chore: add build_table_route_prefix (#1269 )

fix: Adds FileHandle to ChunkStream (#1255 )
* test: Add compaction test * test: Test read during compaction * test: Add s3 object store to test * test: only run compact test * feat: Hold file handle in chunk stream * test: check files still exist after compact * feat: Revert changes to develop.yaml * test: Simplify MockPurgeHandler
2025-12-26 16:10:02 +00:00 · 2023-03-28 16:26:24 +08:00 · 2023-03-28 16:22:07 +08:00 · 2023-03-28 15:21:06 +08:00 · 2023-03-28 14:34:14 +08:00 · 2023-03-28 14:06:13 +08:00
264 changed files with 9787 additions and 3873 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/workflows/release.yml
+++ b/.github/workflows/release.yml
@@ -18,6 +18,9 @@ env:

  CARGO_PROFILE: nightly

+  ## FIXME(zyy17): Enable it after the tests are stabled.
+  DISABLE_RUN_TESTS: true
+
 jobs:
  build:
    name: Build binary
@@ -32,15 +35,35 @@ jobs:
          - arch: aarch64-unknown-linux-gnu
            os: ubuntu-2004-16-cores
            file: greptime-linux-arm64
-            continue-on-error: true
+            continue-on-error: false
          - arch: aarch64-apple-darwin
            os: macos-latest
            file: greptime-darwin-arm64
-            continue-on-error: true
+            continue-on-error: false
          - arch: x86_64-apple-darwin
            os: macos-latest
            file: greptime-darwin-amd64
-            continue-on-error: true
+            continue-on-error: false
+          - arch: x86_64-unknown-linux-gnu
+            os: ubuntu-2004-16-cores
+            file: greptime-linux-amd64-pyo3
+            continue-on-error: false
+            opts: "-F pyo3_backend"
+          - arch: aarch64-unknown-linux-gnu
+            os: ubuntu-2004-16-cores
+            file: greptime-linux-arm64-pyo3
+            continue-on-error: false
+            opts: "-F pyo3_backend"
+          - arch: aarch64-apple-darwin
+            os: macos-latest
+            file: greptime-darwin-arm64-pyo3
+            continue-on-error: false
+            opts: "-F pyo3_backend"
+          - arch: x86_64-apple-darwin
+            os: macos-latest
+            file: greptime-darwin-amd64-pyo3
+            continue-on-error: false
+            opts: "-F pyo3_backend"
    runs-on: ${{ matrix.os }}
    continue-on-error: ${{ matrix.continue-on-error }}
    if: github.repository == 'GreptimeTeam/greptimedb'
@@ -98,13 +121,12 @@ jobs:
          sudo apt-get -y update
          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')
+      # FIXME(zyy17): Should we specify the version of python when building binary for darwin?
+      - name: Compile Python 3.10.10 from source for linux
+        if: contains(matrix.arch, 'linux') && contains(matrix.opts, 'pyo3_backend')
        run: |
          sudo chmod +x ./docker/aarch64/compile-python.sh
-          sudo ./docker/aarch64/compile-python.sh
-          export PYO3_CROSS_LIB_DIR=${PWD}/python310-aarch64/lib
-          echo $PYO3_CROSS_LIB_DIR
+          sudo ./docker/aarch64/compile-python.sh ${{ matrix.arch }}

      - name: Install rust toolchain
        uses: dtolnay/rust-toolchain@master
@@ -116,10 +138,55 @@ 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 with pyo3 for aarch64-linux
+        if: contains(matrix.arch, 'aarch64-unknown-linux-gnu') && contains(matrix.opts, 'pyo3_backend')
+        run: |
+          # TODO(zyy17): We should make PYO3_CROSS_LIB_DIR configurable.
+          export PYTHON_INSTALL_PATH_AMD64=${PWD}/python-3.10.10/amd64
+          export LD_LIBRARY_PATH=$PYTHON_INSTALL_PATH_AMD64/lib:$LD_LIBRARY_PATH
+          export LIBRARY_PATH=$PYTHON_INSTALL_PATH_AMD64/lib:$LIBRARY_PATH
+          export PATH=$PYTHON_INSTALL_PATH_AMD64/bin:$PATH
+
+          export PYO3_CROSS_LIB_DIR=${PWD}/python-3.10.10/aarch64
+          echo "PYO3_CROSS_LIB_DIR: $PYO3_CROSS_LIB_DIR"
+          alias python=$PYTHON_INSTALL_PATH_AMD64/bin/python3
+          alias pip=$PYTHON_INSTALL_PATH_AMD64/bin/python3-pip
+
+          cargo build --profile ${{ env.CARGO_PROFILE }} --locked --target ${{ matrix.arch }} ${{ matrix.opts }}
+
+      - name: Run cargo build with pyo3 for amd64-linux
+        if: contains(matrix.arch, 'x86_64-unknown-linux-gnu') && contains(matrix.opts, 'pyo3_backend')
+        run: |
+          export PYTHON_INSTALL_PATH_AMD64=${PWD}/python-3.10.10/amd64
+          export LD_LIBRARY_PATH=$PYTHON_INSTALL_PATH_AMD64/lib:$LD_LIBRARY_PATH
+          export LIBRARY_PATH=$PYTHON_INSTALL_PATH_AMD64/lib:$LIBRARY_PATH
+          export PATH=$PYTHON_INSTALL_PATH_AMD64/bin:$PATH
+          
+          echo "implementation=CPython" >> pyo3.config
+          echo "version=3.10" >> pyo3.config
+          echo "implementation=CPython" >> pyo3.config
+          echo "shared=true" >> pyo3.config
+          echo "abi3=true" >> pyo3.config
+          echo "lib_name=python3.10" >> pyo3.config
+          echo "lib_dir=$PYTHON_INSTALL_PATH_AMD64/lib" >> pyo3.config
+          echo "executable=$PYTHON_INSTALL_PATH_AMD64/bin/python3" >> pyo3.config
+          echo "pointer_width=64" >> pyo3.config
+          echo "build_flags=" >> pyo3.config
+          echo "suppress_build_script_link_lines=false" >> pyo3.config
+
+          cat pyo3.config
+          export PYO3_CONFIG_FILE=${PWD}/pyo3.config
+          alias python=$PYTHON_INSTALL_PATH_AMD64/bin/python3
+          alias pip=$PYTHON_INSTALL_PATH_AMD64/bin/python3-pip
+
+          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, 'darwin') || contains(matrix.opts, 'pyo3_backend') == false
+        run: cargo build --profile ${{ env.CARGO_PROFILE }} --locked --target ${{ matrix.arch }} ${{ matrix.opts }}

      - name: Calculate checksum and rename binary
        shell: bash
@@ -140,9 +207,96 @@ jobs:
        with:
          name: ${{ matrix.file }}.sha256sum
          path: target/${{ matrix.arch }}/${{ env.CARGO_PROFILE }}/${{ matrix.file }}.sha256sum
+
+  docker:
+    name: Build docker image
+    needs: [build]
+    runs-on: ubuntu-latest
+    if: github.repository == 'GreptimeTeam/greptimedb'
+    steps:
+      - name: Checkout sources
+        uses: actions/checkout@v3
+
+      - name: Login to Dockerhub
+        uses: docker/login-action@v2
+        with:
+          username: ${{ secrets.DOCKERHUB_USERNAME }}
+          password: ${{ secrets.DOCKERHUB_TOKEN }}
+
+      - 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}" >> $GITHUB_ENV
+
+      - name: Set up QEMU
+        uses: docker/setup-qemu-action@v2
+
+      - name: Set up buildx
+        uses: docker/setup-buildx-action@v2
+
+      - name: Download amd64 binary
+        uses: actions/download-artifact@v3
+        with:
+          name: greptime-linux-amd64-pyo3
+          path: amd64
+
+      - name: Unzip the amd64 artifacts
+        run: |
+          tar xvf amd64/greptime-linux-amd64-pyo3.tgz -C amd64/ && rm amd64/greptime-linux-amd64-pyo3.tgz
+          cp -r amd64 docker/ci
+
+      - name: Download arm64 binary
+        id: download-arm64
+        uses: actions/download-artifact@v3
+        with:
+          name: greptime-linux-arm64-pyo3
+          path: arm64
+
+      - name: Unzip the arm64 artifacts
+        id: unzip-arm64
+        if: success() || steps.download-arm64.conclusion == 'success'
+        run: |
+          tar xvf arm64/greptime-linux-arm64-pyo3.tgz -C arm64/ && rm arm64/greptime-linux-arm64-pyo3.tgz
+          cp -r arm64 docker/ci
+
+      - name: Build and push all
+        uses: docker/build-push-action@v3
+        if: success() || steps.unzip-arm64.conclusion == 'success' # Build and push all platform if unzip-arm64 succeeds
+        with:
+          context: ./docker/ci/
+          file: ./docker/ci/Dockerfile
+          push: true
+          platforms: linux/amd64,linux/arm64
+          tags: |
+            greptime/greptimedb:latest
+            greptime/greptimedb:${{ 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: ./docker/ci/
+          file: ./docker/ci/Dockerfile
+          push: true
+          platforms: linux/amd64
+          tags: |
+            greptime/greptimedb:latest
+            greptime/greptimedb:${{ env.IMAGE_TAG }}
+
  release:
    name: Release artifacts
-    needs: [build]
+    # Release artifacts only when all the artifacts are built successfully.
+    needs: [build,docker]
    runs-on: ubuntu-latest
    if: github.repository == 'GreptimeTeam/greptimedb'
    steps:
@@ -183,15 +337,23 @@ jobs:
          files: |
            **/greptime-*

-  docker:
-    name: Build docker image
-    needs: [build]
+  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:
@@ -199,12 +361,6 @@ jobs:
          username: ${{ secrets.UCLOUD_USERNAME }}
          password: ${{ secrets.UCLOUD_PASSWORD }}

-      - name: Login to Dockerhub
-        uses: docker/login-action@v2
-        with:
-          username: ${{ secrets.DOCKERHUB_USERNAME }}
-          password: ${{ secrets.DOCKERHUB_TOKEN }}
-
      - name: Configure scheduled build image tag # the tag would be ${SCHEDULED_BUILD_VERSION_PREFIX}-YYYYMMDD-${SCHEDULED_PERIOD}
        shell: bash
        if: github.event_name == 'schedule'
@@ -220,63 +376,9 @@ jobs:
          VERSION=${{ github.ref_name }}
          echo "IMAGE_TAG=${VERSION:1}" >> $GITHUB_ENV

-      - name: Set up QEMU
-        uses: docker/setup-qemu-action@v2
-
-      - name: Set up buildx
-        uses: docker/setup-buildx-action@v2
-
-      - name: Download amd64 binary
-        uses: actions/download-artifact@v3
-        with:
-          name: greptime-linux-amd64
-          path: amd64
-
-      - name: Unzip the amd64 artifacts
+      - name: Push image to uhub # Use 'docker buildx imagetools create' to create a new image base on source image.
        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
-          push: true
-          platforms: linux/amd64,linux/arm64
-          tags: |
-            greptime/greptimedb:latest
+          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 }}
-            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 }}
-            uhub.service.ucloud.cn/greptime/greptimedb:latest
-            uhub.service.ucloud.cn/greptime/greptimedb:${{ env.IMAGE_TAG }}
--- a/.gitignore
+++ b/.gitignore
@@ -35,3 +35,7 @@ benchmarks/data

 # dotenv
 .env
+
+# dashboard files
+!/src/servers/dashboard/VERSION
+/src/servers/dashboard/*
--- a/Cargo.lock
+++ b/Cargo.lock
--- a/Cargo.toml
+++ b/Cargo.toml
@@ -7,6 +7,7 @@ members = [
    "src/cmd",
    "src/common/base",
    "src/common/catalog",
+    "src/common/datasource",
    "src/common/error",
    "src/common/function",
    "src/common/function-macro",
@@ -50,28 +51,29 @@ edition = "2021"
 license = "Apache-2.0"

 [workspace.dependencies]
-arrow = { version = "33.0" }
-arrow-array = "33.0"
-arrow-flight = "33.0"
-arrow-schema = { version = "33.0", 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"] }
-datafusion = { git = "https://github.com/apache/arrow-datafusion.git", rev = "fad360df0132a2fcb264a7c07b2b02f0b1dfc644" }
-datafusion-common = { git = "https://github.com/apache/arrow-datafusion.git", rev = "fad360df0132a2fcb264a7c07b2b02f0b1dfc644" }
-datafusion-expr = { git = "https://github.com/apache/arrow-datafusion.git", rev = "fad360df0132a2fcb264a7c07b2b02f0b1dfc644" }
-datafusion-optimizer = { git = "https://github.com/apache/arrow-datafusion.git", rev = "fad360df0132a2fcb264a7c07b2b02f0b1dfc644" }
-datafusion-physical-expr = { git = "https://github.com/apache/arrow-datafusion.git", rev = "fad360df0132a2fcb264a7c07b2b02f0b1dfc644" }
-datafusion-sql = { git = "https://github.com/apache/arrow-datafusion.git", rev = "fad360df0132a2fcb264a7c07b2b02f0b1dfc644" }
+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 = "33.0"
+parquet = "34.0"
 paste = "1.0"
 prost = "0.11"
+rand = "0.8"
 serde = { version = "1.0", features = ["derive"] }
 serde_json = "1.0"
 snafu = { version = "0.7", features = ["backtraces"] }
-sqlparser = "0.30"
+sqlparser = "0.32"
 tempfile = "3"
 tokio = { version = "1.24.2", features = ["full"] }
 tokio-util = "0.7"
--- a/README.md
+++ b/README.md
@@ -1,8 +1,8 @@
 <p align="center">
  <picture>
-    <source media="(prefers-color-scheme: light)" srcset="/docs/logo-text-padding.png">
-    <source media="(prefers-color-scheme: dark)" srcset="/docs/logo-text-padding-dark.png">
-    <img alt="GreptimeDB Logo" src="/docs/logo-text-padding.png" width="400px">
+    <source media="(prefers-color-scheme: light)" srcset="https://cdn.jsdelivr.net/gh/GreptimeTeam/greptimedb@develop/docs/logo-text-padding.png">
+    <source media="(prefers-color-scheme: dark)" srcset="https://cdn.jsdelivr.net/gh/GreptimeTeam/greptimedb@develop/docs/logo-text-padding-dark.png">
+    <img alt="GreptimeDB Logo" src="https://cdn.jsdelivr.net/gh/GreptimeTeam/greptimedb@develop/docs/logo-text-padding.png" width="400px">
  </picture>
 </p>

@@ -61,12 +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
+- python3-dev or python3-devel(Optional feature, only needed if you want to run scripts
+  in CPython, and also need to enable `pyo3_backend` feature when compiling(by `cargo run -F pyo3_backend` or add `pyo3_backend` to src/script/Cargo.toml 's `features.default` like `default = ["python", "pyo3_backend]`)): 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.
+  `Python3` package should have this shared library by default. More detail for compiling with PyO3 can be found in [PyO3](https://pyo3.rs/v0.18.1/building_and_distribution#configuring-the-python-version)'s documentation.

 #### Build with Docker

@@ -147,9 +147,9 @@ You can always cleanup test database by removing `/tmp/greptimedb`.
 ### Installation

 - [Pre-built Binaries](https://github.com/GreptimeTeam/greptimedb/releases):
-  downloadable pre-built binaries for Linux and MacOS
- [Docker Images](https://hub.docker.com/r/greptime/greptimedb): pre-built
-  Docker images
+  For Linux and macOS, you can easily download pre-built binaries that are ready to use. In most cases, downloading the version without PyO3 is sufficient. However, if you plan to run scripts in CPython (and use Python packages like NumPy and Pandas), you will need to download the version with PyO3 and install a Python with the same version as the Python in the PyO3 version. We recommend using virtualenv for the installation process to manage multiple Python versions.
+- [Docker Images](https://hub.docker.com/r/greptime/greptimedb)(**recommended**): pre-built
+  Docker images, this is the easiest way to try GreptimeDB. By default it runs CPython script with `pyo3_backend` enabled.
 - [`gtctl`](https://github.com/GreptimeTeam/gtctl): the command-line tool for
  Kubernetes deployment

@@ -158,6 +158,7 @@ You can always cleanup test database by removing `/tmp/greptimedb`.
 - GreptimeDB [User Guide](https://docs.greptime.com/user-guide/concepts.html)
 - GreptimeDB [Developer
  Guide](https://docs.greptime.com/developer-guide/overview.html)
+- GreptimeDB [internal code document](https://greptimedb.rs)

 ### Dashboard
 - [The dashboard UI for GreptimeDB](https://github.com/GreptimeTeam/dashboard)
--- a/benchmarks/src/bin/nyc-taxi.rs
+++ b/benchmarks/src/bin/nyc-taxi.rs
@@ -21,12 +21,12 @@ use std::collections::HashMap;
 use std::path::{Path, PathBuf};
 use std::time::Instant;

-use arrow::array::{ArrayRef, PrimitiveArray, StringArray, TimestampNanosecondArray};
+use arrow::array::{ArrayRef, PrimitiveArray, StringArray, TimestampMicrosecondArray};
 use arrow::datatypes::{DataType, Float64Type, Int64Type};
 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::api::v1::{Column, ColumnDataType, ColumnDef, CreateTableExpr, InsertRequest};
 use client::{Client, Database, DEFAULT_CATALOG_NAME, DEFAULT_SCHEMA_NAME};
 use indicatif::{MultiProgress, ProgressBar, ProgressStyle};
 use parquet::arrow::arrow_reader::ParquetRecordBatchReaderBuilder;
@@ -61,7 +61,7 @@ struct Args {
    #[arg(long = "skip-read")]
    skip_read: bool,

-    #[arg(short, long, default_value_t = String::from("127.0.0.1:3001"))]
+    #[arg(short, long, default_value_t = String::from("127.0.0.1:4001"))]
    endpoint: String,
 }

@@ -97,6 +97,9 @@ async fn write_data(

    for record_batch in record_batch_reader {
        let record_batch = record_batch.unwrap();
+        if !is_record_batch_full(&record_batch) {
+            continue;
+        }
        let (columns, row_count) = convert_record_batch(record_batch);
        let request = InsertRequest {
            table_name: TABLE_NAME.to_string(),
@@ -122,11 +125,16 @@ fn convert_record_batch(record_batch: RecordBatch) -> (Vec<Column>, u32) {
    let mut columns = vec![];

    for (array, field) in record_batch.columns().iter().zip(fields.iter()) {
-        let values = build_values(array);
+        let (values, datatype) = build_values(array);
        let column = Column {
            column_name: field.name().to_owned(),
            values: Some(values),
-            null_mask: vec![],
+            null_mask: array
+                .data()
+                .null_bitmap()
+                .map(|bitmap| bitmap.buffer().as_slice().to_vec())
+                .unwrap_or_default(),
+            datatype: datatype.into(),
            // datatype and semantic_type are set to default
            ..Default::default()
        };
@@ -136,7 +144,7 @@ fn convert_record_batch(record_batch: RecordBatch) -> (Vec<Column>, u32) {
    (columns, row_count as _)
 }

-fn build_values(column: &ArrayRef) -> Values {
+fn build_values(column: &ArrayRef) -> (Values, ColumnDataType) {
    match column.data_type() {
        DataType::Int64 => {
            let array = column
@@ -144,10 +152,13 @@ fn build_values(column: &ArrayRef) -> Values {
                .downcast_ref::<PrimitiveArray<Int64Type>>()
                .unwrap();
            let values = array.values();
-            Values {
-                i64_values: values.to_vec(),
-                ..Default::default()
-            }
+            (
+                Values {
+                    i64_values: values.to_vec(),
+                    ..Default::default()
+                },
+                ColumnDataType::Int64,
+            )
        }
        DataType::Float64 => {
            let array = column
@@ -155,29 +166,38 @@ fn build_values(column: &ArrayRef) -> Values {
                .downcast_ref::<PrimitiveArray<Float64Type>>()
                .unwrap();
            let values = array.values();
-            Values {
-                f64_values: values.to_vec(),
-                ..Default::default()
-            }
+            (
+                Values {
+                    f64_values: values.to_vec(),
+                    ..Default::default()
+                },
+                ColumnDataType::Float64,
+            )
        }
        DataType::Timestamp(_, _) => {
            let array = column
                .as_any()
-                .downcast_ref::<TimestampNanosecondArray>()
+                .downcast_ref::<TimestampMicrosecondArray>()
                .unwrap();
            let values = array.values();
-            Values {
-                i64_values: values.to_vec(),
-                ..Default::default()
-            }
+            (
+                Values {
+                    i64_values: values.to_vec(),
+                    ..Default::default()
+                },
+                ColumnDataType::Int64,
+            )
        }
        DataType::Utf8 => {
            let array = column.as_any().downcast_ref::<StringArray>().unwrap();
            let values = array.iter().filter_map(|s| s.map(String::from)).collect();
-            Values {
-                string_values: values,
-                ..Default::default()
-            }
+            (
+                Values {
+                    string_values: values,
+                    ..Default::default()
+                },
+                ColumnDataType::String,
+            )
        }
        DataType::Null
        | DataType::Boolean
@@ -213,6 +233,10 @@ fn build_values(column: &ArrayRef) -> Values {
    }
 }

+fn is_record_batch_full(batch: &RecordBatch) -> bool {
+    batch.columns().iter().all(|col| col.null_count() == 0)
+}
+
 fn create_table_expr() -> CreateTableExpr {
    CreateTableExpr {
        catalog_name: CATALOG_NAME.to_string(),
@@ -340,7 +364,7 @@ fn create_table_expr() -> CreateTableExpr {
        create_if_not_exists: false,
        table_options: Default::default(),
        region_ids: vec![0],
-        table_id: Some(TableId { id: 0 }),
+        table_id: None,
    }
 }

--- a/config/datanode.example.toml
+++ b/config/datanode.example.toml
@@ -44,5 +44,7 @@ max_purge_tasks = 32

 # Procedure storage options, see `standalone.example.toml`.
 # [procedure.store]
-# type = 'File'
-# data_dir = '/tmp/greptimedb/procedure/'
+# type = "File"
+# data_dir = "/tmp/greptimedb/procedure/"
+# max_retry_times = 3
+# retry_delay = "500ms"
--- a/config/standalone.example.toml
+++ b/config/standalone.example.toml
@@ -114,3 +114,7 @@ max_purge_tasks = 32
 # 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/aarch64/Dockerfile
+++ b/docker/aarch64/Dockerfile
@@ -22,20 +22,27 @@ 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=${PWD}/python_arm64_build
-RUN chmod +x ./docker/aarch64/compile-python.sh && \
-    ./docker/aarch64/compile-python.sh
-# Install rustup target for cross compiling.
-RUN rustup target add aarch64-unknown-linux-gnu
+ENV PY_INSTALL_PATH=/greptimedb/python_arm64_build
+
 # Set the environment variable for cross compiling and compile it
-# Build the project in release mode. Set Net fetch with git cli to true to avoid git error.
+# cross compiled python is `python3` in path, but pyo3 need `python` in path so alias it
+# Build the project in release mode.
 RUN export PYO3_CROSS_LIB_DIR=$PY_INSTALL_PATH/lib && \ 
    alias python=python3 && \
-    CARGO_NET_GIT_FETCH_WITH_CLI=1 && \
    cargo build --target aarch64-unknown-linux-gnu --release -F pyo3_backend

 # Exporting the binary to the clean image
--- a/docker/aarch64/compile-python.sh
+++ b/docker/aarch64/compile-python.sh
@@ -1,9 +1,36 @@
+#!/usr/bin/env bash
+
+set -e
+
 # 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
+ARCH=$1
+PYTHON_VERSION=3.10.10
+PYTHON_SOURCE_DIR=Python-${PYTHON_VERSION}
+PYTHON_INSTALL_PATH_AMD64=${PWD}/python-${PYTHON_VERSION}/amd64
+PYTHON_INSTALL_PATH_AARCH64=${PWD}/python-${PYTHON_VERSION}/aarch64
+
+function download_python_source_code() {
+  wget https://www.python.org/ftp/python/$PYTHON_VERSION/Python-$PYTHON_VERSION.tgz
+  tar -xvf Python-$PYTHON_VERSION.tgz
+}
+
+function compile_for_amd64_platform() {
+  mkdir -p "$PYTHON_INSTALL_PATH_AMD64"
+
+  echo "Compiling for amd64 platform..."
+
+  ./configure \
+    --prefix="$PYTHON_INSTALL_PATH_AMD64" \
+    --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
+}

-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
@@ -14,33 +41,47 @@ cd Python-3.10.10
 # 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
+function compile_for_aarch64_platform() {
+  export LD_LIBRARY_PATH=$PYTHON_INSTALL_PATH_AMD64/lib:$LD_LIBRARY_PATH
+  export LIBRARY_PATH=$PYTHON_INSTALL_PATH_AMD64/lib:$LIBRARY_PATH
+  export PATH=$PYTHON_INSTALL_PATH_AMD64/bin:$PATH
+
+  mkdir -p "$PYTHON_INSTALL_PATH_AARCH64"
+
+  echo "Compiling for aarch64 platform..."
+  echo "LD_LIBRARY_PATH: $LD_LIBRARY_PATH"
+  echo "LIBRARY_PATH: $LIBRARY_PATH"
+  echo "PATH: $PATH"
+
+  ./configure --build=x86_64-linux-gnu --host=aarch64-linux-gnu \
+    --prefix="$PYTHON_INSTALL_PATH_AARCH64" --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
+}
+
+# Main script starts here.
+download_python_source_code
+
+# Enter the python source code directory.
+cd $PYTHON_SOURCE_DIR || exit 1

 # 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 ..
+compile_for_amd64_platform
+
+# Clean the build directory.
+make clean && make distclean
+
+# Cross compile python for aarch64.
+if [ "$ARCH" = "aarch64-unknown-linux-gnu" ]; then
+  compile_for_aarch64_platform
+fi
--- a/docker/ci/Dockerfile
+++ b/docker/ci/Dockerfile
@@ -1,6 +1,14 @@
 FROM ubuntu:22.04

-RUN apt-get update && DEBIAN_FRONTEND=noninteractive apt-get -y install ca-certificates
+RUN apt-get update && DEBIAN_FRONTEND=noninteractive apt-get install -y \
+    ca-certificates \
+    python3.10 \
+    python3.10-dev \
+    python3-pip
+
+COPY requirements.txt /etc/greptime/requirements.txt
+
+RUN python3 -m pip install -r /etc/greptime/requirements.txt

 ARG TARGETARCH

--- a/docker/ci/requirements.txt
+++ b/docker/ci/requirements.txt
@@ -0,0 +1,5 @@
+numpy>=1.24.2
+pandas>=1.5.3
+pyarrow>=11.0.0
+requests>=2.28.2
+scipy>=1.10.1
--- 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/scripts/fetch-dashboard-assets.sh
+++ b/scripts/fetch-dashboard-assets.sh
@@ -0,0 +1,39 @@
+#!/usr/bin/env bash
+
+# This script is used to download built dashboard assets from the "GreptimeTeam/dashboard" repository.
+
+set -e
+
+declare -r SCRIPT_DIR=$(cd $(dirname ${0}) >/dev/null 2>&1 && pwd)
+declare -r ROOT_DIR=$(dirname ${SCRIPT_DIR})
+declare -r STATIC_DIR="$ROOT_DIR/src/servers/dashboard"
+
+RELEASE_VERSION="$(cat $STATIC_DIR/VERSION)"
+
+# Download the SHA256 checksum attached to the release. To verify the integrity
+# of the download, this checksum will be used to check the download tar file
+# containing the built dashboard assets.
+curl -Ls https://github.com/GreptimeTeam/dashboard/releases/download/$RELEASE_VERSION/sha256.txt --output sha256.txt
+
+# Download the tar file containing the built dashboard assets.
+curl -L https://github.com/GreptimeTeam/dashboard/releases/download/$RELEASE_VERSION/build.tar.gz --output build.tar.gz
+
+# Verify the checksums match; exit if they don't.
+case "$(uname -s)" in
+    FreeBSD | Darwin)
+        echo "$(cat sha256.txt)" | shasum --algorithm 256 --check \
+            || { echo "Checksums did not match for downloaded dashboard assets!"; exit 1; } ;;
+    Linux)
+        echo "$(cat sha256.txt)" | sha256sum --check -- \
+            || { echo "Checksums did not match for downloaded dashboard assets!"; exit 1; } ;;
+    *)
+        echo "The '$(uname -s)' operating system is not supported as a build host for the dashboard" >&2
+        exit 1
+esac
+
+# Extract the assets and clean up.
+tar -xzf build.tar.gz -C "$STATIC_DIR"
+rm sha256.txt
+rm build.tar.gz
+
+echo "Successfully download dashboard assets to $STATIC_DIR"
--- 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 = "3a715150563b89d5dfc81a5838eac1f66a5658a1" }
+greptime-proto = { git = "https://github.com/GreptimeTeam/greptime-proto.git", rev = "d3861c34f7920238869d0d4e50dc1e6b189d2a6b" }
 prost.workspace = true
 snafu = { version = "0.7", features = ["backtraces"] }
 tonic.workspace = true
--- a/src/catalog/src/error.rs
+++ b/src/catalog/src/error.rs
@@ -204,6 +204,21 @@ pub enum 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>;
@@ -238,7 +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(),
--- a/src/catalog/src/lib.rs
+++ b/src/catalog/src/lib.rs
@@ -18,8 +18,9 @@ use std::any::Any;
 use std::fmt::{Debug, Formatter};
 use std::sync::Arc;

-use common_telemetry::info;
-use snafu::{OptionExt, ResultExt};
+use api::v1::meta::{RegionStat, TableName};
+use common_telemetry::{info, warn};
+use snafu::ResultExt;
 use table::engine::{EngineContext, TableEngineRef};
 use table::metadata::TableId;
 use table::requests::CreateTableRequest;
@@ -225,39 +226,52 @@ pub(crate) async fn handle_system_table_request<'a, M: CatalogManager>(
    Ok(())
 }

-/// The number of regions in the datanode node.
-pub async fn region_number(catalog_manager: &CatalogManagerRef) -> Result<u64> {
+/// The stat of regions in the datanode node.
+/// The number of regions can be got from len of vec.
+///
+/// Ignores any errors occurred during iterating regions. The intention of this method is to
+/// collect region stats that will be carried in Datanode's heartbeat to Metasrv, so it's a
+/// "try our best" job.
+pub async fn datanode_stat(catalog_manager: &CatalogManagerRef) -> (u64, Vec<RegionStat>) {
    let mut region_number: u64 = 0;
+    let mut region_stats = Vec::new();

-    for catalog_name in catalog_manager.catalog_names()? {
-        let catalog =
-            catalog_manager
-                .catalog(&catalog_name)?
-                .context(error::CatalogNotFoundSnafu {
-                    catalog_name: &catalog_name,
-                })?;
+    let Ok(catalog_names) = catalog_manager.catalog_names() else { return (region_number, region_stats) };
+    for catalog_name in catalog_names {
+        let Ok(Some(catalog)) = catalog_manager.catalog(&catalog_name) else { continue };

-        for schema_name in catalog.schema_names()? {
-            let schema = catalog
-                .schema(&schema_name)?
-                .context(error::SchemaNotFoundSnafu {
-                    catalog: &catalog_name,
-                    schema: &schema_name,
-                })?;
+        let Ok(schema_names) = catalog.schema_names() else { continue };
+        for schema_name in schema_names {
+            let Ok(Some(schema)) = catalog.schema(&schema_name) else { continue };

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

                let region_numbers = &table.table_info().meta.region_numbers;
                region_number += region_numbers.len() as u64;
+
+                match table.region_stats() {
+                    Ok(stats) => {
+                        let stats = stats.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()
+                        });
+
+                        region_stats.extend(stats);
+                    }
+                    Err(e) => {
+                        warn!("Failed to get region status, err: {:?}", e);
+                    }
+                };
            }
        }
    }
-    Ok(region_number)
+    (region_number, region_stats)
 }
--- a/src/catalog/src/system.rs
+++ b/src/catalog/src/system.rs
@@ -400,7 +400,7 @@ mod tests {
    use log_store::NoopLogStore;
    use mito::config::EngineConfig;
    use mito::engine::MitoEngine;
-    use object_store::{ObjectStore, ObjectStoreBuilder};
+    use object_store::ObjectStore;
    use storage::compaction::noop::NoopCompactionScheduler;
    use storage::config::EngineConfig as StorageEngineConfig;
    use storage::EngineImpl;
@@ -482,11 +482,9 @@ mod tests {
    pub async fn prepare_table_engine() -> (TempDir, TableEngineRef) {
        let dir = create_temp_dir("system-table-test");
        let store_dir = dir.path().to_string_lossy();
-        let accessor = object_store::services::Fs::default()
-            .root(&store_dir)
-            .build()
-            .unwrap();
-        let object_store = ObjectStore::new(accessor).finish();
+        let mut builder = object_store::services::Fs::default();
+        builder.root(&store_dir);
+        let object_store = ObjectStore::new(builder).unwrap().finish();
        let noop_compaction_scheduler = Arc::new(NoopCompactionScheduler::default());
        let table_engine = Arc::new(MitoEngine::new(
            EngineConfig::default(),
--- a/src/client/Cargo.toml
+++ b/src/client/Cargo.toml
@@ -23,7 +23,7 @@ enum_dispatch = "0.3"
 futures-util.workspace = true
 parking_lot = "0.12"
 prost.workspace = true
-rand = "0.8"
+rand.workspace = true
 snafu.workspace = true
 tonic.workspace = true

--- a/src/client/src/client.rs
+++ b/src/client/src/client.rs
@@ -14,6 +14,9 @@

 use std::sync::Arc;

+use api::v1::greptime_database_client::GreptimeDatabaseClient;
+use api::v1::health_check_client::HealthCheckClient;
+use api::v1::HealthCheckRequest;
 use arrow_flight::flight_service_client::FlightServiceClient;
 use common_grpc::channel_manager::ChannelManager;
 use parking_lot::RwLock;
@@ -23,6 +26,10 @@ use tonic::transport::Channel;
 use crate::load_balance::{LoadBalance, Loadbalancer};
 use crate::{error, Result};

+pub(crate) struct DatabaseClient {
+    pub(crate) inner: GreptimeDatabaseClient<Channel>,
+}
+
 pub(crate) struct FlightClient {
    addr: String,
    client: FlightServiceClient<Channel>,
@@ -118,7 +125,7 @@ impl Client {
        self.inner.set_peers(urls);
    }

-    pub(crate) fn make_client(&self) -> Result<FlightClient> {
+    fn find_channel(&self) -> Result<(String, Channel)> {
        let addr = self
            .inner
            .get_peer()
@@ -131,11 +138,30 @@ impl Client {
            .channel_manager
            .get(&addr)
            .context(error::CreateChannelSnafu { addr: &addr })?;
+        Ok((addr, channel))
+    }
+
+    pub(crate) fn make_flight_client(&self) -> Result<FlightClient> {
+        let (addr, channel) = self.find_channel()?;
        Ok(FlightClient {
            addr,
            client: FlightServiceClient::new(channel),
        })
    }
+
+    pub(crate) fn make_database_client(&self) -> Result<DatabaseClient> {
+        let (_, channel) = self.find_channel()?;
+        Ok(DatabaseClient {
+            inner: GreptimeDatabaseClient::new(channel),
+        })
+    }
+
+    pub async fn health_check(&self) -> Result<()> {
+        let (_, channel) = self.find_channel()?;
+        let mut client = HealthCheckClient::new(channel);
+        client.health_check(HealthCheckRequest {}).await?;
+        Ok(())
+    }
 }

 #[cfg(test)]
--- a/src/client/src/database.rs
+++ b/src/client/src/database.rs
@@ -12,15 +12,14 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.

