chore: print root cause in opendal logging interceptor (#7183 ) (#7557 )

* chore: print root cause in opendal * refactor: extract a function root_source() to get the cause --------- Signed-off-by: evenyag <realevenyag@gmail.com>
fix: fix SQL table identifier quoting for election (#7541 ) (#7556 )
2026-01-14 01:02:55 +00:00 · 2026-01-12 17:16:04 +08:00 · 2026-01-12 15:03:55 +08:00 · 2026-01-08 12:29:33 +00:00 · 2026-01-08 16:10:06 +08:00 · 2025-12-09 17:34:42 +08:00
141 changed files with 2765 additions and 1285 deletions
--- a/.github/actions/setup-etcd-cluster/action.yml
+++ b/.github/actions/setup-etcd-cluster/action.yml
@@ -12,7 +12,7 @@ runs:
  steps:
  - name: Install Etcd cluster
    shell: bash
-    run: | 
+    run: |
      helm upgrade \
        --install etcd oci://registry-1.docker.io/bitnamicharts/etcd \
        --set replicaCount=${{ inputs.etcd-replicas }} \
@@ -24,4 +24,9 @@ runs:
        --set auth.rbac.token.enabled=false \
        --set persistence.size=2Gi \
        --create-namespace \
+        --set global.security.allowInsecureImages=true \
+        --set image.registry=docker.io \
+        --set image.repository=greptime/etcd \
+        --set image.tag=3.6.1-debian-12-r3 \
+        --version 12.0.8 \
        -n ${{ inputs.namespace }}
--- a/.github/actions/setup-greptimedb-cluster/action.yml
+++ b/.github/actions/setup-greptimedb-cluster/action.yml
@@ -51,7 +51,7 @@ runs:
    run: |
      helm upgrade \
        --install my-greptimedb \
-        --set meta.backendStorage.etcd.endpoints=${{ inputs.etcd-endpoints }} \
+        --set 'meta.backendStorage.etcd.endpoints[0]=${{ inputs.etcd-endpoints }}' \
        --set meta.enableRegionFailover=${{ inputs.enable-region-failover }} \
        --set image.registry=${{ inputs.image-registry }} \
        --set image.repository=${{ inputs.image-repository }}  \
--- a/.github/actions/setup-kafka-cluster/action.yml
+++ b/.github/actions/setup-kafka-cluster/action.yml
@@ -12,7 +12,7 @@ runs:
  steps:
  - name: Install Kafka cluster
    shell: bash
-    run: | 
+    run: |
      helm upgrade \
        --install kafka oci://registry-1.docker.io/bitnamicharts/kafka \
        --set controller.replicaCount=${{ inputs.controller-replicas }} \
@@ -23,4 +23,8 @@ runs:
        --set listeners.controller.protocol=PLAINTEXT \
        --set listeners.client.protocol=PLAINTEXT \
        --create-namespace \
+        --set image.registry=docker.io \
+        --set image.repository=greptime/kafka \
+        --set image.tag=3.9.0-debian-12-r1 \
+        --version 31.0.0 \
        -n ${{ inputs.namespace }}
--- a/.github/actions/setup-postgres-cluster/action.yml
+++ b/.github/actions/setup-postgres-cluster/action.yml
