_NAME, DEFAULT_SCHEMA_NAME};
+use prost::Message;
+use substrait_proto::proto::plan_rel::RelType as PlanRelType;
+use substrait_proto::proto::read_rel::{NamedTable, ReadType};
+use substrait_proto::proto::rel::RelType;
+use substrait_proto::proto::{PlanRel, ReadRel, Rel};
 use tracing::{event, Level};

 fn main() {
@@ -65,7 +66,7 @@ async fn run() {
        region_ids: vec![0],
    };

-    let db = Database::with_client(client);
+    let db = Database::new(DEFAULT_CATALOG_NAME, DEFAULT_SCHEMA_NAME, client);
    let result = db.create(create_table_expr).await.unwrap();
    event!(Level::INFO, "create table result: {:#?}", result);

@@ -88,12 +89,8 @@ fn mock_logical_plan() -> Vec<u8> {
    let read_type = ReadType::NamedTable(named_table);

    let read_rel = ReadRel {
-        common: None,
-        base_schema: None,
-        filter: None,
-        projection: None,
-        advanced_extension: None,
        read_type: Some(read_type),
+        ..Default::default()
    };

    let mut buf = vec![];
--- a/src/client/src/database.rs
+++ b/src/client/src/database.rs
@@ -14,18 +14,19 @@

 use std::str::FromStr;

+use api::v1::auth_header::AuthScheme;
 use api::v1::ddl_request::Expr as DdlExpr;
 use api::v1::greptime_request::Request;
 use api::v1::query_request::Query;
 use api::v1::{
-    AlterExpr, CreateTableExpr, DdlRequest, DropTableExpr, GreptimeRequest, InsertRequest,
-    QueryRequest, RequestHeader,
+    AlterExpr, AuthHeader, CreateTableExpr, DdlRequest, DropTableExpr, FlushTableExpr,
+    GreptimeRequest, InsertRequest, PromRangeQuery, QueryRequest, RequestHeader,
 };
 use arrow_flight::{FlightData, Ticket};
-use common_catalog::consts::{DEFAULT_CATALOG_NAME, DEFAULT_SCHEMA_NAME};
 use common_error::prelude::*;
 use common_grpc::flight::{flight_messages_to_recordbatches, FlightDecoder, FlightMessage};
 use common_query::Output;
+use common_telemetry::logging;
 use futures_util::{TryFutureExt, TryStreamExt};
 use prost::Message;
 use snafu::{ensure, ResultExt};
@@ -42,6 +43,7 @@ pub struct Database {
    schema: String,

    client: Client,
+    ctx: FlightContext,
 }

 impl Database {
@@ -50,17 +52,32 @@ impl Database {
            catalog: catalog.into(),
            schema: schema.into(),
            client,
+            ctx: FlightContext::default(),
        }
    }

-    pub fn with_client(client: Client) -> Self {
-        Self::new(DEFAULT_CATALOG_NAME, DEFAULT_SCHEMA_NAME, client)
+    pub fn catalog(&self) -> &String {
+        &self.catalog
+    }
+
+    pub fn set_catalog(&mut self, catalog: impl Into<String>) {
+        self.catalog = catalog.into();
+    }
+
+    pub fn schema(&self) -> &String {
+        &self.schema
    }

    pub fn set_schema(&mut self, schema: impl Into<String>) {
        self.schema = schema.into();
    }

+    pub fn set_auth(&mut self, auth: AuthScheme) {
+        self.ctx.auth_header = Some(AuthHeader {
+            auth_scheme: Some(auth),
+        });
+    }
+
    pub async fn insert(&self, request: InsertRequest) -> Result<Output> {
        self.do_get(Request::Insert(request)).await
    }
@@ -79,6 +96,24 @@ impl Database {
        .await
    }

+    pub async fn prom_range_query(
+        &self,
+        promql: &str,
+        start: &str,
+        end: &str,
+        step: &str,
+    ) -> Result<Output> {
+        self.do_get(Request::Query(QueryRequest {
+            query: Some(Query::PromRangeQuery(PromRangeQuery {
+                query: promql.to_string(),
+                start: start.to_string(),
+                end: end.to_string(),
+                step: step.to_string(),
+            })),
+        }))
+        .await
+    }
+
    pub async fn create(&self, expr: CreateTableExpr) -> Result<Output> {
        self.do_get(Request::Ddl(DdlRequest {
            expr: Some(DdlExpr::CreateTable(expr)),
@@ -100,16 +135,24 @@ impl Database {
        .await
    }

+    pub async fn flush_table(&self, expr: FlushTableExpr) -> Result<Output> {
+        self.do_get(Request::Ddl(DdlRequest {
+            expr: Some(DdlExpr::FlushTable(expr)),
+        }))
+        .await
+    }
+
    async fn do_get(&self, request: Request) -> Result<Output> {
        let request = GreptimeRequest {
            header: Some(RequestHeader {
                catalog: self.catalog.clone(),
                schema: self.schema.clone(),
+                authorization: self.ctx.auth_header.clone(),
            }),
            request: Some(request),
        };
        let request = Ticket {
-            ticket: request.encode_to_vec(),
+            ticket: request.encode_to_vec().into(),
        };

        let mut client = self.client.make_client()?;
@@ -132,6 +175,15 @@ impl Database {
                        tonic_code: e.code(),
                        addr: client.addr(),
                    })
+                    .map_err(|error| {
+                        logging::error!(
+                            "Failed to do Flight get, addr: {}, code: {}, source: {}",
+                            client.addr(),
+                            e.code(),
+                            error
+                        );
+                        error
+                    })
                    .unwrap_err()
            })?;

@@ -164,12 +216,18 @@ fn get_metadata_value(e: &tonic::Status, key: &str) -> Option<String> {
        .and_then(|v| String::from_utf8(v.as_bytes().to_vec()).ok())
 }

+#[derive(Default, Debug, Clone)]
+pub struct FlightContext {
+    auth_header: Option<AuthHeader>,
+}
+
 #[cfg(test)]
 mod tests {
    use std::sync::Arc;

    use api::helper::ColumnDataTypeWrapper;
-    use api::v1::Column;
+    use api::v1::auth_header::AuthScheme;
+    use api::v1::{AuthHeader, Basic, Column};
    use common_grpc::select::{null_mask, values};
    use common_grpc_expr::column_to_vector;
    use datatypes::prelude::{Vector, VectorRef};
@@ -179,6 +237,8 @@ mod tests {
        UInt32Vector, UInt64Vector, UInt8Vector,
    };

+    use crate::database::FlightContext;
+
    #[test]
    fn test_column_to_vector() {
        let mut column = create_test_column(Arc::new(BooleanVector::from(vec![true])));
@@ -262,4 +322,26 @@ mod tests {
            datatype: wrapper.datatype() as i32,
        }
    }
+
+    #[test]
+    fn test_flight_ctx() {
+        let mut ctx = FlightContext::default();
+        assert!(ctx.auth_header.is_none());
+
+        let basic = AuthScheme::Basic(Basic {
+            username: "u".to_string(),
+            password: "p".to_string(),
+        });
+
+        ctx.auth_header = Some(AuthHeader {
+            auth_scheme: Some(basic),
+        });
+
+        assert!(matches!(
+            ctx.auth_header,
+            Some(AuthHeader {
+                auth_scheme: Some(AuthScheme::Basic(_)),
+            })
+        ))
+    }
 }
--- a/src/client/src/error.rs
+++ b/src/client/src/error.rs
@@ -26,12 +26,7 @@ pub enum Error {
        backtrace: Backtrace,
    },

-    #[snafu(display(
-        "Failed to do Flight get, addr: {}, code: {}, source: {}",
-        addr,
-        tonic_code,
-        source
-    ))]
+    #[snafu(display("Failed to do Flight get, code: {}, source: {}", tonic_code, source))]
    FlightGet {
        addr: String,
        tonic_code: Code,
--- a/src/client/src/lib.rs
+++ b/src/client/src/lib.rs
@@ -18,6 +18,7 @@ mod error;
 pub mod load_balance;

 pub use api;
+pub use common_catalog::consts::{DEFAULT_CATALOG_NAME, DEFAULT_SCHEMA_NAME};

 pub use self::client::Client;
 pub use self::database::Database;
--- a/src/cmd/Cargo.toml
+++ b/src/cmd/Cargo.toml
@@ -9,28 +9,46 @@ default-run = "greptime"
 name = "greptime"
 path = "src/bin/greptime.rs"

+[features]
+mem-prof = ["tikv-jemallocator", "tikv-jemalloc-ctl"]
+
 [dependencies]
 anymap = "1.0.0-beta.2"
+catalog = { path = "../catalog" }
 clap = { version = "3.1", features = ["derive"] }
+client = { path = "../client" }
 common-base = { path = "../common/base" }
 common-error = { path = "../common/error" }
+common-query = { path = "../common/query" }
+common-recordbatch = { path = "../common/recordbatch" }
 common-telemetry = { path = "../common/telemetry", features = [
    "deadlock_detection",
 ] }
 datanode = { path = "../datanode" }
+either = "1.8"
 frontend = { path = "../frontend" }
 futures.workspace = true
 meta-client = { path = "../meta-client" }
 meta-srv = { path = "../meta-srv" }
+nu-ansi-term = "0.46"
+partition = { path = "../partition" }
+query = { path = "../query" }
+rustyline = "10.1"
 serde.workspace = true
 servers = { path = "../servers" }
+session = { path = "../session" }
 snafu.workspace = true
+substrait = { path = "../common/substrait" }
+tikv-jemalloc-ctl = { version = "0.5", optional = true }
+tikv-jemallocator = { version = "0.5", optional = true }
 tokio.workspace = true
 toml = "0.5"

+
 [dev-dependencies]
+common-test-util = { path = "../common/test-util" }
+rexpect = "0.5"
 serde.workspace = true
-tempdir = "0.3"

 [build-dependencies]
 build-data = "0.1.3"
--- a/src/cmd/src/bin/greptime.rs
+++ b/src/cmd/src/bin/greptime.rs
@@ -16,7 +16,7 @@ use std::fmt;

 use clap::Parser;
 use cmd::error::Result;
-use cmd::{datanode, frontend, metasrv, standalone};
+use cmd::{cli, datanode, frontend, metasrv, standalone};
 use common_telemetry::logging::{error, info};

 #[derive(Parser)]
@@ -30,9 +30,39 @@ struct Command {
    subcmd: SubCommand,
 }

+pub enum Application {
+    Datanode(datanode::Instance),
+    Frontend(frontend::Instance),
+    Metasrv(metasrv::Instance),
+    Standalone(standalone::Instance),
+    Cli(cli::Instance),
+}
+
+impl Application {
+    async fn run(&mut self) -> Result<()> {
+        match self {
+            Application::Datanode(instance) => instance.run().await,
+            Application::Frontend(instance) => instance.run().await,
+            Application::Metasrv(instance) => instance.run().await,
+            Application::Standalone(instance) => instance.run().await,
+            Application::Cli(instance) => instance.run().await,
+        }
+    }
+
+    async fn stop(&self) -> Result<()> {
+        match self {
+            Application::Datanode(instance) => instance.stop().await,
+            Application::Frontend(instance) => instance.stop().await,
+            Application::Metasrv(instance) => instance.stop().await,
+            Application::Standalone(instance) => instance.stop().await,
+            Application::Cli(instance) => instance.stop().await,
+        }
+    }
+}
+
 impl Command {
-    async fn run(self) -> Result<()> {
-        self.subcmd.run().await
+    async fn build(self) -> Result<Application> {
+        self.subcmd.build().await
    }
 }

@@ -46,15 +76,33 @@ enum SubCommand {
    Metasrv(metasrv::Command),
    #[clap(name = "standalone")]
    Standalone(standalone::Command),
+    #[clap(name = "cli")]
+    Cli(cli::Command),
 }

 impl SubCommand {
-    async fn run(self) -> Result<()> {
+    async fn build(self) -> Result<Application> {
        match self {
-            SubCommand::Datanode(cmd) => cmd.run().await,
-            SubCommand::Frontend(cmd) => cmd.run().await,
-            SubCommand::Metasrv(cmd) => cmd.run().await,
-            SubCommand::Standalone(cmd) => cmd.run().await,
+            SubCommand::Datanode(cmd) => {
+                let app = cmd.build().await?;
+                Ok(Application::Datanode(app))
+            }
+            SubCommand::Frontend(cmd) => {
+                let app = cmd.build().await?;
+                Ok(Application::Frontend(app))
+            }
+            SubCommand::Metasrv(cmd) => {
+                let app = cmd.build().await?;
+                Ok(Application::Metasrv(app))
+            }
+            SubCommand::Standalone(cmd) => {
+                let app = cmd.build().await?;
+                Ok(Application::Standalone(app))
+            }
+            SubCommand::Cli(cmd) => {
+                let app = cmd.build().await?;
+                Ok(Application::Cli(app))
+            }
        }
    }
 }
@@ -66,6 +114,7 @@ impl fmt::Display for SubCommand {
            SubCommand::Frontend(..) => write!(f, "greptime-frontend"),
            SubCommand::Metasrv(..) => write!(f, "greptime-metasrv"),
            SubCommand::Standalone(..) => write!(f, "greptime-standalone"),
+            SubCommand::Cli(_) => write!(f, "greptime-cli"),
        }
    }
 }
@@ -83,6 +132,10 @@ fn print_version() -> &'static str {
    )
 }

+#[cfg(feature = "mem-prof")]
+#[global_allocator]
+static ALLOC: tikv_jemallocator::Jemalloc = tikv_jemallocator::Jemalloc;
+
 #[tokio::main]
 async fn main() -> Result<()> {
    let cmd = Command::parse();
@@ -96,13 +149,18 @@ async fn main() -> Result<()> {
    common_telemetry::init_default_metrics_recorder();
    let _guard = common_telemetry::init_global_logging(app_name, log_dir, log_level, false);