-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, AuthHeader, CreateTableExpr, DdlRequest, DropTableExpr, GreptimeRequest,
-    InsertRequest, PromRangeQuery, QueryRequest, RequestHeader,
+    greptime_response, AffectedRows, AlterExpr, AuthHeader, CreateTableExpr, DdlRequest,
+    DropTableExpr, FlushTableExpr, GreptimeRequest, InsertRequest, PromRangeQuery, QueryRequest,
+    RequestHeader,
 };
 use arrow_flight::{FlightData, Ticket};
 use common_error::prelude::*;
@@ -31,28 +30,49 @@ use futures_util::{TryFutureExt, TryStreamExt};
 use prost::Message;
 use snafu::{ensure, ResultExt};

-use crate::error::{ConvertFlightDataSnafu, IllegalFlightMessagesSnafu};
+use crate::error::{
+    ConvertFlightDataSnafu, IllegalDatabaseResponseSnafu, IllegalFlightMessagesSnafu,
+};
 use crate::{error, Client, Result};

-#[derive(Clone, Debug)]
+#[derive(Clone, Debug, Default)]
 pub struct Database {
    // The "catalog" and "schema" to be used in processing the requests at the server side.
    // They are the "hint" or "context", just like how the "database" in "USE" statement is treated in MySQL.
    // They will be carried in the request header.
    catalog: String,
    schema: String,
+    // The dbname follows naming rule as out mysql, postgres and http
+    // protocol. The server treat dbname in priority of catalog/schema.
+    dbname: String,

    client: Client,
    ctx: FlightContext,
 }

 impl Database {
+    /// Create database service client using catalog and schema
    pub fn new(catalog: impl Into<String>, schema: impl Into<String>, client: Client) -> Self {
        Self {
            catalog: catalog.into(),
            schema: schema.into(),
            client,
-            ctx: FlightContext::default(),
+            ..Default::default()
+        }
+    }
+
+    /// Create database service client using dbname.
+    ///
+    /// This API is designed for external usage. `dbname` is:
+    ///
+    /// - the name of database when using GreptimeDB standalone or cluster
+    /// - the name provided by GreptimeCloud or other multi-tenant GreptimeDB
+    /// environment
+    pub fn new_with_dbname(dbname: impl Into<String>, client: Client) -> Self {
+        Self {
+            dbname: dbname.into(),
+            client,
+            ..Default::default()
        }
    }

@@ -72,14 +92,41 @@ impl Database {
        self.schema = schema.into();
    }

+    pub fn dbname(&self) -> &String {
+        &self.dbname
+    }
+
+    pub fn set_dbname(&mut self, dbname: impl Into<String>) {
+        self.dbname = dbname.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
+    pub async fn insert(&self, request: InsertRequest) -> Result<u32> {
+        let mut client = self.client.make_database_client()?.inner;
+        let request = GreptimeRequest {
+            header: Some(RequestHeader {
+                catalog: self.catalog.clone(),
+                schema: self.schema.clone(),
+                authorization: self.ctx.auth_header.clone(),
+                dbname: self.dbname.clone(),
+            }),
+            request: Some(Request::Insert(request)),
+        };
+        let response = client
+            .handle(request)
+            .await?
+            .into_inner()
+            .response
+            .context(IllegalDatabaseResponseSnafu {
+                err_msg: "GreptimeResponse is empty",
+            })?;
+        let greptime_response::Response::AffectedRows(AffectedRows { value }) = response;
+        Ok(value)
    }

    pub async fn sql(&self, sql: &str) -> Result<Output> {
@@ -135,12 +182,20 @@ 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(),
+                dbname: self.dbname.clone(),
            }),
            request: Some(request),
        };
@@ -148,7 +203,7 @@ impl Database {
            ticket: request.encode_to_vec().into(),
        };

-        let mut client = self.client.make_client()?;
+        let mut client = self.client.make_flight_client()?;

        // TODO(LFC): Streaming get flight data.
        let flight_data: Vec<FlightData> = client
@@ -157,22 +212,22 @@ impl Database {
            .and_then(|response| response.into_inner().try_collect())
            .await
            .map_err(|e| {
-                let code = get_metadata_value(&e, INNER_ERROR_CODE)
-                    .and_then(|s| StatusCode::from_str(&s).ok())
-                    .unwrap_or(StatusCode::Unknown);
-                let msg = get_metadata_value(&e, INNER_ERROR_MSG).unwrap_or(e.to_string());
-                error::ExternalSnafu { code, msg }
+                let tonic_code = e.code();
+                let e: error::Error = e.into();
+                let code = e.status_code();
+                let msg = e.to_string();
+                error::ServerSnafu { code, msg }
                    .fail::<()>()
                    .map_err(BoxedError::new)
                    .context(error::FlightGetSnafu {
-                        tonic_code: e.code(),
+                        tonic_code,
                        addr: client.addr(),
                    })
                    .map_err(|error| {
                        logging::error!(
                            "Failed to do Flight get, addr: {}, code: {}, source: {}",
                            client.addr(),
-                            e.code(),
+                            tonic_code,
                            error
                        );
                        error
@@ -203,12 +258,6 @@ impl Database {
    }
 }

-fn get_metadata_value(e: &tonic::Status, key: &str) -> Option<String> {
-    e.metadata()
-        .get(key)
-        .and_then(|v| String::from_utf8(v.as_bytes().to_vec()).ok())
-}
-
 #[derive(Default, Debug, Clone)]
 pub struct FlightContext {
    auth_header: Option<AuthHeader>,
--- a/src/client/src/error.rs
+++ b/src/client/src/error.rs
@@ -13,9 +13,10 @@
 // limitations under the License.

 use std::any::Any;
+use std::str::FromStr;

 use common_error::prelude::*;
-use tonic::Code;
+use tonic::{Code, Status};

 #[derive(Debug, Snafu)]
 #[snafu(visibility(pub))]
@@ -65,9 +66,12 @@ pub enum Error {
        source: common_grpc::error::Error,
    },

-    /// Error deserialized from gRPC metadata
+    // Server error carried in Tonic Status's metadata.
    #[snafu(display("{}", msg))]
-    ExternalError { code: StatusCode, msg: String },
+    Server { code: StatusCode, msg: String },
+
+    #[snafu(display("Illegal Database response: {err_msg}"))]
+    IllegalDatabaseResponse { err_msg: String },
 }

 pub type Result<T> = std::result::Result<T, Error>;
@@ -77,13 +81,15 @@ impl ErrorExt for Error {
        match self {
            Error::IllegalFlightMessages { .. }
            | Error::ColumnDataType { .. }
-            | Error::MissingField { .. } => StatusCode::Internal,
+            | Error::MissingField { .. }
+            | Error::IllegalDatabaseResponse { .. } => StatusCode::Internal,
+
+            Error::Server { code, .. } => *code,
            Error::FlightGet { source, .. } => source.status_code(),
            Error::CreateChannel { source, .. } | Error::ConvertFlightData { source } => {
                source.status_code()
            }
            Error::IllegalGrpcClientState { .. } => StatusCode::Unexpected,
-            Error::ExternalError { code, .. } => *code,
        }
    }

@@ -95,3 +101,21 @@ impl ErrorExt for Error {
        self
    }
 }
+
+impl From<Status> for Error {
+    fn from(e: Status) -> Self {
+        fn get_metadata_value(e: &Status, key: &str) -> Option<String> {
+            e.metadata()
+                .get(key)
+                .and_then(|v| String::from_utf8(v.as_bytes().to_vec()).ok())
+        }
+
+        let code = get_metadata_value(&e, INNER_ERROR_CODE)
+            .and_then(|s| StatusCode::from_str(&s).ok())
+            .unwrap_or(StatusCode::Unknown);
+
+        let msg = get_metadata_value(&e, INNER_ERROR_MSG).unwrap_or(e.to_string());
+
+        Self::Server { code, msg }
+    }
+}
--- a/src/cmd/src/bin/greptime.rs
+++ b/src/cmd/src/bin/greptime.rs
@@ -12,6 +12,8 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.

+#![doc = include_str!("../../../../README.md")]
+
 use std::fmt;

 use clap::Parser;
@@ -30,9 +32,39 @@ 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
    }
 }

@@ -51,13 +83,28 @@ enum SubCommand {
 }

 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::Cli(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))
+            }
        }
    }
 }
@@ -104,13 +151,18 @@ 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
@@ -17,10 +17,24 @@ mod helper;
 mod repl;

 use clap::Parser;
-use repl::Repl;
+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<()> {
+        Ok(())
+    }
+}
+
 #[derive(Parser)]
 pub struct Command {
    #[clap(subcommand)]
@@ -28,8 +42,8 @@ pub struct Command {
 }

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

@@ -39,9 +53,9 @@ enum SubCommand {
 }

 impl SubCommand {
-    async fn run(self) -> Result<()> {
+    async fn build(self) -> Result<Instance> {
        match self {
-            SubCommand::Attach(cmd) => cmd.run().await,
+            SubCommand::Attach(cmd) => cmd.build().await,
        }
    }
 }
@@ -57,8 +71,8 @@ pub(crate) struct AttachCommand {
 }

 impl AttachCommand {
-    async fn run(self) -> Result<()> {
-        let mut repl = Repl::try_new(&self).await?;
-        repl.run().await
+    async fn build(self) -> Result<Instance> {
+        let repl = Repl::try_new(&self).await?;
+        Ok(Instance { repl })
    }
 }
--- a/src/cmd/src/cli/repl.rs
+++ b/src/cmd/src/cli/repl.rs
@@ -50,7 +50,7 @@ use crate::error::{
 };

 /// Captures the state of the repl, gathers commands and executes them one by one
-pub(crate) struct Repl {
+pub struct Repl {
    /// Rustyline editor for interacting with user on command line
    rl: Editor<RustylineHelper>,

--- a/src/cmd/src/datanode.rs
+++ b/src/cmd/src/datanode.rs
@@ -21,9 +21,26 @@ use meta_client::MetaClientOptions;
 use servers::Mode;
 use snafu::ResultExt;

-use crate::error::{Error, MissingConfigSnafu, Result, StartDatanodeSnafu};
+use crate::error::{Error, MissingConfigSnafu, Result, ShutdownDatanodeSnafu, 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<()> {
+        self.datanode
+            .shutdown()
+            .await
+            .context(ShutdownDatanodeSnafu)
+    }
+}
+
 #[derive(Parser)]
 pub struct Command {
    #[clap(subcommand)]
@@ -31,8 +48,8 @@ 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
    }
 }

@@ -42,9 +59,9 @@ 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,
        }
    }
 }
@@ -72,19 +89,16 @@ struct StartCommand {
 }

 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 })
    }
 }

@@ -138,7 +152,6 @@ 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));
        }
--- a/src/cmd/src/error.rs
+++ b/src/cmd/src/error.rs
@@ -26,18 +26,42 @@ pub enum Error {
        source: datanode::error::Error,
    },

+    #[snafu(display("Failed to shutdown datanode, source: {}", source))]
+    ShutdownDatanode {
+        #[snafu(backtrace)]
+        source: datanode::error::Error,
+    },
+
    #[snafu(display("Failed to start frontend, source: {}", source))]
    StartFrontend {
        #[snafu(backtrace)]
        source: frontend::error::Error,
    },

+    #[snafu(display("Failed to shutdown frontend, source: {}", source))]
+    ShutdownFrontend {
+        #[snafu(backtrace)]
+        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)]
        source: meta_srv::error::Error,
    },

+    #[snafu(display("Failed to shutdown meta server, source: {}", source))]
+    ShutdownMetaServer {
+        #[snafu(backtrace)]
+        source: meta_srv::error::Error,
+    },
+
    #[snafu(display("Failed to read config file: {}, source: {}", path, source))]
    ReadConfig {
        path: String,
@@ -137,7 +161,11 @@ impl ErrorExt for Error {
        match self {
            Error::StartDatanode { source } => source.status_code(),
            Error::StartFrontend { source } => source.status_code(),
+            Error::ShutdownDatanode { source } => source.status_code(),
+            Error::ShutdownFrontend { source } => source.status_code(),
            Error::StartMetaServer { source } => source.status_code(),
+            Error::ShutdownMetaServer { source } => source.status_code(),
+            Error::BuildMetaServer { source } => source.status_code(),
            Error::UnsupportedSelectorType { source, .. } => source.status_code(),
            Error::ReadConfig { .. } | Error::ParseConfig { .. } | Error::MissingConfig { .. } => {
                StatusCode::InvalidArguments
--- a/src/cmd/src/frontend.rs
+++ b/src/cmd/src/frontend.rs
@@ -19,7 +19,7 @@ use common_base::Plugins;
 use frontend::frontend::FrontendOptions;
 use frontend::grpc::GrpcOptions;
 use frontend::influxdb::InfluxdbOptions;
-use frontend::instance::{FrontendInstance, Instance};
+use frontend::instance::{FrontendInstance, Instance as FeInstance};
 use frontend::mysql::MysqlOptions;
 use frontend::opentsdb::OpentsdbOptions;
 use frontend::postgres::PostgresOptions;
@@ -34,6 +34,26 @@ 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<()> {
+        self.frontend
+            .shutdown()
+            .await
+            .context(error::ShutdownFrontendSnafu)
+    }
+}
+
 #[derive(Parser)]
 pub struct Command {
    #[clap(subcommand)]
@@ -41,8 +61,8 @@ 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
    }
 }

@@ -52,9 +72,9 @@ 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,
        }
    }
 }
@@ -90,11 +110,11 @@ 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 mut instance = Instance::try_new_distributed(&opts, plugins.clone())
+        let mut instance = FeInstance::try_new_distributed(&opts, plugins.clone())
            .await
            .context(error::StartFrontendSnafu)?;

@@ -103,7 +123,7 @@ impl StartCommand {
            .await
            .context(error::StartFrontendSnafu)?;

-        instance.start().await.context(error::StartFrontendSnafu)
+        Ok(Instance { frontend: instance })
    }
 }

--- a/src/cmd/src/metasrv.rs
+++ b/src/cmd/src/metasrv.rs
@@ -14,13 +14,33 @@

 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)
+    }
+
+    pub async fn stop(&self) -> Result<()> {
+        self.instance
+            .shutdown()
+            .await
+            .context(error::ShutdownMetaServerSnafu)
+    }
+}
+
 #[derive(Parser)]
 pub struct Command {
    #[clap(subcommand)]
@@ -28,8 +48,8 @@ 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 +59,9 @@ 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 +83,17 @@ 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 })
    }
 }

--- a/src/cmd/src/standalone.rs
+++ b/src/cmd/src/standalone.rs
@@ -36,7 +36,10 @@ use servers::tls::{TlsMode, TlsOption};
 use servers::Mode;
 use snafu::ResultExt;

-use crate::error::{Error, IllegalConfigSnafu, Result, StartDatanodeSnafu, StartFrontendSnafu};
+use crate::error::{
+    Error, IllegalConfigSnafu, Result, ShutdownDatanodeSnafu, ShutdownFrontendSnafu,
+    StartDatanodeSnafu, StartFrontendSnafu,
+};
 use crate::frontend::load_frontend_plugins;
 use crate::toml_loader;

@@ -47,8 +50,8 @@ 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
    }
 }

@@ -58,9 +61,9 @@ 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,
        }
    }
 }
@@ -133,6 +136,40 @@ impl StandaloneOptions {
    }
 }

+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<()> {
+        self.frontend
+            .shutdown()
+            .await
+            .context(ShutdownFrontendSnafu)?;
+
+        self.datanode
+            .shutdown_instance()
+            .await
+            .context(ShutdownDatanodeSnafu)?;
+        info!("Datanode instance stopped.");
+
+        Ok(())
+    }
+}
+
 #[derive(Debug, Parser)]
 struct StartCommand {
    #[clap(long)]
@@ -164,7 +201,7 @@ 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)?);
@@ -184,25 +221,18 @@ 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(plugins.clone(), datanode.get_instance()).await?;

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

        frontend
            .build_servers(&fe_opts, plugins)
            .await
            .context(StartFrontendSnafu)?;

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

--- a/src/common/datasource/Cargo.toml
+++ b/src/common/datasource/Cargo.toml
@@ -0,0 +1,13 @@
+[package]
+name = "common-datasource"
+version.workspace = true
+edition.workspace = true
+license.workspace = true
+
+[dependencies]
+common-error = { path = "../error" }
+futures.workspace = true
+object-store = { path = "../../object-store" }
+regex = "1.7"
+snafu.workspace = true
+url = "2.3"
--- a/src/common/datasource/src/error.rs
+++ b/src/common/datasource/src/error.rs
@@ -0,0 +1,75 @@
+// 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 common_error::prelude::*;
+use url::ParseError;
+
+#[derive(Debug, Snafu)]
+#[snafu(visibility(pub))]
+pub enum Error {
+    #[snafu(display("Unsupported backend protocol: {}", protocol))]
+    UnsupportedBackendProtocol { protocol: String },
+
+    #[snafu(display("empty host: {}", url))]
+    EmptyHostPath { url: String },
+
+    #[snafu(display("Invalid path: {}", path))]
+    InvalidPath { path: String },
+
+    #[snafu(display("Invalid url: {}, error :{}", url, source))]
+    InvalidUrl { url: String, source: ParseError },
+
+    #[snafu(display("Failed to build backend, source: {}", source))]
+    BuildBackend {
+        source: object_store::Error,
+        backtrace: Backtrace,
+    },
+
+    #[snafu(display("Failed to list object in path: {}, source: {}", path, source))]
+    ListObjects {
+        path: String,
+        backtrace: Backtrace,
+        source: object_store::Error,
+    },
+
+    #[snafu(display("Invalid connection: {}", msg))]
+    InvalidConnection { msg: String },
+}
+
+pub type Result<T> = std::result::Result<T, Error>;
+
+impl ErrorExt for Error {
+    fn status_code(&self) -> StatusCode {
+        use Error::*;
+        match self {
+            BuildBackend { .. } | ListObjects { .. } => StatusCode::StorageUnavailable,
+
+            UnsupportedBackendProtocol { .. }
+            | InvalidConnection { .. }
+            | InvalidUrl { .. }
+            | EmptyHostPath { .. }
+            | InvalidPath { .. } => StatusCode::InvalidArguments,
+        }
+    }
+
+    fn backtrace_opt(&self) -> Option<&Backtrace> {
+        ErrorCompat::backtrace(self)
+    }
+
+    fn as_any(&self) -> &dyn Any {
+        self
+    }
+}
--- a/src/promql/src/engine/functions.rs
+++ b/src/promql/src/engine/functions.rs
@@ -12,4 +12,7 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.

-//! PromQL functions
+pub mod error;
+pub mod lister;
+pub mod object_store;
+pub mod util;
--- a/src/common/datasource/src/lister.rs
+++ b/src/common/datasource/src/lister.rs
@@ -0,0 +1,83 @@
+// 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 futures::{future, TryStreamExt};
+use object_store::{Entry, ObjectStore};
+use regex::Regex;
+use snafu::ResultExt;
+
+use crate::error::{self, Result};
+#[derive(Debug, Clone, PartialEq, Eq)]
+pub enum Source {
+    Filename(String),
+    Dir,
+}
+
+pub struct Lister {
+    object_store: ObjectStore,
+    source: Source,
+    path: String,
+    regex: Option<Regex>,
+}
+
+impl Lister {
+    pub fn new(
+        object_store: ObjectStore,
+        source: Source,
+        path: String,
+        regex: Option<Regex>,
+    ) -> Self {
+        Lister {
+            object_store,
+            source,
+            path,
+            regex,
+        }
+    }
+
+    pub async fn list(&self) -> Result<Vec<Entry>> {
+        match &self.source {
+            Source::Dir => {
+                let streamer = self
+                    .object_store
+                    .list(&self.path)
+                    .await
+                    .context(error::ListObjectsSnafu { path: &self.path })?;
+
+                streamer
+                    .try_filter(|f| {
+                        let res = self
+                            .regex
+                            .as_ref()
+                            .map(|x| x.is_match(f.name()))
+                            .unwrap_or(true);
+                        future::ready(res)
+                    })
+                    .try_collect::<Vec<_>>()
+                    .await
+                    .context(error::ListObjectsSnafu { path: &self.path })
+            }
+            Source::Filename(filename) => {
+                // make sure this file exists
+                let file_full_path = format!("{}{}", self.path, filename);
+                let _ = self.object_store.stat(&file_full_path).await.context(
+                    error::ListObjectsSnafu {
+                        path: &file_full_path,
+                    },
+                )?;
+                Ok(vec![Entry::new(&file_full_path)])
+            }
+        }
+    }
+}
--- a/src/common/datasource/src/object_store.rs
+++ b/src/common/datasource/src/object_store.rs
@@ -0,0 +1,60 @@
+// 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 fs;
+pub mod s3;
+use std::collections::HashMap;
+
+use object_store::ObjectStore;
+use snafu::{OptionExt, ResultExt};
+use url::{ParseError, Url};
+
+use self::fs::build_fs_backend;
+use self::s3::build_s3_backend;
+use crate::error::{self, Result};
+
+pub const FS_SCHEMA: &str = "FS";
+pub const S3_SCHEMA: &str = "S3";
+
+/// parse url returns (schema,Option<host>,path)
+pub fn parse_url(url: &str) -> Result<(String, Option<String>, String)> {
+    let parsed_url = Url::parse(url);
+    match parsed_url {
+        Ok(url) => Ok((
+            url.scheme().to_string(),
+            url.host_str().map(|s| s.to_string()),
+            url.path().to_string(),
+        )),
+        Err(ParseError::RelativeUrlWithoutBase) => {
+            Ok((FS_SCHEMA.to_string(), None, url.to_string()))
+        }
+        Err(err) => Err(err).context(error::InvalidUrlSnafu { url }),
+    }
+}
+
+pub fn build_backend(url: &str, connection: HashMap<String, String>) -> Result<ObjectStore> {
+    let (schema, host, _path) = parse_url(url)?;
+
+    match schema.to_uppercase().as_str() {
+        S3_SCHEMA => {
+            let host = host.context(error::EmptyHostPathSnafu {
+                url: url.to_string(),
+            })?;
+            Ok(build_s3_backend(&host, "/", connection)?)
+        }
+        FS_SCHEMA => Ok(build_fs_backend("/")?),
+
+        _ => error::UnsupportedBackendProtocolSnafu { protocol: schema }.fail(),
+    }
+}
--- a/src/common/datasource/src/object_store/fs.rs
+++ b/src/common/datasource/src/object_store/fs.rs
@@ -0,0 +1,28 @@
+// 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 object_store::services::Fs;
+use object_store::ObjectStore;
+use snafu::ResultExt;
+
+use crate::error::{BuildBackendSnafu, Result};
+
+pub fn build_fs_backend(root: &str) -> Result<ObjectStore> {
+    let mut builder = Fs::default();
+    builder.root(root);
+    let object_store = ObjectStore::new(builder)
+        .context(BuildBackendSnafu)?
+        .finish();
+    Ok(object_store)
+}
--- a/src/common/datasource/src/object_store/s3.rs
+++ b/src/common/datasource/src/object_store/s3.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.
+
+use std::collections::HashMap;
+
+use object_store::services::S3;
+use object_store::ObjectStore;
+use snafu::ResultExt;
+
+use crate::error::{self, Result};
+
+const ENDPOINT_URL: &str = "ENDPOINT_URL";
+const ACCESS_KEY_ID: &str = "ACCESS_KEY_ID";
+const SECRET_ACCESS_KEY: &str = "SECRET_ACCESS_KEY";
+const SESSION_TOKEN: &str = "SESSION_TOKEN";
+const REGION: &str = "REGION";
+const ENABLE_VIRTUAL_HOST_STYLE: &str = "ENABLE_VIRTUAL_HOST_STYLE";
+
+pub fn build_s3_backend(
+    host: &str,
+    path: &str,
+    connection: HashMap<String, String>,
+) -> Result<ObjectStore> {
+    let mut builder = S3::default();
+
+    builder.root(path);
+
+    builder.bucket(host);
+
+    if let Some(endpoint) = connection.get(ENDPOINT_URL) {
+        builder.endpoint(endpoint);
+    }
+
+    if let Some(region) = connection.get(REGION) {
+        builder.region(region);
+    }
+
+    if let Some(key_id) = connection.get(ACCESS_KEY_ID) {
+        builder.access_key_id(key_id);
+    }
+
+    if let Some(key) = connection.get(SECRET_ACCESS_KEY) {
+        builder.secret_access_key(key);
+    }
+
+    if let Some(session_token) = connection.get(SESSION_TOKEN) {
+        builder.security_token(session_token);
+    }
+
+    if let Some(enable_str) = connection.get(ENABLE_VIRTUAL_HOST_STYLE) {
+        let enable = enable_str.as_str().parse::<bool>().map_err(|e| {
+            error::InvalidConnectionSnafu {
+                msg: format!(
+                    "failed to parse the option {}={}, {}",
+                    ENABLE_VIRTUAL_HOST_STYLE, enable_str, e
+                ),
+            }
+            .build()
+        })?;
+        if enable {
+            builder.enable_virtual_host_style();
+        }
+    }
+
+    Ok(ObjectStore::new(builder)
+        .context(error::BuildBackendSnafu)?
+        .finish())
+}
--- a/src/common/datasource/src/util.rs
+++ b/src/common/datasource/src/util.rs
@@ -0,0 +1,125 @@
+// 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 fn find_dir_and_filename(path: &str) -> (String, Option<String>) {
+    if path.is_empty() {
+        ("/".to_string(), None)
+    } else if path.ends_with('/') {
+        (path.to_string(), None)
+    } else if let Some(idx) = path.rfind('/') {
+        (
+            path[..idx + 1].to_string(),
+            Some(path[idx + 1..].to_string()),
+        )
+    } else {
+        ("/".to_string(), Some(path.to_string()))
+    }
+}
+
+#[cfg(test)]
+mod tests {
+
+    use url::Url;
+
+    use super::*;
+
+    #[test]
+    fn test_parse_uri() {
+        struct Test<'a> {
+            uri: &'a str,
+            expected_path: &'a str,
+            expected_schema: &'a str,
+        }
+
+        let tests = [
+            Test {
+                uri: "s3://bucket/to/path/",
+                expected_path: "/to/path/",
+                expected_schema: "s3",
+            },
+            Test {
+                uri: "fs:///to/path/",
+                expected_path: "/to/path/",
+                expected_schema: "fs",
+            },
+            Test {
+                uri: "fs:///to/path/file",
+                expected_path: "/to/path/file",
+                expected_schema: "fs",
+            },
+        ];
+        for test in tests {
+            let parsed_uri = Url::parse(test.uri).unwrap();
+            assert_eq!(parsed_uri.path(), test.expected_path);
+            assert_eq!(parsed_uri.scheme(), test.expected_schema);
+        }
+    }
+
+    #[test]
+    fn test_parse_path_and_dir() {
+        let parsed = Url::from_file_path("/to/path/file").unwrap();
+        assert_eq!(parsed.path(), "/to/path/file");
+
+        let parsed = Url::from_directory_path("/to/path/").unwrap();
+        assert_eq!(parsed.path(), "/to/path/");
+    }
+
+    #[test]
+    fn test_find_dir_and_filename() {
+        struct Test<'a> {
+            path: &'a str,
+            expected_dir: &'a str,
+            expected_filename: Option<String>,
+        }
+
+        let tests = [
+            Test {
+                path: "to/path/",
+                expected_dir: "to/path/",
+                expected_filename: None,
+            },
+            Test {
+                path: "to/path/filename",
+                expected_dir: "to/path/",
+                expected_filename: Some("filename".into()),
+            },
+            Test {
+                path: "/to/path/filename",
+                expected_dir: "/to/path/",
+                expected_filename: Some("filename".into()),
+            },
+            Test {
+                path: "/",
+                expected_dir: "/",
+                expected_filename: None,
+            },
+            Test {
+                path: "filename",
+                expected_dir: "/",
+                expected_filename: Some("filename".into()),
+            },
+            Test {
+                path: "",
+                expected_dir: "/",
+                expected_filename: None,
+            },
+        ];
+
+        for test in tests {
+            let (path, filename) = find_dir_and_filename(test.path);
+            assert_eq!(test.expected_dir, path);
+            assert_eq!(test.expected_filename, filename)
+        }
+    }
+}
--- a/src/common/function/src/scalars/timestamp.rs
+++ b/src/common/function/src/scalars/timestamp.rs
@@ -11,11 +11,10 @@
 // WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 // See the License for the specific language governing permissions and
 // limitations under the License.
-
 use std::sync::Arc;
-mod from_unixtime;
+mod to_unixtime;

-use from_unixtime::FromUnixtimeFunction;
+use to_unixtime::ToUnixtimeFunction;

 use crate::scalars::function_registry::FunctionRegistry;