@@ -6,9 +6,7 @@ inputs:
    description: "Number of PostgreSQL replicas"
  namespace:
    default: "postgres-namespace"
-  postgres-version:
-    default: "14.2"
-    description: "PostgreSQL version"
+    description: "The PostgreSQL namespace"
  storage-size:
    default: "1Gi"
    description: "Storage size for PostgreSQL"
@@ -22,7 +20,11 @@ runs:
      helm upgrade \
        --install postgresql oci://registry-1.docker.io/bitnamicharts/postgresql \
        --set replicaCount=${{ inputs.postgres-replicas }} \
-        --set image.tag=${{ inputs.postgres-version }} \
+        --set global.security.allowInsecureImages=true \
+        --set image.registry=docker.io \
+        --set image.repository=greptime/postgresql \
+        --set image.tag=17.5.0-debian-12-r3 \
+        --version 16.7.4 \
        --set persistence.size=${{ inputs.storage-size }} \
        --set postgresql.username=greptimedb \
        --set postgresql.password=admin \
--- a/.github/scripts/check-version.sh
+++ b/.github/scripts/check-version.sh
@@ -0,0 +1,42 @@
+#!/bin/bash
+
+# Get current version
+CURRENT_VERSION=$1
+if [ -z "$CURRENT_VERSION" ]; then
+  echo "Error: Failed to get current version"
+  exit 1
+fi
+
+# Get the latest version from GitHub Releases
+API_RESPONSE=$(curl -s "https://api.github.com/repos/GreptimeTeam/greptimedb/releases/latest")
+
+if [ -z "$API_RESPONSE" ] || [ "$(echo "$API_RESPONSE" | jq -r '.message')" = "Not Found" ]; then
+  echo "Error: Failed to fetch latest version from GitHub"
+  exit 1
+fi
+
+# Get the latest version
+LATEST_VERSION=$(echo "$API_RESPONSE" | jq -r '.tag_name')
+
+if [ -z "$LATEST_VERSION" ] || [ "$LATEST_VERSION" = "null" ]; then
+  echo "Error: No valid version found in GitHub releases"
+  exit 1
+fi
+
+# Cleaned up version number format (removed possible 'v' prefix and -nightly suffix)
+CLEAN_CURRENT=$(echo "$CURRENT_VERSION" | sed 's/^v//' | sed 's/-nightly-.*//')
+CLEAN_LATEST=$(echo "$LATEST_VERSION" | sed 's/^v//' | sed 's/-nightly-.*//')
+
+echo "Current version: $CLEAN_CURRENT"
+echo "Latest release version: $CLEAN_LATEST"
+
+# Use sort -V to compare versions
+HIGHER_VERSION=$(printf "%s\n%s" "$CLEAN_CURRENT" "$CLEAN_LATEST" | sort -V | tail -n1)
+
+if [ "$HIGHER_VERSION" = "$CLEAN_CURRENT" ]; then
+  echo "Current version ($CLEAN_CURRENT) is NEWER than or EQUAL to latest ($CLEAN_LATEST)"
+  echo "should-push-latest-tag=true" >> $GITHUB_OUTPUT
+else
+  echo "Current version ($CLEAN_CURRENT) is OLDER than latest ($CLEAN_LATEST)"
+  echo "should-push-latest-tag=false" >> $GITHUB_OUTPUT
+fi
--- a/.github/scripts/deploy-greptimedb.sh
+++ b/.github/scripts/deploy-greptimedb.sh
@@ -68,7 +68,7 @@ function deploy_greptimedb_cluster() {