+    let mut app = cmd.build().await?;
+
    tokio::select! {
-        result = cmd.run() => {
+        result = app.run() => {
            if let Err(err) = result {
                error!(err; "Fatal error occurs!");
            }
        }
        _ = tokio::signal::ctrl_c() => {
+            if let Err(err) = app.stop().await {
+                error!(err; "Fatal error occurs!");
+            }
            info!("Goodbye!");
        }
    }
--- a/src/cmd/src/cli.rs
+++ b/src/cmd/src/cli.rs
@@ -0,0 +1,79 @@
+// 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.
+
+mod cmd;
+mod helper;
+mod repl;
+
+use clap::Parser;
+pub use repl::Repl;
+
+use crate::error::Result;
+
+pub struct Instance {
+    repl: Repl,
+}
+
+impl Instance {
+    pub async fn run(&mut self) -> Result<()> {
+        self.repl.run().await
+    }
+
+    pub async fn stop(&self) -> Result<()> {
+        // TODO: handle cli shutdown
+        Ok(())
+    }
+}
+
+#[derive(Parser)]
+pub struct Command {
+    #[clap(subcommand)]
+    cmd: SubCommand,
+}
+
+impl Command {
+    pub async fn build(self) -> Result<Instance> {
+        self.cmd.build().await
+    }
+}
+
+#[derive(Parser)]
+enum SubCommand {
+    Attach(AttachCommand),
+}
+
+impl SubCommand {
+    async fn build(self) -> Result<Instance> {
+        match self {
+            SubCommand::Attach(cmd) => cmd.build().await,
+        }
+    }
+}
+
+#[derive(Debug, Parser)]
+pub(crate) struct AttachCommand {
+    #[clap(long)]
+    pub(crate) grpc_addr: String,
+    #[clap(long)]
+    pub(crate) meta_addr: Option<String>,
+    #[clap(long, action)]
+    pub(crate) disable_helper: bool,
+}
+
+impl AttachCommand {
+    async fn build(self) -> Result<Instance> {
+        let repl = Repl::try_new(&self).await?;
+        Ok(Instance { repl })
+    }
+}
--- a/src/cmd/src/cli/cmd.rs
+++ b/src/cmd/src/cli/cmd.rs
@@ -0,0 +1,154 @@
+// 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 crate::error::{Error, InvalidReplCommandSnafu, Result};
+
+/// Represents the parsed command from the user (which may be over many lines)
+#[derive(Debug, PartialEq)]
+pub(crate) enum ReplCommand {
+    Help,
+    UseDatabase { db_name: String },
+    Sql { sql: String },
+    Exit,
+}
+
+impl TryFrom<&str> for ReplCommand {
+    type Error = Error;
+
+    fn try_from(input: &str) -> Result<Self> {
+        let input = input.trim();
+        if input.is_empty() {
+            return InvalidReplCommandSnafu {
+                reason: "No command specified".to_string(),
+            }
+            .fail();
+        }
+
+        // If line ends with ';', it must be treated as a complete input.
+        // However, the opposite is not true.
+        let input_is_completed = input.ends_with(';');
+
+        let input = input.strip_suffix(';').map(|x| x.trim()).unwrap_or(input);
+        let lowercase = input.to_lowercase();
+        match lowercase.as_str() {
+            "help" => Ok(Self::Help),
+            "exit" | "quit" => Ok(Self::Exit),
+            _ => match input.split_once(' ') {
+                Some((maybe_use, database)) if maybe_use.to_lowercase() == "use" => {
+                    Ok(Self::UseDatabase {
+                        db_name: database.trim().to_string(),
+                    })
+                }
+                // Any valid SQL must contains at least one whitespace.
+                Some(_) if input_is_completed => Ok(Self::Sql {
+                    sql: input.to_string(),
+                }),
+                _ => InvalidReplCommandSnafu {
+                    reason: format!("unknown command '{input}', maybe input is not completed"),
+                }
+                .fail(),
+            },
+        }
+    }
+}
+
+impl ReplCommand {
+    pub fn help() -> &'static str {
+        r#"
+Available commands (case insensitive):
+- 'help': print this help
+- 'exit' or 'quit': exit the REPL
+- 'use <your database name>': switch to another database/schema context
+- Other typed in text will be treated as SQL.
+  You can enter new line while typing, just remember to end it with ';'.
+"#
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use crate::error::Error::InvalidReplCommand;
+
+    #[test]
+    fn test_from_str() {
+        fn test_ok(s: &str, expected: ReplCommand) {
+            let actual: ReplCommand = s.try_into().unwrap();
+            assert_eq!(expected, actual, "'{}'", s);
+        }
+
+        fn test_err(s: &str) {
+            let result: Result<ReplCommand> = s.try_into();
+            assert!(matches!(result, Err(InvalidReplCommand { .. })))
+        }
+
+        test_err("");
+        test_err("  ");
+        test_err("\t");
+
+        test_ok("help", ReplCommand::Help);
+        test_ok("help", ReplCommand::Help);
+        test_ok("  help", ReplCommand::Help);
+        test_ok("  help  ", ReplCommand::Help);
+        test_ok("  HELP  ", ReplCommand::Help);
+        test_ok("  Help;  ", ReplCommand::Help);
+        test_ok("  help  ; ", ReplCommand::Help);
+
+        test_ok("exit", ReplCommand::Exit);
+        test_ok("exit;", ReplCommand::Exit);
+        test_ok("exit ;", ReplCommand::Exit);
+        test_ok("EXIT", ReplCommand::Exit);
+
+        test_ok("quit", ReplCommand::Exit);
+        test_ok("quit;", ReplCommand::Exit);
+        test_ok("quit ;", ReplCommand::Exit);
+        test_ok("QUIT", ReplCommand::Exit);
+
+        test_ok(
+            "use Foo",
+            ReplCommand::UseDatabase {
+                db_name: "Foo".to_string(),
+            },
+        );
+        test_ok(
+            "  use Foo ;  ",
+            ReplCommand::UseDatabase {
+                db_name: "Foo".to_string(),
+            },
+        );
+        // ensure that database name is case sensitive
+        test_ok(
+            "  use FOO ;  ",
+            ReplCommand::UseDatabase {
+                db_name: "FOO".to_string(),
+            },
+        );
+
+        // ensure that we aren't messing with capitalization
+        test_ok(
+            "SELECT * from foo;",
+            ReplCommand::Sql {
+                sql: "SELECT * from foo".to_string(),
+            },
+        );
+        // Input line (that don't belong to any other cases above) must ends with ';' to make it a valid SQL.
+        test_err("insert blah");
+        test_ok(
+            "insert blah;",
+            ReplCommand::Sql {
+                sql: "insert blah".to_string(),
+            },
+        );
+    }
+}
--- a/src/cmd/src/cli/helper.rs
+++ b/src/cmd/src/cli/helper.rs
@@ -0,0 +1,112 @@
+// 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::borrow::Cow;
+
+use rustyline::completion::Completer;
+use rustyline::highlight::{Highlighter, MatchingBracketHighlighter};
+use rustyline::hint::{Hinter, HistoryHinter};
+use rustyline::validate::{ValidationContext, ValidationResult, Validator};
+
+use crate::cli::cmd::ReplCommand;
+
+pub(crate) struct RustylineHelper {
+    hinter: HistoryHinter,
+    highlighter: MatchingBracketHighlighter,
+}
+
+impl Default for RustylineHelper {
+    fn default() -> Self {
+        Self {
+            hinter: HistoryHinter {},
+            highlighter: MatchingBracketHighlighter::default(),
+        }
+    }
+}
+
+impl rustyline::Helper for RustylineHelper {}
+
+impl Validator for RustylineHelper {
+    fn validate(&self, ctx: &mut ValidationContext<'_>) -> rustyline::Result<ValidationResult> {
+        let input = ctx.input();
+        match ReplCommand::try_from(input) {
+            Ok(_) => Ok(ValidationResult::Valid(None)),
+            Err(e) => {
+                if input.trim_end().ends_with(';') {
+                    // If line ends with ';', it HAS to be a valid command.
+                    Ok(ValidationResult::Invalid(Some(e.to_string())))
+                } else {
+                    Ok(ValidationResult::Incomplete)
+                }
+            }
+        }
+    }
+}
+
+impl Hinter for RustylineHelper {
+    type Hint = String;
+
+    fn hint(&self, line: &str, pos: usize, ctx: &rustyline::Context<'_>) -> Option<Self::Hint> {
+        self.hinter.hint(line, pos, ctx)
+    }
+}
+
+impl Highlighter for RustylineHelper {
+    fn highlight<'l>(&self, line: &'l str, pos: usize) -> Cow<'l, str> {
+        self.highlighter.highlight(line, pos)
+    }
+
+    fn highlight_prompt<'b, 's: 'b, 'p: 'b>(
+        &'s self,
+        prompt: &'p str,
+        default: bool,
+    ) -> Cow<'b, str> {
+        self.highlighter.highlight_prompt(prompt, default)
+    }
+
+    fn highlight_hint<'h>(&self, hint: &'h str) -> Cow<'h, str> {
+        use nu_ansi_term::Style;
+        Cow::Owned(Style::new().dimmed().paint(hint).to_string())
+    }
+
+    fn highlight_candidate<'c>(
+        &self,
+        candidate: &'c str,
+        completion: rustyline::CompletionType,
+    ) -> Cow<'c, str> {
+        self.highlighter.highlight_candidate(candidate, completion)
+    }
+
+    fn highlight_char(&self, line: &str, pos: usize) -> bool {
+        self.highlighter.highlight_char(line, pos)
+    }
+}
+
+impl Completer for RustylineHelper {
+    type Candidate = String;
+
+    fn complete(
+        &self,
+        line: &str,
+        pos: usize,
+        ctx: &rustyline::Context<'_>,
+    ) -> rustyline::Result<(usize, Vec<Self::Candidate>)> {
+        // If there is a hint, use that as the auto-complete when user hits `tab`
+        if let Some(hint) = self.hinter.hint(line, pos, ctx) {
+            Ok((pos, vec![hint]))
+        } else {
+            Ok((0, vec![]))
+        }
+    }
+}
--- a/src/cmd/src/cli/repl.rs
+++ b/src/cmd/src/cli/repl.rs
@@ -0,0 +1,273 @@
+// 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::PathBuf;
+use std::sync::Arc;
+use std::time::Instant;
+
+use catalog::remote::MetaKvBackend;
+use client::{Client, Database, DEFAULT_CATALOG_NAME, DEFAULT_SCHEMA_NAME};
+use common_error::prelude::ErrorExt;
+use common_query::Output;
+use common_recordbatch::RecordBatches;
+use common_telemetry::logging;
+use either::Either;
+use frontend::catalog::FrontendCatalogManager;
+use frontend::datanode::DatanodeClients;
+use meta_client::client::MetaClientBuilder;
+use partition::manager::PartitionRuleManager;
+use partition::route::TableRoutes;
+use query::datafusion::DatafusionQueryEngine;
+use query::logical_optimizer::LogicalOptimizer;
+use query::parser::QueryLanguageParser;
+use query::plan::LogicalPlan;
+use query::query_engine::QueryEngineState;
+use query::QueryEngine;
+use rustyline::error::ReadlineError;
+use rustyline::Editor;
+use session::context::QueryContext;
+use snafu::{ErrorCompat, ResultExt};
+use substrait::{DFLogicalSubstraitConvertor, SubstraitPlan};
+
+use crate::cli::cmd::ReplCommand;
+use crate::cli::helper::RustylineHelper;
+use crate::cli::AttachCommand;
+use crate::error::{
+    CollectRecordBatchesSnafu, ParseSqlSnafu, PlanStatementSnafu, PrettyPrintRecordBatchesSnafu,
+    ReadlineSnafu, ReplCreationSnafu, RequestDatabaseSnafu, Result, StartMetaClientSnafu,
+    SubstraitEncodeLogicalPlanSnafu,
+};
+
+/// Captures the state of the repl, gathers commands and executes them one by one
+pub struct Repl {
+    /// Rustyline editor for interacting with user on command line
+    rl: Editor<RustylineHelper>,
+
+    /// Current prompt
+    prompt: String,
+
+    /// Client for interacting with GreptimeDB
+    database: Database,
+
+    query_engine: Option<DatafusionQueryEngine>,
+}
+
+#[allow(clippy::print_stdout)]
+impl Repl {
+    fn print_help(&self) {
+        println!("{}", ReplCommand::help())
+    }
+
+    pub(crate) async fn try_new(cmd: &AttachCommand) -> Result<Self> {
+        let mut rl = Editor::new().context(ReplCreationSnafu)?;
+
+        if !cmd.disable_helper {
+            rl.set_helper(Some(RustylineHelper::default()));
+
+            let history_file = history_file();
+            if let Err(e) = rl.load_history(&history_file) {
+                logging::debug!(
+                    "failed to load history file on {}, error: {e}",
+                    history_file.display()
+                );
+            }
+        }
+
+        let client = Client::with_urls([&cmd.grpc_addr]);
+        let database = Database::new(DEFAULT_CATALOG_NAME, DEFAULT_SCHEMA_NAME, client);
+
+        let query_engine = if let Some(meta_addr) = &cmd.meta_addr {
+            create_query_engine(meta_addr).await.map(Some)?
+        } else {
+            None
+        };
+
+        Ok(Self {
+            rl,
+            prompt: "> ".to_string(),
+            database,
+            query_engine,
+        })
+    }
+
+    /// Parse the next command
+    fn next_command(&mut self) -> Result<ReplCommand> {
+        match self.rl.readline(&self.prompt) {
+            Ok(ref line) => {
+                let request = line.trim();
+
+                self.rl.add_history_entry(request.to_string());
+
+                request.try_into()
+            }
+            Err(ReadlineError::Eof) | Err(ReadlineError::Interrupted) => Ok(ReplCommand::Exit),
+            // Some sort of real underlying error
+            Err(e) => Err(e).context(ReadlineSnafu),
+        }
+    }
+
+    /// Read Evaluate Print Loop (interactive command line) for GreptimeDB
+    ///
+    /// Inspired / based on repl.rs from InfluxDB IOX
+    pub(crate) async fn run(&mut self) -> Result<()> {
+        println!("Ready for commands. (Hint: try 'help')");
+
+        loop {
+            match self.next_command()? {
+                ReplCommand::Help => {
+                    self.print_help();
+                }
+                ReplCommand::UseDatabase { db_name } => {
+                    if self.execute_sql(format!("USE {db_name}")).await {
+                        println!("Using {db_name}");
+                        self.database.set_schema(&db_name);
+                        self.prompt = format!("[{db_name}] > ");
+                    }
+                }
+                ReplCommand::Sql { sql } => {
+                    self.execute_sql(sql).await;
+                }
+                ReplCommand::Exit => {
+                    return Ok(());
+                }
+            }
+        }
+    }
+
+    async fn execute_sql(&self, sql: String) -> bool {
+        self.do_execute_sql(sql)
+            .await
+            .map_err(|e| {
+                let status_code = e.status_code();
+                let root_cause = e.iter_chain().last().unwrap();
+                println!("Error: {}({status_code}), {root_cause}", status_code as u32)
+            })
+            .is_ok()
+    }
+
+    async fn do_execute_sql(&self, sql: String) -> Result<()> {
+        let start = Instant::now();
+
+        let output = if let Some(query_engine) = &self.query_engine {
+            let stmt = QueryLanguageParser::parse_sql(&sql)
+                .with_context(|_| ParseSqlSnafu { sql: sql.clone() })?;
+
+            let query_ctx = Arc::new(QueryContext::with(
+                self.database.catalog(),
+                self.database.schema(),
+            ));
+
+            let plan = query_engine
+                .planner()
+                .plan(stmt, query_ctx)
+                .await
+                .context(PlanStatementSnafu)?;
+
+            let LogicalPlan::DfPlan(plan) =
+                query_engine.optimize(&plan).context(PlanStatementSnafu)?;
+
+            let plan = DFLogicalSubstraitConvertor {}
+                .encode(plan)
+                .context(SubstraitEncodeLogicalPlanSnafu)?;
+
+            self.database.logical_plan(plan.to_vec()).await
+        } else {
+            self.database.sql(&sql).await
+        }
+        .context(RequestDatabaseSnafu { sql: &sql })?;
+
+        let either = match output {
+            Output::Stream(s) => {
+                let x = RecordBatches::try_collect(s)
+                    .await
+                    .context(CollectRecordBatchesSnafu)?;
+                Either::Left(x)
+            }
+            Output::RecordBatches(x) => Either::Left(x),
+            Output::AffectedRows(rows) => Either::Right(rows),
+        };
+
+        let end = Instant::now();
+
+        match either {
+            Either::Left(recordbatches) => {
+                let total_rows: usize = recordbatches.iter().map(|x| x.num_rows()).sum();
+                if total_rows > 0 {
+                    println!(
+                        "{}",
+                        recordbatches
+                            .pretty_print()
+                            .context(PrettyPrintRecordBatchesSnafu)?
+                    );
+                }
+                println!("Total Rows: {total_rows}")
+            }
+            Either::Right(rows) => println!("Affected Rows: {rows}"),
+        };
+
+        println!("Cost {} ms", (end - start).as_millis());
+        Ok(())
+    }
+}
+
+impl Drop for Repl {
+    fn drop(&mut self) {
+        if self.rl.helper().is_some() {
+            let history_file = history_file();
+            if let Err(e) = self.rl.save_history(&history_file) {
+                logging::debug!(
+                    "failed to save history file on {}, error: {e}",
+                    history_file.display()
+                );
+            }
+        }
+    }
+}
+
+/// Return the location of the history file (defaults to $HOME/".greptimedb_cli_history")
+fn history_file() -> PathBuf {
+    let mut buf = match std::env::var("HOME") {
+        Ok(home) => PathBuf::from(home),
+        Err(_) => PathBuf::new(),
+    };
+    buf.push(".greptimedb_cli_history");
+    buf
+}
+
+async fn create_query_engine(meta_addr: &str) -> Result<DatafusionQueryEngine> {
+    let mut meta_client = MetaClientBuilder::default().enable_store().build();
+    meta_client
+        .start([meta_addr])
+        .await
+        .context(StartMetaClientSnafu)?;
+    let meta_client = Arc::new(meta_client);
+
+    let backend = Arc::new(MetaKvBackend {
+        client: meta_client.clone(),
+    });
+
+    let table_routes = Arc::new(TableRoutes::new(meta_client));
+    let partition_manager = Arc::new(PartitionRuleManager::new(table_routes));
+
+    let datanode_clients = Arc::new(DatanodeClients::default());
+
+    let catalog_list = Arc::new(FrontendCatalogManager::new(
+        backend,
+        partition_manager,
+        datanode_clients,
+    ));
+    let state = Arc::new(QueryEngineState::new(catalog_list, Default::default()));
+
+    Ok(DatafusionQueryEngine::new(state))
+}
--- a/src/cmd/src/datanode.rs
+++ b/src/cmd/src/datanode.rs
@@ -14,14 +14,31 @@

 use clap::Parser;
 use common_telemetry::logging;
-use datanode::datanode::{Datanode, DatanodeOptions, FileConfig, ObjectStoreConfig};
-use meta_client::MetaClientOpts;
+use datanode::datanode::{
+    Datanode, DatanodeOptions, FileConfig, ObjectStoreConfig, ProcedureConfig,
+};
+use meta_client::MetaClientOptions;
 use servers::Mode;
 use snafu::ResultExt;

 use crate::error::{Error, MissingConfigSnafu, Result, StartDatanodeSnafu};
 use crate::toml_loader;

+pub struct Instance {
+    datanode: Datanode,
+}
+
+impl Instance {
+    pub async fn run(&mut self) -> Result<()> {
+        self.datanode.start().await.context(StartDatanodeSnafu)
+    }
+
+    pub async fn stop(&self) -> Result<()> {
+        // TODO: handle datanode shutdown
+        Ok(())
+    }
+}
+
 #[derive(Parser)]
 pub struct Command {
    #[clap(subcommand)]
@@ -29,8 +46,8 @@ pub struct Command {
 }

 impl Command {
-    pub async fn run(self) -> Result<()> {
-        self.subcmd.run().await
+    pub async fn build(self) -> Result<Instance> {
+        self.subcmd.build().await
    }
 }

@@ -40,9 +57,9 @@ enum SubCommand {
 }

 impl SubCommand {
-    async fn run(self) -> Result<()> {
+    async fn build(self) -> Result<Instance> {
        match self {
-            SubCommand::Start(cmd) => cmd.run().await,
+            SubCommand::Start(cmd) => cmd.build().await,
        }
    }
 }
@@ -65,22 +82,21 @@ struct StartCommand {
    data_dir: Option<String>,
    #[clap(long)]
    wal_dir: Option<String>,
+    #[clap(long)]
+    procedure_dir: Option<String>,
 }

 impl StartCommand {
-    async fn run(self) -> Result<()> {
+    async fn build(self) -> Result<Instance> {
        logging::info!("Datanode start command: {:#?}", self);

        let opts: DatanodeOptions = self.try_into()?;

        logging::info!("Datanode options: {:#?}", opts);

-        Datanode::new(opts)
-            .await
-            .context(StartDatanodeSnafu)?
-            .start()
-            .await
-            .context(StartDatanodeSnafu)
+        let datanode = Datanode::new(opts).await.context(StartDatanodeSnafu)?;
+
+        Ok(Instance { datanode })
    }
 }

@@ -110,8 +126,8 @@ impl TryFrom<StartCommand> for DatanodeOptions {
        }

        if let Some(meta_addr) = cmd.metasrv_addr {
-            opts.meta_client_opts
-                .get_or_insert_with(MetaClientOpts::default)
+            opts.meta_client_options
+                .get_or_insert_with(MetaClientOptions::default)
                .metasrv_addrs = meta_addr
                .split(',')
                .map(&str::trim)
@@ -134,6 +150,10 @@ impl TryFrom<StartCommand> for DatanodeOptions {
        if let Some(wal_dir) = cmd.wal_dir {
            opts.wal.dir = wal_dir;
        }
+        if let Some(procedure_dir) = cmd.procedure_dir {
+            opts.procedure = Some(ProcedureConfig::from_file_path(procedure_dir));
+        }
+
        Ok(opts)
    }
 }
@@ -141,8 +161,10 @@ impl TryFrom<StartCommand> for DatanodeOptions {
 #[cfg(test)]
 mod tests {
    use std::assert_matches::assert_matches;
+    use std::io::Write;
    use std::time::Duration;

+    use common_test_util::temp_dir::create_named_temp_file;
    use datanode::datanode::{CompactionConfig, ObjectStoreConfig};
    use servers::Mode;

@@ -150,29 +172,68 @@ mod tests {

    #[test]
    fn test_read_from_config_file() {
+        let mut file = create_named_temp_file();
+        let toml_str = r#"
+            mode = "distributed"
+            enable_memory_catalog = false
+            node_id = 42
+            rpc_addr = "127.0.0.1:3001"
+            rpc_hostname = "127.0.0.1"
+            rpc_runtime_size = 8
+            mysql_addr = "127.0.0.1:4406"
+            mysql_runtime_size = 2
+
+            [meta_client_options]
+            metasrv_addrs = ["127.0.0.1:3002"]
+            timeout_millis = 3000
+            connect_timeout_millis = 5000
+            tcp_nodelay = true
+
+            [wal]
+            dir = "/tmp/greptimedb/wal"
+            file_size = "1GB"
+            purge_threshold = "50GB"
+            purge_interval = "10m"
+            read_batch_size = 128
+            sync_write = false
+
+            [storage]
+            type = "File"
+            data_dir = "/tmp/greptimedb/data/"
+
+            [compaction]
+            max_inflight_tasks = 4
+            max_files_in_level0 = 8
+            max_purge_tasks = 32
+        "#;
+        write!(file, "{}", toml_str).unwrap();
+
        let cmd = StartCommand {
-            config_file: Some(format!(
-                "{}/../../config/datanode.example.toml",
-                std::env::current_dir().unwrap().as_path().to_str().unwrap()
-            )),
+            config_file: Some(file.path().to_str().unwrap().to_string()),
            ..Default::default()
        };
        let options: DatanodeOptions = cmd.try_into().unwrap();
        assert_eq!("127.0.0.1:3001".to_string(), options.rpc_addr);
-        assert_eq!("/tmp/greptimedb/wal".to_string(), options.wal.dir);
        assert_eq!("127.0.0.1:4406".to_string(), options.mysql_addr);
-        assert_eq!(4, options.mysql_runtime_size);
-        let MetaClientOpts {
+        assert_eq!(2, options.mysql_runtime_size);
+        assert_eq!(Some(42), options.node_id);
+
+        assert_eq!(Duration::from_secs(600), options.wal.purge_interval);
+        assert_eq!(1024 * 1024 * 1024, options.wal.file_size.0);
+        assert_eq!(1024 * 1024 * 1024 * 50, options.wal.purge_threshold.0);
+        assert!(!options.wal.sync_write);
+
+        let MetaClientOptions {
            metasrv_addrs: metasrv_addr,
            timeout_millis,
            connect_timeout_millis,
            tcp_nodelay,
-        } = options.meta_client_opts.unwrap();
+        } = options.meta_client_options.unwrap();

        assert_eq!(vec!["127.0.0.1:3002".to_string()], metasrv_addr);
        assert_eq!(5000, connect_timeout_millis);
        assert_eq!(3000, timeout_millis);
-        assert!(!tcp_nodelay);
+        assert!(tcp_nodelay);

        match options.storage {
            ObjectStoreConfig::File(FileConfig { data_dir }) => {
@@ -185,7 +246,7 @@ mod tests {
        assert_eq!(
            CompactionConfig {
                max_inflight_tasks: 4,
-                max_files_in_level0: 16,
+                max_files_in_level0: 8,
                max_purge_tasks: 32,
            },
            options.compaction
@@ -223,32 +284,4 @@ mod tests {
        })
        .unwrap();
    }
-
-    #[test]
-    fn test_merge_config() {
-        let dn_opts = DatanodeOptions::try_from(StartCommand {
-            config_file: Some(format!(
-                "{}/../../config/datanode.example.toml",
-                std::env::current_dir().unwrap().as_path().to_str().unwrap()
-            )),
-            ..Default::default()
-        })
-        .unwrap();
-        assert_eq!("/tmp/greptimedb/wal", dn_opts.wal.dir);
-        assert_eq!(Duration::from_secs(600), dn_opts.wal.purge_interval);
-        assert_eq!(1024 * 1024 * 1024, dn_opts.wal.file_size.0);
-        assert_eq!(1024 * 1024 * 1024 * 50, dn_opts.wal.purge_threshold.0);
-        assert!(!dn_opts.wal.sync_write);
-        assert_eq!(Some(42), dn_opts.node_id);
-        let MetaClientOpts {
-            metasrv_addrs: metasrv_addr,
-            timeout_millis,
-            connect_timeout_millis,
-            tcp_nodelay,
-        } = dn_opts.meta_client_opts.unwrap();
-        assert_eq!(vec!["127.0.0.1:3002".to_string()], metasrv_addr);
-        assert_eq!(3000, timeout_millis);
-        assert_eq!(5000, connect_timeout_millis);
-        assert!(!tcp_nodelay);
-    }
 }
--- a/src/cmd/src/error.rs
+++ b/src/cmd/src/error.rs
@@ -15,6 +15,7 @@
 use std::any::Any;

 use common_error::prelude::*;
+use rustyline::error::ReadlineError;

 #[derive(Debug, Snafu)]
 #[snafu(visibility(pub))]
@@ -31,6 +32,12 @@ pub enum Error {
        source: frontend::error::Error,
    },

+    #[snafu(display("Failed to build meta server, source: {}", source))]
+    BuildMetaServer {
+        #[snafu(backtrace)]
+        source: meta_srv::error::Error,
+    },
+
    #[snafu(display("Failed to start meta server, source: {}", source))]
    StartMetaServer {
        #[snafu(backtrace)]
@@ -68,6 +75,65 @@ pub enum Error {
        #[snafu(backtrace)]
        source: meta_srv::error::Error,
    },
+
+    #[snafu(display("Invalid REPL command: {reason}"))]
+    InvalidReplCommand { reason: String },
+
+    #[snafu(display("Cannot create REPL: {}", source))]
+    ReplCreation {
+        source: ReadlineError,
+        backtrace: Backtrace,
+    },
+
+    #[snafu(display("Error reading command: {}", source))]
+    Readline {
+        source: ReadlineError,
+        backtrace: Backtrace,
+    },
+
+    #[snafu(display("Failed to request database, sql: {sql}, source: {source}"))]
+    RequestDatabase {
+        sql: String,
+        #[snafu(backtrace)]
+        source: client::Error,
+    },
+
+    #[snafu(display("Failed to collect RecordBatches, source: {source}"))]
+    CollectRecordBatches {
+        #[snafu(backtrace)]
+        source: common_recordbatch::error::Error,
+    },
+
+    #[snafu(display("Failed to pretty print Recordbatches, source: {source}"))]
+    PrettyPrintRecordBatches {
+        #[snafu(backtrace)]
+        source: common_recordbatch::error::Error,
+    },
+
+    #[snafu(display("Failed to start Meta client, source: {}", source))]
+    StartMetaClient {
+        #[snafu(backtrace)]
+        source: meta_client::error::Error,
+    },
+
+    #[snafu(display("Failed to parse SQL: {}, source: {}", sql, source))]
+    ParseSql {
+        sql: String,
+        #[snafu(backtrace)]
+        source: query::error::Error,
+    },
+
+    #[snafu(display("Failed to plan statement, source: {}", source))]
+    PlanStatement {
+        #[snafu(backtrace)]
+        source: query::error::Error,
+    },
+
+    #[snafu(display("Failed to encode logical plan in substrait, source: {}", source))]
+    SubstraitEncodeLogicalPlan {
+        #[snafu(backtrace)]
+        source: substrait::error::Error,
+    },
 }

 pub type Result<T> = std::result::Result<T, Error>;
@@ -78,12 +144,25 @@ impl ErrorExt for Error {
            Error::StartDatanode { source } => source.status_code(),
            Error::StartFrontend { source } => source.status_code(),
            Error::StartMetaServer { source } => source.status_code(),
+            Error::BuildMetaServer { source } => source.status_code(),
            Error::UnsupportedSelectorType { source, .. } => source.status_code(),
            Error::ReadConfig { .. } | Error::ParseConfig { .. } | Error::MissingConfig { .. } => {
                StatusCode::InvalidArguments
            }
-            Error::IllegalConfig { .. } => StatusCode::InvalidArguments,
+            Error::IllegalConfig { .. } | Error::InvalidReplCommand { .. } => {
+                StatusCode::InvalidArguments
+            }
            Error::IllegalAuthConfig { .. } => StatusCode::InvalidArguments,
+            Error::ReplCreation { .. } | Error::Readline { .. } => StatusCode::Internal,
+            Error::RequestDatabase { source, .. } => source.status_code(),
+            Error::CollectRecordBatches { source } | Error::PrettyPrintRecordBatches { source } => {
+                source.status_code()
+            }
+            Error::StartMetaClient { source } => source.status_code(),
+            Error::ParseSql { source, .. } | Error::PlanStatement { source } => {
+                source.status_code()
+            }
+            Error::SubstraitEncodeLogicalPlan { source } => source.status_code(),
        }
    }