@@ -23,6 +22,6 @@ pub(crate) struct TimestampFunction;

 impl TimestampFunction {
    pub fn register(registry: &FunctionRegistry) {
-        registry.register(Arc::new(FromUnixtimeFunction::default()));
+        registry.register(Arc::new(ToUnixtimeFunction::default()));
    }
 }
--- 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/timestamp/to_unixtime.rs
+++ b/src/common/function/src/scalars/timestamp/to_unixtime.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.
+
+use std::fmt;
+use std::str::FromStr;
+use std::sync::Arc;
+
+use common_query::error::{self, Result, UnsupportedInputDataTypeSnafu};
+use common_query::prelude::{Signature, Volatility};
+use common_time::timestamp::TimeUnit;
+use common_time::Timestamp;
+use datatypes::prelude::ConcreteDataType;
+use datatypes::types::StringType;
+use datatypes::vectors::{Int64Vector, StringVector, Vector, VectorRef};
+use snafu::ensure;
+
+use crate::scalars::function::{Function, FunctionContext};
+
+#[derive(Clone, Debug, Default)]
+pub struct ToUnixtimeFunction;
+
+const NAME: &str = "to_unixtime";
+
+fn convert_to_seconds(arg: &str) -> Option<i64> {
+    match Timestamp::from_str(arg) {
+        Ok(ts) => {
+            let sec_mul = (TimeUnit::Second.factor() / ts.unit().factor()) as i64;
+            Some(ts.value().div_euclid(sec_mul))
+        }
+        Err(_err) => None,
+    }
+}
+
+impl Function for ToUnixtimeFunction {
+    fn name(&self) -> &str {
+        NAME
+    }
+
+    fn return_type(&self, _input_types: &[ConcreteDataType]) -> Result<ConcreteDataType> {
+        Ok(ConcreteDataType::timestamp_second_datatype())
+    }
+
+    fn signature(&self) -> Signature {
+        Signature::exact(
+            vec![ConcreteDataType::String(StringType)],
+            Volatility::Immutable,
+        )
+    }
+
+    fn eval(&self, _func_ctx: FunctionContext, columns: &[VectorRef]) -> Result<VectorRef> {
+        ensure!(
+            columns.len() == 1,
+            error::InvalidFuncArgsSnafu {
+                err_msg: format!(
+                    "The length of the args is not correct, expect exactly one, have: {}",
+                    columns.len()
+                ),
+            }
+        );
+
+        match columns[0].data_type() {
+            ConcreteDataType::String(_) => {
+                let array = columns[0].to_arrow_array();
+                let vector = StringVector::try_from_arrow_array(&array).unwrap();
+                Ok(Arc::new(Int64Vector::from(
+                    (0..vector.len())
+                        .map(|i| convert_to_seconds(&vector.get(i).to_string()))
+                        .collect::<Vec<_>>(),
+                )))
+            }
+            _ => UnsupportedInputDataTypeSnafu {
+                function: NAME,
+                datatypes: columns.iter().map(|c| c.data_type()).collect::<Vec<_>>(),
+            }
+            .fail(),
+        }
+    }
+}
+
+impl fmt::Display for ToUnixtimeFunction {
+    fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
+        write!(f, "TO_UNIXTIME")
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use common_query::prelude::TypeSignature;
+    use datatypes::prelude::ConcreteDataType;
+    use datatypes::types::StringType;
+    use datatypes::value::Value;
+    use datatypes::vectors::StringVector;
+
+    use super::{ToUnixtimeFunction, *};
+    use crate::scalars::Function;
+
+    #[test]
+    fn test_to_unixtime() {
+        let f = ToUnixtimeFunction::default();
+        assert_eq!("to_unixtime", f.name());
+        assert_eq!(
+            ConcreteDataType::timestamp_second_datatype(),
+            f.return_type(&[]).unwrap()
+        );
+
+        assert!(matches!(f.signature(),
+                         Signature {
+                            type_signature: TypeSignature::Exact(valid_types),
+                            volatility: Volatility::Immutable
+                         } if  valid_types == vec![ConcreteDataType::String(StringType)]
+        ));
+
+        let times = vec![
+            Some("2023-03-01T06:35:02Z"),
+            None,
+            Some("2022-06-30T23:59:60Z"),
+            Some("invalid_time_stamp"),
+        ];
+        let results = vec![Some(1677652502), None, Some(1656633600), None];
+        let args: Vec<VectorRef> = vec![Arc::new(StringVector::from(times.clone()))];
+        let vector = f.eval(FunctionContext::default(), &args).unwrap();
+        assert_eq!(4, vector.len());
+        for (i, _t) in times.iter().enumerate() {
+            let v = vector.get(i);
+            if i == 1 || i == 3 {
+                assert_eq!(Value::Null, v);
+                continue;
+            }
+            match v {
+                Value::Int64(ts) => {
+                    assert_eq!(ts, (*results.get(i).unwrap()).unwrap());
+                }
+                _ => unreachable!(),
+            }
+        }
+    }
+}
--- a/src/common/grpc-expr/src/error.rs
+++ b/src/common/grpc-expr/src/error.rs
@@ -32,7 +32,7 @@ pub enum Error {
    DecodeInsert { source: DecodeError },

    #[snafu(display("Illegal insert data"))]
-    IllegalInsertData,
+    IllegalInsertData { backtrace: Backtrace },

    #[snafu(display("Column datatype error, source: {}", source))]
    ColumnDataType {
--- a/src/common/grpc/Cargo.toml
+++ b/src/common/grpc/Cargo.toml
@@ -26,7 +26,7 @@ tower = "0.4"

 [dev-dependencies]
 criterion = "0.4"
-rand = "0.8"
+rand.workspace = true

 [[bench]]
 name = "bench_main"
--- a/src/common/procedure/Cargo.toml
+++ b/src/common/procedure/Cargo.toml
@@ -6,6 +6,7 @@ license.workspace = true

 [dependencies]
 async-trait.workspace = true
+async-stream.workspace = true
 common-error = { path = "../error" }
 common-runtime = { path = "../runtime" }
 common-telemetry = { path = "../telemetry" }
@@ -14,6 +15,7 @@ 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
--- a/src/common/procedure/src/error.rs
+++ b/src/common/procedure/src/error.rs
@@ -97,6 +97,16 @@ pub enum Error {
        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>;
@@ -111,6 +121,7 @@ impl ErrorExt for Error {
            | Error::ListState { .. }
            | Error::ReadState { .. }
            | Error::FromJson { .. }
+            | Error::RetryTimesExceeded { .. }
            | Error::RetryLater { .. }
            | Error::WaitWatcher { .. } => StatusCode::Internal,
            Error::LoaderConflict { .. } | Error::DuplicateProcedure { .. } => {
--- a/src/common/procedure/src/local.rs
+++ b/src/common/procedure/src/local.rs
@@ -17,8 +17,10 @@ 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;
@@ -291,12 +293,16 @@ impl ManagerContext {
 pub struct ManagerConfig {
    /// Object store
    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 {
@@ -305,6 +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,
        }
    }

@@ -321,7 +329,11 @@ impl LocalManager {
            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();
@@ -411,7 +423,6 @@ impl ProcedureManager for LocalManager {
 mod test_util {
    use common_test_util::temp_dir::TempDir;
    use object_store::services::Fs as Builder;
-    use object_store::ObjectStoreBuilder;

    use super::*;

@@ -421,8 +432,9 @@ mod test_util {

    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).finish()
+        let mut builder = Builder::default();
+        builder.root(store_dir);
+        ObjectStore::new(builder).unwrap().finish()
    }
 }

@@ -543,6 +555,8 @@ mod tests {
        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);

@@ -562,6 +576,8 @@ mod tests {
        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);

@@ -606,6 +622,8 @@ mod tests {
        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);

@@ -652,6 +670,8 @@ mod tests {
        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);

--- a/src/common/procedure/src/local/runner.rs
+++ b/src/common/procedure/src/local/runner.rs
@@ -15,15 +15,15 @@
 use std::sync::Arc;
 use std::time::Duration;

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

 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 {
@@ -108,7 +108,9 @@ 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 {
@@ -164,18 +166,56 @@ impl Runner {
            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 {
-                ExecResult::Continue => (),
+            match self.execute_once(ctx).await {
                ExecResult::Done | ExecResult::Failed => return,
+                ExecResult::Continue => (),
                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;
+                    }
                }
            }
        }
    }

+    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!(
@@ -186,8 +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 {
@@ -196,7 +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;
                        }

@@ -217,17 +261,12 @@ impl Runner {
                );

                if e.is_retry_later() {
+                    self.meta.set_state(ProcedureState::retrying(Arc::new(e)));
                    return ExecResult::RetryLater;
                }

-                self.meta.set_state(ProcedureState::failed(Arc::new(e)));
-
                // Write rollback key so we can skip this procedure while recovering procedures.
-                if self.rollback_procedure().await.is_err() {
-                    return ExecResult::RetryLater;
-                }
-
-                ExecResult::Failed
+                self.rollback(Arc::new(e)).await
            }
        }
    }
@@ -261,7 +300,9 @@ impl Runner {
            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
@@ -285,8 +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>) {
@@ -416,14 +465,15 @@ mod tests {
            procedure,
            manager_ctx: Arc::new(ManagerContext::new()),
            step: 0,
+            exponential_builder: ExponentialBuilder::default(),
            store,
+            rolling_back: false,
        }
    }

    async fn check_files(object_store: &ObjectStore, procedure_id: ProcedureId, files: &[&str]) {
        let dir = format!("{procedure_id}/");
-        let object = object_store.object(&dir);
-        let lister = object.list().await.unwrap();
+        let lister = object_store.list(&dir).await.unwrap();
        let mut files_in_dir: Vec<_> = lister
            .map_ok(|de| de.name().to_string())
            .try_collect()
@@ -744,7 +794,7 @@ mod tests {

        let res = runner.execute_once(&ctx).await;
        assert!(res.is_retry_later(), "{res:?}");
-        assert!(meta.state().is_running());
+        assert!(meta.state().is_retrying());

        let res = runner.execute_once(&ctx).await;
        assert!(res.is_done(), "{res:?}");
@@ -752,6 +802,36 @@ mod tests {
        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;
@@ -819,7 +899,7 @@ mod tests {
        // Replace the manager ctx.
        runner.manager_ctx = manager_ctx;

-        // Run the runer and execute the procedure.
+        // 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
@@ -206,6 +206,8 @@ pub enum ProcedureState {
    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 { error: Arc<Error> },
 }
@@ -216,6 +218,11 @@ impl 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)
@@ -231,10 +238,16 @@ impl ProcedureState {
        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,
        }
    }
--- a/src/common/procedure/src/store.rs
+++ b/src/common/procedure/src/store.rs
@@ -248,15 +248,15 @@ mod tests {
    use async_trait::async_trait;
    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};

    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).finish();
+        let mut builder = Builder::default();
+        builder.root(store_dir);
+        let object_store = ObjectStore::new(builder).unwrap().finish();

        ProcedureStore::from(object_store)
    }
--- a/src/common/procedure/src/store/state_store.rs
+++ b/src/common/procedure/src/store/state_store.rs
@@ -15,12 +15,13 @@
 use std::pin::Pin;
 use std::sync::Arc;

+use async_stream::try_stream;
 use async_trait::async_trait;
-use futures::{Stream, TryStreamExt};
-use object_store::{ObjectMode, ObjectStore};
+use futures::{Stream, StreamExt};
+use object_store::{EntryMode, Metakey, ObjectStore};
 use snafu::ResultExt;

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

 /// Key value from state store.
 type KeyValue = (String, Vec<u8>);
@@ -64,49 +65,49 @@ impl ObjectStateStore {
 #[async_trait]
 impl StateStore for ObjectStateStore {
    async fn put(&self, key: &str, value: Vec<u8>) -> Result<()> {
-        let object = self.store.object(key);
-        object.write(value).await.context(PutStateSnafu { key })
+        self.store
+            .write(key, value)
+            .await
+            .context(PutStateSnafu { key })
    }

    async fn walk_top_down(&self, path: &str) -> Result<KeyValueStream> {
        let path_string = path.to_string();

-        let lister = self
-            .store
-            .object(path)
-            .scan()
-            .await
-            .map_err(|e| Error::ListState {
-                path: path_string.clone(),
-                source: e,
-            })?;
+        let mut lister = self.store.scan(path).await.map_err(|e| Error::ListState {
+            path: path_string.clone(),
+            source: e,
+        })?;

-        let stream = lister
-            .try_filter_map(|entry| async move {
+        let store = self.store.clone();
+
+        let stream = try_stream!({
+            while let Some(res) = lister.next().await {
+                let entry = res.context(ListStateSnafu { path: &path_string })?;
                let key = entry.path();
-                let key_value = match entry.mode().await? {
-                    ObjectMode::FILE => {
-                        let value = entry.read().await?;
-
-                        Some((key.to_string(), value))
-                    }
-                    ObjectMode::DIR | ObjectMode::Unknown => None,
-                };
-
-                Ok(key_value)
-            })
-            .map_err(move |e| Error::ListState {
-                path: path_string.clone(),
-                source: e,
-            });
+                let metadata = store
+                    .metadata(&entry, Metakey::Mode)
+                    .await
+                    .context(ListStateSnafu { path: key })?;
+                if let EntryMode::FILE = metadata.mode() {
+                    let value = store
+                        .read(key)
+                        .await
+                        .context(ListStateSnafu { path: key })?;
+                    yield (key.to_string(), value);
+                }
+            }
+        });

        Ok(Box::pin(stream))
    }

    async fn delete(&self, keys: &[String]) -> Result<()> {
        for key in keys {
-            let object = self.store.object(key);
-            object.delete().await.context(DeleteStateSnafu { key })?;
+            self.store
+                .delete(key)
+                .await
+                .context(DeleteStateSnafu { key })?;
        }

        Ok(())
@@ -116,8 +117,8 @@ impl StateStore for ObjectStateStore {
 #[cfg(test)]
 mod tests {
    use common_test_util::temp_dir::create_temp_dir;
+    use futures_util::TryStreamExt;
    use object_store::services::Fs as Builder;
-    use object_store::ObjectStoreBuilder;

    use super::*;

@@ -125,8 +126,10 @@ mod tests {
    async fn test_object_state_store() {
        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).finish();
+        let mut builder = Builder::default();
+        builder.root(store_dir);
+
+        let object_store = ObjectStore::new(builder).unwrap().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
@@ -33,6 +33,9 @@ pub async fn wait(watcher: &mut Watcher) -> Result<()> {
            ProcedureState::Failed { error } => {
                return Err(error.clone()).context(ProcedureExecSnafu);
            }
+            ProcedureState::Retrying { error } => {
+                return Err(error.clone()).context(ProcedureExecSnafu);
+            }
        }
    }
 }
--- a/src/common/recordbatch/src/error.rs
+++ b/src/common/recordbatch/src/error.rs
@@ -17,6 +17,7 @@ use std::any::Any;

 use common_error::ext::BoxedError;
 use common_error::prelude::*;
+use datatypes::prelude::ConcreteDataType;

 pub type Result<T> = std::result::Result<T, Error>;

@@ -70,6 +71,26 @@ pub enum Error {
        source: datafusion_common::DataFusionError,
        backtrace: Backtrace,
    },
+
+    #[snafu(display("Column {} not exists in table {}", column_name, table_name))]
+    ColumnNotExists {
+        column_name: String,
+        table_name: String,
+        backtrace: Backtrace,
+    },
+
+    #[snafu(display(
+        "Failed to cast vector of type '{:?}' to type '{:?}', source: {}",
+        from_type,
+        to_type,
+        source
+    ))]
+    CastVector {
+        from_type: ConcreteDataType,
+        to_type: ConcreteDataType,
+        #[snafu(backtrace)]
+        source: datatypes::error::Error,
+    },
 }

 impl ErrorExt for Error {
@@ -81,11 +102,14 @@ impl ErrorExt for Error {
            | Error::CreateRecordBatches { .. }
            | Error::PollStream { .. }
            | Error::Format { .. }
-            | Error::InitRecordbatchStream { .. } => StatusCode::Internal,
+            | Error::InitRecordbatchStream { .. }
+            | Error::ColumnNotExists { .. } => StatusCode::Internal,

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

-            Error::SchemaConversion { source, .. } => source.status_code(),
+            Error::SchemaConversion { source, .. } | Error::CastVector { source, .. } => {
+                source.status_code()
+            }
        }
    }

--- a/src/common/recordbatch/src/recordbatch.rs
+++ b/src/common/recordbatch/src/recordbatch.rs
@@ -12,14 +12,16 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.

+use std::collections::HashMap;
+
 use datatypes::schema::SchemaRef;
 use datatypes::value::Value;
 use datatypes::vectors::{Helper, VectorRef};
 use serde::ser::{Error, SerializeStruct};
 use serde::{Serialize, Serializer};
-use snafu::ResultExt;
+use snafu::{OptionExt, ResultExt};

-use crate::error::{self, Result};
+use crate::error::{self, CastVectorSnafu, ColumnNotExistsSnafu, Result};
 use crate::DfRecordBatch;

 /// A two-dimensional batch of column-oriented data with a defined schema.
@@ -108,6 +110,41 @@ impl RecordBatch {
    pub fn rows(&self) -> RecordBatchRowIterator<'_> {
        RecordBatchRowIterator::new(self)
    }
+
+    pub fn column_vectors(
+        &self,
+        table_name: &str,
+        table_schema: SchemaRef,
+    ) -> Result<HashMap<String, VectorRef>> {
+        let mut vectors = HashMap::with_capacity(self.num_columns());
+
+        // column schemas in recordbatch must match its vectors, otherwise it's corrupted
+        for (vector_schema, vector) in self.schema.column_schemas().iter().zip(self.columns.iter())
+        {
+            let column_name = &vector_schema.name;
+            let column_schema =
+                table_schema
+                    .column_schema_by_name(column_name)
+                    .context(ColumnNotExistsSnafu {
+                        table_name,
+                        column_name,
+                    })?;
+            let vector = if vector_schema.data_type != column_schema.data_type {
+                vector
+                    .cast(&column_schema.data_type)
+                    .with_context(|_| CastVectorSnafu {
+                        from_type: vector.data_type(),
+                        to_type: column_schema.data_type.clone(),
+                    })?
+            } else {
+                vector.clone()
+            };
+
+            vectors.insert(column_name.clone(), vector);
+        }
+
+        Ok(vectors)
+    }
 }

 impl Serialize for RecordBatch {
--- a/src/common/time/Cargo.toml
+++ b/src/common/time/Cargo.toml
@@ -12,4 +12,4 @@ serde_json = "1.0"
 snafu = { version = "0.7", features = ["backtraces"] }

 [dev-dependencies]
-rand = "0.8"
+rand.workspace = true
--- a/src/common/time/src/lib.rs
+++ b/src/common/time/src/lib.rs
@@ -1,4 +1,3 @@
-#![feature(int_roundings)]
 // Copyright 2023 Greptime Team
 //
 // Licensed under the Apache License, Version 2.0 (the "License");
--- a/src/common/time/src/timestamp.rs
+++ b/src/common/time/src/timestamp.rs
@@ -26,6 +26,7 @@ use snafu::{OptionExt, ResultExt};

 use crate::error;
 use crate::error::{ArithmeticOverflowSnafu, Error, ParseTimestampSnafu, TimestampOverflowSnafu};
+use crate::util::div_ceil;

 #[derive(Debug, Clone, Default, Copy, Serialize, Deserialize)]
 pub struct Timestamp {
@@ -143,7 +144,7 @@ impl Timestamp {
            Some(Timestamp::new(value, unit))
        } else {
            let mul = unit.factor() / self.unit().factor();
-            Some(Timestamp::new(self.value.div_ceil(mul as i64), unit))
+            Some(Timestamp::new(div_ceil(self.value, mul as i64), unit))
        }
    }

--- a/src/common/time/src/util.rs
+++ b/src/common/time/src/util.rs
@@ -17,6 +17,17 @@ pub fn current_time_millis() -> i64 {
    chrono::Utc::now().timestamp_millis()
 }

+/// Port of rust unstable features `int_roundings`.
+pub(crate) fn div_ceil(this: i64, rhs: i64) -> i64 {
+    let d = this / rhs;
+    let r = this % rhs;
+    if r > 0 && rhs > 0 {
+        d + 1
+    } else {
+        d
+    }
+}
+
 #[cfg(test)]
 mod tests {
    use std::time::{self, SystemTime};
@@ -42,4 +53,10 @@ mod tests {
        assert_eq!(datetime_std.hour(), datetime_now.hour());
        assert_eq!(datetime_std.minute(), datetime_now.minute());
    }
+
+    #[test]
+    fn test_div_ceil() {
+        let v0 = 9223372036854676001;
+        assert_eq!(9223372036854677, div_ceil(v0, 1000));
+    }
 }
--- a/src/datanode/Cargo.toml
+++ b/src/datanode/Cargo.toml
@@ -20,6 +20,8 @@ catalog = { path = "../catalog" }
 common-base = { path = "../common/base" }
 common-catalog = { path = "../common/catalog" }
 common-error = { path = "../common/error" }
+common-datasource = { path = "../common/datasource" }
+common-function = { path = "../common/function" }
 common-grpc = { path = "../common/grpc" }
 common-grpc-expr = { path = "../common/grpc-expr" }
 common-procedure = { path = "../common/procedure" }
@@ -64,6 +66,7 @@ tonic.workspace = true
 tower = { version = "0.4", features = ["full"] }
 tower-http = { version = "0.3", features = ["full"] }
 url = "2.3.1"
+uuid.workspace = true

 [dev-dependencies]
 axum-test-helper = { git = "https://github.com/sunng87/axum-test-helper.git", branch = "patch-1" }
--- a/src/datanode/src/datanode.rs
+++ b/src/datanode/src/datanode.rs
@@ -77,6 +77,7 @@ impl Default for ObjectStoreConfig {
 }

 #[derive(Debug, Clone, Serialize, Deserialize)]
+#[serde(default)]
 pub struct WalConfig {
    // wal directory
    pub dir: String,
@@ -108,6 +109,7 @@ impl Default for WalConfig {

 /// Options for table compaction
 #[derive(Debug, Clone, Serialize, Deserialize, Eq, PartialEq)]
+#[serde(default)]
 pub struct CompactionConfig {
    /// Max task number that can concurrently run.
    pub max_inflight_tasks: usize,
@@ -149,11 +151,22 @@ impl From<&DatanodeOptions> for StorageEngineConfig {
 pub struct ProcedureConfig {
    /// Storage config for procedure manager.
    pub store: ObjectStoreConfig,
+    /// Max retry times of procedure.
+    pub max_retry_times: usize,
+    /// Initial retry delay of procedures, increases exponentially.
+    #[serde(with = "humantime_serde")]
+    pub retry_delay: Duration,
 }

 impl Default for ProcedureConfig {
    fn default() -> ProcedureConfig {
-        ProcedureConfig::from_file_path("/tmp/greptimedb/procedure/".to_string())
+        ProcedureConfig {
+            store: ObjectStoreConfig::File(FileConfig {
+                data_dir: "/tmp/greptimedb/procedure/".to_string(),
+            }),
+            max_retry_times: 3,
+            retry_delay: Duration::from_millis(500),
+        }
    }
 }

@@ -161,6 +174,7 @@ impl ProcedureConfig {
    pub fn from_file_path(path: String) -> ProcedureConfig {
        ProcedureConfig {
            store: ObjectStoreConfig::File(FileConfig { data_dir: path }),
+            ..Default::default()
        }
    }
 }
@@ -241,7 +255,7 @@ impl Datanode {
        self.instance.clone()
    }

-    async fn shutdown_instance(&self) -> Result<()> {
+    pub async fn shutdown_instance(&self) -> Result<()> {
        self.instance.shutdown().await
    }

--- a/src/datanode/src/error.rs
+++ b/src/datanode/src/error.rs
@@ -14,11 +14,10 @@

 use std::any::Any;

+use common_datasource::error::Error as DataSourceError;
 use common_error::prelude::*;
 use common_procedure::ProcedureId;
-use common_recordbatch::error::Error as RecordBatchError;
 use datafusion::parquet;
-use datatypes::prelude::ConcreteDataType;
 use storage::error::Error as StorageError;
 use table::error::Error as TableError;
 use url::ParseError;
@@ -124,24 +123,6 @@ pub enum Error {
    ))]
    ColumnValuesNumberMismatch { columns: usize, values: usize },

-    #[snafu(display(
-        "Column type mismatch, column: {}, expected type: {:?}, actual: {:?}",
-        column,
-        expected,
-        actual,
-    ))]
-    ColumnTypeMismatch {
-        column: String,
-        expected: ConcreteDataType,
-        actual: ConcreteDataType,
-    },
-
-    #[snafu(display("Failed to collect record batch, source: {}", source))]
-    CollectRecords {
-        #[snafu(backtrace)]
-        source: RecordBatchError,
-    },
-
    #[snafu(display("Failed to parse sql value, source: {}", source))]
    ParseSqlValue {
        #[snafu(backtrace)]
@@ -169,6 +150,13 @@ pub enum Error {
        source: TableError,
    },

+    #[snafu(display("Failed to flush table: {}, source: {}", table_name, source))]
+    FlushTable {
+        table_name: String,
+        #[snafu(backtrace)]
+        source: TableError,
+    },
+
    #[snafu(display("Failed to start server, source: {}", source))]
    StartServer {
        #[snafu(backtrace)]
@@ -211,7 +199,13 @@ pub enum Error {

    #[snafu(display("Failed to build backend, source: {}", source))]
    BuildBackend {
-        source: object_store::Error,
+        #[snafu(backtrace)]
+        source: DataSourceError,
+    },
+
+    #[snafu(display("Failed to parse url, source: {}", source))]
+    ParseUrl {
+        source: DataSourceError,
        backtrace: Backtrace,
    },

@@ -242,6 +236,12 @@ pub enum Error {
        source: regex::Error,
    },

+    #[snafu(display("Failed to list objects, source: {}", source))]
+    ListObjects {
+        #[snafu(backtrace)]
+        source: DataSourceError,
+    },
+
    #[snafu(display("Failed to parse the data, source: {}", source))]
    ParseDataTypes {
        #[snafu(backtrace)]
@@ -468,13 +468,6 @@ pub enum Error {
        source: object_store::Error,
    },

-    #[snafu(display("Failed to lists object in path: {}, source: {}", path, source))]
-    ListObjects {
-        path: String,
-        backtrace: Backtrace,
-        source: object_store::Error,
-    },
-
    #[snafu(display("Unrecognized table option: {}", source))]
    UnrecognizedTableOption {
        #[snafu(backtrace)]
@@ -539,11 +532,10 @@ impl ErrorExt for Error {
                source.status_code()
            }
            DropTable { source, .. } => source.status_code(),
+            FlushTable { source, .. } => source.status_code(),

            Insert { source, .. } => source.status_code(),
            Delete { source, .. } => source.status_code(),
-            CollectRecords { source, .. } => source.status_code(),
-
            TableNotFound { .. } => StatusCode::TableNotFound,
            ColumnNotFound { .. } => StatusCode::TableColumnNotFound,

@@ -556,7 +548,6 @@ impl ErrorExt for Error {
            ConvertSchema { source, .. } | VectorComputation { source } => source.status_code(),

            ColumnValuesNumberMismatch { .. }
-            | ColumnTypeMismatch { .. }
            | InvalidSql { .. }
            | InvalidUrl { .. }
            | InvalidPath { .. }
@@ -576,7 +567,8 @@ impl ErrorExt for Error {
            | DatabaseNotFound { .. }
            | MissingNodeId { .. }
            | MissingMetasrvOpts { .. }
-            | ColumnNoneDefaultValue { .. } => StatusCode::InvalidArguments,
+            | ColumnNoneDefaultValue { .. }
+            | ParseUrl { .. } => StatusCode::InvalidArguments,

            // TODO(yingwen): Further categorize http error.
            StartServer { .. }
--- a/src/datanode/src/heartbeat.rs
+++ b/src/datanode/src/heartbeat.rs
@@ -17,7 +17,7 @@ use std::sync::Arc;
 use std::time::Duration;

 use api::v1::meta::{HeartbeatRequest, HeartbeatResponse, NodeStat, Peer};
-use catalog::{region_number, CatalogManagerRef};
+use catalog::{datanode_stat, CatalogManagerRef};
 use common_telemetry::{error, info, warn};
 use meta_client::client::{HeartbeatSender, MetaClient};
 use snafu::ResultExt;
@@ -106,13 +106,7 @@ impl HeartbeatTask {
        let mut tx = Self::create_streams(&meta_client, running.clone()).await?;
        common_runtime::spawn_bg(async move {
            while running.load(Ordering::Acquire) {
-                let region_num = match region_number(&catalog_manager_clone).await {
-                    Ok(region_num) => region_num as i64,
-                    Err(e) => {
-                        error!("failed to get region number, err: {e:?}");
-                        -1
-                    }
-                };
+                let (region_num, region_stats) = datanode_stat(&catalog_manager_clone).await;

                let req = HeartbeatRequest {
                    peer: Some(Peer {
@@ -120,9 +114,10 @@ impl HeartbeatTask {
                        addr: addr.clone(),
                    }),
                    node_stat: Some(NodeStat {
-                        region_num,
+                        region_num: region_num as _,
                        ..Default::default()
                    }),
+                    region_stats,
                    ..Default::default()
                };

--- a/src/datanode/src/instance.rs
+++ b/src/datanode/src/instance.rs
@@ -37,12 +37,14 @@ use object_store::services::{Fs as FsBuilder, Oss as OSSBuilder, S3 as S3Builder
 use object_store::{util, ObjectStore, ObjectStoreBuilder};
 use query::query_engine::{QueryEngineFactory, QueryEngineRef};
 use servers::Mode;
+use session::context::QueryContext;
 use snafu::prelude::*;
 use storage::compaction::{CompactionHandler, CompactionSchedulerRef, SimplePicker};
 use storage::config::EngineConfig as StorageEngineConfig;
 use storage::scheduler::{LocalScheduler, SchedulerConfig};
 use storage::EngineImpl;
 use store_api::logstore::LogStore;
+use table::requests::FlushTableRequest;
 use table::table::numbers::NumbersTable;
 use table::table::TableIdProviderRef;
 use table::Table;
@@ -56,7 +58,7 @@ use crate::error::{
 };
 use crate::heartbeat::HeartbeatTask;
 use crate::script::ScriptExecutor;
-use crate::sql::SqlHandler;
+use crate::sql::{SqlHandler, SqlRequest};

 mod grpc;
 mod script;
@@ -202,7 +204,6 @@ impl Instance {
            sql_handler: SqlHandler::new(
                table_engine.clone(),
                catalog_manager.clone(),
-                query_engine.clone(),
                table_engine,
                procedure_manager,
            ),
@@ -233,6 +234,8 @@ impl Instance {
                .context(ShutdownInstanceSnafu)?;
        }

+        self.flush_tables().await?;
+
        self.sql_handler
            .close()
            .await
@@ -240,6 +243,43 @@ impl Instance {
            .context(ShutdownInstanceSnafu)
    }

+    pub async fn flush_tables(&self) -> Result<()> {
+        info!("going to flush all schemas");
+        let schema_list = self
+            .catalog_manager
+            .catalog(DEFAULT_CATALOG_NAME)
+            .map_err(BoxedError::new)
+            .context(ShutdownInstanceSnafu)?
+            .expect("Default schema not found")
+            .schema_names()
+            .map_err(BoxedError::new)
+            .context(ShutdownInstanceSnafu)?;
+        let flush_requests = schema_list
+            .into_iter()
+            .map(|schema_name| {
+                SqlRequest::FlushTable(FlushTableRequest {
+                    catalog_name: DEFAULT_CATALOG_NAME.to_string(),
+                    schema_name,
+                    table_name: None,
+                    region_number: None,
+                    wait: Some(true),
+                })
+            })
+            .collect::<Vec<_>>();
+        let flush_result = futures::future::try_join_all(
+            flush_requests
+                .into_iter()
+                .map(|request| self.sql_handler.execute(request, QueryContext::arc())),
+        )
+        .await
+        .map_err(BoxedError::new)
+        .context(ShutdownInstanceSnafu);
+        info!("Flushed all tables result: {}", flush_result.is_ok());
+        flush_result?;
+
+        Ok(())
+    }
+
    pub fn sql_handler(&self) -> &SqlHandler {
        &self.sql_handler
    }
@@ -290,18 +330,20 @@ pub(crate) async fn new_oss_object_store(store_config: &ObjectStoreConfig) -> Re
    );

    let mut builder = OSSBuilder::default();
-    let builder = builder
+    builder
        .root(&root)
        .bucket(&oss_config.bucket)
        .endpoint(&oss_config.endpoint)
        .access_key_id(&oss_config.access_key_id)
        .access_key_secret(&oss_config.access_key_secret);

-    let accessor = builder.build().with_context(|_| error::InitBackendSnafu {
-        config: store_config.clone(),
-    })?;
+    let object_store = ObjectStore::new(builder)
+        .with_context(|_| error::InitBackendSnafu {
+            config: store_config.clone(),
+        })?
+        .finish();