  helm install "$cluster_name" greptime/greptimedb-cluster \
    --set image.tag="$GREPTIMEDB_IMAGE_TAG" \
-    --set meta.backendStorage.etcd.endpoints="etcd.$install_namespace:2379" \
+    --set 'meta.backendStorage.etcd.endpoints[0]="etcd.$install_namespace:2379"' \
    -n "$install_namespace"

  # Wait for greptimedb cluster to be ready.
@@ -103,7 +103,7 @@ function deploy_greptimedb_cluster_with_s3_storage() {

  helm install "$cluster_name" greptime/greptimedb-cluster -n "$install_namespace" \
    --set image.tag="$GREPTIMEDB_IMAGE_TAG" \
-    --set meta.backendStorage.etcd.endpoints="etcd.$install_namespace:2379" \
+    --set 'meta.backendStorage.etcd.endpoints[0]="etcd.$install_namespace:2379"' \
    --set storage.s3.bucket="$AWS_CI_TEST_BUCKET" \
    --set storage.s3.region="$AWS_REGION" \
    --set storage.s3.root="$DATA_ROOT" \
--- a/.github/scripts/pull-test-deps-images.sh
+++ b/.github/scripts/pull-test-deps-images.sh
@@ -0,0 +1,34 @@
+#!/bin/bash
+
+# This script is used to pull the test dependency images that are stored in public ECR one by one to avoid rate limiting.
+
+set -e
+
+MAX_RETRIES=3
+
+IMAGES=(
+  "greptime/zookeeper:3.7"
+  "greptime/kafka:3.9.0-debian-12-r1"
+  "greptime/etcd:3.6.1-debian-12-r3"
+  "greptime/minio:2024"
+  "greptime/mysql:5.7"
+)
+
+for image in "${IMAGES[@]}"; do
+  for ((attempt=1; attempt<=MAX_RETRIES; attempt++)); do
+    if docker pull "$image"; then
+      # Successfully pulled the image.
+      break
+    else
+      # Use some simple exponential backoff to avoid rate limiting.
+      if [ $attempt -lt $MAX_RETRIES ]; then
+        sleep_seconds=$((attempt * 5))
+        echo "Attempt $attempt failed for $image, waiting $sleep_seconds seconds"
+        sleep $sleep_seconds  # 5s, 10s delays
+      else
+        echo "Failed to pull $image after $MAX_RETRIES attempts"
+        exit 1
+      fi
+    fi
+  done
+done
--- a/.github/workflows/develop.yml
+++ b/.github/workflows/develop.yml
@@ -719,6 +719,10 @@ jobs:
          save-if: ${{ github.ref == 'refs/heads/main' }}
      - name: Install latest nextest release
        uses: taiki-e/install-action@nextest
+
+      - name: Pull test dependencies images
+        run: ./.github/scripts/pull-test-deps-images.sh
+
      - name: Setup external services
        working-directory: tests-integration/fixtures
        run: docker compose up -d --wait
--- a/.github/workflows/release.yml
+++ b/.github/workflows/release.yml
@@ -110,6 +110,8 @@ jobs:

      # The 'version' use as the global tag name of the release workflow.
      version: ${{ steps.create-version.outputs.version }}
+
+      should-push-latest-tag: ${{ steps.check-version.outputs.should-push-latest-tag }}
    steps:
      - name: Checkout
        uses: actions/checkout@v4
@@ -135,6 +137,11 @@ jobs:
          GITHUB_REF_NAME: ${{ github.ref_name }}
          NIGHTLY_RELEASE_PREFIX: ${{ env.NIGHTLY_RELEASE_PREFIX }}

+      - name: Check version
+        id: check-version
+        run: |
+          ./.github/scripts/check-version.sh "${{ steps.create-version.outputs.version }}"
+
      - name: Allocate linux-amd64 runner
        if: ${{ inputs.build_linux_amd64_artifacts || github.event_name == 'push' || github.event_name == 'schedule' }}
        uses: ./.github/actions/start-runner
@@ -314,7 +321,7 @@ jobs:
          image-registry-username: ${{ secrets.DOCKERHUB_USERNAME }}
          image-registry-password: ${{ secrets.DOCKERHUB_TOKEN }}
          version: ${{ needs.allocate-runners.outputs.version }}
-          push-latest-tag: ${{ github.ref_type == 'tag' && !contains(github.ref_name, 'nightly') && github.event_name != 'schedule' }}
+          push-latest-tag: ${{ needs.allocate-runners.outputs.should-push-latest-tag == 'true' && github.ref_type == 'tag' && !contains(github.ref_name, 'nightly') && github.event_name != 'schedule' }}

      - name: Set build image result
        id: set-build-image-result
@@ -361,7 +368,7 @@ jobs:
          dev-mode: false
          upload-to-s3: true
          update-version-info: true
-          push-latest-tag: ${{ github.ref_type == 'tag' && !contains(github.ref_name, 'nightly') && github.event_name != 'schedule' }}
+          push-latest-tag: ${{ needs.allocate-runners.outputs.should-push-latest-tag == 'true' && github.ref_type == 'tag' && !contains(github.ref_name, 'nightly') && github.event_name != 'schedule' }}

  publish-github-release:
    name: Create GitHub release and upload artifacts
--- a/Cargo.lock
+++ b/Cargo.lock
@@ -211,7 +211,7 @@ checksum = "d301b3b94cb4b2f23d7917810addbbaff90738e0ca2be692bd027e70d7e0330c"

 [[package]]
 name = "api"
-version = "0.15.4"
+version = "0.15.5"
 dependencies = [
 "common-base",
 "common-decimal",
@@ -944,7 +944,7 @@ dependencies = [