--- a/src/cmd/src/frontend.rs
+++ b/src/cmd/src/frontend.rs
@@ -16,14 +16,15 @@ use std::sync::Arc;

 use clap::Parser;
 use common_base::Plugins;
-use frontend::frontend::{Frontend, FrontendOptions};
+use frontend::frontend::FrontendOptions;
 use frontend::grpc::GrpcOptions;
 use frontend::influxdb::InfluxdbOptions;
-use frontend::instance::Instance;
+use frontend::instance::{FrontendInstance, Instance as FeInstance};
 use frontend::mysql::MysqlOptions;
 use frontend::opentsdb::OpentsdbOptions;
 use frontend::postgres::PostgresOptions;
-use meta_client::MetaClientOpts;
+use frontend::prom::PromOptions;
+use meta_client::MetaClientOptions;
 use servers::auth::UserProviderRef;
 use servers::http::HttpOptions;
 use servers::tls::{TlsMode, TlsOption};
@@ -33,6 +34,24 @@ use snafu::ResultExt;
 use crate::error::{self, IllegalAuthConfigSnafu, Result};
 use crate::toml_loader;

+pub struct Instance {
+    frontend: FeInstance,
+}
+
+impl Instance {
+    pub async fn run(&mut self) -> Result<()> {
+        self.frontend
+            .start()
+            .await
+            .context(error::StartFrontendSnafu)
+    }
+
+    pub async fn stop(&self) -> Result<()> {
+        // TODO: handle frontend shutdown
+        Ok(())
+    }
+}
+
 #[derive(Parser)]
 pub struct Command {
    #[clap(subcommand)]
@@ -40,8 +59,8 @@ pub struct Command {
 }

 impl Command {
-    pub async fn run(self) -> Result<()> {
-        self.subcmd.run().await
+    pub async fn build(self) -> Result<Instance> {
+        self.subcmd.build().await
    }
 }

@@ -51,9 +70,9 @@ enum SubCommand {
 }

 impl SubCommand {
-    async fn run(self) -> Result<()> {
+    async fn build(self) -> Result<Instance> {
        match self {
-            SubCommand::Start(cmd) => cmd.run().await,
+            SubCommand::Start(cmd) => cmd.build().await,
        }
    }
 }
@@ -67,6 +86,8 @@ pub struct StartCommand {
    #[clap(long)]
    mysql_addr: Option<String>,
    #[clap(long)]
+    prom_addr: Option<String>,
+    #[clap(long)]
    postgres_addr: Option<String>,
    #[clap(long)]
    opentsdb_addr: Option<String>,
@@ -87,16 +108,20 @@ pub struct StartCommand {
 }

 impl StartCommand {
-    async fn run(self) -> Result<()> {
+    async fn build(self) -> Result<Instance> {
        let plugins = Arc::new(load_frontend_plugins(&self.user_provider)?);
        let opts: FrontendOptions = self.try_into()?;

-        let instance = Instance::try_new_distributed(&opts, plugins.clone())
+        let mut instance = FeInstance::try_new_distributed(&opts, plugins.clone())
            .await
            .context(error::StartFrontendSnafu)?;

-        let mut frontend = Frontend::new(opts, instance, plugins);
-        frontend.start().await.context(error::StartFrontendSnafu)
+        instance
+            .build_servers(&opts, plugins)
+            .await
+            .context(error::StartFrontendSnafu)?;
+
+        Ok(Instance { frontend: instance })
    }
 }

@@ -141,6 +166,9 @@ impl TryFrom<StartCommand> for FrontendOptions {
                ..Default::default()
            });
        }
+        if let Some(addr) = cmd.prom_addr {
+            opts.prom_options = Some(PromOptions { addr });
+        }
        if let Some(addr) = cmd.postgres_addr {
            opts.postgres_options = Some(PostgresOptions {
                addr,
@@ -158,8 +186,8 @@ impl TryFrom<StartCommand> for FrontendOptions {
            opts.influxdb_options = Some(InfluxdbOptions { enable });
        }
        if let Some(metasrv_addr) = cmd.metasrv_addr {
-            opts.meta_client_opts
-                .get_or_insert_with(MetaClientOpts::default)
+            opts.meta_client_options
+                .get_or_insert_with(MetaClientOptions::default)
                .metasrv_addrs = metasrv_addr
                .split(',')
                .map(&str::trim)
@@ -173,8 +201,10 @@ impl TryFrom<StartCommand> for FrontendOptions {

 #[cfg(test)]
 mod tests {
+    use std::io::Write;
    use std::time::Duration;

+    use common_test_util::temp_dir::create_named_temp_file;
    use servers::auth::{Identity, Password, UserProviderRef};

    use super::*;
@@ -184,6 +214,7 @@ mod tests {
        let command = StartCommand {
            http_addr: Some("127.0.0.1:1234".to_string()),
            grpc_addr: None,
+            prom_addr: Some("127.0.0.1:4444".to_string()),
            mysql_addr: Some("127.0.0.1:5678".to_string()),
            postgres_addr: Some("127.0.0.1:5432".to_string()),
            opentsdb_addr: Some("127.0.0.1:4321".to_string()),
@@ -207,6 +238,7 @@ mod tests {
            opts.opentsdb_options.as_ref().unwrap().addr,
            "127.0.0.1:4321"
        );
+        assert_eq!(opts.prom_options.as_ref().unwrap().addr, "127.0.0.1:4444");

        let default_opts = FrontendOptions::default();
        assert_eq!(
@@ -231,17 +263,25 @@ mod tests {

    #[test]
    fn test_read_from_config_file() {
+        let mut file = create_named_temp_file();
+        let toml_str = r#"
+            mode = "distributed"
+
+            [http_options]
+            addr = "127.0.0.1:4000"
+            timeout = "30s"
+        "#;
+        write!(file, "{}", toml_str).unwrap();
+
        let command = StartCommand {
            http_addr: None,
            grpc_addr: None,
            mysql_addr: None,
+            prom_addr: None,
            postgres_addr: None,
            opentsdb_addr: None,
            influxdb_enable: None,
-            config_file: Some(format!(
-                "{}/../../config/frontend.example.toml",
-                std::env::current_dir().unwrap().as_path().to_str().unwrap()
-            )),
+            config_file: Some(file.path().to_str().unwrap().to_string()),
            metasrv_addr: None,
            tls_mode: None,
            tls_cert_path: None,
@@ -267,6 +307,7 @@ mod tests {
            http_addr: None,
            grpc_addr: None,
            mysql_addr: None,
+            prom_addr: None,
            postgres_addr: None,
            opentsdb_addr: None,
            influxdb_enable: None,
--- a/src/cmd/src/lib.rs
+++ b/src/cmd/src/lib.rs
@@ -14,6 +14,7 @@

 #![feature(assert_matches)]

+pub mod cli;
 pub mod datanode;
 pub mod error;
 pub mod frontend;
--- a/src/cmd/src/metasrv.rs
+++ b/src/cmd/src/metasrv.rs
@@ -14,13 +14,32 @@

 use clap::Parser;
 use common_telemetry::{info, logging, warn};
-use meta_srv::bootstrap;
+use meta_srv::bootstrap::MetaSrvInstance;
 use meta_srv::metasrv::MetaSrvOptions;
 use snafu::ResultExt;

 use crate::error::{Error, Result};
 use crate::{error, toml_loader};

+pub struct Instance {
+    instance: MetaSrvInstance,
+}
+
+impl Instance {
+    pub async fn run(&mut self) -> Result<()> {
+        self.instance
+            .start()
+            .await
+            .context(error::StartMetaServerSnafu)?;
+        Ok(())
+    }
+
+    pub async fn stop(&self) -> Result<()> {
+        // TODO: handle metasrv shutdown
+        Ok(())
+    }
+}
+
 #[derive(Parser)]
 pub struct Command {
    #[clap(subcommand)]
@@ -28,8 +47,8 @@ pub struct Command {
 }

 impl Command {
-    pub async fn run(self) -> Result<()> {
-        self.subcmd.run().await
+    pub async fn build(self) -> Result<Instance> {
+        self.subcmd.build().await
    }
 }

@@ -39,9 +58,9 @@ enum SubCommand {
 }

 impl SubCommand {
-    async fn run(self) -> Result<()> {
+    async fn build(self) -> Result<Instance> {
        match self {
-            SubCommand::Start(cmd) => cmd.run().await,
+            SubCommand::Start(cmd) => cmd.build().await,
        }
    }
 }
@@ -63,16 +82,17 @@ struct StartCommand {
 }

 impl StartCommand {
-    async fn run(self) -> Result<()> {
+    async fn build(self) -> Result<Instance> {
        logging::info!("MetaSrv start command: {:#?}", self);

        let opts: MetaSrvOptions = self.try_into()?;

        logging::info!("MetaSrv options: {:#?}", opts);
-
-        bootstrap::bootstrap_meta_srv(opts)
+        let instance = MetaSrvInstance::new(opts)
            .await
-            .context(error::StartMetaServerSnafu)
+            .context(error::BuildMetaServerSnafu)?;
+
+        Ok(Instance { instance })
    }
 }

@@ -113,6 +133,9 @@ impl TryFrom<StartCommand> for MetaSrvOptions {

 #[cfg(test)]
 mod tests {
+    use std::io::Write;
+
+    use common_test_util::temp_dir::create_named_temp_file;
    use meta_srv::selector::SelectorType;

    use super::*;
@@ -136,15 +159,23 @@ mod tests {

    #[test]
    fn test_read_from_config_file() {
+        let mut file = create_named_temp_file();
+        let toml_str = r#"
+            bind_addr = "127.0.0.1:3002"
+            server_addr = "127.0.0.1:3002"
+            store_addr = "127.0.0.1:2379"
+            datanode_lease_secs = 15
+            selector = "LeaseBased"
+            use_memory_store = false
+        "#;
+        write!(file, "{}", toml_str).unwrap();
+
        let cmd = StartCommand {
            bind_addr: None,
            server_addr: None,
            store_addr: None,
            selector: None,
-            config_file: Some(format!(
-                "{}/../../config/metasrv.example.toml",
-                std::env::current_dir().unwrap().as_path().to_str().unwrap()
-            )),
+            config_file: Some(file.path().to_str().unwrap().to_string()),
            use_memory_store: false,
        };
        let options: MetaSrvOptions = cmd.try_into().unwrap();
--- a/src/cmd/src/standalone.rs
+++ b/src/cmd/src/standalone.rs
@@ -17,17 +17,19 @@ use std::sync::Arc;
 use clap::Parser;
 use common_base::Plugins;
 use common_telemetry::info;
-use datanode::datanode::{Datanode, DatanodeOptions, ObjectStoreConfig, WalConfig};
+use datanode::datanode::{
+    CompactionConfig, Datanode, DatanodeOptions, ObjectStoreConfig, ProcedureConfig, WalConfig,
+};
 use datanode::instance::InstanceRef;
-use frontend::frontend::{Frontend, FrontendOptions};
+use frontend::frontend::FrontendOptions;
 use frontend::grpc::GrpcOptions;
 use frontend::influxdb::InfluxdbOptions;
-use frontend::instance::Instance as FeInstance;
+use frontend::instance::{FrontendInstance, Instance as FeInstance};
 use frontend::mysql::MysqlOptions;
 use frontend::opentsdb::OpentsdbOptions;
 use frontend::postgres::PostgresOptions;
+use frontend::prom::PromOptions;
 use frontend::prometheus::PrometheusOptions;
-use frontend::promql::PromqlOptions;
 use serde::{Deserialize, Serialize};
 use servers::http::HttpOptions;
 use servers::tls::{TlsMode, TlsOption};
@@ -45,8 +47,8 @@ pub struct Command {
 }

 impl Command {
-    pub async fn run(self) -> Result<()> {
-        self.subcmd.run().await
+    pub async fn build(self) -> Result<Instance> {
+        self.subcmd.build().await
    }
 }

@@ -56,9 +58,9 @@ enum SubCommand {
 }

 impl SubCommand {
-    async fn run(self) -> Result<()> {
+    async fn build(self) -> Result<Instance> {
        match self {
-            SubCommand::Start(cmd) => cmd.run().await,
+            SubCommand::Start(cmd) => cmd.build().await,
        }
    }
 }
@@ -66,6 +68,8 @@ impl SubCommand {
 #[derive(Clone, Debug, Serialize, Deserialize)]
 #[serde(default)]
 pub struct StandaloneOptions {
+    pub mode: Mode,
+    pub enable_memory_catalog: bool,
    pub http_options: Option<HttpOptions>,
    pub grpc_options: Option<GrpcOptions>,
    pub mysql_options: Option<MysqlOptions>,
@@ -73,16 +77,18 @@ pub struct StandaloneOptions {
    pub opentsdb_options: Option<OpentsdbOptions>,
    pub influxdb_options: Option<InfluxdbOptions>,
    pub prometheus_options: Option<PrometheusOptions>,
-    pub promql_options: Option<PromqlOptions>,
-    pub mode: Mode,
+    pub prom_options: Option<PromOptions>,
    pub wal: WalConfig,
    pub storage: ObjectStoreConfig,
-    pub enable_memory_catalog: bool,
+    pub compaction: CompactionConfig,
+    pub procedure: Option<ProcedureConfig>,
 }

 impl Default for StandaloneOptions {
    fn default() -> Self {
        Self {
+            mode: Mode::Standalone,
+            enable_memory_catalog: false,
            http_options: Some(HttpOptions::default()),
            grpc_options: Some(GrpcOptions::default()),
            mysql_options: Some(MysqlOptions::default()),
@@ -90,11 +96,11 @@ impl Default for StandaloneOptions {
            opentsdb_options: Some(OpentsdbOptions::default()),
            influxdb_options: Some(InfluxdbOptions::default()),
            prometheus_options: Some(PrometheusOptions::default()),
-            promql_options: Some(PromqlOptions::default()),
-            mode: Mode::Standalone,
+            prom_options: Some(PromOptions::default()),
            wal: WalConfig::default(),
            storage: ObjectStoreConfig::default(),
-            enable_memory_catalog: false,
+            compaction: CompactionConfig::default(),
+            procedure: None,
        }
    }
 }
@@ -102,6 +108,7 @@ impl Default for StandaloneOptions {
 impl StandaloneOptions {
    fn frontend_options(self) -> FrontendOptions {
        FrontendOptions {
+            mode: self.mode,
            http_options: self.http_options,
            grpc_options: self.grpc_options,
            mysql_options: self.mysql_options,
@@ -109,22 +116,47 @@ impl StandaloneOptions {
            opentsdb_options: self.opentsdb_options,
            influxdb_options: self.influxdb_options,
            prometheus_options: self.prometheus_options,
-            promql_options: self.promql_options,
-            mode: self.mode,
-            meta_client_opts: None,
+            prom_options: self.prom_options,
+            meta_client_options: None,
        }
    }

    fn datanode_options(self) -> DatanodeOptions {
        DatanodeOptions {
+            enable_memory_catalog: self.enable_memory_catalog,
            wal: self.wal,
            storage: self.storage,
-            enable_memory_catalog: self.enable_memory_catalog,
+            compaction: self.compaction,
+            procedure: self.procedure,
            ..Default::default()
        }
    }
 }

+pub struct Instance {
+    datanode: Datanode,
+    frontend: FeInstance,
+}
+
+impl Instance {
+    pub async fn run(&mut self) -> Result<()> {
+        // Start datanode instance before starting services, to avoid requests come in before internal components are started.
+        self.datanode
+            .start_instance()
+            .await
+            .context(StartDatanodeSnafu)?;
+        info!("Datanode instance started");
+
+        self.frontend.start().await.context(StartFrontendSnafu)?;
+        Ok(())
+    }
+
+    pub async fn stop(&self) -> Result<()> {
+        // TODO: handle standalone shutdown
+        Ok(())
+    }
+}
+
 #[derive(Debug, Parser)]
 struct StartCommand {
    #[clap(long)]
@@ -134,6 +166,8 @@ struct StartCommand {
    #[clap(long)]
    mysql_addr: Option<String>,
    #[clap(long)]
+    prom_addr: Option<String>,
+    #[clap(long)]
    postgres_addr: Option<String>,
    #[clap(long)]
    opentsdb_addr: Option<String>,
@@ -154,7 +188,7 @@ struct StartCommand {
 }

 impl StartCommand {
-    async fn run(self) -> Result<()> {
+    async fn build(self) -> Result<Instance> {
        let enable_memory_catalog = self.enable_memory_catalog;
        let config_file = self.config_file.clone();
        let plugins = Arc::new(load_frontend_plugins(&self.user_provider)?);
@@ -174,33 +208,30 @@ impl StartCommand {
            fe_opts, dn_opts
        );

-        let mut datanode = Datanode::new(dn_opts.clone())
+        let datanode = Datanode::new(dn_opts.clone())
            .await
            .context(StartDatanodeSnafu)?;
-        let mut frontend = build_frontend(fe_opts, plugins, datanode.get_instance()).await?;

-        // Start datanode instance before starting services, to avoid requests come in before internal components are started.
-        datanode
-            .start_instance()
+        let mut frontend = build_frontend(plugins.clone(), datanode.get_instance()).await?;
+
+        frontend
+            .build_servers(&fe_opts, plugins)
            .await
-            .context(StartDatanodeSnafu)?;
-        info!("Datanode instance started");
+            .context(StartFrontendSnafu)?;

-        frontend.start().await.context(StartFrontendSnafu)?;
-        Ok(())
+        Ok(Instance { datanode, frontend })
    }
 }

 /// Build frontend instance in standalone mode
 async fn build_frontend(
-    fe_opts: FrontendOptions,
    plugins: Arc<Plugins>,
    datanode_instance: InstanceRef,
-) -> Result<Frontend<FeInstance>> {
+) -> Result<FeInstance> {
    let mut frontend_instance = FeInstance::new_standalone(datanode_instance.clone());
    frontend_instance.set_script_handler(datanode_instance);
    frontend_instance.set_plugins(plugins.clone());
-    Ok(Frontend::new(fe_opts, frontend_instance, plugins))
+    Ok(frontend_instance)
 }

 impl TryFrom<StartCommand> for FrontendOptions {
@@ -246,6 +277,11 @@ impl TryFrom<StartCommand> for FrontendOptions {
                ..Default::default()
            })
        }
+
+        if let Some(addr) = cmd.prom_addr {
+            opts.prom_options = Some(PromOptions { addr })
+        }
+
        if let Some(addr) = cmd.postgres_addr {
            opts.postgres_options = Some(PostgresOptions {
                addr,
@@ -294,6 +330,7 @@ mod tests {
            http_addr: None,
            rpc_addr: None,
            mysql_addr: None,
+            prom_addr: None,
            postgres_addr: None,
            opentsdb_addr: None,
            config_file: Some(format!(
@@ -339,6 +376,7 @@ mod tests {
        let command = StartCommand {
            http_addr: None,
            rpc_addr: None,
+            prom_addr: None,
            mysql_addr: None,
            postgres_addr: None,
            opentsdb_addr: None,
@@ -362,4 +400,11 @@ mod tests {
            .await;
        assert!(result.is_ok());
    }
+
+    #[test]
+    fn test_toml() {
+        let opts = StandaloneOptions::default();
+        let toml_string = toml::to_string(&opts).unwrap();
+        let _parsed: StandaloneOptions = toml::from_str(&toml_string).unwrap();
+    }
 }
--- a/src/cmd/src/toml_loader.rs
+++ b/src/cmd/src/toml_loader.rs
@@ -29,9 +29,9 @@ mod tests {
    use std::fs::File;
    use std::io::Write;

+    use common_test_util::temp_dir::create_temp_dir;
    use serde::{Deserialize, Serialize};
    use snafu::ResultExt;
-    use tempdir::TempDir;

    use super::*;
    use crate::error::Result;
@@ -62,7 +62,7 @@ mod tests {
            host: "greptime.test".to_string(),
        };

-        let dir = TempDir::new("test_from_file").unwrap();
+        let dir = create_temp_dir("test_from_file");
        let test_file = format!("{}/test.toml", dir.path().to_str().unwrap());

        let s = toml::to_string(&config).unwrap();
--- a/src/cmd/tests/cli.rs
+++ b/src/cmd/tests/cli.rs
@@ -0,0 +1,148 @@
+// Copyright 2023 Greptime Team
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#[cfg(target_os = "macos")]
+mod tests {
+    use std::path::PathBuf;
+    use std::process::{Command, Stdio};
+    use std::time::Duration;
+
+    use common_test_util::temp_dir::create_temp_dir;
+    use rexpect::session::PtyReplSession;
+
+    struct Repl {
+        repl: PtyReplSession,
+    }
+
+    impl Repl {
+        fn send_line(&mut self, line: &str) {
+            self.repl.send_line(line).unwrap();
+
+            // read a line to consume the prompt
+            self.read_line();
+        }
+
+        fn read_line(&mut self) -> String {
+            self.repl.read_line().unwrap()
+        }
+
+        fn read_expect(&mut self, expect: &str) {
+            assert_eq!(self.read_line(), expect);
+        }
+
+        fn read_contains(&mut self, pat: &str) {
+            assert!(self.read_line().contains(pat));
+        }
+    }
+
+    // TODO(LFC): Un-ignore this REPL test.
+    // Ignore this REPL test because some logical plans like create database are not supported yet in Datanode.
+    #[ignore]
+    #[test]
+    fn test_repl() {
+        let data_dir = create_temp_dir("data");
+        let wal_dir = create_temp_dir("wal");
+
+        let mut bin_path = PathBuf::from(env!("CARGO_MANIFEST_DIR"));
+        bin_path.push("../../target/debug");
+        let bin_path = bin_path.to_str().unwrap();
+
+        let mut datanode = Command::new("./greptime")
+            .current_dir(bin_path)
+            .args([
+                "datanode",
+                "start",
+                "--rpc-addr=0.0.0.0:4321",
+                "--node-id=1",
+                &format!("--data-dir={}", data_dir.path().display()),
+                &format!("--wal-dir={}", wal_dir.path().display()),
+            ])
+            .stdout(Stdio::null())
+            .spawn()
+            .unwrap();
+
+        // wait for Datanode actually started
+        std::thread::sleep(Duration::from_secs(3));
+
+        let mut repl_cmd = Command::new("./greptime");
+        repl_cmd.current_dir(bin_path).args([
+            "--log-level=off",
+            "cli",
+            "attach",
+            "--grpc-addr=0.0.0.0:4321",
+            // history commands can sneaky into stdout and mess up our tests, so disable it
+            "--disable-helper",
+        ]);
+        let pty_session = rexpect::session::spawn_command(repl_cmd, Some(5_000)).unwrap();
+        let repl = PtyReplSession {
+            prompt: "> ".to_string(),
+            pty_session,
+            quit_command: None,
+            echo_on: false,
+        };
+        let repl = &mut Repl { repl };
+        repl.read_expect("Ready for commands. (Hint: try 'help')");
+
+        test_create_database(repl);
+
+        test_use_database(repl);
+
+        test_create_table(repl);
+
+        test_insert(repl);
+
+        test_select(repl);
+
+        datanode.kill().unwrap();
+        datanode.wait().unwrap();
+    }
+
+    fn test_create_database(repl: &mut Repl) {
+        repl.send_line("CREATE DATABASE db;");
+        repl.read_expect("Affected Rows: 1");
+        repl.read_contains("Cost");
+    }
+
+    fn test_use_database(repl: &mut Repl) {
+        repl.send_line("USE db");
+        repl.read_expect("Total Rows: 0");
+        repl.read_contains("Cost");
+        repl.read_expect("Using db");
+    }
+
+    fn test_create_table(repl: &mut Repl) {
+        repl.send_line("CREATE TABLE t(x STRING, ts TIMESTAMP TIME INDEX);");
+        repl.read_expect("Affected Rows: 0");
+        repl.read_contains("Cost");
+    }
+
+    fn test_insert(repl: &mut Repl) {
+        repl.send_line("INSERT INTO t(x, ts) VALUES ('hello', 1676895812239);");
+        repl.read_expect("Affected Rows: 1");
+        repl.read_contains("Cost");
+    }
+
+    fn test_select(repl: &mut Repl) {
+        repl.send_line("SELECT * FROM t;");
+
+        repl.read_expect("+-------+-------------------------+");
+        repl.read_expect("| x     | ts                      |");
+        repl.read_expect("+-------+-------------------------+");
+        repl.read_expect("| hello | 2023-02-20T12:23:32.239 |");
+        repl.read_expect("+-------+-------------------------+");
+        repl.read_expect("Total Rows: 1");
+
+        repl.read_contains("Cost");
+    }
+}
--- a/src/common/base/src/bytes.rs
+++ b/src/common/base/src/bytes.rs
@@ -20,6 +20,12 @@ use serde::{Deserialize, Deserializer, Serialize, Serializer};
 #[derive(Debug, Default, Clone, PartialEq, Eq, PartialOrd, Ord, Deserialize, Serialize)]
 pub struct Bytes(bytes::Bytes);