-    create_object_store_with_cache(ObjectStore::new(accessor).finish(), store_config)
+    create_object_store_with_cache(object_store, store_config)
 }

 fn create_object_store_with_cache(
@@ -354,24 +396,27 @@ pub(crate) async fn new_s3_object_store(store_config: &ObjectStoreConfig) -> Res
    );

    let mut builder = S3Builder::default();
-    let mut builder = builder
+    builder
        .root(&root)
        .bucket(&s3_config.bucket)
        .access_key_id(&s3_config.access_key_id)
        .secret_access_key(&s3_config.secret_access_key);

    if s3_config.endpoint.is_some() {
-        builder = builder.endpoint(s3_config.endpoint.as_ref().unwrap());
+        builder.endpoint(s3_config.endpoint.as_ref().unwrap());
    }
    if s3_config.region.is_some() {
-        builder = builder.region(s3_config.region.as_ref().unwrap());
+        builder.region(s3_config.region.as_ref().unwrap());
    }

-    let accessor = builder.build().with_context(|_| error::InitBackendSnafu {
-        config: store_config.clone(),
-    })?;
-
-    create_object_store_with_cache(ObjectStore::new(accessor).finish(), store_config)
+    create_object_store_with_cache(
+        ObjectStore::new(builder)
+            .with_context(|_| error::InitBackendSnafu {
+                config: store_config.clone(),
+            })?
+            .finish(),
+        store_config,
+    )
 }

 pub(crate) async fn new_fs_object_store(store_config: &ObjectStoreConfig) -> Result<ObjectStore> {
@@ -386,15 +431,14 @@ pub(crate) async fn new_fs_object_store(store_config: &ObjectStoreConfig) -> Res

    let atomic_write_dir = format!("{data_dir}/.tmp/");

-    let accessor = FsBuilder::default()
-        .root(&data_dir)
-        .atomic_write_dir(&atomic_write_dir)
-        .build()
+    let mut builder = FsBuilder::default();
+    builder.root(&data_dir).atomic_write_dir(&atomic_write_dir);
+
+    Ok(ObjectStore::new(builder)
        .context(error::InitBackendSnafu {
            config: store_config.clone(),
-        })?;
-
-    Ok(ObjectStore::new(accessor).finish())
+        })?
+        .finish())
 }

 /// Create metasrv client instance and spawn heartbeat loop.
@@ -460,7 +504,11 @@ pub(crate) async fn create_procedure_manager(
    );

    let object_store = new_object_store(&procedure_config.store).await?;
-    let manager_config = ManagerConfig { object_store };
+    let manager_config = ManagerConfig {
+        object_store,
+        max_retry_times: procedure_config.max_retry_times,
+        retry_delay: procedure_config.retry_delay,
+    };

    Ok(Some(Arc::new(LocalManager::new(manager_config))))
 }
--- a/src/datanode/src/instance/grpc.rs
+++ b/src/datanode/src/instance/grpc.rs
@@ -21,7 +21,7 @@ use common_query::Output;
 use query::parser::{PromQuery, QueryLanguageParser, QueryStatement};
 use query::plan::LogicalPlan;
 use servers::query_handler::grpc::GrpcQueryHandler;
-use session::context::QueryContextRef;
+use session::context::{QueryContext, QueryContextRef};
 use snafu::prelude::*;
 use sql::statements::statement::Statement;
 use substrait::{DFLogicalSubstraitConvertor, SubstraitPlan};
@@ -53,7 +53,7 @@ impl Instance {
            .context(DecodeLogicalPlanSnafu)?;

        self.query_engine
-            .execute(&LogicalPlan::DfPlan(logical_plan))
+            .execute(LogicalPlan::DfPlan(logical_plan), QueryContext::arc())
            .await
            .context(ExecuteLogicalPlanSnafu)
    }
@@ -69,11 +69,11 @@ impl Instance {
                        let plan = self
                            .query_engine
                            .planner()
-                            .plan(stmt, ctx)
+                            .plan(stmt, ctx.clone())
                            .await
                            .context(PlanStatementSnafu)?;
                        self.query_engine
-                            .execute(&plan)
+                            .execute(plan, ctx)
                            .await
                            .context(ExecuteLogicalPlanSnafu)
                    }
@@ -127,7 +127,7 @@ impl Instance {
            DdlExpr::Alter(expr) => self.handle_alter(expr).await,
            DdlExpr::CreateDatabase(expr) => self.handle_create_database(expr, query_ctx).await,
            DdlExpr::DropTable(expr) => self.handle_drop_table(expr).await,
-            DdlExpr::FlushTable(_) => todo!(),
+            DdlExpr::FlushTable(expr) => self.handle_flush_table(expr).await,
        }
    }
 }
@@ -175,7 +175,7 @@ mod test {
            .plan(stmt, QueryContext::arc())
            .await
            .unwrap();
-        engine.execute(&plan).await.unwrap()
+        engine.execute(plan, QueryContext::arc()).await.unwrap()
    }

    #[tokio::test(flavor = "multi_thread")]
--- a/src/datanode/src/instance/sql.rs
+++ b/src/datanode/src/instance/sql.rs
@@ -19,7 +19,6 @@ use common_error::prelude::BoxedError;
 use common_query::Output;
 use common_telemetry::logging::info;
 use common_telemetry::timer;
-use futures::StreamExt;
 use query::error::QueryExecutionSnafu;
 use query::parser::{PromQuery, QueryLanguageParser, QueryStatement};
 use query::query_engine::StatementHandler;
@@ -28,12 +27,10 @@ use servers::prom::PromHandler;
 use session::context::{QueryContext, QueryContextRef};
 use snafu::prelude::*;
 use sql::ast::ObjectName;
-use sql::statements::copy::CopyTable;
+use sql::statements::copy::{CopyTable, CopyTableArgument};
 use sql::statements::statement::Statement;
 use table::engine::TableReference;
-use table::requests::{
-    CopyTableFromRequest, CopyTableRequest, CreateDatabaseRequest, DropTableRequest,
-};
+use table::requests::{CopyDirection, CopyTableRequest, CreateDatabaseRequest, DropTableRequest};

 use crate::error::{
    self, BumpTableIdSnafu, ExecuteSqlSnafu, ExecuteStatementSnafu, PlanStatementSnafu, Result,
@@ -41,8 +38,7 @@ use crate::error::{
 };
 use crate::instance::Instance;
 use crate::metric;
-use crate::sql::insert::InsertRequests;
-use crate::sql::SqlRequest;
+use crate::sql::{SqlHandler, SqlRequest};

 impl Instance {
    pub async fn execute_stmt(
@@ -52,37 +48,10 @@ impl Instance {
    ) -> Result<Output> {
        match stmt {
            QueryStatement::Sql(Statement::Insert(insert)) => {
-                let requests = self
-                    .sql_handler
-                    .insert_to_requests(self.catalog_manager.clone(), *insert, query_ctx.clone())
-                    .await?;
-
-                match requests {
-                    InsertRequests::Request(request) => {
-                        self.sql_handler.execute(request, query_ctx.clone()).await
-                    }
-
-                    InsertRequests::Stream(mut s) => {
-                        let mut rows = 0;
-                        while let Some(request) = s.next().await {
-                            match self
-                                .sql_handler
-                                .execute(request?, query_ctx.clone())
-                                .await?
-                            {
-                                Output::AffectedRows(n) => {
-                                    rows += n;
-                                }
-                                _ => unreachable!(),
-                            }
-                        }
-                        Ok(Output::AffectedRows(rows))
-                    }
-                }
-            }
-            QueryStatement::Sql(Statement::Delete(delete)) => {
-                let request = SqlRequest::Delete(*delete);
-                self.sql_handler.execute(request, query_ctx).await
+                let request =
+                    SqlHandler::insert_to_request(self.catalog_manager.clone(), *insert, query_ctx)
+                        .await?;
+                self.sql_handler.insert(request).await
            }
            QueryStatement::Sql(Statement::CreateDatabase(create_database)) => {
                let request = CreateDatabaseRequest {
@@ -160,43 +129,59 @@ impl Instance {
            QueryStatement::Sql(Statement::ShowCreateTable(_show_create_table)) => {
                unimplemented!("SHOW CREATE TABLE is unimplemented yet");
            }
-            QueryStatement::Sql(Statement::Copy(copy_table)) => match copy_table {
-                CopyTable::To(copy_table) => {
-                    let (catalog_name, schema_name, table_name) =
-                        table_idents_to_full_name(copy_table.table_name(), query_ctx.clone())?;
-                    let file_name = copy_table.file_name().to_string();
+            QueryStatement::Sql(Statement::Copy(copy_table)) => {
+                let req = match copy_table {
+                    CopyTable::To(copy_table) => {
+                        let CopyTableArgument {
+                            location,
+                            connection,
+                            pattern,
+                            table_name,
+                            ..
+                        } = copy_table;
+                        let (catalog_name, schema_name, table_name) =
+                            table_idents_to_full_name(&table_name, query_ctx.clone())?;
+                        CopyTableRequest {
+                            catalog_name,
+                            schema_name,
+                            table_name,
+                            location,
+                            connection,
+                            pattern,
+                            direction: CopyDirection::Export,
+                        }
+                    }
+                    CopyTable::From(copy_table) => {
+                        let CopyTableArgument {
+                            location,
+                            connection,
+                            pattern,
+                            table_name,
+                            ..
+                        } = copy_table;
+                        let (catalog_name, schema_name, table_name) =
+                            table_idents_to_full_name(&table_name, query_ctx.clone())?;
+                        CopyTableRequest {
+                            catalog_name,
+                            schema_name,
+                            table_name,
+                            location,
+                            connection,
+                            pattern,
+                            direction: CopyDirection::Import,
+                        }
+                    }
+                };

-                    let req = CopyTableRequest {
-                        catalog_name,
-                        schema_name,
-                        table_name,
-                        file_name,
-                    };
-
-                    self.sql_handler
-                        .execute(SqlRequest::CopyTable(req), query_ctx)
-                        .await
-                }
-                CopyTable::From(copy_table) => {
-                    let (catalog_name, schema_name, table_name) =
-                        table_idents_to_full_name(&copy_table.table_name, query_ctx.clone())?;
-                    let req = CopyTableFromRequest {
-                        catalog_name,
-                        schema_name,
-                        table_name,
-                        connection: copy_table.connection,
-                        pattern: copy_table.pattern,
-                        from: copy_table.from,
-                    };
-                    self.sql_handler
-                        .execute(SqlRequest::CopyTableFrom(req), query_ctx)
-                        .await
-                }
-            },
+                self.sql_handler
+                    .execute(SqlRequest::CopyTable(req), query_ctx)
+                    .await
+            }
            QueryStatement::Sql(Statement::Query(_))
            | QueryStatement::Sql(Statement::Explain(_))
            | QueryStatement::Sql(Statement::Use(_))
            | QueryStatement::Sql(Statement::Tql(_))
+            | QueryStatement::Sql(Statement::Delete(_))
            | QueryStatement::Promql(_) => unreachable!(),
        }
    }
@@ -213,10 +198,13 @@ impl Instance {
        let engine = self.query_engine();
        let plan = engine
            .planner()
-            .plan(stmt, query_ctx)
+            .plan(stmt, query_ctx.clone())
            .await
            .context(PlanStatementSnafu)?;
-        engine.execute(&plan).await.context(ExecuteStatementSnafu)
+        engine
+            .execute(plan, query_ctx)
+            .await
+            .context(ExecuteStatementSnafu)
    }

    // TODO(ruihang): merge this and `execute_promql` after #951 landed
@@ -249,10 +237,13 @@ impl Instance {
        let engine = self.query_engine();
        let plan = engine
            .planner()
-            .plan(stmt, query_ctx)
+            .plan(stmt, query_ctx.clone())
            .await
            .context(PlanStatementSnafu)?;
-        engine.execute(&plan).await.context(ExecuteStatementSnafu)
+        engine
+            .execute(plan, query_ctx)
+            .await
+            .context(ExecuteStatementSnafu)
    }
 }

--- a/src/datanode/src/server/grpc.rs
+++ b/src/datanode/src/server/grpc.rs
@@ -12,13 +12,13 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.

-use api::v1::{AlterExpr, CreateTableExpr, DropTableExpr};
+use api::v1::{AlterExpr, CreateTableExpr, DropTableExpr, FlushTableExpr};
 use common_grpc_expr::{alter_expr_to_request, create_expr_to_request};
 use common_query::Output;
 use common_telemetry::info;
 use session::context::QueryContext;
 use snafu::prelude::*;
-use table::requests::DropTableRequest;
+use table::requests::{DropTableRequest, FlushTableRequest};

 use crate::error::{
    AlterExprToRequestSnafu, BumpTableIdSnafu, CreateExprToRequestSnafu,
@@ -82,6 +82,25 @@ impl Instance {
            .execute(SqlRequest::DropTable(req), QueryContext::arc())
            .await
    }
+
+    pub(crate) async fn handle_flush_table(&self, expr: FlushTableExpr) -> Result<Output> {
+        let table_name = if expr.table_name.trim().is_empty() {
+            None
+        } else {
+            Some(expr.table_name)
+        };
+
+        let req = FlushTableRequest {
+            catalog_name: expr.catalog_name,
+            schema_name: expr.schema_name,
+            table_name,
+            region_number: expr.region_id,
+            wait: None,
+        };
+        self.sql_handler()
+            .execute(SqlRequest::FlushTable(req), QueryContext::arc())
+            .await
+    }
 }

 #[cfg(test)]
@@ -136,7 +155,6 @@ mod tests {
    }

    #[test]
-
    fn test_create_column_schema() {
        let column_def = ColumnDef {
            name: "a".to_string(),
--- a/src/datanode/src/sql.rs
+++ b/src/datanode/src/sql.rs
@@ -17,11 +17,9 @@ use common_error::prelude::BoxedError;
 use common_procedure::ProcedureManagerRef;
 use common_query::Output;
 use common_telemetry::error;
-use query::query_engine::QueryEngineRef;
 use query::sql::{describe_table, show_databases, show_tables};
 use session::context::QueryContextRef;
 use snafu::{OptionExt, ResultExt};
-use sql::statements::delete::Delete;
 use sql::statements::describe::DescribeTable;
 use sql::statements::show::{ShowDatabases, ShowTables};
 use table::engine::{EngineContext, TableEngineProcedureRef, TableEngineRef, TableReference};
@@ -34,33 +32,31 @@ use crate::error::{
 use crate::instance::sql::table_idents_to_full_name;

 mod alter;
-mod copy_table;
 mod copy_table_from;
+mod copy_table_to;
 mod create;
-mod delete;
 mod drop_table;
+mod flush_table;
 pub(crate) mod insert;

 #[derive(Debug)]
 pub enum SqlRequest {
-    Insert(InsertRequest),
    CreateTable(CreateTableRequest),
    CreateDatabase(CreateDatabaseRequest),
    Alter(AlterTableRequest),
    DropTable(DropTableRequest),
+    FlushTable(FlushTableRequest),
    ShowDatabases(ShowDatabases),
    ShowTables(ShowTables),
    DescribeTable(DescribeTable),
-    Delete(Delete),
    CopyTable(CopyTableRequest),
-    CopyTableFrom(CopyTableFromRequest),
 }

 // Handler to execute SQL except query
+#[derive(Clone)]
 pub struct SqlHandler {
    table_engine: TableEngineRef,
    catalog_manager: CatalogManagerRef,
-    query_engine: QueryEngineRef,
    engine_procedure: TableEngineProcedureRef,
    procedure_manager: Option<ProcedureManagerRef>,
 }
@@ -69,14 +65,12 @@ impl SqlHandler {
    pub fn new(
        table_engine: TableEngineRef,
        catalog_manager: CatalogManagerRef,
-        query_engine: QueryEngineRef,
        engine_procedure: TableEngineProcedureRef,
        procedure_manager: Option<ProcedureManagerRef>,
    ) -> Self {
        Self {
            table_engine,
            catalog_manager,
-            query_engine,
            engine_procedure,
            procedure_manager,
        }
@@ -88,14 +82,14 @@ impl SqlHandler {
    // there, instead of executing here in a "static" fashion.
    pub async fn execute(&self, request: SqlRequest, query_ctx: QueryContextRef) -> Result<Output> {
        let result = match request {
-            SqlRequest::Insert(req) => self.insert(req).await,
            SqlRequest::CreateTable(req) => self.create_table(req).await,
            SqlRequest::CreateDatabase(req) => self.create_database(req, query_ctx.clone()).await,
            SqlRequest::Alter(req) => self.alter(req).await,
            SqlRequest::DropTable(req) => self.drop_table(req).await,
-            SqlRequest::Delete(req) => self.delete(query_ctx.clone(), req).await,
-            SqlRequest::CopyTable(req) => self.copy_table(req).await,
-            SqlRequest::CopyTableFrom(req) => self.copy_table_from(req).await,
+            SqlRequest::CopyTable(req) => match req.direction {
+                CopyDirection::Export => self.copy_table_to(req).await,
+                CopyDirection::Import => self.copy_table_from(req).await,
+            },
            SqlRequest::ShowDatabases(req) => {
                show_databases(req, self.catalog_manager.clone()).context(ExecuteSqlSnafu)
            }
@@ -116,6 +110,7 @@ impl SqlHandler {
                    })?;
                describe_table(table).context(ExecuteSqlSnafu)
            }
+            SqlRequest::FlushTable(req) => self.flush_table(req).await,
        };
        if let Err(e) = &result {
            error!(e; "{query_ctx}");
@@ -146,239 +141,3 @@ impl SqlHandler {
            .context(CloseTableEngineSnafu)
    }
 }
-
-#[cfg(test)]
-mod tests {
-    use std::any::Any;
-    use std::sync::Arc;
-
-    use catalog::{CatalogManager, RegisterTableRequest};
-    use common_catalog::consts::{DEFAULT_CATALOG_NAME, DEFAULT_SCHEMA_NAME};
-    use common_query::logical_plan::Expr;
-    use common_query::physical_plan::PhysicalPlanRef;
-    use common_test_util::temp_dir::create_temp_dir;
-    use common_time::timestamp::Timestamp;
-    use datatypes::prelude::ConcreteDataType;
-    use datatypes::schema::{ColumnSchema, SchemaBuilder, SchemaRef};
-    use datatypes::value::Value;
-    use futures::StreamExt;
-    use log_store::NoopLogStore;
-    use mito::config::EngineConfig as TableEngineConfig;
-    use mito::engine::MitoEngine;
-    use object_store::services::Fs as Builder;
-    use object_store::{ObjectStore, ObjectStoreBuilder};
-    use query::parser::{QueryLanguageParser, QueryStatement};
-    use query::QueryEngineFactory;
-    use session::context::QueryContext;
-    use sql::statements::statement::Statement;
-    use storage::compaction::noop::NoopCompactionScheduler;
-    use storage::config::EngineConfig as StorageEngineConfig;
-    use storage::EngineImpl;
-    use table::error::Result as TableResult;
-    use table::metadata::TableInfoRef;
-    use table::Table;
-
-    use super::*;
-    use crate::error::Error;
-    use crate::sql::insert::InsertRequests;
-
-    struct DemoTable;
-
-    #[async_trait::async_trait]
-    impl Table for DemoTable {
-        fn as_any(&self) -> &dyn Any {
-            self
-        }
-
-        fn schema(&self) -> SchemaRef {
-            let column_schemas = vec![
-                ColumnSchema::new("host", ConcreteDataType::string_datatype(), false),
-                ColumnSchema::new("cpu", ConcreteDataType::float64_datatype(), true),
-                ColumnSchema::new("memory", ConcreteDataType::float64_datatype(), true),
-                ColumnSchema::new(
-                    "ts",
-                    ConcreteDataType::timestamp_millisecond_datatype(),
-                    true,
-                )
-                .with_time_index(true),
-            ];
-
-            Arc::new(
-                SchemaBuilder::try_from(column_schemas)
-                    .unwrap()
-                    .build()
-                    .unwrap(),
-            )
-        }
-
-        fn table_info(&self) -> TableInfoRef {
-            unimplemented!()
-        }
-
-        async fn scan(
-            &self,
-            _projection: Option<&Vec<usize>>,
-            _filters: &[Expr],
-            _limit: Option<usize>,
-        ) -> TableResult<PhysicalPlanRef> {
-            unimplemented!();
-        }
-    }
-
-    #[tokio::test]
-    async fn test_statement_to_request() {
-        let dir = create_temp_dir("setup_test_engine_and_table");
-        let store_dir = dir.path().to_string_lossy();
-        let accessor = Builder::default().root(&store_dir).build().unwrap();
-        let object_store = ObjectStore::new(accessor).finish();
-        let compaction_scheduler = Arc::new(NoopCompactionScheduler::default());
-        let sql = r#"insert into demo(host, cpu, memory, ts) values
-                           ('host1', 66.6, 1024, 1655276557000),
-                           ('host2', 88.8,  333.3, 1655276558000)
-                           "#;
-
-        let table_engine = Arc::new(MitoEngine::<EngineImpl<NoopLogStore>>::new(
-            TableEngineConfig::default(),
-            EngineImpl::new(
-                StorageEngineConfig::default(),
-                Arc::new(NoopLogStore::default()),
-                object_store.clone(),
-                compaction_scheduler,
-            ),
-            object_store,
-        ));
-
-        let catalog_list = Arc::new(
-            catalog::local::LocalCatalogManager::try_new(table_engine.clone())
-                .await
-                .unwrap(),
-        );
-        catalog_list.start().await.unwrap();
-        assert!(catalog_list
-            .register_table(RegisterTableRequest {
-                catalog: DEFAULT_CATALOG_NAME.to_string(),
-                schema: DEFAULT_SCHEMA_NAME.to_string(),
-                table_name: "demo".to_string(),
-                table_id: 1,
-                table: Arc::new(DemoTable),
-            })
-            .await
-            .unwrap());
-
-        let factory = QueryEngineFactory::new(catalog_list.clone());
-        let query_engine = factory.query_engine();
-        let sql_handler = SqlHandler::new(
-            table_engine.clone(),
-            catalog_list.clone(),
-            query_engine.clone(),
-            table_engine,
-            None,
-        );
-
-        let stmt = match QueryLanguageParser::parse_sql(sql).unwrap() {
-            QueryStatement::Sql(Statement::Insert(i)) => i,
-            _ => {
-                unreachable!()
-            }
-        };
-        let request = sql_handler
-            .insert_to_requests(catalog_list.clone(), *stmt, QueryContext::arc())
-            .await
-            .unwrap();
-
-        match request {
-            InsertRequests::Request(SqlRequest::Insert(req)) => {
-                assert_eq!(req.table_name, "demo");
-                let columns_values = req.columns_values;
-                assert_eq!(4, columns_values.len());
-
-                let hosts = &columns_values["host"];
-                assert_eq!(2, hosts.len());
-                assert_eq!(Value::from("host1"), hosts.get(0));
-                assert_eq!(Value::from("host2"), hosts.get(1));
-
-                let cpus = &columns_values["cpu"];
-                assert_eq!(2, cpus.len());
-                assert_eq!(Value::from(66.6f64), cpus.get(0));
-                assert_eq!(Value::from(88.8f64), cpus.get(1));
-
-                let memories = &columns_values["memory"];
-                assert_eq!(2, memories.len());
-                assert_eq!(Value::from(1024f64), memories.get(0));
-                assert_eq!(Value::from(333.3f64), memories.get(1));
-
-                let ts = &columns_values["ts"];
-                assert_eq!(2, ts.len());
-                assert_eq!(
-                    Value::from(Timestamp::new_millisecond(1655276557000i64)),
-                    ts.get(0)
-                );
-                assert_eq!(
-                    Value::from(Timestamp::new_millisecond(1655276558000i64)),
-                    ts.get(1)
-                );
-            }
-            _ => {
-                panic!("Not supposed to reach here")
-            }
-        }
-
-        // test inert into select
-
-        // type mismatch
-        let sql = "insert into demo(ts) select number from numbers limit 3";
-
-        let stmt = match QueryLanguageParser::parse_sql(sql).unwrap() {
-            QueryStatement::Sql(Statement::Insert(i)) => i,
-            _ => {
-                unreachable!()
-            }
-        };
-        let request = sql_handler
-            .insert_to_requests(catalog_list.clone(), *stmt, QueryContext::arc())
-            .await
-            .unwrap();
-
-        match request {
-            InsertRequests::Stream(mut stream) => {
-                assert!(matches!(
-                    stream.next().await.unwrap().unwrap_err(),
-                    Error::ColumnTypeMismatch { .. }
-                ));
-            }
-            _ => unreachable!(),
-        }
-
-        let sql = "insert into demo(cpu) select cast(number as double) from numbers limit 3";
-        let stmt = match QueryLanguageParser::parse_sql(sql).unwrap() {
-            QueryStatement::Sql(Statement::Insert(i)) => i,
-            _ => {
-                unreachable!()
-            }
-        };
-        let request = sql_handler
-            .insert_to_requests(catalog_list.clone(), *stmt, QueryContext::arc())
-            .await
-            .unwrap();
-
-        match request {
-            InsertRequests::Stream(mut stream) => {
-                let mut times = 0;
-                while let Some(Ok(SqlRequest::Insert(req))) = stream.next().await {
-                    times += 1;
-                    assert_eq!(req.table_name, "demo");
-                    let columns_values = req.columns_values;
-                    assert_eq!(1, columns_values.len());
-
-                    let memories = &columns_values["cpu"];
-                    assert_eq!(3, memories.len());
-                    assert_eq!(Value::from(0.0f64), memories.get(0));
-                    assert_eq!(Value::from(1.0f64), memories.get(1));
-                    assert_eq!(Value::from(2.0f64), memories.get(2));
-                }
-                assert_eq!(1, times);
-            }
-            _ => unreachable!(),
-        }
-    }
-}
--- a/src/datanode/src/sql/copy_table_from.rs
+++ b/src/datanode/src/sql/copy_table_from.rs
@@ -15,35 +15,26 @@
 use std::collections::HashMap;

 use async_compat::CompatExt;
+use common_datasource::lister::{Lister, Source};
+use common_datasource::object_store::{build_backend, parse_url};
+use common_datasource::util::find_dir_and_filename;
 use common_query::Output;
 use common_recordbatch::error::DataTypesSnafu;
 use datafusion::parquet::arrow::ParquetRecordBatchStreamBuilder;
 use datatypes::arrow::record_batch::RecordBatch;
 use datatypes::vectors::{Helper, VectorRef};
-use futures::future;
 use futures_util::TryStreamExt;
-use object_store::services::{Fs, S3};
-use object_store::{Object, ObjectStore, ObjectStoreBuilder};
 use regex::Regex;
 use snafu::{ensure, ResultExt};
 use table::engine::TableReference;
-use table::requests::{CopyTableFromRequest, InsertRequest};
+use table::requests::{CopyTableRequest, InsertRequest};
 use tokio::io::BufReader;
-use url::{ParseError, Url};

 use crate::error::{self, Result};
 use crate::sql::SqlHandler;

-const S3_SCHEMA: &str = "S3";
-const ENDPOINT_URL: &str = "ENDPOINT_URL";
-const ACCESS_KEY_ID: &str = "ACCESS_KEY_ID";
-const SECRET_ACCESS_KEY: &str = "SECRET_ACCESS_KEY";
-const SESSION_TOKEN: &str = "SESSION_TOKEN";
-const REGION: &str = "REGION";
-const ENABLE_VIRTUAL_HOST_STYLE: &str = "ENABLE_VIRTUAL_HOST_STYLE";
-
 impl SqlHandler {
-    pub(crate) async fn copy_table_from(&self, req: CopyTableFromRequest) -> Result<Output> {
+    pub(crate) async fn copy_table_from(&self, req: CopyTableRequest) -> Result<Output> {
        let table_ref = TableReference {
            catalog: &req.catalog_name,
            schema: &req.schema_name,
@@ -51,16 +42,37 @@ impl SqlHandler {
        };
        let table = self.get_table(&table_ref)?;

-        let datasource = DataSource::new(&req.from, req.pattern, req.connection)?;
+        let (_schema, _host, path) = parse_url(&req.location).context(error::ParseUrlSnafu)?;

-        let objects = datasource.list().await?;
+        let object_store =
+            build_backend(&req.location, req.connection).context(error::BuildBackendSnafu)?;
+
+        let (dir, filename) = find_dir_and_filename(&path);
+        let regex = req
+            .pattern
+            .as_ref()
+            .map(|x| Regex::new(x))
+            .transpose()
+            .context(error::BuildRegexSnafu)?;
+
+        let source = if let Some(filename) = filename {
+            Source::Filename(filename)
+        } else {
+            Source::Dir
+        };
+
+        let lister = Lister::new(object_store.clone(), source, dir, regex);
+
+        let entries = lister.list().await.context(error::ListObjectsSnafu)?;

        let mut buf: Vec<RecordBatch> = Vec::new();

-        for obj in objects.iter() {
-            let reader = obj.reader().await.context(error::ReadObjectSnafu {
-                path: &obj.path().to_string(),
-            })?;
+        for entry in entries.iter() {
+            let path = entry.path();
+            let reader = object_store
+                .reader(path)
+                .await
+                .context(error::ReadObjectSnafu { path })?;

            let buf_reader = BufReader::new(reader.compat());