 [[package]]
 name = "auth"
-version = "0.15.4"
+version = "0.15.5"
 dependencies = [
 "api",
 "async-trait",
@@ -1586,7 +1586,7 @@ dependencies = [

 [[package]]
 name = "cache"
-version = "0.15.4"
+version = "0.15.5"
 dependencies = [
 "catalog",
 "common-error",
@@ -1621,7 +1621,7 @@ checksum = "37b2a672a2cb129a2e41c10b1224bb368f9f37a2b16b612598138befd7b37eb5"

 [[package]]
 name = "catalog"
-version = "0.15.4"
+version = "0.15.5"
 dependencies = [
 "api",
 "arrow 54.2.1",
@@ -1659,6 +1659,8 @@ dependencies = [
 "partition",
 "paste",
 "prometheus",
+ "promql-parser",
+ "rand 0.9.0",
 "rustc-hash 2.0.0",
 "serde_json",
 "session",
@@ -1959,7 +1961,7 @@ checksum = "1462739cb27611015575c0c11df5df7601141071f07518d56fcc1be504cbec97"

 [[package]]
 name = "cli"
-version = "0.15.4"
+version = "0.15.5"
 dependencies = [
 "async-stream",
 "async-trait",
@@ -1986,7 +1988,6 @@ dependencies = [
 "common-version",
 "common-wal",
 "datatypes",
- "either",
 "etcd-client",
 "futures",
 "humantime",
@@ -2004,7 +2005,7 @@ dependencies = [
 "session",
 "snafu 0.8.5",
 "store-api",
- "substrait 0.15.4",
+ "substrait 0.15.5",
 "table",
 "tempfile",
 "tokio",
@@ -2013,7 +2014,7 @@ dependencies = [

 [[package]]
 name = "client"
-version = "0.15.4"
+version = "0.15.5"
 dependencies = [
 "api",
 "arc-swap",
@@ -2043,7 +2044,7 @@ dependencies = [
 "rand 0.9.0",
 "serde_json",
 "snafu 0.8.5",
- "substrait 0.15.4",
+ "substrait 0.15.5",
 "substrait 0.37.3",
 "tokio",
 "tokio-stream",
@@ -2084,7 +2085,7 @@ dependencies = [

 [[package]]
 name = "cmd"
-version = "0.15.4"
+version = "0.15.5"
 dependencies = [
 "async-trait",
 "auth",
@@ -2114,7 +2115,6 @@ dependencies = [
 "common-wal",
 "datanode",
 "datatypes",
- "either",
 "etcd-client",
 "file-engine",
 "flow",
@@ -2145,7 +2145,7 @@ dependencies = [
 "snafu 0.8.5",
 "stat",
 "store-api",
- "substrait 0.15.4",
+ "substrait 0.15.5",
 "table",
 "temp-env",
 "tempfile",
@@ -2192,7 +2192,7 @@ checksum = "55b672471b4e9f9e95499ea597ff64941a309b2cdbffcc46f2cc5e2d971fd335"

 [[package]]
 name = "common-base"
-version = "0.15.4"
+version = "0.15.5"
 dependencies = [
 "anymap2",
 "async-trait",
@@ -2214,11 +2214,11 @@ dependencies = [

 [[package]]
 name = "common-catalog"
-version = "0.15.4"
+version = "0.15.5"

 [[package]]
 name = "common-config"
-version = "0.15.4"
+version = "0.15.5"
 dependencies = [
 "common-base",
 "common-error",
@@ -2243,7 +2243,7 @@ dependencies = [

 [[package]]
 name = "common-datasource"
-version = "0.15.4"
+version = "0.15.5"
 dependencies = [
 "arrow 54.2.1",
 "arrow-schema 54.3.1",
@@ -2280,7 +2280,7 @@ dependencies = [

 [[package]]
 name = "common-decimal"
-version = "0.15.4"
+version = "0.15.5"
 dependencies = [
 "bigdecimal 0.4.8",
 "common-error",
@@ -2293,7 +2293,7 @@ dependencies = [