+impl From<Bytes> for bytes::Bytes {
+    fn from(value: Bytes) -> Self {
+        value.0
+    }
+}
+
 impl From<bytes::Bytes> for Bytes {
    fn from(bytes: bytes::Bytes) -> Bytes {
        Bytes(bytes)
--- a/src/common/catalog/Cargo.toml
+++ b/src/common/catalog/Cargo.toml
@@ -17,5 +17,4 @@ snafu = { version = "0.7", features = ["backtraces"] }

 [dev-dependencies]
 chrono.workspace = true
-tempdir = "0.3"
 tokio.workspace = true
--- a/src/common/error/src/status_code.rs
+++ b/src/common/error/src/status_code.rs
@@ -86,6 +86,34 @@ impl StatusCode {
    pub fn is_success(code: u32) -> bool {
        Self::Success as u32 == code
    }
+
+    pub fn is_retryable(&self) -> bool {
+        match self {
+            StatusCode::StorageUnavailable
+            | StatusCode::RuntimeResourcesExhausted
+            | StatusCode::Internal => true,
+
+            StatusCode::Success
+            | StatusCode::Unknown
+            | StatusCode::Unsupported
+            | StatusCode::Unexpected
+            | StatusCode::InvalidArguments
+            | StatusCode::InvalidSyntax
+            | StatusCode::PlanQuery
+            | StatusCode::EngineExecuteQuery
+            | StatusCode::TableAlreadyExists
+            | StatusCode::TableNotFound
+            | StatusCode::TableColumnNotFound
+            | StatusCode::TableColumnExists
+            | StatusCode::DatabaseNotFound
+            | StatusCode::UserNotFound
+            | StatusCode::UnsupportedPasswordType
+            | StatusCode::UserPasswordMismatch
+            | StatusCode::AuthHeaderNotFound
+            | StatusCode::InvalidAuthHeader
+            | StatusCode::AccessDenied => false,
+        }
+    }
 }

 impl fmt::Display for StatusCode {
--- a/src/common/function-macro/Cargo.toml
+++ b/src/common/function-macro/Cargo.toml
@@ -10,6 +10,7 @@ proc-macro = true
 [dependencies]
 quote = "1.0"
 syn = "1.0"
+proc-macro2 = "1.0"

 [dev-dependencies]
 arc-swap = "1.0"
--- a/src/common/function-macro/src/lib.rs
+++ b/src/common/function-macro/src/lib.rs
@@ -12,8 +12,11 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.

+mod range_fn;
+
 use proc_macro::TokenStream;
 use quote::{quote, quote_spanned};
+use range_fn::process_range_fn;
 use syn::parse::Parser;
 use syn::spanned::Spanned;
 use syn::{parse_macro_input, DeriveInput, ItemStruct};
@@ -83,3 +86,31 @@ pub fn as_aggr_func_creator(_args: TokenStream, input: TokenStream) -> TokenStre
    }
    .into()
 }
+
+/// Attribute macro to convert an arithimetic function to a range function. The annotated function
+/// should accept servaral arrays as input and return a single value as output. This procedure
+/// macro can works on any number of input parameters. Return type can be either primitive type
+/// or wrapped in `Option`.
+///
+/// # Example
+/// Take `count_over_time()` in PromQL as an example:
+/// ```rust, ignore
+/// /// The count of all values in the specified interval.
+/// #[range_fn(
+///     name = "CountOverTime",
+///     ret = "Float64Array",
+///     display_name = "prom_count_over_time"
+/// )]
+/// pub fn count_over_time(_: &TimestampMillisecondArray, values: &Float64Array) -> f64 {
+///      values.len() as f64
+/// }
+/// ```
+///
+/// # Arguments
+/// - `name`: The name of the generated [ScalarUDF] struct.
+/// - `ret`: The return type of the generated UDF function.
+/// - `display_name`: The display name of the generated UDF function.
+#[proc_macro_attribute]
+pub fn range_fn(args: TokenStream, input: TokenStream) -> TokenStream {
+    process_range_fn(args, input)
+}
--- a/src/common/function-macro/src/range_fn.rs
+++ b/src/common/function-macro/src/range_fn.rs
@@ -0,0 +1,230 @@
+// 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 proc_macro::TokenStream;
+use proc_macro2::Span;
+use quote::quote;
+use syn::punctuated::Punctuated;
+use syn::spanned::Spanned;
+use syn::token::Comma;
+use syn::{
+    parse_macro_input, Attribute, AttributeArgs, FnArg, Ident, ItemFn, Meta, MetaNameValue,
+    NestedMeta, Signature, Type, TypeReference, Visibility,
+};
+
+/// Internal util macro to early return on error.
+macro_rules! ok {
+    ($item:expr) => {
+        match $item {
+            Ok(item) => item,
+            Err(e) => return e.into_compile_error().into(),
+        }
+    };
+}
+
+pub(crate) fn process_range_fn(args: TokenStream, input: TokenStream) -> TokenStream {
+    // extract arg map
+    let arg_pairs = parse_macro_input!(args as AttributeArgs);
+    let arg_span = arg_pairs[0].span();
+    let arg_map = ok!(extract_arg_map(arg_pairs));
+
+    // decompose the fn block
+    let compute_fn = parse_macro_input!(input as ItemFn);
+    let ItemFn {
+        attrs,
+        vis,
+        sig,
+        block,
+    } = compute_fn;
+
+    // extract fn arg list
+    let Signature {
+        inputs,
+        ident: fn_name,
+        ..
+    } = &sig;
+    let arg_types = ok!(extract_input_types(inputs));
+
+    // build the struct and its impl block
+    let struct_code = build_struct(
+        attrs,
+        vis,
+        ok!(get_ident(&arg_map, "name", arg_span)),
+        ok!(get_ident(&arg_map, "display_name", arg_span)),
+    );
+    let calc_fn_code = build_calc_fn(
+        ok!(get_ident(&arg_map, "name", arg_span)),
+        arg_types,
+        fn_name.clone(),
+        ok!(get_ident(&arg_map, "ret", arg_span)),
+    );
+    // preserve this fn, but remove its `pub` modifier
+    let input_fn_code: TokenStream = quote! {
+        #sig { #block }
+    }
+    .into();
+
+    let mut result = TokenStream::new();
+    result.extend(struct_code);
+    result.extend(calc_fn_code);
+    result.extend(input_fn_code);
+    result
+}
+
+/// Extract a String <-> Ident map from the attribute args.
+fn extract_arg_map(args: Vec<NestedMeta>) -> Result<HashMap<String, Ident>, syn::Error> {
+    args.into_iter()
+        .map(|meta| {
+            if let NestedMeta::Meta(Meta::NameValue(MetaNameValue { path, lit, .. })) = meta {
+                let name = path.get_ident().unwrap().to_string();
+                let ident = match lit {
+                    syn::Lit::Str(lit_str) => lit_str.parse::<Ident>(),
+                    _ => Err(syn::Error::new(
+                        lit.span(),
+                        "Unexpected attribute format. Expected `name = \"value\"`",
+                    )),
+                }?;
+                Ok((name, ident))
+            } else {
+                Err(syn::Error::new(
+                    meta.span(),
+                    "Unexpected attribute format. Expected `name = \"value\"`",
+                ))
+            }
+        })
+        .collect::<Result<HashMap<String, Ident>, syn::Error>>()
+}
+
+/// Helper function to get an Ident from the previous arg map.
+fn get_ident(map: &HashMap<String, Ident>, key: &str, span: Span) -> Result<Ident, syn::Error> {
+    map.get(key)
+        .cloned()
+        .ok_or_else(|| syn::Error::new(span, format!("Expect attribute {key} but not found")))
+}
+
+/// Extract the argument list from the annotated function.
+fn extract_input_types(inputs: &Punctuated<FnArg, Comma>) -> Result<Vec<Type>, syn::Error> {
+    inputs
+        .iter()
+        .map(|arg| match arg {
+            FnArg::Receiver(receiver) => Err(syn::Error::new(receiver.span(), "expected bool")),
+            FnArg::Typed(pat_type) => Ok(*pat_type.ty.clone()),
+        })
+        .collect()
+}
+
+fn build_struct(
+    attrs: Vec<Attribute>,
+    vis: Visibility,
+    name: Ident,
+    display_name_ident: Ident,
+) -> TokenStream {
+    let display_name = display_name_ident.to_string();
+    quote! {
+        #(#attrs)*
+        #[derive(Debug)]
+        #vis struct #name {}
+
+        impl #name {
+            pub const fn name() -> &'static str {
+                #display_name
+            }
+
+            pub fn scalar_udf() -> ScalarUDF {
+                ScalarUDF {
+                    name: Self::name().to_string(),
+                    signature: Signature::new(
+                        TypeSignature::Exact(Self::input_type()),
+                        Volatility::Immutable,
+                    ),
+                    return_type: Arc::new(|_| Ok(Arc::new(Self::return_type()))),
+                    fun: Arc::new(Self::calc),
+                }
+            }
+
+            // TODO(ruihang): this should be parameterized
+            // time index column and value column
+            fn input_type() -> Vec<DataType> {
+                vec![
+                    RangeArray::convert_data_type(DataType::Timestamp(TimeUnit::Millisecond, None)),
+                    RangeArray::convert_data_type(DataType::Float64),
+                ]
+            }
+
+            // TODO(ruihang): this should be parameterized
+            fn return_type() -> DataType {
+                DataType::Float64
+            }
+        }
+    }
+    .into()
+}
+
+fn build_calc_fn(
+    name: Ident,
+    param_types: Vec<Type>,
+    fn_name: Ident,
+    ret_type: Ident,
+) -> TokenStream {
+    let param_names = param_types
+        .iter()
+        .enumerate()
+        .map(|(i, ty)| Ident::new(&format!("param_{}", i), ty.span()))
+        .collect::<Vec<_>>();
+    let unref_param_types = param_types
+        .iter()
+        .map(|ty| {
+            if let Type::Reference(TypeReference { elem, .. }) = ty {
+                elem.as_ref().clone()
+            } else {
+                ty.clone()
+            }
+        })
+        .collect::<Vec<_>>();
+    let num_params = param_types.len();
+    let param_numbers = (0..num_params).collect::<Vec<_>>();
+    let range_array_names = param_names
+        .iter()
+        .map(|name| Ident::new(&format!("{}_range_array", name), name.span()))
+        .collect::<Vec<_>>();
+    let first_range_array_name = range_array_names.first().unwrap().clone();
+
+    quote! {
+        impl #name {
+            fn calc(input: &[ColumnarValue]) -> Result<ColumnarValue, DataFusionError> {
+                assert_eq!(input.len(), #num_params);
+
+                #( let #range_array_names = RangeArray::try_new(extract_array(&input[#param_numbers])?.data().clone().into())?; )*
+
+                // TODO(ruihang): add ensure!() 
+
+                let mut result_array = Vec::new();
+                for index in 0..#first_range_array_name.len(){
+                    #( let #param_names = #range_array_names.get(index).unwrap().as_any().downcast_ref::<#unref_param_types>().unwrap().clone(); )*
+
+                    // TODO(ruihang): add ensure!() to check length
+
+                    let result = #fn_name(#( &#param_names, )*);
+                    result_array.push(result);
+                }
+
+                let result = ColumnarValue::Array(Arc::new(#ret_type::from_iter(result_array)));
+                Ok(result)
+            }
+        }
+    }
+    .into()
+}
--- a/src/common/function/src/scalars/timestamp.rs
+++ b/src/common/function/src/scalars/timestamp.rs
@@ -12,17 +12,10 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.

-use std::sync::Arc;
-mod from_unixtime;
-
-use from_unixtime::FromUnixtimeFunction;
-
 use crate::scalars::function_registry::FunctionRegistry;

 pub(crate) struct TimestampFunction;

 impl TimestampFunction {
-    pub fn register(registry: &FunctionRegistry) {
-        registry.register(Arc::new(FromUnixtimeFunction::default()));
-    }
+    pub fn register(_registry: &FunctionRegistry) {}
 }
--- a/src/common/function/src/scalars/timestamp/from_unixtime.rs
+++ b/src/common/function/src/scalars/timestamp/from_unixtime.rs
@@ -1,133 +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.
-
-//! from_unixtime function.
-/// TODO(dennis) It can be removed after we upgrade datafusion.
-use std::fmt;
-use std::sync::Arc;
-
-use common_query::error::{
-    ArrowComputeSnafu, IntoVectorSnafu, Result, TypeCastSnafu, UnsupportedInputDataTypeSnafu,
-};
-use common_query::prelude::{Signature, Volatility};
-use datatypes::arrow::compute;
-use datatypes::arrow::datatypes::{DataType as ArrowDatatype, Int64Type};
-use datatypes::data_type::DataType;
-use datatypes::prelude::ConcreteDataType;
-use datatypes::vectors::{TimestampMillisecondVector, VectorRef};
-use snafu::ResultExt;
-
-use crate::scalars::function::{Function, FunctionContext};
-
-#[derive(Clone, Debug, Default)]
-pub struct FromUnixtimeFunction;
-
-const NAME: &str = "from_unixtime";
-
-impl Function for FromUnixtimeFunction {
-    fn name(&self) -> &str {
-        NAME
-    }
-
-    fn return_type(&self, _input_types: &[ConcreteDataType]) -> Result<ConcreteDataType> {
-        Ok(ConcreteDataType::timestamp_millisecond_datatype())
-    }
-
-    fn signature(&self) -> Signature {
-        Signature::uniform(
-            1,
-            vec![ConcreteDataType::int64_datatype()],
-            Volatility::Immutable,
-        )
-    }
-
-    fn eval(&self, _func_ctx: FunctionContext, columns: &[VectorRef]) -> Result<VectorRef> {
-        match columns[0].data_type() {
-            ConcreteDataType::Int64(_) => {
-                let array = columns[0].to_arrow_array();
-                // Our timestamp vector's time unit is millisecond
-                let array = compute::multiply_scalar_dyn::<Int64Type>(&array, 1000i64)
-                    .context(ArrowComputeSnafu)?;
-
-                let arrow_datatype = &self.return_type(&[]).unwrap().as_arrow_type();
-                Ok(Arc::new(
-                    TimestampMillisecondVector::try_from_arrow_array(
-                        compute::cast(&array, arrow_datatype).context(TypeCastSnafu {
-                            typ: ArrowDatatype::Int64,
-                        })?,
-                    )
-                    .context(IntoVectorSnafu {
-                        data_type: arrow_datatype.clone(),
-                    })?,
-                ))
-            }
-            _ => UnsupportedInputDataTypeSnafu {
-                function: NAME,
-                datatypes: columns.iter().map(|c| c.data_type()).collect::<Vec<_>>(),
-            }
-            .fail(),
-        }
-    }
-}
-
-impl fmt::Display for FromUnixtimeFunction {
-    fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
-        write!(f, "FROM_UNIXTIME")
-    }
-}
-
-#[cfg(test)]
-mod tests {
-    use common_query::prelude::TypeSignature;
-    use datatypes::value::Value;
-    use datatypes::vectors::Int64Vector;
-
-    use super::*;
-
-    #[test]
-    fn test_from_unixtime() {
-        let f = FromUnixtimeFunction::default();
-        assert_eq!("from_unixtime", f.name());
-        assert_eq!(
-            ConcreteDataType::timestamp_millisecond_datatype(),
-            f.return_type(&[]).unwrap()
-        );
-
-        assert!(matches!(f.signature(),
-                         Signature {
-                             type_signature: TypeSignature::Uniform(1, valid_types),
-                             volatility: Volatility::Immutable
-                         } if  valid_types == vec![ConcreteDataType::int64_datatype()]
-        ));
-
-        let times = vec![Some(1494410783), None, Some(1494410983)];
-        let args: Vec<VectorRef> = vec![Arc::new(Int64Vector::from(times.clone()))];
-
-        let vector = f.eval(FunctionContext::default(), &args).unwrap();
-        assert_eq!(3, vector.len());
-        for (i, t) in times.iter().enumerate() {
-            let v = vector.get(i);
-            if i == 1 {
-                assert_eq!(Value::Null, v);
-                continue;
-            }
-            match v {
-                Value::Timestamp(ts) => {
-                    assert_eq!(ts.value(), t.unwrap() * 1000);
-                }
-                _ => unreachable!(),
-            }
-        }
-    }
-}
--- a/src/common/function/src/scalars/udf.rs
+++ b/src/common/function/src/scalars/udf.rs
@@ -14,9 +14,9 @@

 use std::sync::Arc;

-use common_query::error::{ExecuteFunctionSnafu, FromScalarValueSnafu};
+use common_query::error::FromScalarValueSnafu;
 use common_query::prelude::{
-    ColumnarValue, ReturnTypeFunction, ScalarFunctionImplementation, ScalarUdf, ScalarValue,
+    ColumnarValue, ReturnTypeFunction, ScalarFunctionImplementation, ScalarUdf,
 };
 use datatypes::error::Error as DataTypeError;
 use datatypes::prelude::*;
@@ -54,16 +54,8 @@ pub fn create_udf(func: FunctionRef) -> ScalarUdf {
            .collect();

        let result = func_cloned.eval(func_ctx, &args.context(FromScalarValueSnafu)?);
-
-        let udf = if len.is_some() {
-            result.map(ColumnarValue::Vector)?
-        } else {
-            ScalarValue::try_from_array(&result?.to_arrow_array(), 0)
-                .map(ColumnarValue::Scalar)
-                .context(ExecuteFunctionSnafu)?
-        };
-
-        Ok(udf)
+        let udf_result = result.map(ColumnarValue::Vector)?;
+        Ok(udf_result)
    });

    ScalarUdf::new(func.name(), &func.signature(), &return_type, &fun)
--- a/src/common/grpc-expr/src/alter.rs
+++ b/src/common/grpc-expr/src/alter.rs
@@ -18,11 +18,13 @@ use common_catalog::consts::{DEFAULT_CATALOG_NAME, DEFAULT_SCHEMA_NAME};
 use datatypes::schema::{ColumnSchema, RawSchema};
 use snafu::{ensure, OptionExt, ResultExt};
 use table::metadata::TableId;
-use table::requests::{AddColumnRequest, AlterKind, AlterTableRequest, CreateTableRequest};
+use table::requests::{
+    AddColumnRequest, AlterKind, AlterTableRequest, CreateTableRequest, TableOptions,
+};

 use crate::error::{
    ColumnNotFoundSnafu, InvalidColumnDefSnafu, MissingFieldSnafu, MissingTimestampColumnSnafu,
-    Result,
+    Result, UnrecognizedTableOptionSnafu,
 };

 /// Convert an [`AlterExpr`] to an [`AlterTableRequest`]
@@ -163,6 +165,8 @@ pub fn create_expr_to_request(
        expr.region_ids
    };

+    let table_options =
+        TableOptions::try_from(&expr.table_options).context(UnrecognizedTableOptionSnafu)?;
    Ok(CreateTableRequest {
        id: table_id,
        catalog_name,
@@ -173,7 +177,7 @@ pub fn create_expr_to_request(
        region_numbers: region_ids,
        primary_key_indices,
        create_if_not_exists: expr.create_if_not_exists,
-        table_options: expr.table_options,
+        table_options,
    })
 }