@@ -131,315 +143,3 @@ impl SqlHandler {
        Ok(Output::AffectedRows(result.iter().sum()))
    }
 }
-
-#[derive(Debug, Clone, PartialEq, Eq)]
-enum Source {
-    Filename(String),
-    Dir,
-}
-
-struct DataSource {
-    object_store: ObjectStore,
-    source: Source,
-    path: String,
-    regex: Option<Regex>,
-}
-
-impl DataSource {
-    fn from_path(url: &str, regex: Option<Regex>) -> Result<DataSource> {
-        let result = if url.ends_with('/') {
-            Url::from_directory_path(url)
-        } else {
-            Url::from_file_path(url)
-        };
-
-        match result {
-            Ok(url) => {
-                let path = url.path();
-
-                let (path, filename) = DataSource::find_dir_and_filename(path);
-
-                let source = if let Some(filename) = filename {
-                    Source::Filename(filename)
-                } else {
-                    Source::Dir
-                };
-
-                let accessor = Fs::default()
-                    .root(&path)
-                    .build()
-                    .context(error::BuildBackendSnafu)?;
-
-                Ok(DataSource {
-                    object_store: ObjectStore::new(accessor).finish(),
-                    source,
-                    path,
-                    regex,
-                })
-            }
-            Err(()) => error::InvalidPathSnafu {
-                path: url.to_string(),
-            }
-            .fail(),
-        }
-    }
-
-    fn build_s3_backend(
-        host: Option<&str>,
-        path: &str,
-        connection: HashMap<String, String>,
-    ) -> Result<ObjectStore> {
-        let mut builder = S3::default();
-
-        builder.root(path);
-
-        if let Some(bucket) = host {
-            builder.bucket(bucket);
-        }
-
-        if let Some(endpoint) = connection.get(ENDPOINT_URL) {
-            builder.endpoint(endpoint);
-        }
-
-        if let Some(region) = connection.get(REGION) {
-            builder.region(region);
-        }
-
-        if let Some(key_id) = connection.get(ACCESS_KEY_ID) {
-            builder.access_key_id(key_id);
-        }
-
-        if let Some(key) = connection.get(SECRET_ACCESS_KEY) {
-            builder.secret_access_key(key);
-        }
-
-        if let Some(session_token) = connection.get(SESSION_TOKEN) {
-            builder.security_token(session_token);
-        }
-
-        if let Some(enable_str) = connection.get(ENABLE_VIRTUAL_HOST_STYLE) {
-            let enable = enable_str.as_str().parse::<bool>().map_err(|e| {
-                error::InvalidConnectionSnafu {
-                    msg: format!(
-                        "failed to parse the option {}={}, {}",
-                        ENABLE_VIRTUAL_HOST_STYLE, enable_str, e
-                    ),
-                }
-                .build()
-            })?;
-            if enable {
-                builder.enable_virtual_host_style();
-            }
-        }
-
-        let accessor = builder.build().context(error::BuildBackendSnafu)?;
-
-        Ok(ObjectStore::new(accessor).finish())
-    }
-
-    fn from_url(
-        url: Url,
-        regex: Option<Regex>,
-        connection: HashMap<String, String>,
-    ) -> Result<DataSource> {
-        let host = url.host_str();
-
-        let path = url.path();
-
-        let schema = url.scheme();
-
-        let (dir, filename) = DataSource::find_dir_and_filename(path);
-
-        let source = if let Some(filename) = filename {
-            Source::Filename(filename)
-        } else {
-            Source::Dir
-        };
-
-        let object_store = match schema.to_uppercase().as_str() {
-            S3_SCHEMA => DataSource::build_s3_backend(host, &dir, connection)?,
-            _ => {
-                return error::UnsupportedBackendProtocolSnafu {
-                    protocol: schema.to_string(),
-                }
-                .fail()
-            }
-        };
-
-        Ok(DataSource {
-            object_store,
-            source,
-            path: dir,
-            regex,
-        })
-    }
-
-    pub fn new(
-        url: &str,
-        pattern: Option<String>,
-        connection: HashMap<String, String>,
-    ) -> Result<DataSource> {
-        let regex = if let Some(pattern) = pattern {
-            let regex = Regex::new(&pattern).context(error::BuildRegexSnafu)?;
-            Some(regex)
-        } else {
-            None
-        };
-        let result = Url::parse(url);
-
-        match result {
-            Ok(url) => DataSource::from_url(url, regex, connection),
-            Err(err) => {
-                if ParseError::RelativeUrlWithoutBase == err {
-                    DataSource::from_path(url, regex)
-                } else {
-                    Err(error::Error::InvalidUrl {
-                        url: url.to_string(),
-                        source: err,
-                    })
-                }
-            }
-        }
-    }
-
-    pub async fn list(&self) -> Result<Vec<Object>> {
-        match &self.source {
-            Source::Dir => {
-                let streamer = self
-                    .object_store
-                    .object("/")
-                    .list()
-                    .await
-                    .context(error::ListObjectsSnafu { path: &self.path })?;
-                streamer
-                    .try_filter(|f| {
-                        let res = if let Some(regex) = &self.regex {
-                            regex.is_match(f.name())
-                        } else {
-                            true
-                        };
-                        future::ready(res)
-                    })
-                    .try_collect::<Vec<_>>()
-                    .await
-                    .context(error::ListObjectsSnafu { path: &self.path })
-            }
-            Source::Filename(filename) => {
-                let obj = self.object_store.object(filename);
-
-                Ok(vec![obj])
-            }
-        }
-    }
-
-    fn find_dir_and_filename(path: &str) -> (String, Option<String>) {
-        if path.is_empty() {
-            ("/".to_string(), None)
-        } else if path.ends_with('/') {
-            (path.to_string(), None)
-        } else if let Some(idx) = path.rfind('/') {
-            (
-                path[..idx + 1].to_string(),
-                Some(path[idx + 1..].to_string()),
-            )
-        } else {
-            ("/".to_string(), Some(path.to_string()))
-        }
-    }
-}
-
-#[cfg(test)]
-mod tests {
-
-    use url::Url;
-
-    use super::*;
-    #[test]
-    fn test_parse_uri() {
-        struct Test<'a> {
-            uri: &'a str,
-            expected_path: &'a str,
-            expected_schema: &'a str,
-        }
-
-        let tests = [
-            Test {
-                uri: "s3://bucket/to/path/",
-                expected_path: "/to/path/",
-                expected_schema: "s3",
-            },
-            Test {
-                uri: "fs:///to/path/",
-                expected_path: "/to/path/",
-                expected_schema: "fs",
-            },
-            Test {
-                uri: "fs:///to/path/file",
-                expected_path: "/to/path/file",
-                expected_schema: "fs",
-            },
-        ];
-        for test in tests {
-            let parsed_uri = Url::parse(test.uri).unwrap();
-            assert_eq!(parsed_uri.path(), test.expected_path);
-            assert_eq!(parsed_uri.scheme(), test.expected_schema);
-        }
-    }
-
-    #[test]
-    fn test_parse_path_and_dir() {
-        let parsed = Url::from_file_path("/to/path/file").unwrap();
-        assert_eq!(parsed.path(), "/to/path/file");
-
-        let parsed = Url::from_directory_path("/to/path/").unwrap();
-        assert_eq!(parsed.path(), "/to/path/");
-    }
-
-    #[test]
-    fn test_find_dir_and_filename() {
-        struct Test<'a> {
-            path: &'a str,
-            expected_dir: &'a str,
-            expected_filename: Option<String>,
-        }
-
-        let tests = [
-            Test {
-                path: "to/path/",
-                expected_dir: "to/path/",
-                expected_filename: None,
-            },
-            Test {
-                path: "to/path/filename",
-                expected_dir: "to/path/",
-                expected_filename: Some("filename".into()),
-            },
-            Test {
-                path: "/to/path/filename",
-                expected_dir: "/to/path/",
-                expected_filename: Some("filename".into()),
-            },
-            Test {
-                path: "/",
-                expected_dir: "/",
-                expected_filename: None,
-            },
-            Test {
-                path: "filename",
-                expected_dir: "/",
-                expected_filename: Some("filename".into()),
-            },
-            Test {
-                path: "",
-                expected_dir: "/",
-                expected_filename: None,
-            },
-        ];
-
-        for test in tests {
-            let (path, filename) = DataSource::find_dir_and_filename(test.path);
-            assert_eq!(test.expected_dir, path);
-            assert_eq!(test.expected_filename, filename)
-        }
-    }
-}
--- a/src/datanode/src/sql/copy_table_to.rs
+++ b/src/datanode/src/sql/copy_table_to.rs
@@ -14,6 +14,8 @@

 use std::pin::Pin;

+use common_datasource;
+use common_datasource::object_store::{build_backend, parse_url};
 use common_query::physical_plan::SessionContext;
 use common_query::Output;
 use common_recordbatch::adapter::DfRecordBatchStreamAdapter;
@@ -22,8 +24,7 @@ use datafusion::parquet::basic::{Compression, Encoding};
 use datafusion::parquet::file::properties::WriterProperties;
 use datafusion::physical_plan::RecordBatchStream;
 use futures::TryStreamExt;
-use object_store::services::Fs as Builder;
-use object_store::{ObjectStore, ObjectStoreBuilder};
+use object_store::ObjectStore;
 use snafu::ResultExt;
 use table::engine::TableReference;
 use table::requests::CopyTableRequest;
@@ -32,7 +33,7 @@ use crate::error::{self, Result};
 use crate::sql::SqlHandler;

 impl SqlHandler {
-    pub(crate) async fn copy_table(&self, req: CopyTableRequest) -> Result<Output> {
+    pub(crate) async fn copy_table_to(&self, req: CopyTableRequest) -> Result<Output> {
        let table_ref = TableReference {
            catalog: &req.catalog_name,
            schema: &req.schema_name,
@@ -52,13 +53,11 @@ impl SqlHandler {
            .context(error::TableScanExecSnafu)?;
        let stream = Box::pin(DfRecordBatchStreamAdapter::new(stream));

-        let accessor = Builder::default()
-            .root("/")
-            .build()
-            .context(error::BuildBackendSnafu)?;
-        let object_store = ObjectStore::new(accessor).finish();
+        let (_schema, _host, path) = parse_url(&req.location).context(error::ParseUrlSnafu)?;
+        let object_store =
+            build_backend(&req.location, req.connection).context(error::BuildBackendSnafu)?;

-        let mut parquet_writer = ParquetWriter::new(req.file_name, stream, object_store);
+        let mut parquet_writer = ParquetWriter::new(path.to_string(), stream, object_store);
        // TODO(jiachun):
        // For now, COPY is implemented synchronously.
        // When copying large table, it will be blocked for a long time.
@@ -137,10 +136,10 @@ impl ParquetWriter {
            // "file_name_1_1000000"        (row num: 1 ~ 1000000),
            // "file_name_1000001_xxx"      (row num: 1000001 ~ xxx)
            let file_name = format!("{}_{}_{}", self.file_name, start_row_num, total_rows);
-            let object = self.object_store.object(&file_name);
-            object.write(buf).await.context(error::WriteObjectSnafu {
-                path: object.path(),
-            })?;
+            self.object_store
+                .write(&file_name, buf)
+                .await
+                .context(error::WriteObjectSnafu { path: file_name })?;

            if end_loop {
                return Ok(total_rows);
--- a/src/datanode/src/sql/delete.rs
+++ b/src/datanode/src/sql/delete.rs
@@ -1,142 +0,0 @@
-// Copyright 2023 Greptime Team
-//
-// Licensed under the Apache License, Version 2.0 (the "License");
-// you may not use this file except in compliance with the License.
-// You may obtain a copy of the License at
-//
-//     http://www.apache.org/licenses/LICENSE-2.0
-//
-// Unless required by applicable law or agreed to in writing, software
-// distributed under the License is distributed on an "AS IS" BASIS,
-// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-// See the License for the specific language governing permissions and
-// limitations under the License.
-
-use std::collections::HashMap;
-use std::sync::Arc;
-
-use common_query::Output;
-use datatypes::data_type::DataType;
-use datatypes::prelude::VectorRef;
-use datatypes::vectors::StringVector;
-use session::context::QueryContextRef;
-use snafu::{OptionExt, ResultExt};
-use sql::ast::{BinaryOperator, Expr, Value};
-use sql::statements::delete::Delete;
-use sql::statements::sql_value_to_value;
-use table::engine::TableReference;
-use table::requests::DeleteRequest;
-use table::TableRef;
-
-use crate::error::{ColumnNotFoundSnafu, DeleteSnafu, InvalidSqlSnafu, NotSupportSqlSnafu, Result};
-use crate::instance::sql::table_idents_to_full_name;
-use crate::sql::SqlHandler;
-
-impl SqlHandler {
-    pub(crate) async fn delete(&self, query_ctx: QueryContextRef, stmt: Delete) -> Result<Output> {
-        let (catalog_name, schema_name, table_name) =
-            table_idents_to_full_name(stmt.table_name(), query_ctx)?;
-        let table_ref = TableReference {
-            catalog: &catalog_name.to_string(),
-            schema: &schema_name.to_string(),
-            table: &table_name.to_string(),
-        };
-
-        let table = self.get_table(&table_ref)?;
-
-        let req = DeleteRequest {
-            key_column_values: parse_selection(stmt.selection(), &table)?,
-        };
-
-        let affected_rows = table.delete(req).await.with_context(|_| DeleteSnafu {
-            table_name: table_ref.to_string(),
-        })?;
-
-        Ok(Output::AffectedRows(affected_rows))
-    }
-}
-
-/// parse selection, currently supported format is `tagkey1 = 'tagvalue1' and 'ts' = 'value'`.
-/// (only uses =, and in the where clause and provides all columns needed by the key.)
-fn parse_selection(
-    selection: &Option<Expr>,
-    table: &TableRef,
-) -> Result<HashMap<String, VectorRef>> {
-    let mut key_column_values = HashMap::new();
-    if let Some(expr) = selection {
-        parse_expr(expr, &mut key_column_values, table)?;
-    }
-    Ok(key_column_values)
-}
-
-fn parse_expr(
-    expr: &Expr,
-    key_column_values: &mut HashMap<String, VectorRef>,
-    table: &TableRef,
-) -> Result<()> {
-    // match BinaryOp
-    if let Expr::BinaryOp { left, op, right } = expr {
-        match (&**left, op, &**right) {
-            // match And operator
-            (Expr::BinaryOp { .. }, BinaryOperator::And, Expr::BinaryOp { .. }) => {
-                parse_expr(left, key_column_values, table)?;
-                parse_expr(right, key_column_values, table)?;
-                return Ok(());
-            }
-            // match Eq operator
-            (Expr::Identifier(column_name), BinaryOperator::Eq, Expr::Value(value)) => {
-                key_column_values.insert(
-                    column_name.to_string(),
-                    value_to_vector(&column_name.to_string(), value, table)?,
-                );
-                return Ok(());
-            }
-            (Expr::Identifier(column_name), BinaryOperator::Eq, Expr::Identifier(value)) => {
-                key_column_values.insert(
-                    column_name.to_string(),
-                    Arc::new(StringVector::from(vec![value.to_string()])),
-                );
-                return Ok(());
-            }
-            _ => {}
-        }
-    }
-    NotSupportSqlSnafu {
-        msg: format!(
-            "Not support sql expr:{expr},correct format is tagkey1 = tagvalue1 and ts = value"
-        ),
-    }
-    .fail()
-}
-
-/// parse value to vector
-fn value_to_vector(column_name: &String, sql_value: &Value, table: &TableRef) -> Result<VectorRef> {
-    let schema = table.schema();
-    let column_schema =
-        schema
-            .column_schema_by_name(column_name)
-            .with_context(|| ColumnNotFoundSnafu {
-                table_name: table.table_info().name.clone(),
-                column_name: column_name.to_string(),
-            })?;
-    let data_type = &column_schema.data_type;
-    let value = sql_value_to_value(column_name, data_type, sql_value);
-    match value {
-        Ok(value) => {
-            let mut vec = data_type.create_mutable_vector(1);
-            if vec.try_push_value_ref(value.as_value_ref()).is_err() {
-                return InvalidSqlSnafu {
-                    msg: format!(
-                        "invalid sql, column name is {column_name}, value is {sql_value}",
-                    ),
-                }
-                    .fail();
-            }
-            Ok(vec.to_vector())
-        }
-        _ => InvalidSqlSnafu {
-            msg: format!("invalid sql, column name is {column_name}, value is {sql_value}",),
-        }
-        .fail(),
-    }
-}
--- a/src/datanode/src/sql/flush_table.rs
+++ b/src/datanode/src/sql/flush_table.rs
@@ -0,0 +1,65 @@
+// 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 catalog::SchemaProviderRef;
+use common_query::Output;
+use snafu::{OptionExt, ResultExt};
+use table::requests::FlushTableRequest;
+
+use crate::error::{self, CatalogSnafu, DatabaseNotFoundSnafu, Result};
+use crate::sql::SqlHandler;
+
+impl SqlHandler {
+    pub(crate) async fn flush_table(&self, req: FlushTableRequest) -> Result<Output> {
+        let schema = self
+            .catalog_manager
+            .schema(&req.catalog_name, &req.schema_name)
+            .context(CatalogSnafu)?
+            .context(DatabaseNotFoundSnafu {
+                catalog: &req.catalog_name,
+                schema: &req.schema_name,
+            })?;
+
+        if let Some(table) = &req.table_name {
+            self.flush_table_inner(schema, table, req.region_number, req.wait)
+                .await?;
+        } else {
+            let all_table_names = schema.table_names().context(CatalogSnafu)?;
+            futures::future::join_all(all_table_names.iter().map(|table| {
+                self.flush_table_inner(schema.clone(), table, req.region_number, req.wait)
+            }))
+            .await
+            .into_iter()
+            .collect::<Result<Vec<_>>>()?;
+        }
+        Ok(Output::AffectedRows(0))
+    }
+
+    async fn flush_table_inner(
+        &self,
+        schema: SchemaProviderRef,
+        table_name: &str,
+        region: Option<u32>,
+        wait: Option<bool>,
+    ) -> Result<()> {
+        schema
+            .table(table_name)
+            .await
+            .context(error::FindTableSnafu { table_name })?
+            .context(error::TableNotFoundSnafu { table_name })?
+            .flush(region, wait)
+            .await
+            .context(error::FlushTableSnafu { table_name })
+    }
+}
--- a/src/datanode/src/sql/insert.rs
+++ b/src/datanode/src/sql/insert.rs
@@ -11,49 +11,31 @@
 // 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::pin::Pin;

 use catalog::CatalogManagerRef;
 use common_catalog::format_full_table_name;
 use common_query::Output;
-use common_recordbatch::RecordBatch;
-use datafusion_expr::type_coercion::binary::coerce_types;
-use datafusion_expr::Operator;
 use datatypes::data_type::DataType;
 use datatypes::schema::ColumnSchema;
 use datatypes::vectors::MutableVector;
-use futures::stream::{self, StreamExt};
-use futures::Stream;
-use query::parser::QueryStatement;
 use session::context::QueryContextRef;
 use snafu::{ensure, OptionExt, ResultExt};
 use sql::ast::Value as SqlValue;
 use sql::statements::insert::Insert;
-use sql::statements::statement::Statement;
 use sql::statements::{self};
 use table::engine::TableReference;
 use table::requests::*;
 use table::TableRef;

 use crate::error::{
-    CatalogSnafu, CollectRecordsSnafu, ColumnDefaultValueSnafu, ColumnNoneDefaultValueSnafu,
-    ColumnNotFoundSnafu, ColumnTypeMismatchSnafu, ColumnValuesNumberMismatchSnafu, Error,
-    ExecuteLogicalPlanSnafu, InsertSnafu, MissingInsertBodySnafu, ParseSqlSnafu,
-    ParseSqlValueSnafu, PlanStatementSnafu, Result, TableNotFoundSnafu,
+    CatalogSnafu, ColumnDefaultValueSnafu, ColumnNoneDefaultValueSnafu, ColumnNotFoundSnafu,
+    ColumnValuesNumberMismatchSnafu, InsertSnafu, MissingInsertBodySnafu, ParseSqlSnafu,
+    ParseSqlValueSnafu, Result, TableNotFoundSnafu,
 };
-use crate::sql::{table_idents_to_full_name, SqlHandler, SqlRequest};
+use crate::sql::{table_idents_to_full_name, SqlHandler};

 const DEFAULT_PLACEHOLDER_VALUE: &str = "default";

-type InsertRequestStream = Pin<Box<dyn Stream<Item = Result<SqlRequest>> + Send>>;
-pub(crate) enum InsertRequests {
-    // Single request
-    Request(SqlRequest),
-    // Streaming requests
-    Stream(InsertRequestStream),
-}
-
 impl SqlHandler {
    pub(crate) async fn insert(&self, req: InsertRequest) -> Result<Output> {
        // FIXME(dennis): table_ref is used in InsertSnafu and the req is consumed
@@ -77,7 +59,7 @@ impl SqlHandler {
        table_ref: TableReference,
        table: &TableRef,
        stmt: Insert,
-    ) -> Result<SqlRequest> {
+    ) -> Result<InsertRequest> {
        let values = stmt
            .values_body()
            .context(ParseSqlValueSnafu)?
@@ -129,7 +111,7 @@ impl SqlHandler {
            }
        }

-        Ok(SqlRequest::Insert(InsertRequest {
+        Ok(InsertRequest {
            catalog_name: table_ref.catalog.to_string(),
            schema_name: table_ref.schema.to_string(),
            table_name: table_ref.table.to_string(),
@@ -138,150 +120,14 @@ impl SqlHandler {
                .map(|(cs, mut b)| (cs.name.to_string(), b.to_vector()))
                .collect(),
            region_number: 0,
-        }))
+        })
    }

-    fn build_request_from_batch(
-        stmt: Insert,
-        table: TableRef,
-        batch: RecordBatch,
-        query_ctx: QueryContextRef,
-    ) -> Result<SqlRequest> {
-        let (catalog_name, schema_name, table_name) =
-            table_idents_to_full_name(stmt.table_name(), query_ctx)?;
-
-        let schema = table.schema();
-        let columns: Vec<_> = if stmt.columns().is_empty() {
-            schema
-                .column_schemas()
-                .iter()
-                .map(|c| c.name.to_string())
-                .collect()
-        } else {
-            stmt.columns().iter().map(|c| (*c).clone()).collect()
-        };
-        let columns_num = columns.len();
-
-        ensure!(
-            batch.num_columns() == columns_num,
-            ColumnValuesNumberMismatchSnafu {
-                columns: columns_num,
-                values: batch.num_columns(),
-            }
-        );
-
-        let batch_schema = &batch.schema;
-        let batch_columns = batch_schema.column_schemas();
-        assert_eq!(batch_columns.len(), columns_num);
-        let mut columns_values = HashMap::with_capacity(columns_num);
-
-        for (i, column_name) in columns.into_iter().enumerate() {
-            let column_schema = schema
-                .column_schema_by_name(&column_name)
-                .with_context(|| ColumnNotFoundSnafu {
-                    table_name: &table_name,
-                    column_name: &column_name,
-                })?;
-            let expect_datatype = column_schema.data_type.as_arrow_type();
-            // It's safe to retrieve the column schema by index, we already
-            // check columns number is the same above.
-            let batch_datatype = batch_columns[i].data_type.as_arrow_type();
-            let coerced_type = coerce_types(&expect_datatype, &Operator::Eq, &batch_datatype)
-                .map_err(|_| Error::ColumnTypeMismatch {
-                    column: column_name.clone(),
-                    expected: column_schema.data_type.clone(),
-                    actual: batch_columns[i].data_type.clone(),
-                })?;
-
-            ensure!(
-                expect_datatype == coerced_type,
-                ColumnTypeMismatchSnafu {
-                    column: column_name,
-                    expected: column_schema.data_type.clone(),
-                    actual: batch_columns[i].data_type.clone(),
-                }
-            );
-            let vector = batch
-                .column(i)
-                .cast(&column_schema.data_type)
-                .map_err(|_| Error::ColumnTypeMismatch {
-                    column: column_name.clone(),
-                    expected: column_schema.data_type.clone(),
-                    actual: batch_columns[i].data_type.clone(),
-                })?;
-
-            columns_values.insert(column_name, vector);
-        }
-
-        Ok(SqlRequest::Insert(InsertRequest {
-            catalog_name,
-            schema_name,
-            table_name,
-            columns_values,
-            region_number: 0,
-        }))
-    }
-
-    // FIXME(dennis): move it to frontend when refactor is done.
-    async fn build_stream_from_query(
-        &self,
-        table: TableRef,
-        stmt: Insert,
-        query_ctx: QueryContextRef,
-    ) -> Result<InsertRequestStream> {
-        let query = stmt
-            .query_body()
-            .context(ParseSqlValueSnafu)?
-            .context(MissingInsertBodySnafu)?;
-
-        let logical_plan = self
-            .query_engine
-            .planner()
-            .plan(
-                QueryStatement::Sql(Statement::Query(Box::new(query))),
-                query_ctx.clone(),
-            )
-            .await
-            .context(PlanStatementSnafu)?;
-
-        let output = self
-            .query_engine
-            .execute(&logical_plan)
-            .await
-            .context(ExecuteLogicalPlanSnafu)?;
-
-        let stream: InsertRequestStream = match output {
-            Output::RecordBatches(batches) => {
-                Box::pin(stream::iter(batches.take()).map(move |batch| {
-                    Self::build_request_from_batch(
-                        stmt.clone(),
-                        table.clone(),
-                        batch,
-                        query_ctx.clone(),
-                    )
-                }))
-            }
-
-            Output::Stream(stream) => Box::pin(stream.map(move |batch| {
-                Self::build_request_from_batch(
-                    stmt.clone(),
-                    table.clone(),
-                    batch.context(CollectRecordsSnafu)?,
-                    query_ctx.clone(),
-                )
-            })),
-            _ => unreachable!(),
-        };
-
-        Ok(stream)
-    }
-
-    pub(crate) async fn insert_to_requests(
-        &self,
+    pub async fn insert_to_request(
        catalog_manager: CatalogManagerRef,
        stmt: Insert,
        query_ctx: QueryContextRef,
-    ) -> Result<InsertRequests> {
+    ) -> Result<InsertRequest> {
        let (catalog_name, schema_name, table_name) =
            table_idents_to_full_name(stmt.table_name(), query_ctx.clone())?;

@@ -293,16 +139,8 @@ impl SqlHandler {
                table_name: format_full_table_name(&catalog_name, &schema_name, &table_name),
            })?;

-        if stmt.is_insert_select() {
-            Ok(InsertRequests::Stream(
-                self.build_stream_from_query(table, stmt, query_ctx).await?,
-            ))
-        } else {
-            let table_ref = TableReference::full(&catalog_name, &schema_name, &table_name);
-            Ok(InsertRequests::Request(Self::build_request_from_values(
-                table_ref, &table, stmt,
-            )?))
-        }
+        let table_ref = TableReference::full(&catalog_name, &schema_name, &table_name);
+        Self::build_request_from_values(table_ref, &table, stmt)
    }
 }

--- a/src/datanode/src/tests/instance_test.rs
+++ b/src/datanode/src/tests/instance_test.rs
@@ -12,15 +12,17 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.

+use std::env;
 use std::sync::Arc;

 use common_catalog::consts::{DEFAULT_CATALOG_NAME, DEFAULT_SCHEMA_NAME};
 use common_query::Output;
 use common_recordbatch::util;
+use common_telemetry::logging;
 use datatypes::data_type::ConcreteDataType;
 use datatypes::vectors::{Int64Vector, StringVector, UInt64Vector, VectorRef};
 use query::parser::{QueryLanguageParser, QueryStatement};
-use session::context::QueryContext;
+use session::context::{QueryContext, QueryContextRef};
 use snafu::ResultExt;
 use sql::statements::statement::Statement;

@@ -215,20 +217,20 @@ async fn test_execute_insert_by_select() {
        try_execute_sql(&instance, "insert into demo2(host) select * from demo1")
            .await
            .unwrap_err(),
-        Error::ColumnValuesNumberMismatch { .. }
+        Error::PlanStatement { .. }
    ));
    assert!(matches!(
        try_execute_sql(&instance, "insert into demo2 select cpu,memory from demo1")
            .await
            .unwrap_err(),
-        Error::ColumnValuesNumberMismatch { .. }
+        Error::PlanStatement { .. }
    ));

    assert!(matches!(
        try_execute_sql(&instance, "insert into demo2(ts) select memory from demo1")
            .await
            .unwrap_err(),
-        Error::ColumnTypeMismatch { .. }
+        Error::PlanStatement { .. }
    ));

    let output = execute_sql(&instance, "insert into demo2 select * from demo1").await;
@@ -749,7 +751,7 @@ async fn test_delete() {

    let output = execute_sql(
        &instance,
-        "delete from test_table where host = host1 and ts = 1655276557000 ",
+        "delete from test_table where host = 'host1' and ts = 1655276557000 ",
    )
    .await;
    assert!(matches!(output, Output::AffectedRows(1)));
@@ -768,117 +770,141 @@ async fn test_delete() {
 }

 #[tokio::test(flavor = "multi_thread")]
-async fn test_execute_copy_to() {
-    let instance = setup_test_instance("test_execute_copy_to").await;
+async fn test_execute_copy_to_s3() {
+    logging::init_default_ut_logging();
+    if let Ok(bucket) = env::var("GT_S3_BUCKET") {
+        if !bucket.is_empty() {
+            let instance = setup_test_instance("test_execute_copy_to_s3").await;

-    // setups
-    execute_sql(
-        &instance,
-        "create table demo(host string, cpu double, memory double, ts timestamp time index);",
-    )
-    .await;
+            // setups
+            execute_sql(
+                &instance,
+                "create table demo(host string, cpu double, memory double, ts timestamp time index);",
+            )
+            .await;

-    let output = execute_sql(
-        &instance,
-        r#"insert into demo(host, cpu, memory, ts) values
+            let output = execute_sql(
+                &instance,
+                r#"insert into demo(host, cpu, memory, ts) values
                            ('host1', 66.6, 1024, 1655276557000),
                            ('host2', 88.8,  333.3, 1655276558000)
                            "#,
-    )
-    .await;
-    assert!(matches!(output, Output::AffectedRows(2)));
+            )
+            .await;
+            assert!(matches!(output, Output::AffectedRows(2)));
+            let key_id = env::var("GT_S3_ACCESS_KEY_ID").unwrap();
+            let key = env::var("GT_S3_ACCESS_KEY").unwrap();
+            let url =
+                env::var("GT_S3_ENDPOINT_URL").unwrap_or("https://s3.amazonaws.com".to_string());

-    // exports
-    let data_dir = instance.data_tmp_dir().path();
+            let root = uuid::Uuid::new_v4().to_string();

-    let copy_to_stmt = format!("Copy demo TO '{}/export/demo.parquet'", data_dir.display());
+            // exports
+            let copy_to_stmt = format!("Copy demo TO 's3://{}/{}/export/demo.parquet' CONNECTION (ACCESS_KEY_ID='{}',SECRET_ACCESS_KEY='{}',ENDPOINT_URL='{}')", bucket, root, key_id, key, url);

-    let output = execute_sql(&instance, &copy_to_stmt).await;
-    assert!(matches!(output, Output::AffectedRows(2)));
+            let output = execute_sql(&instance, &copy_to_stmt).await;
+            assert!(matches!(output, Output::AffectedRows(2)));
+        }
+    }
 }

 #[tokio::test(flavor = "multi_thread")]
-async fn test_execute_copy_from() {
-    let instance = setup_test_instance("test_execute_copy_from").await;
+async fn test_execute_copy_from_s3() {
+    logging::init_default_ut_logging();
+    if let Ok(bucket) = env::var("GT_S3_BUCKET") {
+        if !bucket.is_empty() {
+            let instance = setup_test_instance("test_execute_copy_from_s3").await;

-    // setups
-    execute_sql(
-        &instance,
-        "create table demo(host string, cpu double, memory double, ts timestamp time index);",
-    )
-    .await;
+            // setups
+            execute_sql(
+                &instance,
+                "create table demo(host string, cpu double, memory double, ts timestamp time index);",
+            )
+            .await;

-    let output = execute_sql(
-        &instance,
-        r#"insert into demo(host, cpu, memory, ts) values
+            let output = execute_sql(
+                &instance,
+                r#"insert into demo(host, cpu, memory, ts) values
                            ('host1', 66.6, 1024, 1655276557000),
                            ('host2', 88.8,  333.3, 1655276558000)
                            "#,
-    )
-    .await;
-    assert!(matches!(output, Output::AffectedRows(2)));
+            )
+            .await;
+            assert!(matches!(output, Output::AffectedRows(2)));

-    // export
-    let data_dir = instance.data_tmp_dir().path();
+            // export
+            let root = uuid::Uuid::new_v4().to_string();
+            let key_id = env::var("GT_S3_ACCESS_KEY_ID").unwrap();
+            let key = env::var("GT_S3_ACCESS_KEY").unwrap();
+            let url =
+                env::var("GT_S3_ENDPOINT_URL").unwrap_or("https://s3.amazonaws.com".to_string());

-    let copy_to_stmt = format!("Copy demo TO '{}/export/demo.parquet'", data_dir.display());
+            let copy_to_stmt = format!("Copy demo TO 's3://{}/{}/export/demo.parquet' CONNECTION (ACCESS_KEY_ID='{}',SECRET_ACCESS_KEY='{}',ENDPOINT_URL='{}')", bucket, root, key_id, key, url);
+            logging::info!("Copy table to s3: {}", copy_to_stmt);

-    let output = execute_sql(&instance, &copy_to_stmt).await;
-    assert!(matches!(output, Output::AffectedRows(2)));
+            let output = execute_sql(&instance, &copy_to_stmt).await;
+            assert!(matches!(output, Output::AffectedRows(2)));

-    struct Test<'a> {
-        sql: &'a str,
-        table_name: &'a str,
-    }
-    let tests = [
-        Test {
-            sql: &format!(
-                "Copy with_filename FROM '{}/export/demo.parquet_1_2'",
-                data_dir.display()
-            ),
-            table_name: "with_filename",
-        },
-        Test {
-            sql: &format!("Copy with_path FROM '{}/export/'", data_dir.display()),
-            table_name: "with_path",
-        },
-        Test {
-            sql: &format!(
-                "Copy with_pattern FROM '{}/export/' WITH (PATTERN = 'demo.*')",
-                data_dir.display()
-            ),
-            table_name: "with_pattern",
-        },
-    ];
+            struct Test<'a> {
+                sql: &'a str,
+                table_name: &'a str,
+            }
+            let tests = [
+                Test {
+                    sql: &format!(
+                        "Copy with_filename FROM 's3://{}/{}/export/demo.parquet_1_2'",
+                        bucket, root
+                    ),
+                    table_name: "with_filename",
+                },
+                Test {
+                    sql: &format!("Copy with_path FROM 's3://{}/{}/export/'", bucket, root),
+                    table_name: "with_path",
+                },
+                Test {
+                    sql: &format!(
+                        "Copy with_pattern FROM 's3://{}/{}/export/' WITH (PATTERN = 'demo.*')",
+                        bucket, root
+                    ),
+                    table_name: "with_pattern",
+                },
+            ];

-    for test in tests {
-        // import
-        execute_sql(
-            &instance,
-            &format!(
+            for test in tests {
+                // import
+                execute_sql(
+                    &instance,
+                    &format!(
                "create table {}(host string, cpu double, memory double, ts timestamp time index);",
                test.table_name
            ),
-        )
-        .await;
+                )
+                .await;
+                let sql = format!(
+                    "{} CONNECTION (ACCESS_KEY_ID='{}',SECRET_ACCESS_KEY='{}',ENDPOINT_URL='{}')",
+                    test.sql, key_id, key, url
+                );
+                logging::info!("Running sql: {}", sql);

-        let output = execute_sql(&instance, test.sql).await;
-        assert!(matches!(output, Output::AffectedRows(2)));
+                let output = execute_sql(&instance, &sql).await;
+                assert!(matches!(output, Output::AffectedRows(2)));