 [[package]]
 name = "common-error"
-version = "0.15.4"
+version = "0.15.5"
 dependencies = [
 "common-macro",
 "http 1.1.0",
@@ -2304,7 +2304,7 @@ dependencies = [

 [[package]]
 name = "common-frontend"
-version = "0.15.4"
+version = "0.15.5"
 dependencies = [
 "async-trait",
 "common-error",
@@ -2313,6 +2313,7 @@ dependencies = [
 "common-meta",
 "greptime-proto",
 "meta-client",
+ "session",
 "snafu 0.8.5",
 "tokio",
 "tonic 0.12.3",
@@ -2320,7 +2321,7 @@ dependencies = [

 [[package]]
 name = "common-function"
-version = "0.15.4"
+version = "0.15.5"
 dependencies = [
 "ahash 0.8.11",
 "api",
@@ -2373,7 +2374,7 @@ dependencies = [

 [[package]]
 name = "common-greptimedb-telemetry"
-version = "0.15.4"
+version = "0.15.5"
 dependencies = [
 "async-trait",
 "common-runtime",
@@ -2390,7 +2391,7 @@ dependencies = [

 [[package]]
 name = "common-grpc"
-version = "0.15.4"
+version = "0.15.5"
 dependencies = [
 "api",
 "arrow-flight",
@@ -2422,7 +2423,7 @@ dependencies = [

 [[package]]
 name = "common-grpc-expr"
-version = "0.15.4"
+version = "0.15.5"
 dependencies = [
 "api",
 "common-base",
@@ -2441,7 +2442,7 @@ dependencies = [

 [[package]]
 name = "common-macro"
-version = "0.15.4"
+version = "0.15.5"
 dependencies = [
 "arc-swap",
 "common-query",
@@ -2455,7 +2456,7 @@ dependencies = [

 [[package]]
 name = "common-mem-prof"
-version = "0.15.4"
+version = "0.15.5"
 dependencies = [
 "anyhow",
 "common-error",
@@ -2471,7 +2472,7 @@ dependencies = [

 [[package]]
 name = "common-meta"
-version = "0.15.4"
+version = "0.15.5"
 dependencies = [
 "anymap2",
 "api",
@@ -2537,7 +2538,7 @@ dependencies = [

 [[package]]
 name = "common-options"
-version = "0.15.4"
+version = "0.15.5"
 dependencies = [
 "common-grpc",
 "humantime-serde",
@@ -2546,11 +2547,11 @@ dependencies = [

 [[package]]
 name = "common-plugins"
-version = "0.15.4"
+version = "0.15.5"

 [[package]]
 name = "common-pprof"
-version = "0.15.4"
+version = "0.15.5"
 dependencies = [
 "common-error",
 "common-macro",
@@ -2562,7 +2563,7 @@ dependencies = [

 [[package]]
 name = "common-procedure"
-version = "0.15.4"
+version = "0.15.5"
 dependencies = [
 "async-stream",
 "async-trait",
@@ -2589,7 +2590,7 @@ dependencies = [

 [[package]]
 name = "common-procedure-test"
-version = "0.15.4"
+version = "0.15.5"
 dependencies = [
 "async-trait",
 "common-procedure",
@@ -2599,7 +2600,7 @@ dependencies = [

 [[package]]
 name = "common-query"
-version = "0.15.4"
+version = "0.15.5"
 dependencies = [
 "api",
 "async-trait",
@@ -2625,7 +2626,7 @@ dependencies = [

 [[package]]
 name = "common-recordbatch"
-version = "0.15.4"
+version = "0.15.5"
 dependencies = [
 "arc-swap",
 "common-error",
@@ -2645,7 +2646,7 @@ dependencies = [

 [[package]]
 name = "common-runtime"
-version = "0.15.4"
+version = "0.15.5"
 dependencies = [
 "async-trait",
 "clap 4.5.19",
@@ -2675,14 +2676,14 @@ dependencies = [

 [[package]]
 name = "common-session"
-version = "0.15.4"
+version = "0.15.5"
 dependencies = [
 "strum 0.27.1",
 ]

 [[package]]
 name = "common-telemetry"
-version = "0.15.4"
+version = "0.15.5"
 dependencies = [
 "backtrace",
 "common-error",
@@ -2710,7 +2711,7 @@ dependencies = [