--- a/src/common/grpc-expr/src/error.rs
+++ b/src/common/grpc-expr/src/error.rs
@@ -84,6 +84,12 @@ pub enum Error {
        #[snafu(backtrace)]
        source: api::error::Error,
    },
+
+    #[snafu(display("Unrecognized table option: {}", source))]
+    UnrecognizedTableOption {
+        #[snafu(backtrace)]
+        source: table::error::Error,
+    },
 }

 pub type Result<T> = std::result::Result<T, Error>;
@@ -104,6 +110,7 @@ impl ErrorExt for Error {
            Error::MissingField { .. } => StatusCode::InvalidArguments,
            Error::ColumnDefaultConstraint { source, .. } => source.status_code(),
            Error::InvalidColumnDef { source, .. } => source.status_code(),
+            Error::UnrecognizedTableOption { .. } => StatusCode::InvalidArguments,
        }
    }
    fn backtrace_opt(&self) -> Option<&Backtrace> {
--- a/src/common/grpc-expr/src/insert.rs
+++ b/src/common/grpc-expr/src/insert.rs
@@ -26,6 +26,7 @@ use common_time::{Date, DateTime};
 use datatypes::data_type::{ConcreteDataType, DataType};
 use datatypes::prelude::{ValueRef, VectorRef};
 use datatypes::schema::SchemaRef;
+use datatypes::types::TimestampType;
 use datatypes::value::Value;
 use datatypes::vectors::MutableVector;
 use snafu::{ensure, OptionExt, ResultExt};
@@ -414,13 +415,29 @@ fn convert_values(data_type: &ConcreteDataType, values: Values) -> Vec<Value> {
            .into_iter()
            .map(|v| Value::Date(v.into()))
            .collect(),
-        ConcreteDataType::Timestamp(_) => values
+        ConcreteDataType::Timestamp(TimestampType::Second(_)) => values
+            .ts_second_values
+            .into_iter()
+            .map(|v| Value::Timestamp(Timestamp::new_second(v)))
+            .collect(),
+        ConcreteDataType::Timestamp(TimestampType::Millisecond(_)) => values
            .ts_millisecond_values
            .into_iter()
            .map(|v| Value::Timestamp(Timestamp::new_millisecond(v)))
            .collect(),
-        ConcreteDataType::Null(_) => unreachable!(),
-        ConcreteDataType::List(_) => unreachable!(),
+        ConcreteDataType::Timestamp(TimestampType::Microsecond(_)) => values
+            .ts_microsecond_values
+            .into_iter()
+            .map(|v| Value::Timestamp(Timestamp::new_microsecond(v)))
+            .collect(),
+        ConcreteDataType::Timestamp(TimestampType::Nanosecond(_)) => values
+            .ts_nanosecond_values
+            .into_iter()
+            .map(|v| Value::Timestamp(Timestamp::new_nanosecond(v)))
+            .collect(),
+        ConcreteDataType::Null(_) | ConcreteDataType::List(_) | ConcreteDataType::Dictionary(_) => {
+            unreachable!()
+        }
    }
 }

@@ -443,6 +460,7 @@ mod tests {
    use common_time::timestamp::Timestamp;
    use datatypes::data_type::ConcreteDataType;
    use datatypes::schema::{ColumnSchema, SchemaBuilder, SchemaRef};
+    use datatypes::types::{TimestampMillisecondType, TimestampSecondType, TimestampType};
    use datatypes::value::Value;
    use snafu::ResultExt;
    use table::error::Result as TableResult;
@@ -646,6 +664,39 @@ mod tests {
        );
    }

+    #[test]
+    fn test_convert_timestamp_values() {
+        // second
+        let actual = convert_values(
+            &ConcreteDataType::Timestamp(TimestampType::Second(TimestampSecondType)),
+            Values {
+                ts_second_values: vec![1_i64, 2_i64, 3_i64],
+                ..Default::default()
+            },
+        );
+        let expect = vec![
+            Value::Timestamp(Timestamp::new_second(1_i64)),
+            Value::Timestamp(Timestamp::new_second(2_i64)),
+            Value::Timestamp(Timestamp::new_second(3_i64)),
+        ];
+        assert_eq!(expect, actual);
+
+        // millisecond
+        let actual = convert_values(
+            &ConcreteDataType::Timestamp(TimestampType::Millisecond(TimestampMillisecondType)),
+            Values {
+                ts_millisecond_values: vec![1_i64, 2_i64, 3_i64],
+                ..Default::default()
+            },
+        );
+        let expect = vec![
+            Value::Timestamp(Timestamp::new_millisecond(1_i64)),
+            Value::Timestamp(Timestamp::new_millisecond(2_i64)),
+            Value::Timestamp(Timestamp::new_millisecond(3_i64)),
+        ];
+        assert_eq!(expect, actual);
+    }
+
    #[test]
    fn test_is_null() {
        let null_mask = BitVec::from_slice(&[0b0000_0001, 0b0000_1000]);
--- a/src/common/grpc/Cargo.toml
+++ b/src/common/grpc/Cargo.toml
@@ -16,7 +16,7 @@ common-runtime = { path = "../runtime" }
 dashmap = "5.4"
 datafusion.workspace = true
 datatypes = { path = "../../datatypes" }
-flatbuffers = "22"
+flatbuffers = "23.1"
 futures = "0.3"
 prost.workspace = true
 snafu = { version = "0.7", features = ["backtraces"] }
--- a/src/common/grpc/src/flight.rs
+++ b/src/common/grpc/src/flight.rs
@@ -16,8 +16,9 @@ use std::collections::HashMap;
 use std::sync::Arc;

 use api::v1::{AffectedRows, FlightMetadata};
-use arrow_flight::utils::{flight_data_from_arrow_batch, flight_data_to_arrow_batch};
+use arrow_flight::utils::flight_data_to_arrow_batch;
 use arrow_flight::{FlightData, IpcMessage, SchemaAsIpc};
+use common_base::bytes::Bytes;
 use common_recordbatch::{RecordBatch, RecordBatches};
 use datatypes::arrow;
 use datatypes::arrow::datatypes::Schema as ArrowSchema;
@@ -39,38 +40,58 @@ pub enum FlightMessage {
    AffectedRows(usize),
 }

-#[derive(Default)]
 pub struct FlightEncoder {
    write_options: writer::IpcWriteOptions,
+    data_gen: writer::IpcDataGenerator,
+    dictionary_tracker: writer::DictionaryTracker,
+}
+
+impl Default for FlightEncoder {
+    fn default() -> Self {
+        Self {
+            write_options: writer::IpcWriteOptions::default(),
+            data_gen: writer::IpcDataGenerator::default(),
+            dictionary_tracker: writer::DictionaryTracker::new(false),
+        }
+    }
 }

 impl FlightEncoder {
-    pub fn encode(&self, flight_message: FlightMessage) -> FlightData {
+    pub fn encode(&mut self, flight_message: FlightMessage) -> FlightData {
        match flight_message {
            FlightMessage::Schema(schema) => {
                SchemaAsIpc::new(schema.arrow_schema(), &self.write_options).into()
            }
            FlightMessage::Recordbatch(recordbatch) => {
-                let (flight_dictionaries, flight_batch) = flight_data_from_arrow_batch(
-                    recordbatch.df_record_batch(),
-                    &self.write_options,
-                );
+                let (encoded_dictionaries, encoded_batch) = self
+                    .data_gen
+                    .encoded_batch(
+                        recordbatch.df_record_batch(),
+                        &mut self.dictionary_tracker,
+                        &self.write_options,
+                    )
+                    .expect("DictionaryTracker configured above to not fail on replacement");

                // TODO(LFC): Handle dictionary as FlightData here, when we supported Arrow's Dictionary DataType.
                // Currently we don't have a datatype corresponding to Arrow's Dictionary DataType,
                // so there won't be any "dictionaries" here. Assert to be sure about it, and
                // perform a "testing guard" in case we forgot to handle the possible "dictionaries"
                // here in the future.
-                debug_assert_eq!(flight_dictionaries.len(), 0);
+                debug_assert_eq!(encoded_dictionaries.len(), 0);

-                flight_batch
+                encoded_batch.into()
            }
            FlightMessage::AffectedRows(rows) => {
                let metadata = FlightMetadata {
                    affected_rows: Some(AffectedRows { value: rows as _ }),
                }
                .encode_to_vec();
-                FlightData::new(None, IpcMessage(build_none_flight_msg()), metadata, vec![])
+                FlightData::new(
+                    None,
+                    IpcMessage(build_none_flight_msg().into()),
+                    metadata,
+                    vec![],
+                )
            }
        }
    }
@@ -83,7 +104,8 @@ pub struct FlightDecoder {

 impl FlightDecoder {
    pub fn try_decode(&mut self, flight_data: FlightData) -> Result<FlightMessage> {
-        let message = root_as_message(flight_data.data_header.as_slice()).map_err(|e| {
+        let bytes = flight_data.data_header.slice(..);
+        let message = root_as_message(&bytes).map_err(|e| {
            InvalidFlightDataSnafu {
                reason: e.to_string(),
            }
@@ -91,7 +113,7 @@ impl FlightDecoder {
        })?;
        match message.header_type() {
            MessageHeader::NONE => {
-                let metadata = FlightMetadata::decode(flight_data.app_metadata.as_slice())
+                let metadata = FlightMetadata::decode(flight_data.app_metadata)
                    .context(DecodeFlightDataSnafu)?;
                if let Some(AffectedRows { value }) = metadata.affected_rows {
                    return Ok(FlightMessage::AffectedRows(value as _));
@@ -176,7 +198,7 @@ pub fn flight_messages_to_recordbatches(messages: Vec<FlightMessage>) -> Result<
    }
 }

-fn build_none_flight_msg() -> Vec<u8> {
+fn build_none_flight_msg() -> Bytes {
    let mut builder = FlatBufferBuilder::new();

    let mut message = arrow::ipc::MessageBuilder::new(&mut builder);
@@ -187,7 +209,7 @@ fn build_none_flight_msg() -> Vec<u8> {
    let data = message.finish();
    builder.finish(data, None);

-    builder.finished_data().to_vec()
+    builder.finished_data().into()
 }

 #[cfg(test)]
--- a/src/common/grpc/src/select.rs
+++ b/src/common/grpc/src/select.rs
@@ -67,7 +67,7 @@ macro_rules! convert_arrow_array_to_grpc_vals {
                    return Ok(vals);
                },
            )+
-            ConcreteDataType::Null(_) | ConcreteDataType::List(_) => unreachable!("Should not send {:?} in gRPC", $data_type),
+            ConcreteDataType::Null(_) | ConcreteDataType::List(_) | ConcreteDataType::Dictionary(_) => unreachable!("Should not send {:?} in gRPC", $data_type),
        }
    }};
 }
--- a/src/common/mem-prof/Cargo.toml
+++ b/src/common/mem-prof/Cargo.toml
@@ -0,0 +1,17 @@
+[package]
+name = "common-mem-prof"
+version.workspace = true
+edition.workspace = true
+license.workspace = true
+
+[dependencies]
+common-error = { path = "../error" }
+snafu.workspace = true
+tempfile = "3.4"
+tikv-jemalloc-ctl = { version = "0.5", features = ["use_std"] }
+tikv-jemallocator = "0.5"
+tokio.workspace = true
+
+[dependencies.tikv-jemalloc-sys]
+version = "0.5"
+features = ["stats", "profiling", "unprefixed_malloc_on_supported_platforms"]
--- a/src/common/mem-prof/README.md
+++ b/src/common/mem-prof/README.md
@@ -0,0 +1,50 @@
+# Profile memory usage of GreptimeDB
+
+This crate provides an easy approach to dump memory profiling info.
+
+## Prerequisites
+### jemalloc
+```bash
+# for macOS
+brew install jemalloc
+
+# for Ubuntu
+sudo apt install libjemalloc-dev
+```
+
+### [flamegraph](https://github.com/brendangregg/FlameGraph) 
+
+```bash
+curl https://raw.githubusercontent.com/brendangregg/FlameGraph/master/flamegraph.pl > ./flamegraph.pl 
+```
+
+### Build GreptimeDB with `mem-prof` feature.
+
+```bash
+cargo build --features=mem-prof
+```
+
+## Profiling
+
+Start GreptimeDB instance with environment variables:
+
+```bash
+MALLOC_CONF=prof:true,lg_prof_interval:28 ./target/debug/greptime standalone start
+```
+
+Dump memory profiling data through HTTP API:
+
+```bash
+curl localhost:4000/v1/prof/mem > greptime.hprof
+```
+
+You can periodically dump profiling data and compare them to find the delta memory usage.
+
+## Analyze profiling data with flamegraph
+
+To create flamegraph according to dumped profiling data:
+
+```bash
+jeprof --svg <path_to_greptimedb_binary> --base=<baseline_prof> <profile_data> > output.svg
+```
+
--- a/src/common/mem-prof/src/error.rs
+++ b/src/common/mem-prof/src/error.rs
@@ -0,0 +1,66 @@
+// Copyright 2023 Greptime Team
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+use std::any::Any;
+use std::path::PathBuf;
+
+use common_error::prelude::{ErrorExt, StatusCode};
+use snafu::{Backtrace, Snafu};
+
+pub type Result<T> = std::result::Result<T, Error>;
+
+#[derive(Debug, Snafu)]
+#[snafu(visibility(pub))]
+pub enum Error {
+    #[snafu(display("Failed to read OPT_PROF"))]
+    ReadOptProf { source: tikv_jemalloc_ctl::Error },
+
+    #[snafu(display("Memory profiling is not enabled"))]
+    ProfilingNotEnabled,
+
+    #[snafu(display("Failed to build temp file from given path: {:?}", path))]
+    BuildTempPath { path: PathBuf, backtrace: Backtrace },
+
+    #[snafu(display("Failed to open temp file: {}", path))]
+    OpenTempFile {
+        path: String,
+        source: std::io::Error,
+    },
+
+    #[snafu(display("Failed to dump profiling data to temp file: {:?}", path))]
+    DumpProfileData {
+        path: PathBuf,
+        source: tikv_jemalloc_ctl::Error,
+    },
+}
+
+impl ErrorExt for Error {
+    fn status_code(&self) -> StatusCode {
+        match self {
+            Error::ReadOptProf { .. } => StatusCode::Internal,
+            Error::ProfilingNotEnabled => StatusCode::InvalidArguments,
+            Error::BuildTempPath { .. } => StatusCode::Internal,
+            Error::OpenTempFile { .. } => StatusCode::StorageUnavailable,
+            Error::DumpProfileData { .. } => StatusCode::StorageUnavailable,
+        }
+    }
+
+    fn backtrace_opt(&self) -> Option<&Backtrace> {
+        snafu::ErrorCompat::backtrace(self)
+    }
+
+    fn as_any(&self) -> &dyn Any {
+        self
+    }
+}
--- a/src/common/mem-prof/src/lib.rs
+++ b/src/common/mem-prof/src/lib.rs
@@ -0,0 +1,74 @@
+// Copyright 2023 Greptime Team
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+pub mod error;
+
+use std::ffi::{c_char, CString};
+use std::path::PathBuf;
+
+use snafu::{ensure, ResultExt};
+use tokio::io::AsyncReadExt;
+
+use crate::error::{
+    BuildTempPathSnafu, DumpProfileDataSnafu, OpenTempFileSnafu, ProfilingNotEnabledSnafu,
+    ReadOptProfSnafu,
+};
+
+const PROF_DUMP: &[u8] = b"prof.dump\0";
+const OPT_PROF: &[u8] = b"opt.prof\0";
+
+pub async fn dump_profile() -> error::Result<Vec<u8>> {
+    ensure!(is_prof_enabled()?, ProfilingNotEnabledSnafu);
+    let tmp_path = tempfile::tempdir().map_err(|_| {
+        BuildTempPathSnafu {
+            path: std::env::temp_dir(),
+        }
+        .build()
+    })?;
+
+    let mut path_buf = PathBuf::from(tmp_path.path());
+    path_buf.push("greptimedb.hprof");
+
+    let path = path_buf
+        .to_str()
+        .ok_or_else(|| BuildTempPathSnafu { path: &path_buf }.build())?
+        .to_string();
+
+    let mut bytes = CString::new(path.as_str())
+        .map_err(|_| BuildTempPathSnafu { path: &path_buf }.build())?
+        .into_bytes_with_nul();
+
+    {
+        // #safety: we always expect a valid temp file path to write profiling data to.
+        let ptr = bytes.as_mut_ptr() as *mut c_char;
+        unsafe {
+            tikv_jemalloc_ctl::raw::write(PROF_DUMP, ptr)
+                .context(DumpProfileDataSnafu { path: path_buf })?
+        }
+    }
+
+    let mut f = tokio::fs::File::open(path.as_str())
+        .await
+        .context(OpenTempFileSnafu { path: &path })?;
+    let mut buf = vec![];
+    f.read_to_end(&mut buf)
+        .await
+        .context(OpenTempFileSnafu { path })?;
+    Ok(buf)
+}
+
+fn is_prof_enabled() -> error::Result<bool> {
+    // safety: OPT_PROF variable, if present, is always a boolean value.
+    Ok(unsafe { tikv_jemalloc_ctl::raw::read::<bool>(OPT_PROF).context(ReadOptProfSnafu)? })
+}
--- a/src/common/procedure/Cargo.toml
+++ b/src/common/procedure/Cargo.toml
@@ -14,10 +14,11 @@ object-store = { path = "../../object-store" }
 serde.workspace = true
 serde_json = "1.0"
 smallvec = "1"
+backon = "0.4.0"
 snafu.workspace = true
 tokio.workspace = true
 uuid.workspace = true

 [dev-dependencies]
+common-test-util = { path = "../test-util" }
 futures-util.workspace = true
-tempdir = "0.3"
--- a/src/common/procedure/src/error.rs
+++ b/src/common/procedure/src/error.rs
@@ -13,6 +13,7 @@
 // limitations under the License.

 use std::any::Any;
+use std::sync::Arc;

 use common_error::prelude::*;

@@ -69,6 +70,43 @@ pub enum Error {
        key: String,
        source: object_store::Error,
    },
+
+    #[snafu(display("Failed to deserialize from json, source: {}", source))]
+    FromJson {
+        source: serde_json::Error,
+        backtrace: Backtrace,
+    },
+
+    #[snafu(display("Procedure exec failed, source: {}", source))]
+    RetryLater {
+        #[snafu(backtrace)]
+        source: BoxedError,
+    },
+
+    #[snafu(display("Procedure panics, procedure_id: {}", procedure_id))]
+    ProcedurePanic { procedure_id: ProcedureId },
+
+    #[snafu(display("Failed to wait watcher, source: {}", source))]
+    WaitWatcher {
+        source: tokio::sync::watch::error::RecvError,
+        backtrace: Backtrace,
+    },
+
+    #[snafu(display("Failed to execute procedure, source: {}", source))]
+    ProcedureExec {
+        source: Arc<Error>,
+        backtrace: Backtrace,
+    },
+
+    #[snafu(display(
+        "Procedure retry exceeded max times, procedure_id: {}, source:{}",
+        procedure_id,
+        source
+    ))]
+    RetryTimesExceeded {
+        source: Arc<Error>,
+        procedure_id: ProcedureId,
+    },
 }

 pub type Result<T> = std::result::Result<T, Error>;
@@ -81,10 +119,16 @@ impl ErrorExt for Error {
            | Error::PutState { .. }
            | Error::DeleteState { .. }
            | Error::ListState { .. }
-            | Error::ReadState { .. } => StatusCode::Internal,
+            | Error::ReadState { .. }
+            | Error::FromJson { .. }
+            | Error::RetryTimesExceeded { .. }
+            | Error::RetryLater { .. }
+            | Error::WaitWatcher { .. } => StatusCode::Internal,
            Error::LoaderConflict { .. } | Error::DuplicateProcedure { .. } => {
                StatusCode::InvalidArguments
            }
+            Error::ProcedurePanic { .. } => StatusCode::Unexpected,
+            Error::ProcedureExec { source, .. } => source.status_code(),
        }
    }

@@ -104,4 +148,26 @@ impl Error {
            source: BoxedError::new(err),
        }
    }
+
+    /// Creates a new [Error::RetryLater] error from source `err`.
+    pub fn retry_later<E: ErrorExt + Send + Sync + 'static>(err: E) -> Error {
+        Error::RetryLater {
+            source: BoxedError::new(err),
+        }
+    }
+
+    /// Determine whether it is a retry later type through [StatusCode]
+    pub fn is_retry_later(&self) -> bool {
+        matches!(self, Error::RetryLater { .. })
+    }
+
+    /// Creates a new [Error::RetryLater] or [Error::External] error from source `err` according
+    /// to its [StatusCode].
+    pub fn from_error_ext<E: ErrorExt + Send + Sync + 'static>(err: E) -> Self {
+        if err.status_code().is_retryable() {
+            Error::retry_later(err)
+        } else {
+            Error::external(err)
+        }
+    }
 }
--- a/src/common/procedure/src/lib.rs
+++ b/src/common/procedure/src/lib.rs
@@ -15,15 +15,14 @@
 //! Common traits and structures for the procedure framework.

 pub mod error;
-#[allow(dead_code)]
-mod local;
+pub mod local;
 mod procedure;
-// TODO(yingwen): Remove this attribute once ProcedureManager is implemented.
-#[allow(dead_code)]
 mod store;
+pub mod watcher;

 pub use crate::error::{Error, Result};
 pub use crate::procedure::{
    BoxedProcedure, Context, ContextProvider, LockKey, Procedure, ProcedureId, ProcedureManager,
    ProcedureManagerRef, ProcedureState, ProcedureWithId, Status,
 };
+pub use crate::watcher::Watcher;
--- a/src/common/procedure/src/local.rs
+++ b/src/common/procedure/src/local.rs
@@ -17,11 +17,14 @@ mod runner;

 use std::collections::{HashMap, VecDeque};
 use std::sync::{Arc, Mutex, RwLock};
+use std::time::Duration;

 use async_trait::async_trait;
+use backon::ExponentialBuilder;
 use common_telemetry::logging;
 use object_store::ObjectStore;
 use snafu::ensure;
+use tokio::sync::watch::{self, Receiver, Sender};
 use tokio::sync::Notify;

 use crate::error::{DuplicateProcedureSnafu, LoaderConflictSnafu, Result};
@@ -31,27 +34,9 @@ use crate::procedure::BoxedProcedureLoader;
 use crate::store::{ObjectStateStore, ProcedureMessage, ProcedureStore, StateStoreRef};
 use crate::{
    BoxedProcedure, ContextProvider, LockKey, ProcedureId, ProcedureManager, ProcedureState,
-    ProcedureWithId,
+    ProcedureWithId, Watcher,
 };

-/// Mutable metadata of a procedure during execution.
-#[derive(Debug)]
-struct ExecMeta {
-    /// Current procedure state.
-    state: ProcedureState,
-    /// Id of child procedures.
-    children: Vec<ProcedureId>,
-}
-
-impl Default for ExecMeta {
-    fn default() -> ExecMeta {
-        ExecMeta {
-            state: ProcedureState::Running,
-            children: Vec::new(),
-        }
-    }
-}
-
 /// Shared metadata of a procedure.
 ///
 /// # Note
@@ -72,38 +57,55 @@ pub(crate) struct ProcedureMeta {
    child_notify: Notify,
    /// Lock required by this procedure.
    lock_key: LockKey,
-    /// Mutable status during execution.
-    exec_meta: Mutex<ExecMeta>,
+    /// Sender to notify the procedure state.
+    state_sender: Sender<ProcedureState>,
+    /// Receiver to watch the procedure state.
+    state_receiver: Receiver<ProcedureState>,
+    /// Id of child procedures.
+    children: Mutex<Vec<ProcedureId>>,
 }

 impl ProcedureMeta {
+    fn new(id: ProcedureId, parent_id: Option<ProcedureId>, lock_key: LockKey) -> ProcedureMeta {
+        let (state_sender, state_receiver) = watch::channel(ProcedureState::Running);
+        ProcedureMeta {
+            id,
+            lock_notify: Notify::new(),
+            parent_id,
+            child_notify: Notify::new(),
+            lock_key,
+            state_sender,
+            state_receiver,
+            children: Mutex::new(Vec::new()),
+        }
+    }
+
    /// Returns current [ProcedureState].
    fn state(&self) -> ProcedureState {
-        let meta = self.exec_meta.lock().unwrap();
-        meta.state.clone()
+        self.state_receiver.borrow().clone()
    }