-        let output = execute_sql(
-            &instance,
-            &format!("select * from {} order by ts", test.table_name),
-        )
-        .await;
-        let expected = "\
+                let output = execute_sql(
+                    &instance,
+                    &format!("select * from {} order by ts", test.table_name),
+                )
+                .await;
+                let expected = "\
 +-------+------+--------+---------------------+
 | host  | cpu  | memory | ts                  |
 +-------+------+--------+---------------------+
 | host1 | 66.6 | 1024.0 | 2022-06-15T07:02:37 |
 | host2 | 88.8 | 333.3  | 2022-06-15T07:02:38 |
 +-------+------+--------+---------------------+"
-            .to_string();
-        check_output_stream(output, expected).await;
+                    .to_string();
+                check_output_stream(output, expected).await;
+            }
+        }
    }
 }

@@ -936,16 +962,30 @@ async fn try_execute_sql_in_db(
 ) -> Result<Output, crate::error::Error> {
    let query_ctx = Arc::new(QueryContext::with(DEFAULT_CATALOG_NAME, db));

+    async fn plan_exec(
+        instance: &MockInstance,
+        stmt: QueryStatement,
+        query_ctx: QueryContextRef,
+    ) -> Result<Output, Error> {
+        let engine = instance.inner().query_engine();
+        let plan = engine
+            .planner()
+            .plan(stmt, query_ctx.clone())
+            .await
+            .context(PlanStatementSnafu)?;
+        engine
+            .execute(plan, query_ctx)
+            .await
+            .context(ExecuteLogicalPlanSnafu)
+    }
+
    let stmt = QueryLanguageParser::parse_sql(sql).unwrap();
    match stmt {
-        QueryStatement::Sql(Statement::Query(_)) => {
-            let engine = instance.inner().query_engine();
-            let plan = engine
-                .planner()
-                .plan(stmt, query_ctx)
-                .await
-                .context(PlanStatementSnafu)?;
-            engine.execute(&plan).await.context(ExecuteLogicalPlanSnafu)
+        QueryStatement::Sql(Statement::Query(_)) | QueryStatement::Sql(Statement::Delete(_)) => {
+            plan_exec(instance, stmt, query_ctx).await
+        }
+        QueryStatement::Sql(Statement::Insert(ref insert)) if insert.is_insert_select() => {
+            plan_exec(instance, stmt, query_ctx).await
        }
        _ => instance.inner().execute_stmt(stmt, query_ctx).await,
    }
--- a/src/datanode/src/tests/test_util.rs
+++ b/src/datanode/src/tests/test_util.rs
@@ -13,6 +13,7 @@
 // limitations under the License.

 use std::sync::Arc;
+use std::time::Duration;

 use common_catalog::consts::{DEFAULT_CATALOG_NAME, DEFAULT_SCHEMA_NAME, MIN_USER_TABLE_ID};
 use common_query::Output;
@@ -23,7 +24,6 @@ use datatypes::schema::{ColumnSchema, RawSchema};
 use mito::config::EngineConfig;
 use mito::table::test_util::{new_test_object_store, MockEngine, MockMitoEngine};
 use query::parser::{PromQuery, QueryLanguageParser, QueryStatement};
-use query::QueryEngineFactory;
 use servers::Mode;
 use session::context::QueryContext;
 use snafu::ResultExt;
@@ -64,6 +64,8 @@ impl MockInstance {
            store: ObjectStoreConfig::File(FileConfig {
                data_dir: procedure_dir.path().to_str().unwrap().to_string(),
            }),
+            max_retry_times: 3,
+            retry_delay: Duration::from_millis(500),
        });

        let instance = Instance::with_mock_meta_client(&opts).await.unwrap();
@@ -84,7 +86,7 @@ impl MockInstance {
        match stmt {
            QueryStatement::Sql(Statement::Query(_)) => {
                let plan = planner.plan(stmt, QueryContext::arc()).await.unwrap();
-                engine.execute(&plan).await.unwrap()
+                engine.execute(plan, QueryContext::arc()).await.unwrap()
            }
            QueryStatement::Sql(Statement::Tql(tql)) => {
                let plan = match tql {
@@ -100,7 +102,7 @@ impl MockInstance {
                    }
                    Tql::Explain(_) => unimplemented!(),
                };
-                engine.execute(&plan).await.unwrap()
+                engine.execute(plan, QueryContext::arc()).await.unwrap()
            }
            _ => self
                .inner()
@@ -113,10 +115,6 @@ impl MockInstance {
    pub(crate) fn inner(&self) -> &Instance {
        &self.instance
    }
-
-    pub(crate) fn data_tmp_dir(&self) -> &TempDir {
-        &self._guard._data_tmp_dir
-    }
 }

 struct TestGuard {
@@ -202,17 +200,7 @@ pub async fn create_mock_sql_handler() -> SqlHandler {
            .await
            .unwrap(),
    );
-
-    let catalog_list = catalog::local::new_memory_catalog_list().unwrap();
-    let factory = QueryEngineFactory::new(catalog_list);
-
-    SqlHandler::new(
-        mock_engine.clone(),
-        catalog_manager,
-        factory.query_engine(),
-        mock_engine,
-        None,
-    )
+    SqlHandler::new(mock_engine.clone(), catalog_manager, mock_engine, None)
 }

 pub(crate) async fn setup_test_instance(test_name: &str) -> MockInstance {
--- a/src/frontend/src/instance.rs
+++ b/src/frontend/src/instance.rs
@@ -428,52 +428,63 @@ fn parse_stmt(sql: &str) -> Result<Vec<Statement>> {
 }

 impl Instance {
+    async fn plan_exec(&self, stmt: Statement, query_ctx: QueryContextRef) -> Result<Output> {
+        let planner = self.query_engine.planner();
+        let plan = planner
+            .plan(QueryStatement::Sql(stmt), query_ctx.clone())
+            .await
+            .context(PlanStatementSnafu)?;
+        self.query_engine
+            .execute(plan, query_ctx)
+            .await
+            .context(ExecLogicalPlanSnafu)
+    }
+
+    async fn execute_tql(&self, tql: Tql, query_ctx: QueryContextRef) -> Result<Output> {
+        let plan = match tql {
+            Tql::Eval(eval) => {
+                let promql = PromQuery {
+                    start: eval.start,
+                    end: eval.end,
+                    step: eval.step,
+                    query: eval.query,
+                };
+                let stmt = QueryLanguageParser::parse_promql(&promql).context(ParseQuerySnafu)?;
+                self.query_engine
+                    .planner()
+                    .plan(stmt, query_ctx.clone())
+                    .await
+                    .context(PlanStatementSnafu)?
+            }
+            Tql::Explain(_) => unimplemented!(),
+        };
+        self.query_engine
+            .execute(plan, query_ctx)
+            .await
+            .context(ExecLogicalPlanSnafu)
+    }
+
    async fn query_statement(&self, stmt: Statement, query_ctx: QueryContextRef) -> Result<Output> {
        check_permission(self.plugins.clone(), &stmt, &query_ctx)?;

-        let planner = self.query_engine.planner();
-
        match stmt {
-            Statement::Query(_) | Statement::Explain(_) => {
-                let plan = planner
-                    .plan(QueryStatement::Sql(stmt), query_ctx)
-                    .await
-                    .context(PlanStatementSnafu)?;
-                self.query_engine
-                    .execute(&plan)
-                    .await
-                    .context(ExecLogicalPlanSnafu)
+            Statement::Query(_) | Statement::Explain(_) | Statement::Delete(_) => {
+                self.plan_exec(stmt, query_ctx).await
            }
-            Statement::Tql(tql) => {
-                let plan = match tql {
-                    Tql::Eval(eval) => {
-                        let promql = PromQuery {
-                            start: eval.start,
-                            end: eval.end,
-                            step: eval.step,
-                            query: eval.query,
-                        };
-                        let stmt =
-                            QueryLanguageParser::parse_promql(&promql).context(ParseQuerySnafu)?;
-                        planner
-                            .plan(stmt, query_ctx)
-                            .await
-                            .context(PlanStatementSnafu)?
-                    }
-                    Tql::Explain(_) => unimplemented!(),
-                };
-                self.query_engine
-                    .execute(&plan)
-                    .await
-                    .context(ExecLogicalPlanSnafu)
+
+            // For performance consideration, only "insert with select" is executed by query engine.
+            // Plain insert ("insert with values") is still executed directly in statement.
+            Statement::Insert(ref insert) if insert.is_insert_select() => {
+                self.plan_exec(stmt, query_ctx).await
            }
+
+            Statement::Tql(tql) => self.execute_tql(tql, query_ctx).await,
            Statement::CreateDatabase(_)
            | Statement::ShowDatabases(_)
            | Statement::CreateTable(_)
            | Statement::ShowTables(_)
            | Statement::DescribeTable(_)
            | Statement::Insert(_)
-            | Statement::Delete(_)
            | Statement::Alter(_)
            | Statement::DropTable(_)
            | Statement::Copy(_) => self
@@ -647,8 +658,8 @@ pub fn check_permission(
    }

    match stmt {
-        // query,explain and tql will be checked in QueryEngineState
-        Statement::Query(_) | Statement::Explain(_) | Statement::Tql(_) => {}
+        // These are executed by query engine, and will be checked there.
+        Statement::Query(_) | Statement::Explain(_) | Statement::Tql(_) | Statement::Delete(_) => {}
        // database ops won't be checked
        Statement::CreateDatabase(_) | Statement::ShowDatabases(_) | Statement::Use(_) => {}
        // show create table and alter are not supported yet
@@ -673,11 +684,8 @@ pub fn check_permission(
        Statement::DescribeTable(stmt) => {
            validate_param(stmt.name(), query_ctx)?;
        }
-        Statement::Delete(delete) => {
-            validate_param(delete.table_name(), query_ctx)?;
-        }
        Statement::Copy(stmd) => match stmd {
-            CopyTable::To(copy_table_to) => validate_param(copy_table_to.table_name(), query_ctx)?,
+            CopyTable::To(copy_table_to) => validate_param(&copy_table_to.table_name, query_ctx)?,
            CopyTable::From(copy_table_from) => {
                validate_param(&copy_table_from.table_name, query_ctx)?
            }
@@ -1086,7 +1094,7 @@ mod tests {
                .plan(stmt.clone(), QueryContext::arc())
                .await
                .unwrap();
-            let output = engine.execute(&plan).await.unwrap();
+            let output = engine.execute(plan, QueryContext::arc()).await.unwrap();
            let Output::Stream(stream) = output else { unreachable!() };
            let recordbatches = RecordBatches::try_collect(stream).await.unwrap();
            let actual = recordbatches.pretty_print().unwrap();
--- a/src/frontend/src/instance/distributed.rs
+++ b/src/frontend/src/instance/distributed.rs
@@ -19,8 +19,8 @@ use std::sync::Arc;

 use api::helper::ColumnDataTypeWrapper;
 use api::v1::{
-    column_def, AlterExpr, CreateDatabaseExpr, CreateTableExpr, DropTableExpr, InsertRequest,
-    TableId,
+    column_def, AlterExpr, CreateDatabaseExpr, CreateTableExpr, DropTableExpr, FlushTableExpr,
+    InsertRequest, TableId,
 };
 use async_trait::async_trait;
 use catalog::helper::{SchemaKey, SchemaValue};
@@ -33,13 +33,14 @@ use common_error::prelude::BoxedError;
 use common_query::Output;
 use common_telemetry::{debug, info};
 use datanode::instance::sql::table_idents_to_full_name;
+use datanode::sql::SqlHandler;
 use datatypes::prelude::ConcreteDataType;
 use datatypes::schema::RawSchema;
 use meta_client::client::MetaClient;
 use meta_client::rpc::router::DeleteRequest as MetaDeleteRequest;
 use meta_client::rpc::{
    CompareAndPutRequest, CreateRequest as MetaCreateRequest, Partition as MetaPartition,
-    RouteResponse, TableName,
+    RouteRequest, RouteResponse, TableName,
 };
 use partition::partition::{PartitionBound, PartitionDef};
 use query::error::QueryExecutionSnafu;
@@ -60,13 +61,12 @@ use crate::catalog::FrontendCatalogManager;
 use crate::datanode::DatanodeClients;
 use crate::error::{
    self, AlterExprToRequestSnafu, CatalogEntrySerdeSnafu, CatalogSnafu, ColumnDataTypeSnafu,
-    DeserializePartitionSnafu, NotSupportedSnafu, ParseSqlSnafu, PrimaryKeyNotFoundSnafu,
-    RequestDatanodeSnafu, RequestMetaSnafu, Result, SchemaExistsSnafu, StartMetaClientSnafu,
-    TableAlreadyExistSnafu, TableNotFoundSnafu, TableSnafu, ToTableInsertRequestSnafu,
-    UnrecognizedTableOptionSnafu,
+    DeserializePartitionSnafu, InvokeDatanodeSnafu, NotSupportedSnafu, ParseSqlSnafu,
+    PrimaryKeyNotFoundSnafu, RequestDatanodeSnafu, RequestMetaSnafu, Result, SchemaExistsSnafu,
+    StartMetaClientSnafu, TableAlreadyExistSnafu, TableNotFoundSnafu, TableSnafu,
+    ToTableInsertRequestSnafu, UnrecognizedTableOptionSnafu,
 };
 use crate::expr_factory;
-use crate::sql::insert_to_request;
 use crate::table::DistTable;

 #[derive(Clone)]
@@ -259,6 +259,61 @@ impl DistInstance {
        Ok(Output::AffectedRows(1))
    }

+    async fn flush_table(&self, table_name: TableName, region_id: Option<u32>) -> Result<Output> {
+        let _ = self
+            .catalog_manager
+            .table(
+                &table_name.catalog_name,
+                &table_name.schema_name,
+                &table_name.table_name,
+            )
+            .await
+            .context(CatalogSnafu)?
+            .with_context(|| TableNotFoundSnafu {
+                table_name: table_name.to_string(),
+            })?;
+
+        let route_response = self
+            .meta_client
+            .route(RouteRequest {
+                table_names: vec![table_name.clone()],
+            })
+            .await
+            .context(RequestMetaSnafu)?;
+
+        let expr = FlushTableExpr {
+            catalog_name: table_name.catalog_name.clone(),
+            schema_name: table_name.schema_name.clone(),
+            table_name: table_name.table_name.clone(),
+            region_id,
+        };
+
+        for table_route in &route_response.table_routes {
+            let should_send_rpc = table_route.region_routes.iter().any(|route| {
+                if let Some(region_id) = region_id {
+                    region_id == route.region.id as u32
+                } else {
+                    true
+                }
+            });
+
+            if !should_send_rpc {
+                continue;
+            }
+            for datanode in table_route.find_leaders() {
+                debug!("Flushing table {table_name} on Datanode {datanode:?}");
+
+                let client = self.datanode_clients.get_client(&datanode).await;
+                let client = Database::new(&expr.catalog_name, &expr.schema_name, client);
+                client
+                    .flush_table(expr.clone())
+                    .await
+                    .context(RequestDatanodeSnafu)?;
+            }
+        }
+        Ok(Output::AffectedRows(0))
+    }
+
    async fn handle_statement(
        &self,
        stmt: Statement,
@@ -319,7 +374,10 @@ impl DistInstance {
                    .context(CatalogSnafu)?
                    .context(TableNotFoundSnafu { table_name: table })?;

-                let insert_request = insert_to_request(&table, *insert, query_ctx)?;
+                let insert_request =
+                    SqlHandler::insert_to_request(self.catalog_manager.clone(), *insert, query_ctx)
+                        .await
+                        .context(InvokeDatanodeSnafu)?;

                return Ok(Output::AffectedRows(
                    table.insert(insert_request).await.context(TableSnafu)?,
--- a/src/frontend/src/instance/distributed/grpc.rs
+++ b/src/frontend/src/instance/distributed/grpc.rs
@@ -57,7 +57,11 @@ impl GrpcQueryHandler for DistInstance {
                            TableName::new(&expr.catalog_name, &expr.schema_name, &expr.table_name);
                        self.drop_table(table_name).await
                    }
-                    DdlExpr::FlushTable(_) => todo!(),
+                    DdlExpr::FlushTable(expr) => {
+                        let table_name =
+                            TableName::new(&expr.catalog_name, &expr.schema_name, &expr.table_name);
+                        self.flush_table(table_name, expr.region_id).await
+                    }
                }
            }
        }
--- a/src/frontend/src/instance/grpc.rs
+++ b/src/frontend/src/instance/grpc.rs
@@ -91,14 +91,15 @@ mod test {
    use api::v1::ddl_request::Expr as DdlExpr;
    use api::v1::{
        alter_expr, AddColumn, AddColumns, AlterExpr, Column, ColumnDataType, ColumnDef,
-        CreateDatabaseExpr, CreateTableExpr, DdlRequest, DropTableExpr, InsertRequest,
-        QueryRequest,
+        CreateDatabaseExpr, CreateTableExpr, DdlRequest, DropTableExpr, FlushTableExpr,
+        InsertRequest, QueryRequest,
    };
    use catalog::helper::{TableGlobalKey, TableGlobalValue};
    use common_query::Output;
    use common_recordbatch::RecordBatches;
    use query::parser::QueryLanguageParser;
    use session::context::QueryContext;
+    use tests::{has_parquet_file, test_region_dir};

    use super::*;
    use crate::table::DistTable;
@@ -352,6 +353,108 @@ CREATE TABLE {table_name} (
        test_insert_and_query_on_auto_created_table(instance).await
    }

+    #[tokio::test(flavor = "multi_thread")]
+    async fn test_distributed_flush_table() {
+        common_telemetry::init_default_ut_logging();
+
+        let instance = tests::create_distributed_instance("test_distributed_flush_table").await;
+        let data_tmp_dirs = instance.data_tmp_dirs();
+        let frontend = instance.frontend.as_ref();
+
+        let table_name = "my_dist_table";
+        let sql = format!(
+            r"
+CREATE TABLE {table_name} (
+    a INT,
+    ts TIMESTAMP,
+    TIME INDEX (ts)
+) PARTITION BY RANGE COLUMNS(a) (
+    PARTITION r0 VALUES LESS THAN (10),
+    PARTITION r1 VALUES LESS THAN (20),
+    PARTITION r2 VALUES LESS THAN (50),
+    PARTITION r3 VALUES LESS THAN (MAXVALUE),
+)"
+        );
+        create_table(frontend, sql).await;
+
+        test_insert_and_query_on_existing_table(frontend, table_name).await;
+
+        flush_table(frontend, "greptime", "public", table_name, None).await;
+        // Wait for previous task finished
+        flush_table(frontend, "greptime", "public", table_name, None).await;
+
+        let table_id = 1024;
+
+        let table = instance
+            .frontend
+            .catalog_manager()
+            .table("greptime", "public", table_name)
+            .await
+            .unwrap()
+            .unwrap();
+        let table = table.as_any().downcast_ref::<DistTable>().unwrap();
+
+        let TableGlobalValue { regions_id_map, .. } = table
+            .table_global_value(&TableGlobalKey {
+                catalog_name: "greptime".to_string(),
+                schema_name: "public".to_string(),
+                table_name: table_name.to_string(),
+            })
+            .await
+            .unwrap()
+            .unwrap();
+        let region_to_dn_map = regions_id_map
+            .iter()
+            .map(|(k, v)| (v[0], *k))
+            .collect::<HashMap<u32, u64>>();
+
+        for (region, dn) in region_to_dn_map.iter() {
+            // data_tmp_dirs -> dn: 1..4
+            let data_tmp_dir = data_tmp_dirs.get((*dn - 1) as usize).unwrap();
+            let region_dir = test_region_dir(
+                data_tmp_dir.path().to_str().unwrap(),
+                "greptime",
+                "public",
+                table_id,
+                *region,
+            );
+            has_parquet_file(&region_dir);
+        }
+    }
+
+    #[tokio::test(flavor = "multi_thread")]
+    async fn test_standalone_flush_table() {
+        common_telemetry::init_default_ut_logging();
+
+        let standalone = tests::create_standalone_instance("test_standalone_flush_table").await;
+        let instance = &standalone.instance;
+        let data_tmp_dir = standalone.data_tmp_dir();
+
+        let table_name = "my_table";
+        let sql = format!("CREATE TABLE {table_name} (a INT, ts TIMESTAMP, TIME INDEX (ts))");
+
+        create_table(instance, sql).await;
+
+        test_insert_and_query_on_existing_table(instance, table_name).await;
+
+        let table_id = 1024;
+        let region_id = 0;
+        let region_dir = test_region_dir(
+            data_tmp_dir.path().to_str().unwrap(),
+            "greptime",
+            "public",
+            table_id,
+            region_id,
+        );
+        assert!(!has_parquet_file(&region_dir));
+
+        flush_table(instance, "greptime", "public", "my_table", None).await;
+        // Wait for previous task finished
+        flush_table(instance, "greptime", "public", "my_table", None).await;
+
+        assert!(has_parquet_file(&region_dir));
+    }
+
    async fn create_table(frontend: &Instance, sql: String) {
        let request = Request::Query(QueryRequest {
            query: Some(Query::Sql(sql)),
@@ -360,6 +463,26 @@ CREATE TABLE {table_name} (
        assert!(matches!(output, Output::AffectedRows(0)));
    }

+    async fn flush_table(
+        frontend: &Instance,
+        catalog_name: &str,
+        schema_name: &str,
+        table_name: &str,
+        region_id: Option<u32>,
+    ) {
+        let request = Request::Ddl(DdlRequest {
+            expr: Some(DdlExpr::FlushTable(FlushTableExpr {
+                catalog_name: catalog_name.to_string(),
+                schema_name: schema_name.to_string(),
+                table_name: table_name.to_string(),
+                region_id,
+            })),
+        });
+
+        let output = query(frontend, request).await;
+        assert!(matches!(output, Output::AffectedRows(0)));
+    }
+
    async fn test_insert_and_query_on_existing_table(instance: &Instance, table_name: &str) {
        let insert = InsertRequest {
            table_name: table_name.to_string(),
@@ -467,7 +590,7 @@ CREATE TABLE {table_name} (
                .plan(stmt, QueryContext::arc())
                .await
                .unwrap();
-            let output = engine.execute(&plan).await.unwrap();
+            let output = engine.execute(plan, QueryContext::arc()).await.unwrap();
            let Output::Stream(stream) = output else { unreachable!() };
            let recordbatches = RecordBatches::try_collect(stream).await.unwrap();
            let actual = recordbatches.pretty_print().unwrap();
--- a/src/frontend/src/lib.rs
+++ b/src/frontend/src/lib.rs
@@ -28,7 +28,6 @@ pub mod postgres;
 pub mod prom;
 pub mod prometheus;
 mod server;
-mod sql;
 mod table;
 #[cfg(test)]
 mod tests;
--- a/src/frontend/src/server.rs
+++ b/src/frontend/src/server.rs
@@ -152,6 +152,7 @@ impl Services {

            let mut http_server = HttpServer::new(
                ServerSqlQueryHandlerAdaptor::arc(instance.clone()),
+                ServerGrpcQueryHandlerAdaptor::arc(instance.clone()),
                http_options.clone(),
            );
            if let Some(user_provider) = user_provider.clone() {
--- a/src/frontend/src/sql.rs
+++ b/src/frontend/src/sql.rs
@@ -1,130 +0,0 @@
-// Copyright 2023 Greptime Team
-//
-// Licensed under the Apache License, Version 2.0 (the "License");
-// you may not use this file except in compliance with the License.
-// You may obtain a copy of the License at
-//
-//     http://www.apache.org/licenses/LICENSE-2.0
-//
-// Unless required by applicable law or agreed to in writing, software
-// distributed under the License is distributed on an "AS IS" BASIS,
-// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-// See the License for the specific language governing permissions and
-// limitations under the License.
-
-use common_error::ext::BoxedError;
-use common_error::snafu::ensure;
-use datanode::instance::sql::table_idents_to_full_name;
-use datatypes::data_type::DataType;
-use datatypes::prelude::MutableVector;
-use datatypes::schema::ColumnSchema;
-use session::context::QueryContextRef;
-use snafu::{OptionExt, ResultExt};
-use sql::ast::Value as SqlValue;
-use sql::statements;
-use sql::statements::insert::Insert;
-use table::requests::InsertRequest;
-use table::TableRef;
-
-use crate::error::{self, ExternalSnafu, Result};
-
-const DEFAULT_PLACEHOLDER_VALUE: &str = "default";
-
-// TODO(fys): Extract the common logic in datanode and frontend in the future.
-// This function convert insert statement to an `InsertRequest` to region 0.
-pub(crate) fn insert_to_request(
-    table: &TableRef,
-    stmt: Insert,
-    query_ctx: QueryContextRef,
-) -> Result<InsertRequest> {
-    let columns = stmt.columns();
-    let values = stmt
-        .values_body()
-        .context(error::ParseSqlSnafu)?
-        .context(error::MissingInsertValuesSnafu)?;
-
-    let (catalog_name, schema_name, table_name) =
-        table_idents_to_full_name(stmt.table_name(), query_ctx)
-            .map_err(BoxedError::new)
-            .context(ExternalSnafu)?;
-
-    let schema = table.schema();
-    let columns_num = if columns.is_empty() {
-        schema.column_schemas().len()
-    } else {
-        columns.len()
-    };
-    let rows_num = values.len();
-
-    let mut columns_builders: Vec<(&ColumnSchema, Box<dyn MutableVector>)> =
-        Vec::with_capacity(columns_num);
-
-    if columns.is_empty() {
-        for column_schema in schema.column_schemas() {
-            let data_type = &column_schema.data_type;
-            columns_builders.push((column_schema, data_type.create_mutable_vector(rows_num)));
-        }
-    } else {
-        for column_name in columns {
-            let column_schema = schema.column_schema_by_name(column_name).with_context(|| {
-                error::ColumnNotFoundSnafu {
-                    table_name: &table_name,
-                    column_name: column_name.to_string(),
-                }
-            })?;
-            let data_type = &column_schema.data_type;
-            columns_builders.push((column_schema, data_type.create_mutable_vector(rows_num)));
-        }
-    }
-
-    for row in values {
-        ensure!(
-            row.len() == columns_num,
-            error::ColumnValuesNumberMismatchSnafu {
-                columns: columns_num,
-                values: row.len(),
-            }
-        );
-
-        for (sql_val, (column_schema, builder)) in row.iter().zip(columns_builders.iter_mut()) {
-            add_row_to_vector(column_schema, sql_val, builder)?;
-        }
-    }
-
-    Ok(InsertRequest {
-        catalog_name,
-        schema_name,
-        table_name,
-        columns_values: columns_builders
-            .into_iter()
-            .map(|(cs, mut b)| (cs.name.to_string(), b.to_vector()))
-            .collect(),
-        region_number: 0,
-    })
-}
-
-fn add_row_to_vector(
-    column_schema: &ColumnSchema,
-    sql_val: &SqlValue,
-    builder: &mut Box<dyn MutableVector>,
-) -> Result<()> {
-    let value = if replace_default(sql_val) {
-        column_schema
-            .create_default()
-            .context(error::ColumnDefaultValueSnafu {
-                column: column_schema.name.to_string(),
-            })?
-            .context(error::ColumnNoneDefaultValueSnafu {
-                column: column_schema.name.to_string(),
-            })?
-    } else {
-        statements::sql_value_to_value(&column_schema.name, &column_schema.data_type, sql_val)
-            .context(error::ParseSqlSnafu)?
-    };
-    builder.push_value_ref(value.as_value_ref());
-    Ok(())
-}
-
-fn replace_default(sql_val: &SqlValue) -> bool {
-    matches!(sql_val, SqlValue::Placeholder(s) if s.to_lowercase() == DEFAULT_PLACEHOLDER_VALUE)
-}
--- a/src/frontend/src/table.rs
+++ b/src/frontend/src/table.rs
@@ -140,8 +140,11 @@ impl Table for DistTable {
        Ok(Arc::new(dist_scan))
    }

-    fn supports_filter_pushdown(&self, _filter: &Expr) -> table::Result<FilterPushDownType> {
-        Ok(FilterPushDownType::Inexact)
+    fn supports_filters_pushdown(
+        &self,
+        filters: &[&Expr],
+    ) -> table::Result<Vec<FilterPushDownType>> {
+        Ok(vec![FilterPushDownType::Inexact; filters.len()])
    }

    async fn alter(&self, context: AlterContext, request: &AlterTableRequest) -> table::Result<()> {
--- a/src/frontend/src/table/insert.rs
+++ b/src/frontend/src/table/insert.rs
@@ -74,8 +74,7 @@ impl DistTable {

        let mut success = 0;
        for join in joins {
-            let object_result = join.await.context(error::JoinTaskSnafu)??;
-            let Output::AffectedRows(rows) = object_result else { unreachable!() };
+            let rows = join.await.context(error::JoinTaskSnafu)?? as usize;
            success += rows;
        }
        Ok(Output::AffectedRows(success))
--- a/src/frontend/src/table/scan.rs
+++ b/src/frontend/src/table/scan.rs
@@ -47,7 +47,7 @@ impl DatanodeInstance {
        Self { table, db }
    }

-    pub(crate) async fn grpc_insert(&self, request: InsertRequest) -> client::Result<Output> {
+    pub(crate) async fn grpc_insert(&self, request: InsertRequest) -> client::Result<u32> {
        self.db.insert(request).await
    }

--- a/src/frontend/src/tests.rs
+++ b/src/frontend/src/tests.rs
@@ -34,6 +34,7 @@ use partition::route::TableRoutes;
 use servers::grpc::GrpcServer;
 use servers::query_handler::grpc::ServerGrpcQueryHandlerAdaptor;
 use servers::Mode;
+use table::engine::{region_name, table_dir};
 use tonic::transport::Server;
 use tower::service_fn;

@@ -56,11 +57,23 @@ pub(crate) struct MockDistributedInstance {
    _guards: Vec<TestGuard>,
 }

+impl MockDistributedInstance {
+    pub fn data_tmp_dirs(&self) -> Vec<&TempDir> {
+        self._guards.iter().map(|g| &g._data_tmp_dir).collect()
+    }
+}
+
 pub(crate) struct MockStandaloneInstance {
    pub(crate) instance: Arc<Instance>,
    _guard: TestGuard,
 }

+impl MockStandaloneInstance {
+    pub fn data_tmp_dir(&self) -> &TempDir {
+        &self._guard._data_tmp_dir
+    }
+}
+
 pub(crate) async fn create_standalone_instance(test_name: &str) -> MockStandaloneInstance {
    let (opts, guard) = create_tmp_dir_and_datanode_opts(test_name);
    let datanode_instance = DatanodeInstance::new(&opts).await.unwrap();
@@ -112,15 +125,15 @@ pub(crate) async fn create_datanode_client(

    // create a mock datanode grpc service, see example here:
    // https://github.com/hyperium/tonic/blob/master/examples/src/mock/mock.rs
-    let datanode_service = GrpcServer::new(
+    let grpc_server = GrpcServer::new(
        ServerGrpcQueryHandlerAdaptor::arc(datanode_instance),
        None,
        runtime,
-    )
-    .create_service();
+    );
    tokio::spawn(async move {
        Server::builder()
-            .add_service(datanode_service)
+            .add_service(grpc_server.create_flight_service())
+            .add_service(grpc_server.create_database_service())
            .serve_with_incoming(futures::stream::iter(vec![Ok::<_, std::io::Error>(server)]))
            .await
    });
@@ -269,3 +282,29 @@ pub(crate) async fn create_distributed_instance(test_name: &str) -> MockDistribu
        _guards: test_guards,
    }
 }
+
+pub fn test_region_dir(
+    dir: &str,
+    catalog_name: &str,
+    schema_name: &str,
+    table_id: u32,
+    region_id: u32,
+) -> String {
+    let table_dir = table_dir(catalog_name, schema_name, table_id);
+    let region_name = region_name(table_id, region_id);
+
+    format!("{}/{}/{}", dir, table_dir, region_name)
+}
+
+pub fn has_parquet_file(sst_dir: &str) -> bool {
+    for entry in std::fs::read_dir(sst_dir).unwrap() {
+        let entry = entry.unwrap();
+        let path = entry.path();
+        if !path.is_dir() {
+            assert_eq!("parquet", path.extension().unwrap());
+            return true;
+        }
+    }
+
+    false
+}
--- a/src/log-store/Cargo.toml
+++ b/src/log-store/Cargo.toml
@@ -33,4 +33,4 @@ tokio-util.workspace = true

 [dev-dependencies]
 common-test-util = { path = "../common/test-util" }
-rand = "0.8"
+rand.workspace = true
--- a/src/meta-client/Cargo.toml
+++ b/src/meta-client/Cargo.toml
@@ -12,7 +12,7 @@ common-error = { path = "../common/error" }
 common-grpc = { path = "../common/grpc" }
 common-telemetry = { path = "../common/telemetry" }
 etcd-client = "0.10"
-rand = "0.8"
+rand.workspace = true
 serde.workspace = true
 serde_json.workspace = true
 snafu.workspace = true
--- a/src/meta-client/examples/meta_client.rs
+++ b/src/meta-client/examples/meta_client.rs
@@ -22,8 +22,8 @@ use datatypes::prelude::ConcreteDataType;
 use datatypes::schema::{ColumnSchema, RawSchema};
 use meta_client::client::MetaClientBuilder;
 use meta_client::rpc::{
-    BatchPutRequest, CompareAndPutRequest, CreateRequest, DeleteRangeRequest, Partition,
-    PutRequest, RangeRequest, TableName,
+    BatchDeleteRequest, BatchGetRequest, BatchPutRequest, CompareAndPutRequest, CreateRequest,
+    DeleteRangeRequest, Partition, PutRequest, RangeRequest, TableName,
 };
 use table::metadata::{RawTableInfo, RawTableMeta, TableIdent, TableType};
 use table::requests::TableOptions;
@@ -146,6 +146,30 @@ async fn run() {
    // get none
    let res = meta_client.range(range).await.unwrap();
    event!(Level::INFO, "get range result: {:#?}", res);
+
+    // batch delete
+    // put two
+    let batch_put = BatchPutRequest::new()
+        .add_kv(b"batch_put1".to_vec(), b"batch_put_v1".to_vec())
+        .add_kv(b"batch_put2".to_vec(), b"batch_put_v2".to_vec())
+        .with_prev_kv();
+    let res = meta_client.batch_put(batch_put).await.unwrap();
+    event!(Level::INFO, "batch put result: {:#?}", res);
+
+    // delete one
+    let batch_delete = BatchDeleteRequest::new()
+        .add_key(b"batch_put1".to_vec())
+        .with_prev_kv();
+    let res = meta_client.batch_delete(batch_delete).await.unwrap();
+    event!(Level::INFO, "batch delete result: {:#?}", res);
+
+    // get other one
+    let batch_get = BatchGetRequest::new()
+        .add_key(b"batch_put1".to_vec())
+        .add_key(b"batch_put2".to_vec());
+
+    let res = meta_client.batch_get(batch_get).await.unwrap();
+    event!(Level::INFO, "batch get result: {:#?}", res);
 }

 fn new_table_info() -> RawTableInfo {
--- a/src/meta-client/src/client.rs
+++ b/src/meta-client/src/client.rs
@@ -32,10 +32,10 @@ use crate::error::Result;
 use crate::rpc::lock::{LockRequest, LockResponse, UnlockRequest};
 use crate::rpc::router::DeleteRequest;
 use crate::rpc::{
-    BatchGetRequest, BatchGetResponse, BatchPutRequest, BatchPutResponse, CompareAndPutRequest,
-    CompareAndPutResponse, CreateRequest, DeleteRangeRequest, DeleteRangeResponse,
-    MoveValueRequest, MoveValueResponse, PutRequest, PutResponse, RangeRequest, RangeResponse,
-    RouteRequest, RouteResponse,
+    BatchDeleteRequest, BatchDeleteResponse, BatchGetRequest, BatchGetResponse, BatchPutRequest,
+    BatchPutResponse, CompareAndPutRequest, CompareAndPutResponse, CreateRequest,
+    DeleteRangeRequest, DeleteRangeResponse, MoveValueRequest, MoveValueResponse, PutRequest,
+    PutResponse, RangeRequest, RangeResponse, RouteRequest, RouteResponse,
 };

 pub type Id = (u64, u64);
@@ -256,6 +256,14 @@ impl MetaClient {
        self.store_client()?.batch_put(req.into()).await?.try_into()
    }

+    /// BatchDelete atomically deletes the given keys from the key-value store.
+    pub async fn batch_delete(&self, req: BatchDeleteRequest) -> Result<BatchDeleteResponse> {
+        self.store_client()?
+            .batch_delete(req.into())
+            .await?
+            .try_into()
+    }
+
    /// CompareAndPut atomically puts the value to the given updated
    /// value if the current value == the expected value.
    pub async fn compare_and_put(
--- a/src/meta-client/src/client/store.rs
+++ b/src/meta-client/src/client/store.rs
@@ -17,9 +17,10 @@ use std::sync::Arc;

 use api::v1::meta::store_client::StoreClient;
 use api::v1::meta::{
-    BatchGetRequest, BatchGetResponse, BatchPutRequest, BatchPutResponse, CompareAndPutRequest,
-    CompareAndPutResponse, DeleteRangeRequest, DeleteRangeResponse, MoveValueRequest,
-    MoveValueResponse, PutRequest, PutResponse, RangeRequest, RangeResponse,
+    BatchDeleteRequest, BatchDeleteResponse, BatchGetRequest, BatchGetResponse, BatchPutRequest,
+    BatchPutResponse, CompareAndPutRequest, CompareAndPutResponse, DeleteRangeRequest,
+    DeleteRangeResponse, MoveValueRequest, MoveValueResponse, PutRequest, PutResponse,
+    RangeRequest, RangeResponse,
 };
 use common_grpc::channel_manager::ChannelManager;
 use snafu::{ensure, OptionExt, ResultExt};
@@ -80,6 +81,11 @@ impl Client {
        inner.batch_put(req).await
    }

+    pub async fn batch_delete(&self, req: BatchDeleteRequest) -> Result<BatchDeleteResponse> {
+        let inner = self.inner.read().await;
+        inner.batch_delete(req).await
+    }
+
    pub async fn compare_and_put(
        &self,
        req: CompareAndPutRequest,
@@ -169,6 +175,17 @@ impl Inner {
        Ok(res.into_inner())
    }

+    async fn batch_delete(&self, mut req: BatchDeleteRequest) -> Result<BatchDeleteResponse> {
+        let mut client = self.random_client()?;
+        req.set_header(self.id);
+        let res = client
+            .batch_delete(req)
+            .await
+            .context(error::TonicStatusSnafu)?;
+
+        Ok(res.into_inner())
+    }
+
    async fn compare_and_put(
        &self,
        mut req: CompareAndPutRequest,
--- a/src/meta-client/src/rpc.rs
+++ b/src/meta-client/src/rpc.rs
@@ -28,9 +28,10 @@ pub use router::{
 };
 use serde::{Deserialize, Serialize};
 pub use store::{
-    BatchGetRequest, BatchGetResponse, BatchPutRequest, BatchPutResponse, CompareAndPutRequest,
-    CompareAndPutResponse, DeleteRangeRequest, DeleteRangeResponse, MoveValueRequest,
-    MoveValueResponse, PutRequest, PutResponse, RangeRequest, RangeResponse,
+    BatchDeleteRequest, BatchDeleteResponse, BatchGetRequest, BatchGetResponse, BatchPutRequest,
+    BatchPutResponse, CompareAndPutRequest, CompareAndPutResponse, DeleteRangeRequest,
+    DeleteRangeResponse, MoveValueRequest, MoveValueResponse, PutRequest, PutResponse,
+    RangeRequest, RangeResponse,
 };

 #[derive(Debug, Clone)]
--- a/src/meta-client/src/rpc/store.rs
+++ b/src/meta-client/src/rpc/store.rs
@@ -13,6 +13,7 @@
 // limitations under the License.

 use api::v1::meta::{
+    BatchDeleteRequest as PbBatchDeleteRequest, BatchDeleteResponse as PbBatchDeleteResponse,
    BatchGetRequest as PbBatchGetRequest, BatchGetResponse as PbBatchGetResponse,
    BatchPutRequest as PbBatchPutRequest, BatchPutResponse as PbBatchPutResponse,
    CompareAndPutRequest as PbCompareAndPutRequest,
@@ -377,6 +378,78 @@ impl BatchPutResponse {
    }
 }

+#[derive(Debug, Clone, Default)]
+pub struct BatchDeleteRequest {
+    pub keys: Vec<Vec<u8>>,
+    /// If prev_kv is set, gets the previous key-value pairs before deleting it.
+    /// The previous key-value pairs will be returned in the batch delete response.
+    pub prev_kv: bool,
+}
+
+impl From<BatchDeleteRequest> for PbBatchDeleteRequest {
+    fn from(req: BatchDeleteRequest) -> Self {
+        Self {
+            header: None,
+            keys: req.keys,
+            prev_kv: req.prev_kv,
+        }
+    }
+}
+
+impl BatchDeleteRequest {
+    #[inline]
+    pub fn new() -> Self {
+        Self {
+            keys: vec![],
+            prev_kv: false,
+        }
+    }
+
+    #[inline]
+    pub fn add_key(mut self, key: impl Into<Vec<u8>>) -> Self {
+        self.keys.push(key.into());
+        self
+    }
+
+    /// If prev_kv is set, gets the previous key-value pair before deleting it.
+    /// The previous key-value pair will be returned in the batch delete response.
+    #[inline]
+    pub fn with_prev_kv(mut self) -> Self {
+        self.prev_kv = true;
+        self
+    }
+}
+
+#[derive(Debug, Clone)]
+pub struct BatchDeleteResponse(PbBatchDeleteResponse);
+
+impl TryFrom<PbBatchDeleteResponse> for BatchDeleteResponse {
+    type Error = error::Error;
+
+    fn try_from(pb: PbBatchDeleteResponse) -> Result<Self> {
+        util::check_response_header(pb.header.as_ref())?;
+
+        Ok(Self::new(pb))
+    }
+}
+
+impl BatchDeleteResponse {
+    #[inline]
+    pub fn new(res: PbBatchDeleteResponse) -> Self {
+        Self(res)
+    }
+
+    #[inline]
+    pub fn take_header(&mut self) -> Option<ResponseHeader> {
+        self.0.header.take().map(ResponseHeader::new)
+    }
+
+    #[inline]
+    pub fn take_prev_kvs(&mut self) -> Vec<KeyValue> {
+        self.0.prev_kvs.drain(..).map(KeyValue::new).collect()
+    }
+}
+
 #[derive(Debug, Clone, Default)]
 pub struct CompareAndPutRequest {
    /// key is the key, in bytes, to put into the key-value store.
@@ -832,6 +905,39 @@ mod tests {
        assert_eq!(b"v1".to_vec(), kvs[0].value().to_vec());
    }

+    #[test]
+    fn test_batch_delete_request_trans() {
+        let req = BatchDeleteRequest::new()
+            .add_key(b"test_key1".to_vec())
+            .add_key(b"test_key2".to_vec())
+            .add_key(b"test_key3".to_vec())
+            .with_prev_kv();
+
+        let into_req: PbBatchDeleteRequest = req.into();
+        assert!(into_req.header.is_none());
+        assert_eq!(&b"test_key1".to_vec(), into_req.keys.get(0).unwrap());
+        assert_eq!(&b"test_key2".to_vec(), into_req.keys.get(1).unwrap());
+        assert_eq!(&b"test_key3".to_vec(), into_req.keys.get(2).unwrap());
+        assert!(into_req.prev_kv);
+    }
+
+    #[test]
+    fn test_batch_delete_response_trans() {
+        let pb_res = PbBatchDeleteResponse {
+            header: None,
+            prev_kvs: vec![PbKeyValue {
+                key: b"k1".to_vec(),
+                value: b"v1".to_vec(),
+            }],
+        };
+
+        let mut res = BatchDeleteResponse::new(pb_res);
+        assert!(res.take_header().is_none());
+        let kvs = res.take_prev_kvs();
+        assert_eq!(b"k1".to_vec(), kvs[0].key().to_vec());
+        assert_eq!(b"v1".to_vec(), kvs[0].value().to_vec());
+    }
+
    #[test]
    fn test_compare_and_put_request_trans() {
        let (key, expect, value) = (
--- a/src/meta-srv/Cargo.toml
+++ b/src/meta-srv/Cargo.toml
@@ -28,6 +28,7 @@ http-body = "0.4"
 lazy_static = "1.4"
 parking_lot = "0.12"
 prost.workspace = true
+rand.workspace = true
 regex = "1.6"
 serde = "1.0"
 serde_json = "1.0"
--- a/src/meta-srv/src/bootstrap.rs
+++ b/src/meta-srv/src/bootstrap.rs
@@ -22,6 +22,7 @@ use api::v1::meta::store_server::StoreServer;
 use etcd_client::Client;
 use snafu::ResultExt;
 use tokio::net::TcpListener;
+use tokio::sync::mpsc::{self, Receiver, Sender};
 use tokio_stream::wrappers::TcpListenerStream;
 use tonic::transport::server::Router;

@@ -39,22 +40,70 @@ use crate::service::store::kv::ResettableKvStoreRef;
 use crate::service::store::memory::MemStore;
 use crate::{error, Result};

-// Bootstrap the rpc server to serve incoming request
-pub async fn bootstrap_meta_srv(opts: MetaSrvOptions) -> Result<()> {
-    let meta_srv = make_meta_srv(opts.clone()).await?;
-    bootstrap_meta_srv_with_router(opts, router(meta_srv)).await
+#[derive(Clone)]
+pub struct MetaSrvInstance {
+    meta_srv: MetaSrv,
+
+    opts: MetaSrvOptions,
+
+    signal_sender: Option<Sender<()>>,
 }

-pub async fn bootstrap_meta_srv_with_router(opts: MetaSrvOptions, router: Router) -> Result<()> {
-    let listener = TcpListener::bind(&opts.bind_addr)
+impl MetaSrvInstance {
+    pub async fn new(opts: MetaSrvOptions) -> Result<MetaSrvInstance> {
+        let meta_srv = build_meta_srv(&opts).await?;
+
+        Ok(MetaSrvInstance {
+            meta_srv,
+            opts,
+            signal_sender: None,
+        })
+    }
+
+    pub async fn start(&mut self) -> Result<()> {
+        self.meta_srv.start().await;
+        let (tx, mut rx) = mpsc::channel::<()>(1);
+
+        self.signal_sender = Some(tx);
+
+        bootstrap_meta_srv_with_router(
+            &self.opts.bind_addr,
+            router(self.meta_srv.clone()),
+            &mut rx,
+        )
+        .await?;
+
+        Ok(())
+    }
+
+    pub async fn shutdown(&self) -> Result<()> {
+        if let Some(signal) = &self.signal_sender {
+            signal
+                .send(())
+                .await
+                .context(error::SendShutdownSignalSnafu)?;
+        }
+
+        self.meta_srv.shutdown();
+
+        Ok(())
+    }
+}
+
+pub async fn bootstrap_meta_srv_with_router(
+    bind_addr: &str,
+    router: Router,
+    signal: &mut Receiver<()>,
+) -> Result<()> {
+    let listener = TcpListener::bind(bind_addr)
        .await
-        .context(error::TcpBindSnafu {
-            addr: &opts.bind_addr,
-        })?;
+        .context(error::TcpBindSnafu { addr: bind_addr })?;
    let listener = TcpListenerStream::new(listener);

    router
-        .serve_with_incoming(listener)
+        .serve_with_incoming_shutdown(listener, async {
+            signal.recv().await;
+        })
        .await
        .context(error::StartGrpcSnafu)?;

@@ -72,7 +121,7 @@ pub fn router(meta_srv: MetaSrv) -> Router {
        .add_service(admin::make_admin_service(meta_srv))
 }

-pub async fn make_meta_srv(opts: MetaSrvOptions) -> Result<MetaSrv> {
+pub async fn build_meta_srv(opts: &MetaSrvOptions) -> Result<MetaSrv> {
    let (kv_store, election, lock) = if opts.use_memory_store {
        (Arc::new(MemStore::new()) as _, None, None)
    } else {
@@ -107,7 +156,7 @@ pub async fn make_meta_srv(opts: MetaSrvOptions) -> Result<MetaSrv> {
    };

    let meta_srv = MetaSrvBuilder::new()
-        .options(opts)
+        .options(opts.clone())
        .kv_store(kv_store)
        .in_memory(in_memory)
        .selector(selector)
@@ -117,6 +166,12 @@ pub async fn make_meta_srv(opts: MetaSrvOptions) -> Result<MetaSrv> {
        .build()
        .await;

+    Ok(meta_srv)
+}
+
+pub async fn make_meta_srv(opts: &MetaSrvOptions) -> Result<MetaSrv> {
+    let meta_srv = build_meta_srv(opts).await?;
+
    meta_srv.start().await;

    Ok(meta_srv)
--- a/src/meta-srv/src/cluster.rs
+++ b/src/meta-srv/src/cluster.rs
@@ -261,7 +261,7 @@ mod tests {
        let stat_val = StatValue { stats: vec![stat] }.try_into().unwrap();

        let kv = KeyValue {
-            key: stat_key.clone().into(),
+            key: stat_key.into(),
            value: stat_val,
        };

--- a/src/meta-srv/src/error.rs
+++ b/src/meta-srv/src/error.rs
@@ -15,12 +15,16 @@
 use std::string::FromUtf8Error;

 use common_error::prelude::*;
+use tokio::sync::mpsc::error::SendError;
 use tonic::codegen::http;
 use tonic::{Code, Status};

 #[derive(Debug, Snafu)]
 #[snafu(visibility(pub))]
 pub enum Error {
+    #[snafu(display("Failed to send shutdown signal"))]
+    SendShutdownSignal { source: SendError<()> },
+
    #[snafu(display("Error stream request next is None"))]
    StreamNone { backtrace: Backtrace },

@@ -312,6 +316,7 @@ impl ErrorExt for Error {
            | Error::LeaseGrant { .. }
            | Error::LockNotConfig { .. }
            | Error::ExceededRetryLimit { .. }
+            | Error::SendShutdownSignal { .. }
            | Error::StartGrpc { .. } => StatusCode::Internal,
            Error::EmptyKey { .. }
            | Error::MissingRequiredParameter { .. }
--- a/src/meta-srv/src/failure_detector.rs
+++ b/src/meta-srv/src/failure_detector.rs
@@ -0,0 +1,546 @@
+// 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::VecDeque;
+
+/// This is our port of Akka's "[PhiAccrualFailureDetector](https://github.com/akka/akka/blob/main/akka-remote/src/main/scala/akka/remote/PhiAccrualFailureDetector.scala)"
+/// You can find it's document here:
+/// https://doc.akka.io/docs/akka/current/typed/failure-detector.html
+///
+/// Implementation of 'The Phi Accrual Failure Detector' by Hayashibara et al. as defined in their
+/// paper: [https://oneofus.la/have-emacs-will-hack/files/HDY04.pdf]
+///
+/// The suspicion level of failure is given by a value called φ (phi).
+/// The basic idea of the φ failure detector is to express the value of φ on a scale that
+/// is dynamically adjusted to reflect current network conditions. A configurable
+/// threshold is used to decide if φ is considered to be a failure.
+///
+/// The value of φ is calculated as:
+///
+/// φ = -log10(1 - F(timeSinceLastHeartbeat)
+///
+/// where F is the cumulative distribution function of a normal distribution with mean
+/// and standard deviation estimated from historical heartbeat inter-arrival times.
+#[cfg_attr(test, derive(Clone))]
+pub(crate) struct PhiAccrualFailureDetector {
+    /// A low threshold is prone to generate many wrong suspicions but ensures a quick detection
+    /// in the event of a real crash. Conversely, a high threshold generates fewer mistakes but
+    /// needs more time to detect actual crashes.
+    threshold: f32,
+
+    /// Minimum standard deviation to use for the normal distribution used when calculating phi.
+    /// Too low standard deviation might result in too much sensitivity for sudden, but normal,
+    /// deviations in heartbeat inter arrival times.
+    min_std_deviation_millis: f32,
+
+    /// Duration corresponding to number of potentially lost/delayed heartbeats that will be
+    /// accepted before considering it to be an anomaly.
+    /// This margin is important to be able to survive sudden, occasional, pauses in heartbeat
+    /// arrivals, due to for example network drop.
+    acceptable_heartbeat_pause_millis: u32,
+
+    /// Bootstrap the stats with heartbeats that corresponds to this duration, with a rather high
+    /// standard deviation (since environment is unknown in the beginning).
+    first_heartbeat_estimate_millis: u32,
+
+    heartbeat_history: HeartbeatHistory,
+    last_heartbeat_millis: Option<i64>,
+}
+
+impl Default for PhiAccrualFailureDetector {
+    fn default() -> Self {
+        // default configuration is the same as of Akka:
+        // https://github.com/akka/akka/blob/main/akka-cluster/src/main/resources/reference.conf#L181
+        Self {
+            threshold: 8_f32,
+            min_std_deviation_millis: 100_f32,
+            acceptable_heartbeat_pause_millis: 3000,
+            first_heartbeat_estimate_millis: 1000,
+            heartbeat_history: HeartbeatHistory::new(1000),
+            last_heartbeat_millis: None,
+        }
+    }
+}
+
+impl PhiAccrualFailureDetector {
+    pub(crate) fn heartbeat(&mut self, ts_millis: i64) {
+        if let Some(last_heartbeat_millis) = self.last_heartbeat_millis {
+            if ts_millis < last_heartbeat_millis {
+                return;
+            }
+
+            if self.is_available(ts_millis) {
+                let interval = ts_millis - last_heartbeat_millis;
+                self.heartbeat_history.add(interval)
+            }
+        } else {
+            // guess statistics for first heartbeat,
+            // important so that connections with only one heartbeat becomes unavailable
+            // bootstrap with 2 entries with rather high standard deviation
+            let std_deviation = self.first_heartbeat_estimate_millis / 4;
+            self.heartbeat_history
+                .add((self.first_heartbeat_estimate_millis - std_deviation) as _);
+            self.heartbeat_history
+                .add((self.first_heartbeat_estimate_millis + std_deviation) as _);
+        }
+        let _ = self.last_heartbeat_millis.insert(ts_millis);
+    }
+
+    pub(crate) fn is_available(&self, ts_millis: i64) -> bool {
+        self.phi(ts_millis) < self.threshold as _
+    }
+
+    /// The suspicion level of the accrual failure detector.
+    ///
+    /// If a connection does not have any records in failure detector then it is considered healthy.
+    pub(crate) fn phi(&self, ts_millis: i64) -> f64 {
+        if let Some(last_heartbeat_millis) = self.last_heartbeat_millis {
+            let time_diff = ts_millis - last_heartbeat_millis;
+            let mean = self.heartbeat_history.mean();
+            let std_deviation = self
+                .heartbeat_history
+                .std_deviation()
+                .max(self.min_std_deviation_millis as _);
+
+            phi(
+                time_diff,
+                mean + self.acceptable_heartbeat_pause_millis as f64,
+                std_deviation,
+            )
+        } else {
+            // treat unmanaged connections, e.g. with zero heartbeats, as healthy connections
+            0.0
+        }
+    }
+
+    #[cfg(test)]
+    pub(crate) fn threshold(&self) -> f32 {
+        self.threshold
+    }
+
+    #[cfg(test)]
+    pub(crate) fn acceptable_heartbeat_pause_millis(&self) -> u32 {
+        self.acceptable_heartbeat_pause_millis
+    }
+}
+
+/// Calculation of phi, derived from the Cumulative distribution function for
+/// N(mean, stdDeviation) normal distribution, given by
+/// 1.0 / (1.0 + math.exp(-y * (1.5976 + 0.070566 * y * y)))
+/// where y = (x - mean) / standard_deviation
+/// This is an approximation defined in β Mathematics Handbook (Logistic approximation).
+/// Error is 0.00014 at +- 3.16
+/// The calculated value is equivalent to -log10(1 - CDF(y))
+///
+/// Usually phi = 1 means likeliness that we will make a mistake is about 10%.
+/// The likeliness is about 1% with phi = 2, 0.1% with phi = 3 and so on.
+fn phi(time_diff: i64, mean: f64, std_deviation: f64) -> f64 {
+    assert_ne!(std_deviation, 0.0);
+
+    let time_diff = time_diff as f64;
+    let y = (time_diff - mean) / std_deviation;
+    let e = (-y * (1.5976 + 0.070566 * y * y)).exp();
+    if time_diff > mean {
+        -(e / (1.0 + e)).log10()
+    } else {
+        -(1.0 - 1.0 / (1.0 + e)).log10()
+    }
+}
+
+/// Holds the heartbeat statistics.
+/// It is capped by the number of samples specified in `max_sample_size`.
+///
+/// The stats (mean, variance, std_deviation) are not defined for empty HeartbeatHistory.
+#[derive(Clone)]
+struct HeartbeatHistory {
+    /// Number of samples to use for calculation of mean and standard deviation of inter-arrival
+    /// times.
+    max_sample_size: u32,
+
+    intervals: VecDeque<i64>,
+    interval_sum: i64,
+    squared_interval_sum: i64,
+}
+
+impl HeartbeatHistory {
+    fn new(max_sample_size: u32) -> Self {
+        Self {
+            max_sample_size,
+            intervals: VecDeque::with_capacity(max_sample_size as usize),
+            interval_sum: 0,
+            squared_interval_sum: 0,
+        }
+    }
+
+    fn mean(&self) -> f64 {
+        self.interval_sum as f64 / self.intervals.len() as f64
+    }
+
+    fn variance(&self) -> f64 {
+        let mean = self.mean();
+        self.squared_interval_sum as f64 / self.intervals.len() as f64 - mean * mean
+    }
+
+    fn std_deviation(&self) -> f64 {
+        self.variance().sqrt()
+    }
+
+    fn add(&mut self, interval: i64) {
+        if self.intervals.len() as u32 >= self.max_sample_size {
+            self.drop_oldest();
+        }
+        self.intervals.push_back(interval);
+        self.interval_sum += interval;
+        self.squared_interval_sum += interval * interval;
+    }
+
+    fn drop_oldest(&mut self) {
+        let oldest = self
+            .intervals
+            .pop_front()
+            .expect("intervals must not be empty here");
+        self.interval_sum -= oldest;
+        self.squared_interval_sum -= oldest * oldest;
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use common_time::util::current_time_millis;
+
+    use super::*;
+
+    #[test]
+    fn test_is_available() {
+        let ts_millis = current_time_millis();
+
+        let mut fd = PhiAccrualFailureDetector::default();
+
+        // is available before first heartbeat
+        assert!(fd.is_available(ts_millis));
+
+        fd.heartbeat(ts_millis);
+
+        let acceptable_heartbeat_pause_millis = fd.acceptable_heartbeat_pause_millis as i64;
+        // is available when heartbeat
+        assert!(fd.is_available(ts_millis));
+        // is available before heartbeat timeout
+        assert!(fd.is_available(ts_millis + acceptable_heartbeat_pause_millis / 2));
+        // is not available after heartbeat timeout
+        assert!(!fd.is_available(ts_millis + acceptable_heartbeat_pause_millis * 2));
+    }
+
+    #[test]
+    fn test_last_heartbeat() {
+        let ts_millis = current_time_millis();
+
+        let mut fd = PhiAccrualFailureDetector::default();
+
+        // no heartbeat yet
+        assert!(fd.last_heartbeat_millis.is_none());
+
+        fd.heartbeat(ts_millis);
+        assert_eq!(fd.last_heartbeat_millis, Some(ts_millis));
+    }
+
+    #[test]
+    fn test_phi() {
+        let ts_millis = current_time_millis();
+
+        let mut fd = PhiAccrualFailureDetector::default();
+
+        // phi == 0 before first heartbeat
+        assert_eq!(fd.phi(ts_millis), 0.0);
+
+        fd.heartbeat(ts_millis);
+
+        let acceptable_heartbeat_pause_millis = fd.acceptable_heartbeat_pause_millis as i64;
+        // phi == 0 when heartbeat
+        assert_eq!(fd.phi(ts_millis), 0.0);
+        // phi < threshold before heartbeat timeout
+        let now = ts_millis + acceptable_heartbeat_pause_millis / 2;
+        assert!(fd.phi(now) < fd.threshold as _);
+        // phi >= threshold after heartbeat timeout
+        let now = ts_millis + acceptable_heartbeat_pause_millis * 2;
+        assert!(fd.phi(now) >= fd.threshold as _);
+    }
+
+    // The following test cases are port from Akka's test:
+    // [AccrualFailureDetectorSpec.scala](https://github.com/akka/akka/blob/main/akka-remote/src/test/scala/akka/remote/AccrualFailureDetectorSpec.scala).
+
+    #[test]
+    fn test_use_good_enough_cumulative_distribution_function() {
+        fn cdf(phi: f64) -> f64 {
+            1.0 - 10.0_f64.powf(-phi)
+        }
+
+        assert!((cdf(phi(0, 0.0, 10.0)) - 0.5).abs() < 0.001);
+        assert!((cdf(phi(6, 0.0, 10.0)) - 0.7257).abs() < 0.001);
+        assert!((cdf(phi(15, 0.0, 10.0)) - 0.9332).abs() < 0.001);
+        assert!((cdf(phi(20, 0.0, 10.0)) - 0.97725).abs() < 0.001);
+        assert!((cdf(phi(25, 0.0, 10.0)) - 0.99379).abs() < 0.001);
+        assert!((cdf(phi(35, 0.0, 10.0)) - 0.99977).abs() < 0.001);
+        assert!((cdf(phi(40, 0.0, 10.0)) - 0.99997).abs() < 0.0001);
+
+        for w in (0..40).collect::<Vec<i64>>().windows(2) {
+            assert!(phi(w[0], 0.0, 10.0) < phi(w[1], 0.0, 10.0));
+        }
+
+        assert!((cdf(phi(22, 20.0, 3.0)) - 0.7475).abs() < 0.001);
+    }
+
+    #[test]
+    fn test_handle_outliers_without_losing_precision_or_hitting_exceptions() {
+        assert!((phi(10, 0.0, 1.0) - 38.0).abs() < 1.0);
+        assert_eq!(phi(-25, 0.0, 1.0), 0.0);
+    }
+
+    #[test]
+    fn test_return_realistic_phi_values() {
+        let test = vec![
+            (0, 0.0),
+            (500, 0.1),
+            (1000, 0.3),
+            (1200, 1.6),
+            (1400, 4.7),
+            (1600, 10.8),
+            (1700, 15.3),
+        ];
+        for (time_diff, expected_phi) in test {
+            assert!((phi(time_diff, 1000.0, 100.0) - expected_phi).abs() < 0.1);
+        }
+
+        // larger std_deviation results => lower phi
+        assert!(phi(1100, 1000.0, 500.0) < phi(1100, 1000.0, 100.0));
+    }
+
+    #[test]
+    fn test_return_phi_of_0_on_startup_when_no_heartbeats() {
+        let fd = PhiAccrualFailureDetector {
+            threshold: 8.0,
+            min_std_deviation_millis: 100.0,
+            acceptable_heartbeat_pause_millis: 0,
+            first_heartbeat_estimate_millis: 1000,
+            heartbeat_history: HeartbeatHistory::new(1000),
+            last_heartbeat_millis: None,
+        };
+        assert_eq!(fd.phi(current_time_millis()), 0.0);
+        assert_eq!(fd.phi(current_time_millis()), 0.0);
+    }
+
+    #[test]
+    fn test_return_phi_based_on_guess_when_only_one_heartbeat() {
+        let mut fd = PhiAccrualFailureDetector {
+            threshold: 8.0,
+            min_std_deviation_millis: 100.0,
+            acceptable_heartbeat_pause_millis: 0,
+            first_heartbeat_estimate_millis: 1000,
+            heartbeat_history: HeartbeatHistory::new(1000),
+            last_heartbeat_millis: None,
+        };
+        fd.heartbeat(0);
+        assert!((fd.phi(1000)).abs() - 0.3 < 0.2);
+        assert!((fd.phi(2000)).abs() - 4.5 < 0.3);
+        assert!((fd.phi(3000)).abs() > 15.0);
+    }
+
+    #[test]
+    fn test_return_phi_using_first_interval_after_second_heartbeat() {
+        let mut fd = PhiAccrualFailureDetector {
+            threshold: 8.0,
+            min_std_deviation_millis: 100.0,
+            acceptable_heartbeat_pause_millis: 0,
+            first_heartbeat_estimate_millis: 1000,
+            heartbeat_history: HeartbeatHistory::new(1000),
+            last_heartbeat_millis: None,
+        };
+        fd.heartbeat(0);
+        assert!(fd.phi(100) > 0.0);
+        fd.heartbeat(200);
+        assert!(fd.phi(300) > 0.0);
+    }
+
+    #[test]
+    fn test_is_available_after_a_series_of_successful_heartbeats() {
+        let mut fd = PhiAccrualFailureDetector {
+            threshold: 8.0,
+            min_std_deviation_millis: 100.0,
+            acceptable_heartbeat_pause_millis: 0,
+            first_heartbeat_estimate_millis: 1000,
+            heartbeat_history: HeartbeatHistory::new(1000),
+            last_heartbeat_millis: None,
+        };
+        assert!(fd.last_heartbeat_millis.is_none());
+        fd.heartbeat(0);
+        fd.heartbeat(1000);
+        fd.heartbeat(1100);
+        assert!(fd.last_heartbeat_millis.is_some());
+        assert!(fd.is_available(1200));
+    }
+
+    #[test]
+    fn test_is_not_available_if_heartbeat_are_missed() {
+        let mut fd = PhiAccrualFailureDetector {
+            threshold: 3.0,
+            min_std_deviation_millis: 100.0,
+            acceptable_heartbeat_pause_millis: 0,
+            first_heartbeat_estimate_millis: 1000,
+            heartbeat_history: HeartbeatHistory::new(1000),
+            last_heartbeat_millis: None,
+        };
+        fd.heartbeat(0);
+        fd.heartbeat(1000);
+        fd.heartbeat(1100);
+        assert!(fd.is_available(1200));
+        assert!(!fd.is_available(8200));
+    }
+
+    #[test]
+    fn test_is_available_if_it_starts_heartbeat_again_after_being_marked_dead_due_to_detection_of_failure(
+    ) {
+        let mut fd = PhiAccrualFailureDetector {
+            threshold: 8.0,
+            min_std_deviation_millis: 100.0,
+            acceptable_heartbeat_pause_millis: 3000,
+            first_heartbeat_estimate_millis: 1000,
+            heartbeat_history: HeartbeatHistory::new(1000),
+            last_heartbeat_millis: None,
+        };
+
+        // 1000 regular intervals, 5 minute pause, and then a short pause again that should trigger
+        // unreachable again
+
+        let mut now = 0;
+        for _ in 0..1000 {
+            fd.heartbeat(now);
+            now += 1000;
+        }
+        now += 5 * 60 * 1000;
+        assert!(!fd.is_available(now)); // after the long pause
+        now += 100;
+        fd.heartbeat(now);
+        now += 900;
+        assert!(fd.is_available(now));
+        now += 100;
+        fd.heartbeat(now);
+        now += 7000;
+        assert!(!fd.is_available(now)); // after the 7 seconds pause
+        now += 100;
+        fd.heartbeat(now);
+        now += 900;
+        assert!(fd.is_available(now));
+        now += 100;
+        fd.heartbeat(now);
+        now += 900;
+        assert!(fd.is_available(now));
+    }
+
+    #[test]
+    fn test_accept_some_configured_missing_heartbeats() {
+        let mut fd = PhiAccrualFailureDetector {
+            threshold: 8.0,
+            min_std_deviation_millis: 100.0,
+            acceptable_heartbeat_pause_millis: 3000,
+            first_heartbeat_estimate_millis: 1000,
+            heartbeat_history: HeartbeatHistory::new(1000),
+            last_heartbeat_millis: None,
+        };
+        fd.heartbeat(0);
+        fd.heartbeat(1000);
+        fd.heartbeat(2000);
+        fd.heartbeat(3000);
+        assert!(fd.is_available(7000));
+        fd.heartbeat(8000);
+        assert!(fd.is_available(9000));
+    }
+
+    #[test]
+    fn test_fail_after_configured_acceptable_missing_heartbeats() {
+        let mut fd = PhiAccrualFailureDetector {
+            threshold: 8.0,
+            min_std_deviation_millis: 100.0,
+            acceptable_heartbeat_pause_millis: 3000,
+            first_heartbeat_estimate_millis: 1000,
+            heartbeat_history: HeartbeatHistory::new(1000),
+            last_heartbeat_millis: None,
+        };
+        fd.heartbeat(0);
+        fd.heartbeat(1000);
+        fd.heartbeat(2000);
+        fd.heartbeat(3000);
+        fd.heartbeat(4000);
+        fd.heartbeat(5000);
+        assert!(fd.is_available(5500));
+        fd.heartbeat(6000);
+        assert!(!fd.is_available(11000));
+    }
+
+    #[test]
+    fn test_use_max_sample_size_heartbeats() {
+        let mut fd = PhiAccrualFailureDetector {
+            threshold: 8.0,
+            min_std_deviation_millis: 100.0,
+            acceptable_heartbeat_pause_millis: 0,
+            first_heartbeat_estimate_millis: 1000,
+            heartbeat_history: HeartbeatHistory::new(3),
+            last_heartbeat_millis: None,
+        };
+        // 100 ms interval
+        fd.heartbeat(0);
+        fd.heartbeat(100);
+        fd.heartbeat(200);
+        fd.heartbeat(300);
+        let phi1 = fd.phi(400);
+        // 500 ms interval, should become same phi when 100 ms intervals have been dropped
+        fd.heartbeat(1000);
+        fd.heartbeat(1500);
+        fd.heartbeat(2000);
+        fd.heartbeat(2500);
+        let phi2 = fd.phi(3000);
+        assert_eq!(phi1, phi2);
+    }
+
+    #[test]
+    fn test_heartbeat_history_calculate_correct_mean_and_variance() {
+        let mut history = HeartbeatHistory::new(20);
+        for i in [100, 200, 125, 340, 130] {
+            history.add(i);
+        }
+        assert!((history.mean() - 179.0).abs() < 0.00001);
+        assert!((history.variance() - 7584.0).abs() < 0.00001);
+    }
+
+    #[test]
+    fn test_heartbeat_history_have_0_variance_for_one_sample() {
+        let mut history = HeartbeatHistory::new(600);
+        history.add(1000);
+        assert!((history.variance() - 0.0).abs() < 0.00001);
+    }
+
+    #[test]
+    fn test_heartbeat_history_be_capped_by_the_specified_max_sample_size() {
+        let mut history = HeartbeatHistory::new(3);
+        history.add(100);
+        history.add(110);
+        history.add(90);
+        assert!((history.mean() - 100.0).abs() < 0.00001);
+        assert!((history.variance() - 66.6666667).abs() < 0.00001);
+        history.add(140);
+        assert!((history.mean() - 113.333333).abs() < 0.00001);
+        assert!((history.variance() - 422.222222).abs() < 0.00001);
+        history.add(80);
+        assert!((history.mean() - 103.333333).abs() < 0.00001);
+        assert!((history.variance() - 688.88888889).abs() < 0.00001);
+    }
+}
--- a/src/meta-srv/src/handler.rs
+++ b/src/meta-srv/src/handler.rs
@@ -14,6 +14,7 @@