    /// Update current [ProcedureState].
    fn set_state(&self, state: ProcedureState) {
-        let mut meta = self.exec_meta.lock().unwrap();
-        meta.state = state;
+        // Safety: ProcedureMeta also holds the receiver, so `send()` should never fail.
+        self.state_sender.send(state).unwrap();
    }

    /// Push `procedure_id` of the subprocedure to the metadata.
    fn push_child(&self, procedure_id: ProcedureId) {
-        let mut meta = self.exec_meta.lock().unwrap();
-        meta.children.push(procedure_id);
+        let mut children = self.children.lock().unwrap();
+        children.push(procedure_id);
    }

    /// Append subprocedures to given `buffer`.
    fn list_children(&self, buffer: &mut Vec<ProcedureId>) {
-        let meta = self.exec_meta.lock().unwrap();
-        buffer.extend_from_slice(&meta.children);
+        let children = self.children.lock().unwrap();
+        buffer.extend_from_slice(&children);
    }

    /// Returns the number of subprocedures.
    fn num_children(&self) -> usize {
-        self.exec_meta.lock().unwrap().children.len()
+        self.children.lock().unwrap().len()
    }
 }

@@ -123,8 +125,6 @@ pub(crate) struct ManagerContext {
    loaders: Mutex<HashMap<String, BoxedProcedureLoader>>,
    lock_map: LockMap,
    procedures: RwLock<HashMap<ProcedureId, ProcedureMetaRef>>,
-    // TODO(yingwen): Now we never clean the messages. But when the root procedure is done, we
-    // should be able to remove the its message and all its child messages.
    /// Messages loaded from the procedure store.
    messages: Mutex<HashMap<ProcedureId, ProcedureMessage>>,
 }
@@ -175,6 +175,14 @@ impl ManagerContext {
        procedures.get(&procedure_id).map(|meta| meta.state())
    }

+    /// Returns the [Watcher] of specific `procedure_id`.
+    fn watcher(&self, procedure_id: ProcedureId) -> Option<Watcher> {
+        let procedures = self.procedures.read().unwrap();
+        procedures
+            .get(&procedure_id)
+            .map(|meta| meta.state_receiver.clone())
+    }
+
    /// Notify a suspended parent procedure with specific `procedure_id` by its subprocedure.
    fn notify_by_subprocedure(&self, procedure_id: ProcedureId) {
        let procedures = self.procedures.read().unwrap();
@@ -284,13 +292,17 @@ impl ManagerContext {
 #[derive(Debug)]
 pub struct ManagerConfig {
    /// Object store
-    object_store: ObjectStore,
+    pub object_store: ObjectStore,
+    pub max_retry_times: usize,
+    pub retry_delay: Duration,
 }

 /// A [ProcedureManager] that maintains procedure states locally.
 pub struct LocalManager {
    manager_ctx: Arc<ManagerContext>,
    state_store: StateStoreRef,
+    max_retry_times: usize,
+    retry_delay: Duration,
 }

 impl LocalManager {
@@ -299,6 +311,8 @@ impl LocalManager {
        LocalManager {
            manager_ctx: Arc::new(ManagerContext::new()),
            state_store: Arc::new(ObjectStateStore::new(config.object_store)),
+            max_retry_times: config.max_retry_times,
+            retry_delay: config.retry_delay,
        }
    }

@@ -308,23 +322,22 @@ impl LocalManager {
        procedure_id: ProcedureId,
        step: u32,
        procedure: BoxedProcedure,
-    ) -> Result<()> {
-        let meta = Arc::new(ProcedureMeta {
-            id: procedure_id,
-            lock_notify: Notify::new(),
-            parent_id: None,
-            child_notify: Notify::new(),
-            lock_key: procedure.lock_key(),
-            exec_meta: Mutex::new(ExecMeta::default()),
-        });
+    ) -> Result<Watcher> {
+        let meta = Arc::new(ProcedureMeta::new(procedure_id, None, procedure.lock_key()));
        let runner = Runner {
            meta: meta.clone(),
            procedure,
            manager_ctx: self.manager_ctx.clone(),
            step,
+            exponential_builder: ExponentialBuilder::default()
+                .with_min_delay(self.retry_delay)
+                .with_max_times(self.max_retry_times),
            store: ProcedureStore::new(self.state_store.clone()),
+            rolling_back: false,
        };

+        let watcher = meta.state_receiver.clone();
+
        // Inserts meta into the manager before actually spawnd the runner.
        ensure!(
            self.manager_ctx.try_insert_procedure(meta),
@@ -333,10 +346,10 @@ impl LocalManager {

        common_runtime::spawn_bg(async move {
            // Run the root procedure.
-            let _ = runner.run().await;
+            runner.run().await;
        });

-        Ok(())
+        Ok(watcher)
    }
 }

@@ -351,16 +364,14 @@ impl ProcedureManager for LocalManager {
        Ok(())
    }

-    async fn submit(&self, procedure: ProcedureWithId) -> Result<()> {
+    async fn submit(&self, procedure: ProcedureWithId) -> Result<Watcher> {
        let procedure_id = procedure.id;
        ensure!(
            !self.manager_ctx.contains_procedure(procedure_id),
            DuplicateProcedureSnafu { procedure_id }
        );

-        self.submit_root(procedure.id, 0, procedure.procedure)?;
-
-        Ok(())
+        self.submit_root(procedure.id, 0, procedure.procedure)
    }

    async fn recover(&self) -> Result<()> {
@@ -401,37 +412,37 @@ impl ProcedureManager for LocalManager {
    async fn procedure_state(&self, procedure_id: ProcedureId) -> Result<Option<ProcedureState>> {
        Ok(self.manager_ctx.state(procedure_id))
    }
+
+    fn procedure_watcher(&self, procedure_id: ProcedureId) -> Option<Watcher> {
+        self.manager_ctx.watcher(procedure_id)
+    }
 }

 /// Create a new [ProcedureMeta] for test purpose.
 #[cfg(test)]
 mod test_util {
-    use object_store::services::fs::Builder;
-    use tempdir::TempDir;
+    use common_test_util::temp_dir::TempDir;
+    use object_store::services::Fs as Builder;
+    use object_store::ObjectStoreBuilder;

    use super::*;

    pub(crate) fn procedure_meta_for_test() -> ProcedureMeta {
-        ProcedureMeta {
-            id: ProcedureId::random(),
-            lock_notify: Notify::new(),
-            parent_id: None,
-            child_notify: Notify::new(),
-            lock_key: LockKey::default(),
-            exec_meta: Mutex::new(ExecMeta::default()),
-        }
+        ProcedureMeta::new(ProcedureId::random(), None, LockKey::default())
    }

    pub(crate) fn new_object_store(dir: &TempDir) -> ObjectStore {
        let store_dir = dir.path().to_str().unwrap();
        let accessor = Builder::default().root(store_dir).build().unwrap();
-        ObjectStore::new(accessor)
+        ObjectStore::new(accessor).finish()
    }
 }

 #[cfg(test)]
 mod tests {
-    use tempdir::TempDir;
+    use common_error::mock::MockError;
+    use common_error::prelude::StatusCode;
+    use common_test_util::temp_dir::create_temp_dir;

    use super::*;
    use crate::error::Error;
@@ -448,9 +459,9 @@ mod tests {
        assert!(ctx.try_insert_procedure(meta.clone()));
        assert!(ctx.contains_procedure(meta.id));

-        assert_eq!(ProcedureState::Running, ctx.state(meta.id).unwrap());
+        assert!(ctx.state(meta.id).unwrap().is_running());
        meta.set_state(ProcedureState::Done);
-        assert_eq!(ProcedureState::Done, ctx.state(meta.id).unwrap());
+        assert!(ctx.state(meta.id).unwrap().is_done());
    }

    #[test]
@@ -500,6 +511,7 @@ mod tests {
    #[derive(Debug)]
    struct ProcedureToLoad {
        content: String,
+        lock_key: LockKey,
    }

    #[async_trait]
@@ -517,7 +529,7 @@ mod tests {
        }

        fn lock_key(&self) -> LockKey {
-            LockKey::default()
+            self.lock_key.clone()
        }
    }

@@ -525,6 +537,7 @@ mod tests {
        fn new(content: &str) -> ProcedureToLoad {
            ProcedureToLoad {
                content: content.to_string(),
+                lock_key: LockKey::default(),
            }
        }

@@ -539,9 +552,11 @@ mod tests {

    #[test]
    fn test_register_loader() {
-        let dir = TempDir::new("register").unwrap();
+        let dir = create_temp_dir("register");
        let config = ManagerConfig {
            object_store: test_util::new_object_store(&dir),
+            max_retry_times: 3,
+            retry_delay: Duration::from_millis(500),
        };
        let manager = LocalManager::new(config);

@@ -557,10 +572,12 @@ mod tests {

    #[tokio::test]
    async fn test_recover() {
-        let dir = TempDir::new("recover").unwrap();
+        let dir = create_temp_dir("recover");
        let object_store = test_util::new_object_store(&dir);
        let config = ManagerConfig {
            object_store: object_store.clone(),
+            max_retry_times: 3,
+            retry_delay: Duration::from_millis(500),
        };
        let manager = LocalManager::new(config);

@@ -602,45 +619,28 @@ mod tests {

    #[tokio::test]
    async fn test_submit_procedure() {
-        let dir = TempDir::new("submit").unwrap();
+        let dir = create_temp_dir("submit");
        let config = ManagerConfig {
            object_store: test_util::new_object_store(&dir),
+            max_retry_times: 3,
+            retry_delay: Duration::from_millis(500),
        };
        let manager = LocalManager::new(config);

-        #[derive(Debug)]
-        struct MockProcedure {}
-
-        #[async_trait]
-        impl Procedure for MockProcedure {
-            fn type_name(&self) -> &str {
-                "MockProcedure"
-            }
-
-            async fn execute(&mut self, _ctx: &Context) -> Result<Status> {
-                unimplemented!()
-            }
-
-            fn dump(&self) -> Result<String> {
-                unimplemented!()
-            }
-
-            fn lock_key(&self) -> LockKey {
-                LockKey::single("test.submit")
-            }
-        }
-
        let procedure_id = ProcedureId::random();
        assert!(manager
            .procedure_state(procedure_id)
            .await
            .unwrap()
            .is_none());
+        assert!(manager.procedure_watcher(procedure_id).is_none());

+        let mut procedure = ProcedureToLoad::new("submit");
+        procedure.lock_key = LockKey::single("test.submit");
        manager
            .submit(ProcedureWithId {
                id: procedure_id,
-                procedure: Box::new(MockProcedure {}),
+                procedure: Box::new(procedure),
            })
            .await
            .unwrap();
@@ -649,15 +649,79 @@ mod tests {
            .await
            .unwrap()
            .is_some());
+        // Wait for the procedure done.
+        let mut watcher = manager.procedure_watcher(procedure_id).unwrap();
+        watcher.changed().await.unwrap();
+        assert!(watcher.borrow().is_done());

        // Try to submit procedure with same id again.
        let err = manager
            .submit(ProcedureWithId {
                id: procedure_id,
-                procedure: Box::new(MockProcedure {}),
+                procedure: Box::new(ProcedureToLoad::new("submit")),
            })
            .await
            .unwrap_err();
        assert!(matches!(err, Error::DuplicateProcedure { .. }), "{err}");
    }
+
+    #[tokio::test]
+    async fn test_state_changed_on_err() {
+        let dir = create_temp_dir("on_err");
+        let config = ManagerConfig {
+            object_store: test_util::new_object_store(&dir),
+            max_retry_times: 3,
+            retry_delay: Duration::from_millis(500),
+        };
+        let manager = LocalManager::new(config);
+
+        #[derive(Debug)]
+        struct MockProcedure {
+            panic: bool,
+        }
+
+        #[async_trait]
+        impl Procedure for MockProcedure {
+            fn type_name(&self) -> &str {
+                "MockProcedure"
+            }
+
+            async fn execute(&mut self, _ctx: &Context) -> Result<Status> {
+                if self.panic {
+                    // Test the runner can set the state to failed even the procedure
+                    // panics.
+                    panic!();
+                } else {
+                    Err(Error::external(MockError::new(StatusCode::Unexpected)))
+                }
+            }
+
+            fn dump(&self) -> Result<String> {
+                Ok(String::new())
+            }
+
+            fn lock_key(&self) -> LockKey {
+                LockKey::single("test.submit")
+            }
+        }
+
+        let check_procedure = |procedure| {
+            async {
+                let procedure_id = ProcedureId::random();
+                let mut watcher = manager
+                    .submit(ProcedureWithId {
+                        id: procedure_id,
+                        procedure: Box::new(procedure),
+                    })
+                    .await
+                    .unwrap();
+                // Wait for the notification.
+                watcher.changed().await.unwrap();
+                assert!(watcher.borrow().is_failed());
+            }
+        };
+
+        check_procedure(MockProcedure { panic: false }).await;
+        check_procedure(MockProcedure { panic: true }).await;
+    }
 }
--- a/src/common/procedure/src/local/runner.rs
+++ b/src/common/procedure/src/local/runner.rs
@@ -12,26 +12,25 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.

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

+use backon::{BackoffBuilder, ExponentialBuilder};
 use common_telemetry::logging;
-use tokio::sync::Notify;
 use tokio::time;

-use crate::error::{Error, Result};
-use crate::local::{ExecMeta, ManagerContext, ProcedureMeta, ProcedureMetaRef};
+use crate::error::{ProcedurePanicSnafu, Result};
+use crate::local::{ManagerContext, ProcedureMeta, ProcedureMetaRef};
 use crate::store::ProcedureStore;
-use crate::{BoxedProcedure, Context, ProcedureId, ProcedureState, ProcedureWithId, Status};
-
-const ERR_WAIT_DURATION: Duration = Duration::from_secs(30);
+use crate::ProcedureState::Retrying;
+use crate::{BoxedProcedure, Context, Error, ProcedureId, ProcedureState, ProcedureWithId, Status};

 #[derive(Debug)]
 enum ExecResult {
    Continue,
    Done,
    RetryLater,
-    Failed(Error),
+    Failed,
 }

 #[cfg(test)]
@@ -49,7 +48,57 @@ impl ExecResult {
    }

    fn is_failed(&self) -> bool {
-        matches!(self, ExecResult::Failed(_))
+        matches!(self, ExecResult::Failed)
+    }
+}
+
+/// A guard to cleanup procedure state.
+struct ProcedureGuard {
+    meta: ProcedureMetaRef,
+    manager_ctx: Arc<ManagerContext>,
+    finish: bool,
+}
+
+impl ProcedureGuard {
+    /// Returns a new [ProcedureGuard].
+    fn new(meta: ProcedureMetaRef, manager_ctx: Arc<ManagerContext>) -> ProcedureGuard {
+        ProcedureGuard {
+            meta,
+            manager_ctx,
+            finish: false,
+        }
+    }
+
+    /// The procedure is finished successfully.
+    fn finish(mut self) {
+        self.finish = true;
+    }
+}
+
+impl Drop for ProcedureGuard {
+    fn drop(&mut self) {
+        if !self.finish {
+            logging::error!("Procedure {} exits unexpectedly", self.meta.id);
+
+            // Set state to failed. This is useful in test as runtime may not abort when the runner task panics.
+            // See https://github.com/tokio-rs/tokio/issues/2002 .
+            // We set set_panic_hook() in the application's main function. But our tests don't have this panic hook.
+            let err = ProcedurePanicSnafu {
+                procedure_id: self.meta.id,
+            }
+            .build();
+            self.meta.set_state(ProcedureState::failed(Arc::new(err)));
+        }
+
+        // Notify parent procedure.
+        if let Some(parent_id) = self.meta.parent_id {
+            self.manager_ctx.notify_by_subprocedure(parent_id);
+        }
+
+        // Release lock in reverse order.
+        for key in self.meta.lock_key.keys_to_unlock() {
+            self.manager_ctx.lock_map.release_lock(key, self.meta.id);
+        }
    }
 }

@@ -59,12 +108,17 @@ pub(crate) struct Runner {
    pub(crate) procedure: BoxedProcedure,
    pub(crate) manager_ctx: Arc<ManagerContext>,
    pub(crate) step: u32,
+    pub(crate) exponential_builder: ExponentialBuilder,
    pub(crate) store: ProcedureStore,
+    pub(crate) rolling_back: bool,
 }

 impl Runner {
    /// Run the procedure.
-    pub(crate) async fn run(mut self) -> Result<()> {
+    pub(crate) async fn run(mut self) {
+        // Ensure we can update the procedure state.
+        let guard = ProcedureGuard::new(self.meta.clone(), self.manager_ctx.clone());
+
        logging::info!(
            "Runner {}-{} starts",
            self.procedure.type_name(),
@@ -81,22 +135,17 @@ impl Runner {
                .await;
        }

-        let mut result = Ok(());
        // Execute the procedure. We need to release the lock whenever the the execution
        // is successful or fail.
-        if let Err(e) = self.execute_procedure_in_loop().await {
-            result = Err(e);
-        }
+        self.execute_procedure_in_loop().await;

-        // Notify parent procedure.
-        if let Some(parent_id) = self.meta.parent_id {
-            self.manager_ctx.notify_by_subprocedure(parent_id);
-        }
+        // We can't remove the metadata of the procedure now as users and its parent might
+        // need to query its state.
+        // TODO(yingwen): 1. Add TTL to the metadata; 2. Only keep state in the procedure store
+        // so we don't need to always store the metadata in memory after the procedure is done.

-        // Release lock in reverse order.
-        for key in self.meta.lock_key.keys_to_unlock() {
-            self.manager_ctx.lock_map.release_lock(key, self.meta.id);
-        }
+        // Release locks and notify parent procedure.
+        guard.finish();

        // If this is the root procedure, clean up message cache.
        if self.meta.parent_id.is_none() {
@@ -104,39 +153,69 @@ impl Runner {
            self.manager_ctx.remove_messages(&procedure_ids);
        }

-        // We can't remove the metadata of the procedure now as users and its parent might
-        // need to query its state.
-        // TODO(yingwen): 1. Add TTL to the metadata; 2. Only keep state in the procedure store
-        // so we don't need to always store the metadata in memory after the procedure is done.
-
        logging::info!(
            "Runner {}-{} exits",
            self.procedure.type_name(),
            self.meta.id
        );
-
-        result
    }

-    async fn execute_procedure_in_loop(&mut self) -> Result<()> {
+    async fn execute_procedure_in_loop(&mut self) {
        let ctx = Context {
            procedure_id: self.meta.id,
            provider: self.manager_ctx.clone(),
        };

+        self.rolling_back = false;
+        self.execute_once_with_retry(&ctx).await;
+    }
+
+    async fn execute_once_with_retry(&mut self, ctx: &Context) {
+        let mut retry = self.exponential_builder.build();
+        let mut retry_times = 0;
        loop {
-            match self.execute_once(&ctx).await {
+            match self.execute_once(ctx).await {
+                ExecResult::Done | ExecResult::Failed => return,
                ExecResult::Continue => (),
-                ExecResult::Done => return Ok(()),
                ExecResult::RetryLater => {
-                    self.wait_on_err().await;
+                    retry_times += 1;
+                    if let Some(d) = retry.next() {
+                        self.wait_on_err(d, retry_times).await;
+                    } else {
+                        assert!(self.meta.state().is_retrying());
+                        if let Retrying { error } = self.meta.state() {
+                            self.meta.set_state(ProcedureState::failed(Arc::new(
+                                Error::RetryTimesExceeded {
+                                    source: error,
+                                    procedure_id: self.meta.id,
+                                },
+                            )))
+                        }
+                        return;
+                    }
                }
-                ExecResult::Failed(e) => return Err(e),
            }
        }
    }

+    async fn rollback(&mut self, error: Arc<Error>) -> ExecResult {
+        if let Err(e) = self.rollback_procedure().await {
+            self.rolling_back = true;
+            self.meta.set_state(ProcedureState::retrying(Arc::new(e)));
+            return ExecResult::RetryLater;
+        }
+        self.meta.set_state(ProcedureState::failed(error));
+        ExecResult::Failed
+    }
+
    async fn execute_once(&mut self, ctx: &Context) -> ExecResult {
+        // if rolling_back, there is no need to execute again.
+        if self.rolling_back {
+            // We can definitely get the previous error here.
+            let state = self.meta.state();
+            let err = state.error().unwrap();
+            return self.rollback(err.clone()).await;
+        }
        match self.procedure.execute(ctx).await {
            Ok(status) => {
                logging::debug!(
@@ -147,8 +226,11 @@ impl Runner {
                    status.need_persist(),
                );

-                if status.need_persist() && self.persist_procedure().await.is_err() {
-                    return ExecResult::RetryLater;
+                if status.need_persist() {
+                    if let Err(err) = self.persist_procedure().await {
+                        self.meta.set_state(ProcedureState::retrying(Arc::new(err)));
+                        return ExecResult::RetryLater;
+                    }
                }

                match status {
@@ -157,7 +239,8 @@ impl Runner {
                        self.on_suspended(subprocedures).await;
                    }
                    Status::Done => {
-                        if self.commit_procedure().await.is_err() {
+                        if let Err(e) = self.commit_procedure().await {
+                            self.meta.set_state(ProcedureState::retrying(Arc::new(e)));
                            return ExecResult::RetryLater;
                        }

@@ -171,19 +254,19 @@ impl Runner {
            Err(e) => {
                logging::error!(
                    e;
-                    "Failed to execute procedure {}-{}",
+                    "Failed to execute procedure {}-{}, retry: {}",
                    self.procedure.type_name(),
-                    self.meta.id
+                    self.meta.id,
+                    e.is_retry_later(),
                );

-                self.meta.set_state(ProcedureState::Failed);
-
-                // Write rollback key so we can skip this procedure while recovering procedures.
-                if self.rollback_procedure().await.is_err() {
+                if e.is_retry_later() {
+                    self.meta.set_state(ProcedureState::retrying(Arc::new(e)));
                    return ExecResult::RetryLater;
                }

-                ExecResult::Failed(e)
+                // Write rollback key so we can skip this procedure while recovering procedures.
+                self.rollback(Arc::new(e)).await
            }
        }
    }
@@ -207,20 +290,19 @@ impl Runner {
            step = loaded_procedure.step;
        }

-        let meta = Arc::new(ProcedureMeta {
-            id: procedure_id,
-            lock_notify: Notify::new(),
-            parent_id: Some(self.meta.id),
-            child_notify: Notify::new(),
-            lock_key: procedure.lock_key(),
-            exec_meta: Mutex::new(ExecMeta::default()),
-        });
+        let meta = Arc::new(ProcedureMeta::new(
+            procedure_id,
+            Some(self.meta.id),
+            procedure.lock_key(),
+        ));
        let runner = Runner {
            meta: meta.clone(),
            procedure,
            manager_ctx: self.manager_ctx.clone(),
            step,
+            exponential_builder: self.exponential_builder.clone(),
            store: self.store.clone(),
+            rolling_back: false,
        };

        // Insert the procedure. We already check the procedure existence before inserting
@@ -244,8 +326,16 @@ impl Runner {
        });
    }

-    async fn wait_on_err(&self) {
-        time::sleep(ERR_WAIT_DURATION).await;
+    /// Extend the retry time to wait for the next retry.
+    async fn wait_on_err(&self, d: Duration, i: u64) {
+        logging::info!(
+            "Procedure {}-{} retry for the {} times after {} millis",
+            self.procedure.type_name(),
+            self.meta.id,
+            i,
+            d.as_millis(),
+        );
+        time::sleep(d).await;
    }

    async fn on_suspended(&self, subprocedures: Vec<ProcedureWithId>) {
@@ -256,7 +346,7 @@ impl Runner {
                self.procedure.type_name(),
                self.meta.id,
                subprocedure.procedure.type_name(),
-                subprocedure.id
+                subprocedure.id,
            );

            self.submit_subprocedure(subprocedure.id, subprocedure.procedure);
@@ -338,7 +428,7 @@ impl Runner {
        logging::info!(
            "Procedure {}-{} done",
            self.procedure.type_name(),
-            self.meta.id
+            self.meta.id,
        );

        // Mark the state of this procedure to done.
@@ -354,14 +444,14 @@ mod tests {
    use common_error::ext::PlainError;
    use common_error::mock::MockError;
    use common_error::prelude::StatusCode;
+    use common_test_util::temp_dir::create_temp_dir;
    use futures_util::future::BoxFuture;
    use futures_util::{FutureExt, TryStreamExt};
    use object_store::ObjectStore;
-    use tempdir::TempDir;

    use super::*;
    use crate::local::test_util;
-    use crate::{ContextProvider, LockKey, Procedure};
+    use crate::{ContextProvider, Error, LockKey, Procedure};

    const ROOT_ID: &str = "9f805a1f-05f7-490c-9f91-bd56e3cc54c1";

@@ -375,7 +465,9 @@ mod tests {
            procedure,
            manager_ctx: Arc::new(ManagerContext::new()),
            step: 0,
+            exponential_builder: ExponentialBuilder::default(),
            store,
+            rolling_back: false,
        }
    }

@@ -470,7 +562,7 @@ mod tests {
            exec_fn,
        };

-        let dir = TempDir::new("normal").unwrap();
+        let dir = create_temp_dir("normal");
        let meta = normal.new_meta(ROOT_ID);
        let ctx = context_without_provider(meta.id);
        let object_store = test_util::new_object_store(&dir);
@@ -518,7 +610,7 @@ mod tests {
            exec_fn,
        };

-        let dir = TempDir::new("suspend").unwrap();
+        let dir = create_temp_dir("suspend");
        let meta = suspend.new_meta(ROOT_ID);
        let ctx = context_without_provider(meta.id);
        let object_store = test_util::new_object_store(&dir);
@@ -587,9 +679,14 @@ mod tests {
                    // Wait for subprocedures.
                    let mut all_child_done = true;
                    for id in children_ids {
-                        if ctx.provider.procedure_state(id).await.unwrap()
-                            != Some(ProcedureState::Done)
-                        {
+                        let is_not_done = ctx
+                            .provider
+                            .procedure_state(id)
+                            .await
+                            .unwrap()
+                            .map(|s| !s.is_done())
+                            .unwrap_or(true);
+                        if is_not_done {
                            all_child_done = false;
                        }
                    }
@@ -612,7 +709,7 @@ mod tests {
            exec_fn,
        };

-        let dir = TempDir::new("parent").unwrap();
+        let dir = create_temp_dir("parent");
        let meta = parent.new_meta(ROOT_ID);
        let procedure_id = meta.id;

@@ -625,7 +722,7 @@ mod tests {
        // Replace the manager ctx.
        runner.manager_ctx = manager_ctx;

-        runner.run().await.unwrap();
+        runner.run().await;

        // Check files on store.
        for child_id in children_ids {
@@ -654,7 +751,7 @@ mod tests {
            exec_fn,
        };

-        let dir = TempDir::new("fail").unwrap();
+        let dir = create_temp_dir("fail");
        let meta = fail.new_meta(ROOT_ID);
        let ctx = context_without_provider(meta.id);
        let object_store = test_util::new_object_store(&dir);
@@ -663,10 +760,79 @@ mod tests {

        let res = runner.execute_once(&ctx).await;
        assert!(res.is_failed(), "{res:?}");
-        assert_eq!(ProcedureState::Failed, meta.state());
+        assert!(meta.state().is_failed());
        check_files(&object_store, ctx.procedure_id, &["0000000000.rollback"]).await;
    }

+    #[tokio::test]
+    async fn test_execute_on_retry_later_error() {
+        let mut times = 0;
+
+        let exec_fn = move |_| {
+            times += 1;
+            async move {
+                if times == 1 {
+                    Err(Error::retry_later(MockError::new(StatusCode::Unexpected)))
+                } else {
+                    Ok(Status::Done)
+                }
+            }
+            .boxed()
+        };
+
+        let retry_later = ProcedureAdapter {
+            data: "retry_later".to_string(),
+            lock_key: LockKey::single("catalog.schema.table"),
+            exec_fn,
+        };
+
+        let dir = create_temp_dir("retry_later");
+        let meta = retry_later.new_meta(ROOT_ID);
+        let ctx = context_without_provider(meta.id);
+        let object_store = test_util::new_object_store(&dir);
+        let procedure_store = ProcedureStore::from(object_store.clone());
+        let mut runner = new_runner(meta.clone(), Box::new(retry_later), procedure_store);
+
+        let res = runner.execute_once(&ctx).await;
+        assert!(res.is_retry_later(), "{res:?}");
+        assert!(meta.state().is_retrying());
+
+        let res = runner.execute_once(&ctx).await;
+        assert!(res.is_done(), "{res:?}");
+        assert!(meta.state().is_done());
+        check_files(&object_store, ctx.procedure_id, &["0000000000.commit"]).await;
+    }
+
+    #[tokio::test]
+    async fn test_execute_exceed_max_retry_later() {
+        let exec_fn =
+            |_| async { Err(Error::retry_later(MockError::new(StatusCode::Unexpected))) }.boxed();
+
+        let exceed_max_retry_later = ProcedureAdapter {
+            data: "exceed_max_retry_later".to_string(),
+            lock_key: LockKey::single("catalog.schema.table"),
+            exec_fn,
+        };
+
+        let dir = create_temp_dir("exceed_max_retry_later");
+        let meta = exceed_max_retry_later.new_meta(ROOT_ID);
+        let object_store = test_util::new_object_store(&dir);
+        let procedure_store = ProcedureStore::from(object_store.clone());
+        let mut runner = new_runner(
+            meta.clone(),
+            Box::new(exceed_max_retry_later),
+            procedure_store,
+        );
+        runner.exponential_builder = ExponentialBuilder::default()
+            .with_min_delay(Duration::from_millis(1))
+            .with_max_times(3);
+
+        // Run the runner and execute the procedure.
+        runner.execute_procedure_in_loop().await;
+        let err = meta.state().error().unwrap().to_string();
+        assert!(err.contains("Procedure retry exceeded max times"));
+    }
+
    #[tokio::test]
    async fn test_child_error() {
        let mut times = 0;
@@ -697,9 +863,10 @@ mod tests {
                } else {
                    // Wait for subprocedures.
                    let state = ctx.provider.procedure_state(child_id).await.unwrap();
-                    if state == Some(ProcedureState::Failed) {
+                    let is_failed = state.map(|s| s.is_failed()).unwrap_or(false);
+                    if is_failed {
                        // The parent procedure to abort itself if child procedure is failed.
-                        Err(Error::external(PlainError::new(
+                        Err(Error::from_error_ext(PlainError::new(
                            "subprocedure failed".to_string(),
                            StatusCode::Unexpected,
                        )))
@@ -720,7 +887,7 @@ mod tests {
            exec_fn,
        };

-        let dir = TempDir::new("child_err").unwrap();
+        let dir = create_temp_dir("child_err");
        let meta = parent.new_meta(ROOT_ID);

        let object_store = test_util::new_object_store(&dir);
@@ -729,12 +896,13 @@ mod tests {

        let manager_ctx = Arc::new(ManagerContext::new());
        // Manually add this procedure to the manager ctx.
-        assert!(manager_ctx.try_insert_procedure(meta));
+        assert!(manager_ctx.try_insert_procedure(meta.clone()));
        // Replace the manager ctx.
        runner.manager_ctx = manager_ctx;

-        // Run the runer and execute the procedure.
-        let err = runner.run().await.unwrap_err();
-        assert!(err.to_string().contains("subprocedure failed"), "{err}");
+        // Run the runner and execute the procedure.
+        runner.run().await;
+        let err = meta.state().error().unwrap().to_string();
+        assert!(err.contains("subprocedure failed"), "{err}");
    }
 }
--- a/src/common/procedure/src/procedure.rs
+++ b/src/common/procedure/src/procedure.rs
@@ -22,7 +22,8 @@ use smallvec::{smallvec, SmallVec};
 use snafu::{ResultExt, Snafu};
 use uuid::Uuid;

-use crate::error::Result;
+use crate::error::{Error, Result};
+use crate::watcher::Watcher;

 /// Procedure execution status.
 #[derive(Debug)]
@@ -197,16 +198,59 @@ impl FromStr for ProcedureId {
 /// Loader to recover the [Procedure] instance from serialized data.
 pub type BoxedProcedureLoader = Box<dyn Fn(&str) -> Result<BoxedProcedure> + Send>;

-// TODO(yingwen): Find a way to return the error message if the procedure is failed.
 /// State of a submitted procedure.
-#[derive(Debug, Clone, PartialEq, Eq)]
+#[derive(Debug, Default, Clone)]
 pub enum ProcedureState {
    /// The procedure is running.
+    #[default]
    Running,
    /// The procedure is finished.
    Done,
+    /// The procedure is failed and can be retried.
+    Retrying { error: Arc<Error> },
    /// The procedure is failed and cannot proceed anymore.
-    Failed,
+    Failed { error: Arc<Error> },
+}
+
+impl ProcedureState {
+    /// Returns a [ProcedureState] with failed state.
+    pub fn failed(error: Arc<Error>) -> ProcedureState {
+        ProcedureState::Failed { error }
+    }
+
+    /// Returns a [ProcedureState] with retrying state.
+    pub fn retrying(error: Arc<Error>) -> ProcedureState {
+        ProcedureState::Retrying { error }
+    }
+
+    /// Returns true if the procedure state is running.
+    pub fn is_running(&self) -> bool {
+        matches!(self, ProcedureState::Running)
+    }
+
+    /// Returns true if the procedure state is done.
+    pub fn is_done(&self) -> bool {
+        matches!(self, ProcedureState::Done)
+    }
+
+    /// Returns true if the procedure state failed.
+    pub fn is_failed(&self) -> bool {
+        matches!(self, ProcedureState::Failed { .. })
+    }
+
+    /// Returns true if the procedure state is retrying.
+    pub fn is_retrying(&self) -> bool {
+        matches!(self, ProcedureState::Retrying { .. })
+    }
+
+    /// Returns the error.
+    pub fn error(&self) -> Option<&Arc<Error>> {
+        match self {
+            ProcedureState::Failed { error } => Some(error),
+            ProcedureState::Retrying { error } => Some(error),
+            _ => None,
+        }
+    }
 }

 // TODO(yingwen): Shutdown
@@ -217,7 +261,9 @@ pub trait ProcedureManager: Send + Sync + 'static {
    fn register_loader(&self, name: &str, loader: BoxedProcedureLoader) -> Result<()>;

    /// Submits a procedure to execute.
-    async fn submit(&self, procedure: ProcedureWithId) -> Result<()>;
+    ///
+    /// Returns a [Watcher] to watch the created procedure.
+    async fn submit(&self, procedure: ProcedureWithId) -> Result<Watcher>;

    /// Recovers unfinished procedures and reruns them.
    ///
@@ -228,6 +274,9 @@ pub trait ProcedureManager: Send + Sync + 'static {
    ///
    /// Returns `Ok(None)` if the procedure doesn't exist.
    async fn procedure_state(&self, procedure_id: ProcedureId) -> Result<Option<ProcedureState>>;
+
+    /// Returns a [Watcher] to watch [ProcedureState] of specific procedure.
+    fn procedure_watcher(&self, procedure_id: ProcedureId) -> Option<Watcher>;
 }

 /// Ref-counted pointer to the [ProcedureManager].
@@ -235,6 +284,9 @@ pub type ProcedureManagerRef = Arc<dyn ProcedureManager>;

 #[cfg(test)]
 mod tests {
+    use common_error::mock::MockError;
+    use common_error::prelude::StatusCode;
+
    use super::*;

    #[test]
@@ -302,4 +354,17 @@ mod tests {
        let parsed = serde_json::from_str(&json).unwrap();
        assert_eq!(id, parsed);
    }
+
+    #[test]
+    fn test_procedure_state() {
+        assert!(ProcedureState::Running.is_running());
+        assert!(ProcedureState::Running.error().is_none());
+        assert!(ProcedureState::Done.is_done());
+
+        let state = ProcedureState::failed(Arc::new(Error::external(MockError::new(
+            StatusCode::Unexpected,
+        ))));
+        assert!(state.is_failed());
+        assert!(state.error().is_some());
+    }
 }
--- a/src/common/procedure/src/store.rs
+++ b/src/common/procedure/src/store.rs
@@ -246,8 +246,9 @@ impl ParsedKey {
 #[cfg(test)]
 mod tests {
    use async_trait::async_trait;
-    use object_store::services::fs::Builder;
-    use tempdir::TempDir;
+    use common_test_util::temp_dir::{create_temp_dir, TempDir};
+    use object_store::services::Fs as Builder;
+    use object_store::ObjectStoreBuilder;

    use super::*;
    use crate::{Context, LockKey, Procedure, Status};
@@ -255,7 +256,7 @@ mod tests {
    fn procedure_store_for_test(dir: &TempDir) -> ProcedureStore {
        let store_dir = dir.path().to_str().unwrap();
        let accessor = Builder::default().root(store_dir).build().unwrap();
-        let object_store = ObjectStore::new(accessor);
+        let object_store = ObjectStore::new(accessor).finish();

        ProcedureStore::from(object_store)
    }
@@ -372,7 +373,7 @@ mod tests {

    #[tokio::test]
    async fn test_store_procedure() {
-        let dir = TempDir::new("store_procedure").unwrap();
+        let dir = create_temp_dir("store_procedure");
        let store = procedure_store_for_test(&dir);

        let procedure_id = ProcedureId::random();
@@ -397,7 +398,7 @@ mod tests {

    #[tokio::test]
    async fn test_commit_procedure() {
-        let dir = TempDir::new("commit_procedure").unwrap();
+        let dir = create_temp_dir("commit_procedure");
        let store = procedure_store_for_test(&dir);

        let procedure_id = ProcedureId::random();
@@ -415,7 +416,7 @@ mod tests {

    #[tokio::test]
    async fn test_rollback_procedure() {
-        let dir = TempDir::new("rollback_procedure").unwrap();
+        let dir = create_temp_dir("rollback_procedure");
        let store = procedure_store_for_test(&dir);

        let procedure_id = ProcedureId::random();
@@ -433,7 +434,7 @@ mod tests {

    #[tokio::test]
    async fn test_load_messages() {
-        let dir = TempDir::new("load_messages").unwrap();
+        let dir = create_temp_dir("load_messages");
        let store = procedure_store_for_test(&dir);

        // store 3 steps
--- a/src/common/procedure/src/store/state_store.rs
+++ b/src/common/procedure/src/store/state_store.rs
@@ -20,9 +20,7 @@ use futures::{Stream, TryStreamExt};
 use object_store::{ObjectMode, ObjectStore};
 use snafu::ResultExt;

-use crate::error::{
-    DeleteStateSnafu, Error, ListStateSnafu, PutStateSnafu, ReadStateSnafu, Result,
-};
+use crate::error::{DeleteStateSnafu, Error, PutStateSnafu, Result};

 /// Key value from state store.
 type KeyValue = (String, Vec<u8>);
@@ -72,22 +70,23 @@ impl StateStore for ObjectStateStore {

    async fn walk_top_down(&self, path: &str) -> Result<KeyValueStream> {
        let path_string = path.to_string();
-        let op = self.store.batch();
-        // Note that there is no guarantee about the order between files and dirs
-        // at the same level.
-        // See https://docs.rs/opendal/0.25.2/opendal/raw/struct.TopDownWalker.html#note
-        let stream = op
-            .walk_top_down(path)
-            .context(ListStateSnafu { path })?
-            .map_err(move |e| Error::ListState {
+
+        let lister = self
+            .store
+            .object(path)
+            .scan()
+            .await
+            .map_err(|e| Error::ListState {
                path: path_string.clone(),
                source: e,
-            })
+            })?;
+
+        let stream = lister
            .try_filter_map(|entry| async move {
                let key = entry.path();
-                let key_value = match entry.mode().await.context(ReadStateSnafu { key })? {
+                let key_value = match entry.mode().await? {
                    ObjectMode::FILE => {
-                        let value = entry.read().await.context(ReadStateSnafu { key })?;
+                        let value = entry.read().await?;

                        Some((key.to_string(), value))
                    }
@@ -95,6 +94,10 @@ impl StateStore for ObjectStateStore {
                };

                Ok(key_value)
+            })
+            .map_err(move |e| Error::ListState {
+                path: path_string.clone(),
+                source: e,
            });

        Ok(Box::pin(stream))
@@ -112,17 +115,18 @@ impl StateStore for ObjectStateStore {

 #[cfg(test)]
 mod tests {
-    use object_store::services::fs::Builder;
-    use tempdir::TempDir;
+    use common_test_util::temp_dir::create_temp_dir;
+    use object_store::services::Fs as Builder;
+    use object_store::ObjectStoreBuilder;

    use super::*;

    #[tokio::test]
    async fn test_object_state_store() {
-        let dir = TempDir::new("state_store").unwrap();
+        let dir = create_temp_dir("state_store");
        let store_dir = dir.path().to_str().unwrap();
        let accessor = Builder::default().root(store_dir).build().unwrap();
-        let object_store = ObjectStore::new(accessor);
+        let object_store = ObjectStore::new(accessor).finish();
        let state_store = ObjectStateStore::new(object_store);

        let data: Vec<_> = state_store
--- a/src/common/procedure/src/watcher.rs
+++ b/src/common/procedure/src/watcher.rs
@@ -0,0 +1,41 @@
+// Copyright 2023 Greptime Team
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+use snafu::ResultExt;
+use tokio::sync::watch::Receiver;
+
+use crate::error::{ProcedureExecSnafu, Result, WaitWatcherSnafu};
+use crate::procedure::ProcedureState;
+
+/// Watcher to watch procedure state.
+pub type Watcher = Receiver<ProcedureState>;
+
+/// Wait the [Watcher] until the [ProcedureState] is done.
+pub async fn wait(watcher: &mut Watcher) -> Result<()> {
+    loop {
+        watcher.changed().await.context(WaitWatcherSnafu)?;
+        match &*watcher.borrow() {
+            ProcedureState::Running => (),
+            ProcedureState::Done => {
+                return Ok(());
+            }
+            ProcedureState::Failed { error } => {
+                return Err(error.clone()).context(ProcedureExecSnafu);
+            }
+            ProcedureState::Retrying { error } => {
+                return Err(error.clone()).context(ProcedureExecSnafu);
+            }
+        }
+    }
+}
--- a/src/common/query/src/error.rs
+++ b/src/common/query/src/error.rs
@@ -239,7 +239,6 @@ impl From<BoxedError> for Error {

 #[cfg(test)]
 mod tests {
-    use datatypes::arrow::error::ArrowError;
    use snafu::GenerateImplicitData;

    use super::*;
@@ -286,7 +285,7 @@ mod tests {
    fn test_convert_df_recordbatch_stream_error() {
        let result: std::result::Result<i32, common_recordbatch::error::Error> =
            Err(common_recordbatch::error::Error::PollStream {
-                source: ArrowError::DivideByZero,
+                source: DataFusionError::Internal("blabla".to_string()),
                backtrace: Backtrace::generate(),
            });
        let error = result
--- a/src/common/query/src/physical_plan.rs
+++ b/src/common/query/src/physical_plan.rs
@@ -315,7 +315,11 @@ mod test {
                .unwrap()
                .build()
                .unwrap();
-        let physical_plan = ctx.create_physical_plan(&logical_plan).await.unwrap();
+        let physical_plan = ctx
+            .state()
+            .create_physical_plan(&logical_plan)
+            .await
+            .unwrap();
        let df_recordbatches = collect(physical_plan, Arc::new(TaskContext::from(&ctx)))
            .await
            .unwrap();
--- a/src/common/recordbatch/src/adapter.rs
+++ b/src/common/recordbatch/src/adapter.rs
@@ -18,9 +18,9 @@ use std::sync::Arc;
 use std::task::{Context, Poll};

 use datafusion::arrow::datatypes::SchemaRef as DfSchemaRef;
+use datafusion::error::Result as DfResult;
 use datafusion::physical_plan::RecordBatchStream as DfRecordBatchStream;
 use datafusion_common::DataFusionError;
-use datatypes::arrow::error::{ArrowError, Result as ArrowResult};
 use datatypes::schema::{Schema, SchemaRef};
 use futures::ready;
 use snafu::ResultExt;
@@ -57,14 +57,14 @@ impl DfRecordBatchStream for DfRecordBatchStreamAdapter {
 }

 impl Stream for DfRecordBatchStreamAdapter {
-    type Item = ArrowResult<DfRecordBatch>;
+    type Item = DfResult<DfRecordBatch>;

    fn poll_next(mut self: Pin<&mut Self>, cx: &mut Context<'_>) -> Poll<Option<Self::Item>> {
        match Pin::new(&mut self.stream).poll_next(cx) {
            Poll::Pending => Poll::Pending,
            Poll::Ready(Some(recordbatch)) => match recordbatch {
                Ok(recordbatch) => Poll::Ready(Some(Ok(recordbatch.into_df_record_batch()))),
-                Err(e) => Poll::Ready(Some(Err(ArrowError::ExternalError(Box::new(e))))),
+                Err(e) => Poll::Ready(Some(Err(DataFusionError::External(Box::new(e))))),
            },
            Poll::Ready(None) => Poll::Ready(None),
        }
@@ -242,12 +242,12 @@ mod test {
        )]));
        let batch1 = RecordBatch::new(
            schema.clone(),
-            vec![Arc::new(Int32Vector::from_slice(&[1])) as _],
+            vec![Arc::new(Int32Vector::from_slice([1])) as _],
        )
        .unwrap();
        let batch2 = RecordBatch::new(
            schema.clone(),
-            vec![Arc::new(Int32Vector::from_slice(&[2])) as _],
+            vec![Arc::new(Int32Vector::from_slice([2])) as _],
        )
        .unwrap();