 pub use check_leader_handler::CheckLeaderHandler;
 pub use collect_stats_handler::CollectStatsHandler;
+pub use failure_handler::RegionFailureHandler;
 pub use keep_lease_handler::KeepLeaseHandler;
 pub use on_leader_start::OnLeaderStartHandler;
 pub use persist_stats_handler::PersistStatsHandler;
@@ -21,6 +22,7 @@ pub use response_header_handler::ResponseHeaderHandler;

 mod check_leader_handler;
 mod collect_stats_handler;
+mod failure_handler;
 mod instruction;
 mod keep_lease_handler;
 pub mod node_stat;
@@ -54,8 +56,8 @@ pub trait HeartbeatHandler: Send + Sync {
 #[derive(Debug, Default)]
 pub struct HeartbeatAccumulator {
    pub header: Option<ResponseHeader>,
-    pub stats: Vec<Stat>,
    pub instructions: Vec<Instruction>,
+    pub stat: Option<Stat>,
 }

 impl HeartbeatAccumulator {
--- a/src/meta-srv/src/handler/collect_stats_handler.rs
+++ b/src/meta-srv/src/handler/collect_stats_handler.rs
@@ -12,39 +12,15 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.

-use std::collections::VecDeque;
-
 use api::v1::meta::HeartbeatRequest;
 use common_telemetry::debug;
-use dashmap::mapref::entry::Entry;
-use dashmap::DashMap;

 use super::node_stat::Stat;
 use crate::error::Result;
 use crate::handler::{HeartbeatAccumulator, HeartbeatHandler};
 use crate::metasrv::Context;

-type StatKey = (u64, u64);
-
-pub struct CollectStatsHandler {
-    max_cached_stats_per_key: usize,
-    cache: DashMap<StatKey, VecDeque<Stat>>,
-}
-
-impl Default for CollectStatsHandler {
-    fn default() -> Self {
-        Self::new(10)
-    }
-}
-
-impl CollectStatsHandler {
-    pub fn new(max_cached_stats_per_key: usize) -> Self {
-        Self {
-            max_cached_stats_per_key,
-            cache: DashMap::new(),
-        }
-    }
-}
+pub struct CollectStatsHandler;

 #[async_trait::async_trait]
 impl HeartbeatHandler for CollectStatsHandler {
@@ -60,21 +36,7 @@ impl HeartbeatHandler for CollectStatsHandler {

        match Stat::try_from(req.clone()) {
            Ok(stat) => {
-                let key = (stat.cluster_id, stat.id);
-                match self.cache.entry(key) {
-                    Entry::Occupied(mut e) => {
-                        let deque = e.get_mut();
-                        deque.push_front(stat);
-                        if deque.len() >= self.max_cached_stats_per_key {
-                            acc.stats = deque.drain(..).collect();
-                        }
-                    }
-                    Entry::Vacant(e) => {
-                        let mut stat_vec = VecDeque::with_capacity(self.max_cached_stats_per_key);
-                        stat_vec.push_front(stat);
-                        e.insert(stat_vec);
-                    }
-                }
+                let _ = acc.stat.insert(stat);
            }
            Err(_) => {
                debug!("Incomplete heartbeat data: {:?}", req);
--- a/src/meta-srv/src/handler/failure_handler.rs
+++ b/src/meta-srv/src/handler/failure_handler.rs
@@ -0,0 +1,151 @@
+// 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 runner;
+
+use api::v1::meta::HeartbeatRequest;
+use async_trait::async_trait;
+
+use crate::error::Result;
+use crate::handler::failure_handler::runner::{FailureDetectControl, FailureDetectRunner};
+use crate::handler::{HeartbeatAccumulator, HeartbeatHandler};
+use crate::metasrv::{Context, ElectionRef};
+
+#[derive(Eq, Hash, PartialEq, Clone)]
+pub(crate) struct RegionIdent {
+    catalog: String,
+    schema: String,
+    table: String,
+    region_id: u64,
+}
+
+// TODO(LFC): TBC
+pub(crate) struct DatanodeHeartbeat {
+    #[allow(dead_code)]
+    cluster_id: u64,
+    #[allow(dead_code)]
+    node_id: u64,
+    region_idents: Vec<RegionIdent>,
+    heartbeat_time: i64,
+}
+
+pub struct RegionFailureHandler {
+    failure_detect_runner: FailureDetectRunner,
+}
+
+impl RegionFailureHandler {
+    pub fn new(election: Option<ElectionRef>) -> Self {
+        Self {
+            failure_detect_runner: FailureDetectRunner::new(election),
+        }
+    }
+
+    pub async fn start(&mut self) {
+        self.failure_detect_runner.start().await;
+    }
+}
+
+#[async_trait]
+impl HeartbeatHandler for RegionFailureHandler {
+    async fn handle(
+        &self,
+        _: &HeartbeatRequest,
+        ctx: &mut Context,
+        acc: &mut HeartbeatAccumulator,
+    ) -> Result<()> {
+        if ctx.is_infancy {
+            self.failure_detect_runner
+                .send_control(FailureDetectControl::Purge)
+                .await;
+        }
+
+        if ctx.is_skip_all() {
+            return Ok(());
+        }
+
+        let Some(stat) = acc.stat.as_ref() else { return Ok(()) };
+
+        let heartbeat = DatanodeHeartbeat {
+            cluster_id: stat.cluster_id,
+            node_id: stat.id,
+            region_idents: stat
+                .region_stats
+                .iter()
+                .map(|x| RegionIdent {
+                    catalog: x.catalog.clone(),
+                    schema: x.schema.clone(),
+                    table: x.table.clone(),
+                    region_id: x.id,
+                })
+                .collect(),
+            heartbeat_time: stat.timestamp_millis,
+        };
+
+        self.failure_detect_runner.send_heartbeat(heartbeat).await;
+        Ok(())
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use crate::handler::node_stat::{RegionStat, Stat};
+    use crate::metasrv::builder::MetaSrvBuilder;
+
+    #[tokio::test(flavor = "multi_thread")]
+    async fn test_handle_heartbeat() {
+        let mut handler = RegionFailureHandler::new(None);
+        handler.start().await;
+
+        let req = &HeartbeatRequest::default();
+
+        let builder = MetaSrvBuilder::new();
+        let metasrv = builder.build().await;
+        let mut ctx = metasrv.new_ctx();
+        ctx.is_infancy = false;
+
+        let acc = &mut HeartbeatAccumulator::default();
+        fn new_region_stat(region_id: u64) -> RegionStat {
+            RegionStat {
+                id: region_id,
+                catalog: "a".to_string(),
+                schema: "b".to_string(),
+                table: "c".to_string(),
+                rcus: 0,
+                wcus: 0,
+                approximate_bytes: 0,
+                approximate_rows: 0,
+            }
+        }
+        acc.stat = Some(Stat {
+            cluster_id: 1,
+            id: 42,
+            region_stats: vec![new_region_stat(1), new_region_stat(2), new_region_stat(3)],
+            timestamp_millis: 1000,
+            ..Default::default()
+        });
+
+        handler.handle(req, &mut ctx, acc).await.unwrap();
+
+        let dump = handler.failure_detect_runner.dump().await;
+        assert_eq!(dump.iter().collect::<Vec<_>>().len(), 3);
+
+        // infancy makes heartbeats re-accumulated
+        ctx.is_infancy = true;
+        acc.stat = None;
+        handler.handle(req, &mut ctx, acc).await.unwrap();
+        let dump = handler.failure_detect_runner.dump().await;
+        assert_eq!(dump.iter().collect::<Vec<_>>().len(), 0);
+    }
+}
--- a/Show More
+++ b/Show More
Author	SHA1	Message	Date
shuiyisong	509d07b798	chore: add `build_table_route_prefix` (#1269 )	2023-03-28 16:26:24 +08:00
Yingwen	e72ce5eaa9	fix: Adds FileHandle to ChunkStream (#1255 ) * test: Add compaction test * test: Test read during compaction * test: Add s3 object store to test * test: only run compact test * feat: Hold file handle in chunk stream * test: check files still exist after compact * feat: Revert changes to develop.yaml * test: Simplify MockPurgeHandler	2023-03-28 16:22:07 +08:00
Ruihang Xia	f491a040f5	feat: implelemt rate, increase and delta in PromQL (#1258 ) * clean up Signed-off-by: Ruihang Xia <waynestxia@gmail.com> * fix increase fn Signed-off-by: Ruihang Xia <waynestxia@gmail.com> * impl rate and delta Signed-off-by: Ruihang Xia <waynestxia@gmail.com> * fix typo Signed-off-by: Ruihang Xia <waynestxia@gmail.com> * fix IS_RATE condition Signed-off-by: Ruihang Xia <waynestxia@gmail.com> * more tests about rate and delta Signed-off-by: Ruihang Xia <waynestxia@gmail.com> * ensure range_length is not zero Signed-off-by: Ruihang Xia <waynestxia@gmail.com> --------- Signed-off-by: Ruihang Xia <waynestxia@gmail.com>	2023-03-28 15:21:06 +08:00
Yingwen	47179a7812	feat: Support sending multiple affected rows (#1203 ) * feat: Support sending multiple affected rows * feat: Skip federated check if query starts with insert * style: Fix clippy	2023-03-28 14:34:14 +08:00
shuiyisong	995a28a27d	feat: impl `BatchDelete` (#1253 ) * chore: impl `BatchDelete` * chore: add `batch_delete` to meta-client * fix: auth param length check * fix: auth param length check * chore: rebase develop * chore: use `filter_map` Co-authored-by: LFC <bayinamine@gmail.com> * chore: update error msg Co-authored-by: LFC <bayinamine@gmail.com> * fix: pre-allocate vec length --------- Co-authored-by: LFC <bayinamine@gmail.com>	2023-03-28 14:06:13 +08:00
LFC	ed1cb73ffc	fix: a minor misuse of tokio::select (#1266 )	2023-03-28 13:50:35 +08:00
dennis zhuang	0ffa628c22	refactor: scripts perf and metrics (#1261 ) * refactor: retrieve pyvector datatype by inner vector * perf: replace all ok_or to ok_or_else * feat: adds metrics for scripts execution	2023-03-28 10:07:21 +08:00
Lei, HUANG	5edd2a3dbe	feat: upgrade opendal (#1245 ) * chore: upgrade opendal * chore: finish upgrading opendal * fix: clippy complaints * fix some tests * fix: all unit tests * chore: rebase develop * fix: sqlness tests * optimize imports * chore: rebase develop * doc: add todo	2023-03-28 09:47:33 +08:00
Ning Sun	e63b28bff1	feat: add dbname and health check for grpc api (#1220 ) * feat: add dbname and health check for grpc api * refactor: move health check to dedicated service * chore: switch to merged proto rev * feat: implement healthcheck on server-side	2023-03-28 09:46:30 +08:00
zyy17	8140d4e3e5	ci: modify the copy path of binary artifacts (#1257 )	2023-03-27 21:49:42 +08:00
shuiyisong	6825459c75	chore: ignore dashboard files (#1260 )	2023-03-27 19:11:31 +08:00
Ning Sun	7eb4d81929	feat: adopt pgwire 0.12 and simplify encoding apis (#1250 ) * feat: adopt pgwire 0.12 and simplify encoding apis * refactor: remove duplicated format match clause	2023-03-27 18:16:43 +08:00
discord9	8ba0741c81	fix: set locals to main.dict too (#1242 )	2023-03-27 15:23:52 +08:00
zyy17	0eeb5b460c	ci: install python requests lib in release container image (#1241 ) * ci: install python requests lib in release container image * refactor: add requirements.txt	2023-03-27 15:20:31 +08:00
LFC	65ea6fd85f	feat: embed dashboard into GreptimeDB binary (#1239 ) * feat: embed dashboard into GreptimeDB binary * fix: resolve PR comments	2023-03-27 15:08:44 +08:00
dennis zhuang	4f15b26b28	feat: region manifest checkpoint (#1202 ) * chore: adds log when manifest protocol is changed * chore: refactor region manifest * temp commit * feat: impl region manifest checkpoint * feat: recover region version from manifest snapshot * test: adds region snapshot test * test: region manifest checkpoint * test: alter region with manifest checkpoint * fix: revert storage api * feat: delete old snapshot * refactor: manifest log storage * Update src/storage/src/version.rs Co-authored-by: Ruihang Xia <waynestxia@gmail.com> * Update src/storage/src/manifest/checkpoint.rs Co-authored-by: Ruihang Xia <waynestxia@gmail.com> * Update src/storage/src/manifest/region.rs Co-authored-by: Ruihang Xia <waynestxia@gmail.com> * Update src/storage/src/manifest/region.rs Co-authored-by: Ruihang Xia <waynestxia@gmail.com> * chore: by CR comments * refactor: by CR comments * fix: typo * chore: tweak start_version --------- Co-authored-by: Ruihang Xia <waynestxia@gmail.com>	2023-03-27 11:15:52 +08:00
Lei, HUANG	15ee4ac729	fix: noop flush impl for numbers table (#1247 ) * fix: noop flush impl for numbers table * fix: clippy	2023-03-27 10:54:07 +08:00
dennis zhuang	b4fc8c5b78	refactor: make sql function in scripts return a list of column vectors (#1243 )	2023-03-27 08:50:19 +08:00
Lei, HUANG	6f81717866	fix: skip empty parquet (#1236 ) * fix: returns None if parquet file does not contain any rows * fix: skip empty parquet file * chore: add doc * rebase develop * fix: use flatten instead of filter_map with identity	2023-03-26 09:39:15 +08:00
Lei, HUANG	77f9383daf	fix: allow larger compaction window to reduce parallel task num (#1223 ) fix: unit tests	2023-03-24 17:12:13 +08:00
discord9	c788b7fc26	feat: slicing PyVector&Create DataFrame from sql (#1190 ) * chore: some typos * feat: slicing for pyo3 vector * feat: slice tests * feat: from_sql * feat: from_sql for dataframe * test: df tests * feat: `from_sql` for rspython * test: tweak a bit * test: and CR advices * typos: ordered points * chore: update error msg * test: add more `slicing` testcase	2023-03-24 15:37:45 +08:00
LFC	0f160a73be	feat: metasrv collects datanode heartbeats for region failure detection (#1214 ) * feat: metasrv collects datanode heartbeats for region failure detection * chore: change visibility * fix: fragile tests * Update src/meta-srv/src/handler/persist_stats_handler.rs Co-authored-by: fys <40801205+Fengys123@users.noreply.github.com> * Update src/meta-srv/src/handler/failure_handler.rs Co-authored-by: fys <40801205+Fengys123@users.noreply.github.com> * fix: resolve PR comments * fix: resolve PR comments * fix: resolve PR comments --------- Co-authored-by: shuiyisong <xixing.sys@gmail.com> Co-authored-by: fys <40801205+Fengys123@users.noreply.github.com>	2023-03-24 04:28:34 +00:00
LFC	92963b9614	feat: execute "delete" in query engine (in the form of "LogicalPlan") (#1222 ) fix: execute "delete" in query engine (in the form of "LogicalPlan")	2023-03-24 12:11:58 +08:00
Yingwen	f1139fba59	fix: Holds FileHandle in ParquetReader to avoid the purger purges it (#1224 )	2023-03-23 14:24:25 +00:00
Ruihang Xia	4e552245b1	fix: range func tests (#1221 ) * remove ignore on range fn tests Signed-off-by: Ruihang Xia <waynestxia@gmail.com> * placeholder for changes, deriv and resets Signed-off-by: Ruihang Xia <waynestxia@gmail.com> --------- Signed-off-by: Ruihang Xia <waynestxia@gmail.com>	2023-03-23 17:33:11 +08:00
Ruihang Xia	3126bbc1c7	docs: use CDN for logos (#1219 ) Signed-off-by: Ruihang Xia <waynestxia@gmail.com>	2023-03-23 11:39:24 +08:00
LFC	b77b561bc8	refactor: execute insert with select in query engine (#1181 ) * refactor: execute insert with select in query engine * fix: resolve PR comments	2023-03-23 10:38:26 +08:00
dennis zhuang	501faad8ab	chore: rename params in flush api (#1213 )	2023-03-22 14:07:23 +08:00
Eugene Tolbakov	5397a9bbe6	feat(to_unixtime): add initial implementation (#1186 ) * feat(to_unixtime): add initial implementation * feat(to_unixtime): use Timestamp for conversion * feat(to_unixtime): implement conversion to Result<VectorRef> * feat(to_unixtime): make unit test pass * feat(to_unixtime): preserve None for invalid timestamps * feat(to_unixtime): address code review suggestions * feat(to_unixtime): add an sqlness test * feat(to_unixtime): adjust the assertion for the sqlness test * Update tests/cases/standalone/common/select/dummy.sql --------- Co-authored-by: Ruihang Xia <waynestxia@gmail.com>	2023-03-21 12:41:07 +00:00
Ruihang Xia	f351ee7042	docs: update document string and site (#1211 ) Signed-off-by: Ruihang Xia <waynestxia@gmail.com>	2023-03-21 07:01:08 +00:00
Ruihang Xia	e0493e0b8f	feat: flush all tables on shutdown (#1185 ) * feat: impl flush on shutdown (#14) * feat: impl flush on shutdown Signed-off-by: Ruihang Xia <waynestxia@gmail.com> * powerful if-else! Signed-off-by: Ruihang Xia <waynestxia@gmail.com> --------- Signed-off-by: Ruihang Xia <waynestxia@gmail.com> * retrieve table handler from schema provider Signed-off-by: Ruihang Xia <waynestxia@gmail.com> * feat: impl flush on shutdown * feat: impl flush on shutdown Signed-off-by: Ruihang Xia <waynestxia@gmail.com> * powerful if-else! Signed-off-by: Ruihang Xia <waynestxia@gmail.com> --------- Signed-off-by: Ruihang Xia <waynestxia@gmail.com> * retrieve table handler from schema provider Signed-off-by: Ruihang Xia <waynestxia@gmail.com> * Update src/datanode/src/instance.rs Co-authored-by: dennis zhuang <killme2008@gmail.com> * fix: uncommitted merge change Signed-off-by: Ruihang Xia <waynestxia@gmail.com> --------- Signed-off-by: Ruihang Xia <waynestxia@gmail.com> Co-authored-by: dennis zhuang <killme2008@gmail.com>	2023-03-21 14:36:30 +08:00
LFC	b2a09c888a	feat: phi accrual failure detector (#1200 )	2023-03-21 11:47:47 +08:00
LFC	af101480b3	feat: add gRPC reflection service (#1208 ) * feat: add gRPC reflection service * feat: add gRPC reflection service	2023-03-21 11:23:29 +08:00
Weny Xu	b8f7f603cf	test: add copy clause sqlness tests (#1198 )	2023-03-21 11:22:26 +08:00
dennis zhuang	8fb97ea1d8	fix: losing region numbers after altering table (#1209 )	2023-03-21 11:19:43 +08:00
discord9	21ce9c1163	docs: more explain in readme (#1195 ) * docs: more explain in readme * fix: typos * fix: CR advices	2023-03-20 21:56:34 +08:00
Ruihang Xia	0a22375ac1	fix: nyc-taxi bench suite (#1204 ) Signed-off-by: Ruihang Xia <waynestxia@gmail.com>	2023-03-20 21:53:01 +08:00
fys	0596d20a3b	fix: can not create table in the local distributed environment (#1207 ) fix: create table in local distribute env	2023-03-20 20:12:35 +08:00
Weny Xu	e19c8fa2b6	refactor: combine Copy To and Copy From (#1197 ) * refactor: combine Copy To and Copy From * Apply suggestions from code review Co-authored-by: LFC <bayinamine@gmail.com> * Apply suggestions from code review Co-authored-by: Lei, HUANG <6406592+v0y4g3r@users.noreply.github.com> --------- Co-authored-by: LFC <bayinamine@gmail.com> Co-authored-by: Lei, HUANG <6406592+v0y4g3r@users.noreply.github.com>	2023-03-20 19:23:25 +08:00
LFC	ad886f5b3e	feat: GRPC client stream interface for insertion (#1206 ) * feat: GRPC client stream interface for insertion * feat: GRPC client stream interface for insertion	2023-03-20 18:45:37 +08:00
LFC	f6669a8201	feat: add GRPC unary call service to our GreptimeDB (#1196 ) * feat: add GRPC unary call service to our GreptimeDB	2023-03-20 14:27:32 +08:00
Yingwen	ad5c47185d	feat: wait flush until the flush is done (#1188 ) * feat: Add wait argument to flush * test(storage): Fix flush tests	2023-03-20 11:25:19 +08:00
zyy17	64441616db	ci: refactor compile-python.sh and use the python310 to build amd64 binary (#1199 )	2023-03-18 16:16:15 +08:00
zyy17	09491d6aee	ci: release the standalone binaries with pyo3 and install python utils in images (#1194 ) * ci: install python3 and python3-dev in CI Dockerfile * ci: release the standalone binaries with pyo3 support for multiple platforms * refactor: install pip and pyarrow * refactor: specify the python version	2023-03-17 15:42:13 +08:00
Weny Xu	7cfa30b2ab	feat: add shutdown for standalone and metasrv (#1174 )	2023-03-17 11:35:17 +08:00
Ning Sun	a7676d8860	refactor: port div_ceil from stdlib to avoid unstable features (#1191 ) * refactor: use float div&ceil to avoid unstable features * refactor: port div_ceil from rust stdlib	2023-03-16 22:55:35 +08:00
zyy17	62e2a60b7b	ci: release artifacts after binary and container is ready (#1192 ) ci: release artifacts before binary and container is ready	2023-03-16 09:20:03 +00:00
zyy17	128c5cabe1	ci: disable run tests temporarily (#1187 )	2023-03-16 14:12:19 +08:00
Yingwen	9a001d3392	chore(datanode): derive serde default for Wal/CompactionConfig (#1173 )	2023-03-16 11:56:28 +08:00
Weny Xu	facdda4d9f	feat: implement CONNECTION clause of Copy To (#1163 ) * feat: implement CONNECTION clause of Copy To * test: add tests for s3 backend * Apply suggestions from code review Co-authored-by: Yingwen <realevenyag@gmail.com> --------- Co-authored-by: Yingwen <realevenyag@gmail.com>	2023-03-16 11:36:38 +08:00
Lei, HUANG	17eb99bc52	feat: allow manual table flush through HTTP API (#1184 )	2023-03-15 20:15:34 +08:00
Xieqijun	cd8be77968	feat(procedure): Max retry time (#1095 ) * feat: procedure config * fix: modify config * feat: add retry logic * feat: add error * feat: add it * feat: add it * feat: add it * feat: rm retry from runner * feat: use backon * feat: add retry_interval * feat: add retry_interval * fix: conflict * fix: cr * feat: add retry error and id * feat: rename * refactor: execute * feat: use config dir * fix: cr * fix: cr * fix: fmt * fix: fmt * fix: pr * fix: it * fix: rm unless cmd params * feat: add toml * fix: ut * feat: add rolling back * fix: cr * fix: cr * fix: cr * fix: ci * fix: ci * fix: ci * chore: Apply suggestions from code review --------- Co-authored-by: Yingwen <realevenyag@gmail.com>	2023-03-15 08:28:08 +00:00
Eugene Tolbakov	b530ac9e60	chore(from_unixtime): remove UDF from_unixtime (#1179 ) * chore(from_unixtime): remove UDF from_unixtime * chore(from_unixtime): restore timestamp.rs for further usage * chore(from_unixtime): address fmt issue	2023-03-15 16:27:09 +08:00
zyy17	76f1a79f1b	ci: set 'continue-on-error' to false since the problem of compiling binary was resolved (#1182 ) Signed-off-by: zyy17 <zyylsxm@gmail.com>	2023-03-15 15:41:36 +08:00
LFC	4705245d60	docs: region failover RFC (#1139 ) * docs: region failover RFC * fix: resolve PR comments	2023-03-15 15:21:58 +08:00
Zheming Li	f712f978cf	feat: Report disk usage stats to metasrv thru heartbeat (#1167 ) * feat: Report disk usage stats to metasrv thru heartbeat Signed-off-by: Zheming Li <nkdudu@126.com> * Update src/catalog/src/error.rs Co-authored-by: fys <40801205+Fengys123@users.noreply.github.com> * Update src/catalog/src/lib.rs Co-authored-by: fys <40801205+Fengys123@users.noreply.github.com> * Update src/mito/src/table.rs Co-authored-by: fys <40801205+Fengys123@users.noreply.github.com> --------- Signed-off-by: Zheming Li <nkdudu@126.com> Co-authored-by: fys <40801205+Fengys123@users.noreply.github.com>	2023-03-15 03:11:32 +00:00
discord9	cbf64e65b9	refactor: put `dataframe` & `query` into `greptime` module (#1172 ) * feat: impl getitem for `vector` * feat: mv `query`&`dataframe` into `greptime` for PyO3 * refactor: allow call dataframe&query * refactor: pyo3 query&dataframe * chore: CR advices	2023-03-15 11:01:43 +08:00
zyy17	242ce5c2aa	ci: add pyo3 options for mac (#1178 )	2023-03-14 13:51:58 +00:00
Ruihang Xia	e8d2e82335	fix: ambiguous column reference (#1177 ) Signed-off-by: Ruihang Xia <waynestxia@gmail.com>	2023-03-14 13:18:43 +00:00
zyy17	0086cc2d3d	fix: export 'PYO3_CROSS_LIB_DIR' when cargo build for aarch64-linux and refactor matrix opts (#1171 )	2023-03-14 15:35:29 +08:00
Weny Xu	cdc111b607	refactor: make the cmd hold the application instance (#1159 )	2023-03-14 15:18:50 +08:00
zyy17	81ca1d8399	refactor: add the separate GitHub Action job to push the image to the UCloud registry (#1170 )	2023-03-14 11:35:18 +08:00
LFC	8d3999df5f	fix: failed to run subquery wrapped in two parentheses (#1157 )	2023-03-14 10:59:43 +08:00
discord9	a60788e92e	fix: use correct env var (#1166 ) * fix: use correct env var * fix: move COPY up so rustup know it's nightly * fix: add `pyo3_backend` in GHA yml * chore: name for `TODO` * temp: not set `pyo3_backend` before find DSO * fix: release linux with pyo3_backend	2023-03-14 10:57:13 +08:00
Weny Xu	296c6dfcbf	feat: implement table flush (#1121 ) * feat: add flush method for trait * feat: implement flush via grpc * chore: move table_dir/region_name/region_id to table crate * chore: Update src/mito/src/table.rs --------- Co-authored-by: Yingwen <realevenyag@gmail.com>	2023-03-13 20:10:37 +08:00