--- a/src/common/recordbatch/src/error.rs
+++ b/src/common/recordbatch/src/error.rs
@@ -55,7 +55,7 @@ pub enum Error {

    #[snafu(display("Failed to poll stream, source: {}", source))]
    PollStream {
-        source: datatypes::arrow::error::ArrowError,
+        source: datafusion::error::DataFusionError,
        backtrace: Backtrace,
    },

--- a/src/common/recordbatch/src/lib.rs
+++ b/src/common/recordbatch/src/lib.rs
@@ -204,7 +204,7 @@ mod tests {
        );
        assert!(result.is_err());

-        let v: VectorRef = Arc::new(Int32Vector::from_slice(&[1, 2]));
+        let v: VectorRef = Arc::new(Int32Vector::from_slice([1, 2]));
        let expected = vec![RecordBatch::new(schema.clone(), vec![v.clone()]).unwrap()];
        let r = RecordBatches::try_from_columns(schema, vec![v]).unwrap();
        assert_eq!(r.take(), expected);
@@ -216,7 +216,7 @@ mod tests {
        let column_b = ColumnSchema::new("b", ConcreteDataType::string_datatype(), false);
        let column_c = ColumnSchema::new("c", ConcreteDataType::boolean_datatype(), false);

-        let va: VectorRef = Arc::new(Int32Vector::from_slice(&[1, 2]));
+        let va: VectorRef = Arc::new(Int32Vector::from_slice([1, 2]));
        let vb: VectorRef = Arc::new(StringVector::from(vec!["hello", "world"]));
        let vc: VectorRef = Arc::new(BooleanVector::from(vec![true, false]));

@@ -255,11 +255,11 @@ mod tests {
        let column_b = ColumnSchema::new("b", ConcreteDataType::string_datatype(), false);
        let schema = Arc::new(Schema::new(vec![column_a, column_b]));

-        let va1: VectorRef = Arc::new(Int32Vector::from_slice(&[1, 2]));
+        let va1: VectorRef = Arc::new(Int32Vector::from_slice([1, 2]));
        let vb1: VectorRef = Arc::new(StringVector::from(vec!["a", "b"]));
        let batch1 = RecordBatch::new(schema.clone(), vec![va1, vb1]).unwrap();

-        let va2: VectorRef = Arc::new(Int32Vector::from_slice(&[3, 4, 5]));
+        let va2: VectorRef = Arc::new(Int32Vector::from_slice([3, 4, 5]));
        let vb2: VectorRef = Arc::new(StringVector::from(vec!["c", "d", "e"]));
        let batch2 = RecordBatch::new(schema.clone(), vec![va2, vb2]).unwrap();

--- a/src/common/recordbatch/src/recordbatch.rs
+++ b/src/common/recordbatch/src/recordbatch.rs
@@ -189,8 +189,8 @@ mod tests {
        ]));
        let schema = Arc::new(Schema::try_from(arrow_schema).unwrap());

-        let c1 = Arc::new(UInt32Vector::from_slice(&[1, 2, 3]));
-        let c2 = Arc::new(UInt32Vector::from_slice(&[4, 5, 6]));
+        let c1 = Arc::new(UInt32Vector::from_slice([1, 2, 3]));
+        let c2 = Arc::new(UInt32Vector::from_slice([4, 5, 6]));
        let columns: Vec<VectorRef> = vec![c1, c2];

        let batch = RecordBatch::new(schema.clone(), columns.clone()).unwrap();
@@ -222,7 +222,7 @@ mod tests {
        let schema = Arc::new(Schema::try_new(column_schemas).unwrap());

        let numbers: Vec<u32> = (0..10).collect();
-        let columns = vec![Arc::new(UInt32Vector::from_slice(&numbers)) as VectorRef];
+        let columns = vec![Arc::new(UInt32Vector::from_slice(numbers)) as VectorRef];
        let batch = RecordBatch::new(schema, columns).unwrap();

        let output = serde_json::to_string(&batch).unwrap();
--- a/src/common/substrait/Cargo.toml
+++ b/src/common/substrait/Cargo.toml
@@ -5,6 +5,8 @@ edition.workspace = true
 license.workspace = true

 [dependencies]
+async-recursion = "1.0"
+async-trait.workspace = true
 bytes = "1.1"
 catalog = { path = "../../catalog" }
 common-catalog = { path = "../catalog" }
@@ -14,13 +16,14 @@ datafusion.workspace = true
 datafusion-expr.workspace = true
 datatypes = { path = "../../datatypes" }
 futures = "0.3"
-prost = "0.9"
+prost.workspace = true
+session = { path = "../../session" }
 snafu.workspace = true
 table = { path = "../../table" }

 [dependencies.substrait_proto]
 package = "substrait"
-version = "0.2"
+version = "0.4"

 [dev-dependencies]
 datatypes = { path = "../../datatypes" }
--- a/src/common/substrait/src/context.rs
+++ b/src/common/substrait/src/context.rs
@@ -15,10 +15,10 @@
 use std::collections::HashMap;

 use datafusion::common::DFSchemaRef;
-use substrait_proto::protobuf::extensions::simple_extension_declaration::{
+use substrait_proto::proto::extensions::simple_extension_declaration::{
    ExtensionFunction, MappingType,
 };
-use substrait_proto::protobuf::extensions::SimpleExtensionDeclaration;
+use substrait_proto::proto::extensions::SimpleExtensionDeclaration;

 #[derive(Default)]
 pub struct ConvertorContext {
--- a/src/common/substrait/src/df_expr.rs
+++ b/src/common/substrait/src/df_expr.rs
@@ -20,15 +20,15 @@ use datafusion_expr::expr::Sort;
 use datafusion_expr::{expr_fn, lit, Between, BinaryExpr, BuiltinScalarFunction, Expr, Operator};
 use datatypes::schema::Schema;
 use snafu::{ensure, OptionExt};
-use substrait_proto::protobuf::expression::field_reference::ReferenceType as FieldReferenceType;
-use substrait_proto::protobuf::expression::reference_segment::{
+use substrait_proto::proto::expression::field_reference::ReferenceType as FieldReferenceType;
+use substrait_proto::proto::expression::reference_segment::{
    ReferenceType as SegReferenceType, StructField,
 };
-use substrait_proto::protobuf::expression::{
+use substrait_proto::proto::expression::{
    FieldReference, Literal, ReferenceSegment, RexType, ScalarFunction,
 };
-use substrait_proto::protobuf::function_argument::ArgType;
-use substrait_proto::protobuf::Expression;
+use substrait_proto::proto::function_argument::ArgType;
+use substrait_proto::proto::Expression;

 use crate::context::ConvertorContext;
 use crate::error::{
@@ -61,6 +61,7 @@ pub(crate) fn to_df_expr(
        | RexType::MultiOrList(_)
        | RexType::Cast(_)
        | RexType::Subquery(_)
+        | RexType::Nested(_)
        | RexType::Enum(_) => UnsupportedExprSnafu {
            name: format!("substrait expression {expr_rex_type:?}"),
        }
@@ -615,9 +616,9 @@ pub fn convert_column(column: &Column, schema: &Schema) -> Result<FieldReference
 /// Some utils special for this `DataFusion::Expr` and `Substrait::Expression` conversion.
 mod utils {
    use datafusion_expr::{BuiltinScalarFunction, Operator};
-    use substrait_proto::protobuf::expression::{RexType, ScalarFunction};
-    use substrait_proto::protobuf::function_argument::ArgType;
-    use substrait_proto::protobuf::{Expression, FunctionArgument};
+    use substrait_proto::proto::expression::{RexType, ScalarFunction};
+    use substrait_proto::proto::function_argument::ArgType;
+    use substrait_proto::proto::{Expression, FunctionArgument};

    pub(crate) fn name_df_operator(op: &Operator) -> &str {
        match op {
@@ -634,8 +635,6 @@ mod utils {
            Operator::Modulo => "modulo",
            Operator::And => "and",
            Operator::Or => "or",
-            Operator::Like => "like",
-            Operator::NotLike => "not_like",
            Operator::IsDistinctFrom => "is_distinct_from",
            Operator::IsNotDistinctFrom => "is_not_distinct_from",
            Operator::RegexMatch => "regex_match",
@@ -648,8 +647,6 @@ mod utils {
            Operator::BitwiseShiftRight => "bitwise_shift_right",
            Operator::BitwiseShiftLeft => "bitwise_shift_left",
            Operator::StringConcat => "string_concat",
-            Operator::ILike => "i_like",
-            Operator::NotILike => "not_i_like",
        }
    }

--- a/src/common/substrait/src/df_logical.rs
+++ b/src/common/substrait/src/df_logical.rs
@@ -14,31 +14,35 @@

 use std::sync::Arc;

+use async_recursion::async_recursion;
+use async_trait::async_trait;
 use bytes::{Buf, Bytes, BytesMut};
+use catalog::table_source::DfTableSourceProvider;
 use catalog::CatalogManagerRef;
-use common_error::prelude::BoxedError;
+use common_catalog::format_full_table_name;
 use common_telemetry::debug;
 use datafusion::arrow::datatypes::SchemaRef as ArrowSchemaRef;
-use datafusion::common::{DFField, DFSchema};
+use datafusion::common::{DFField, DFSchema, OwnedTableReference};
 use datafusion::datasource::DefaultTableSource;
 use datafusion::physical_plan::project_schema;
-use datafusion_expr::{Filter, LogicalPlan, TableScan, TableSource};
+use datafusion_expr::{Filter, LogicalPlan, TableScan};
 use prost::Message;
+use session::context::QueryContext;
 use snafu::{ensure, OptionExt, ResultExt};
-use substrait_proto::protobuf::expression::mask_expression::{StructItem, StructSelect};
-use substrait_proto::protobuf::expression::MaskExpression;
-use substrait_proto::protobuf::extensions::simple_extension_declaration::MappingType;
-use substrait_proto::protobuf::plan_rel::RelType as PlanRelType;
-use substrait_proto::protobuf::read_rel::{NamedTable, ReadType};
-use substrait_proto::protobuf::rel::RelType;
-use substrait_proto::protobuf::{FilterRel, Plan, PlanRel, ReadRel, Rel};
+use substrait_proto::proto::expression::mask_expression::{StructItem, StructSelect};
+use substrait_proto::proto::expression::MaskExpression;
+use substrait_proto::proto::extensions::simple_extension_declaration::MappingType;
+use substrait_proto::proto::plan_rel::RelType as PlanRelType;
+use substrait_proto::proto::read_rel::{NamedTable, ReadType};
+use substrait_proto::proto::rel::RelType;
+use substrait_proto::proto::{FilterRel, Plan, PlanRel, ReadRel, Rel};
 use table::table::adapter::DfTableProviderAdapter;

 use crate::context::ConvertorContext;
 use crate::df_expr::{expression_from_df_expr, to_df_expr};
 use crate::error::{
-    self, DFInternalSnafu, DecodeRelSnafu, EmptyPlanSnafu, EncodeRelSnafu, Error, InternalSnafu,
-    InvalidParametersSnafu, MissingFieldSnafu, SchemaNotMatchSnafu, TableNotFoundSnafu,
+    self, DFInternalSnafu, DecodeRelSnafu, EmptyPlanSnafu, EncodeRelSnafu, Error,
+    InvalidParametersSnafu, MissingFieldSnafu, ResolveTableSnafu, SchemaNotMatchSnafu,
    UnknownPlanSnafu, UnsupportedExprSnafu, UnsupportedPlanSnafu,
 };
 use crate::schema::{from_schema, to_schema};
@@ -46,18 +50,19 @@ use crate::SubstraitPlan;

 pub struct DFLogicalSubstraitConvertor;

+#[async_trait]
 impl SubstraitPlan for DFLogicalSubstraitConvertor {
    type Error = Error;

    type Plan = LogicalPlan;

-    fn decode<B: Buf + Send>(
+    async fn decode<B: Buf + Send>(
        &self,
        message: B,
        catalog_manager: CatalogManagerRef,
    ) -> Result<Self::Plan, Self::Error> {
        let plan = Plan::decode(message).context(DecodeRelSnafu)?;
-        self.convert_plan(plan, catalog_manager)
+        self.convert_plan(plan, catalog_manager).await
    }

    fn encode(&self, plan: Self::Plan) -> Result<Bytes, Self::Error> {
@@ -71,7 +76,7 @@ impl SubstraitPlan for DFLogicalSubstraitConvertor {
 }

 impl DFLogicalSubstraitConvertor {
-    fn convert_plan(
+    async fn convert_plan(
        &self,
        mut plan: Plan,
        catalog_manager: CatalogManagerRef,
@@ -102,20 +107,25 @@ impl DFLogicalSubstraitConvertor {
            .fail()?
        };

-        self.rel_to_logical_plan(&mut ctx, Box::new(rel), catalog_manager)
+        // TODO(LFC): Create table provider from outside, respect "disallow_cross_schema_query" option in query engine state.
+        let mut table_provider =
+            DfTableSourceProvider::new(catalog_manager, false, &QueryContext::new());
+        self.rel_to_logical_plan(&mut ctx, Box::new(rel), &mut table_provider)
+            .await
    }

-    fn rel_to_logical_plan(
+    #[async_recursion]
+    async fn rel_to_logical_plan(
        &self,
        ctx: &mut ConvertorContext,
        rel: Box<Rel>,
-        catalog_manager: CatalogManagerRef,
+        table_provider: &mut DfTableSourceProvider,
    ) -> Result<LogicalPlan, Error> {
        let rel_type = rel.rel_type.context(EmptyPlanSnafu)?;

        // build logical plan
        let logical_plan = match rel_type {
-            RelType::Read(read_rel) => self.convert_read_rel(ctx, read_rel, catalog_manager)?,
+            RelType::Read(read_rel) => self.convert_read_rel(ctx, read_rel, table_provider).await?,
            RelType::Filter(filter) => {
                let FilterRel {
                    common: _,
@@ -128,7 +138,7 @@ impl DFLogicalSubstraitConvertor {
                    field: "input",
                    plan: "Filter",
                })?;
-                let input = Arc::new(self.rel_to_logical_plan(ctx, input, catalog_manager)?);
+                let input = Arc::new(self.rel_to_logical_plan(ctx, input, table_provider).await?);

                let condition = condition.context(MissingFieldSnafu {
                    field: "condition",
@@ -191,11 +201,11 @@ impl DFLogicalSubstraitConvertor {
        Ok(logical_plan)
    }

-    fn convert_read_rel(
+    async fn convert_read_rel(
        &self,
        ctx: &mut ConvertorContext,
        read_rel: Box<ReadRel>,
-        catalog_manager: CatalogManagerRef,
+        table_provider: &mut DfTableSourceProvider,
    ) -> Result<LogicalPlan, Error> {
        // Extract the catalog, schema and table name from NamedTable. Assume the first three are those names.
        let read_type = read_rel.read_type.context(MissingFieldSnafu {
@@ -230,17 +240,17 @@ impl DFLogicalSubstraitConvertor {
            .projection
            .map(|mask_expr| self.convert_mask_expression(mask_expr));

-        // Get table handle from catalog manager
-        let table_ref = catalog_manager
-            .table(&catalog_name, &schema_name, &table_name)
-            .map_err(BoxedError::new)
-            .context(InternalSnafu)?
-            .context(TableNotFoundSnafu {
-                name: format!("{catalog_name}.{schema_name}.{table_name}"),
+        let table_ref = OwnedTableReference::Full {
+            catalog: catalog_name.clone(),
+            schema: schema_name.clone(),
+            table: table_name.clone(),
+        };
+        let adapter = table_provider
+            .resolve_table(table_ref)
+            .await
+            .with_context(|_| ResolveTableSnafu {
+                table_name: format_full_table_name(&catalog_name, &schema_name, &table_name),
            })?;
-        let adapter = Arc::new(DefaultTableSource::new(Arc::new(
-            DfTableProviderAdapter::new(table_ref),
-        )));

        // Get schema directly from the table, and compare it with the schema retrieved from substrait proto.
        let stored_schema = adapter.schema();
@@ -262,7 +272,7 @@ impl DFLogicalSubstraitConvertor {
        };

        // Calculate the projected schema
-        let qualified = &format!("{catalog_name}.{schema_name}.{table_name}");
+        let qualified = &format_full_table_name(&catalog_name, &schema_name, &table_name);
        let projected_schema = Arc::new(
            project_schema(&stored_schema, projection.as_ref())
                .and_then(|x| {
@@ -281,7 +291,7 @@ impl DFLogicalSubstraitConvertor {

        // TODO(ruihang): Support limit(fetch)
        Ok(LogicalPlan::TableScan(TableScan {
-            table_name: format!("{catalog_name}.{schema_name}.{table_name}"),
+            table_name: qualified.to_string(),
            source: adapter,
            projection,
            projected_schema,
@@ -314,7 +324,7 @@ impl DFLogicalSubstraitConvertor {
            .fail()?,
            LogicalPlan::Filter(filter) => {
                let input = Some(Box::new(
-                    self.logical_plan_to_rel(ctx, filter.input().clone())?,
+                    self.logical_plan_to_rel(ctx, filter.input.clone())?,
                ));

                let schema = plan
@@ -324,7 +334,7 @@ impl DFLogicalSubstraitConvertor {
                    .context(error::ConvertDfSchemaSnafu)?;
                let condition = Some(Box::new(expression_from_df_expr(
                    ctx,
-                    filter.predicate(),
+                    &filter.predicate,
                    &schema,
                )?));

@@ -396,7 +406,10 @@ impl DFLogicalSubstraitConvertor {
            | LogicalPlan::Explain(_)
            | LogicalPlan::Analyze(_)
            | LogicalPlan::Extension(_)
-            | LogicalPlan::Prepare(_) => InvalidParametersSnafu {
+            | LogicalPlan::Prepare(_)
+            | LogicalPlan::Dml(_)
+            | LogicalPlan::DescribeTable(_)
+            | LogicalPlan::Unnest(_) => InvalidParametersSnafu {
                reason: format!(
                    "Trying to convert DDL/DML plan to substrait proto, plan: {plan:?}",
                ),
@@ -424,6 +437,7 @@ impl DFLogicalSubstraitConvertor {
            relations: vec![plan_rel],
            advanced_extensions: None,
            expected_type_urls: vec![],
+            ..Default::default()
        })
    }

@@ -485,6 +499,7 @@ impl DFLogicalSubstraitConvertor {
            projection,
            advanced_extension: None,
            read_type: Some(read_type),
+            ..Default::default()
        };

        Ok(read_rel)
@@ -522,6 +537,7 @@ mod test {
    use catalog::{CatalogList, CatalogProvider, RegisterTableRequest};
    use common_catalog::consts::{DEFAULT_CATALOG_NAME, DEFAULT_SCHEMA_NAME};
    use datafusion::common::{DFSchema, ToDFSchema};
+    use datafusion_expr::TableSource;
    use datatypes::schema::RawSchema;
    use table::requests::CreateTableRequest;
    use table::test_util::{EmptyTable, MockTableEngine};
@@ -570,7 +586,7 @@ mod test {
        let convertor = DFLogicalSubstraitConvertor;

        let proto = convertor.encode(plan.clone()).unwrap();
-        let tripped_plan = convertor.decode(proto, catalog).unwrap();
+        let tripped_plan = convertor.decode(proto, catalog).await.unwrap();

        assert_eq!(format!("{plan:?}"), format!("{tripped_plan:?}"));
    }
--- a/src/common/substrait/src/error.rs
+++ b/src/common/substrait/src/error.rs
@@ -105,6 +105,13 @@ pub enum Error {
        #[snafu(backtrace)]
        source: datatypes::error::Error,
    },
+
+    #[snafu(display("Unable to resolve table: {table_name}, error: {source}"))]
+    ResolveTable {
+        table_name: String,
+        #[snafu(backtrace)]
+        source: catalog::error::Error,
+    },
 }

 pub type Result<T> = std::result::Result<T, Error>;
@@ -127,6 +134,7 @@ impl ErrorExt for Error {
            | Error::SchemaNotMatch { .. } => StatusCode::InvalidArguments,
            Error::DFInternal { .. } | Error::Internal { .. } => StatusCode::Internal,
            Error::ConvertDfSchema { source } => source.status_code(),
+            Error::ResolveTable { source, .. } => source.status_code(),
        }
    }

--- a/src/common/substrait/src/lib.rs
+++ b/src/common/substrait/src/lib.rs
@@ -13,6 +13,7 @@
 // limitations under the License.

 #![feature(let_chains)]
+#![feature(trait_upcasting)]

 mod context;
 mod df_expr;
@@ -21,17 +22,19 @@ pub mod error;
 mod schema;
 mod types;

+use async_trait::async_trait;
 use bytes::{Buf, Bytes};
 use catalog::CatalogManagerRef;

 pub use crate::df_logical::DFLogicalSubstraitConvertor;

+#[async_trait]
 pub trait SubstraitPlan {
    type Error: std::error::Error;

    type Plan;

-    fn decode<B: Buf + Send>(
+    async fn decode<B: Buf + Send>(
        &self,
        message: B,
        catalog_manager: CatalogManagerRef,
--- a/Show More
+++ b/Show More