 [[package]]
 name = "common-test-util"
-version = "0.15.4"
+version = "0.15.5"
 dependencies = [
 "client",
 "common-grpc",
@@ -2723,7 +2724,7 @@ dependencies = [

 [[package]]
 name = "common-time"
-version = "0.15.4"
+version = "0.15.5"
 dependencies = [
 "arrow 54.2.1",
 "chrono",
@@ -2741,7 +2742,7 @@ dependencies = [

 [[package]]
 name = "common-version"
-version = "0.15.4"
+version = "0.15.5"
 dependencies = [
 "build-data",
 "cargo-manifest",
@@ -2752,7 +2753,7 @@ dependencies = [

 [[package]]
 name = "common-wal"
-version = "0.15.4"
+version = "0.15.5"
 dependencies = [
 "common-base",
 "common-error",
@@ -2775,7 +2776,7 @@ dependencies = [

 [[package]]
 name = "common-workload"
-version = "0.15.4"
+version = "0.15.5"
 dependencies = [
 "api",
 "common-telemetry",
@@ -3731,7 +3732,7 @@ dependencies = [

 [[package]]
 name = "datanode"
-version = "0.15.4"
+version = "0.15.5"
 dependencies = [
 "api",
 "arrow-flight",
@@ -3784,7 +3785,7 @@ dependencies = [
 "session",
 "snafu 0.8.5",
 "store-api",
- "substrait 0.15.4",
+ "substrait 0.15.5",
 "table",
 "tokio",
 "toml 0.8.19",
@@ -3793,7 +3794,7 @@ dependencies = [

 [[package]]
 name = "datatypes"
-version = "0.15.4"
+version = "0.15.5"
 dependencies = [
 "arrow 54.2.1",
 "arrow-array 54.2.1",
@@ -4213,9 +4214,9 @@ dependencies = [

 [[package]]
 name = "either"
-version = "1.13.0"
+version = "1.15.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "60b1af1c220855b6ceac025d3f6ecdd2b7c4894bfe9cd9bda4fbb4bc7c0d4cf0"
+checksum = "48c757948c5ede0e46177b7add2e67155f70e33c07fea8284df6576da70b3719"
 dependencies = [
 "serde",
 ]
@@ -4453,7 +4454,7 @@ checksum = "e8c02a5121d4ea3eb16a80748c74f5549a5665e4c21333c6098f283870fbdea6"

 [[package]]
 name = "file-engine"
-version = "0.15.4"
+version = "0.15.5"
 dependencies = [
 "api",
 "async-trait",
@@ -4590,7 +4591,7 @@ checksum = "8bf7cc16383c4b8d58b9905a8509f02926ce3058053c056376248d958c9df1e8"

 [[package]]
 name = "flow"
-version = "0.15.4"
+version = "0.15.5"
 dependencies = [
 "api",
 "arrow 54.2.1",
@@ -4655,7 +4656,7 @@ dependencies = [
 "sql",
 "store-api",
 "strum 0.27.1",
- "substrait 0.15.4",
+ "substrait 0.15.5",
 "table",
 "tokio",
 "tonic 0.12.3",
@@ -4710,7 +4711,7 @@ checksum = "6c2141d6d6c8512188a7891b4b01590a45f6dac67afb4f255c4124dbb86d4eaa"

 [[package]]
 name = "frontend"
-version = "0.15.4"
+version = "0.15.5"
 dependencies = [
 "api",
 "arc-swap",
@@ -4770,7 +4771,7 @@ dependencies = [
 "sqlparser 0.54.0 (git+https://github.com/GreptimeTeam/sqlparser-rs.git?rev=0cf6c04490d59435ee965edd2078e8855bd8471e)",
 "store-api",
 "strfmt",
- "substrait 0.15.4",
+ "substrait 0.15.5",
 "table",
 "tokio",
 "tokio-util",
@@ -5160,7 +5161,7 @@ dependencies = [
 [[package]]
 name = "greptime-proto"
 version = "0.1.0"
-source = "git+https://github.com/GreptimeTeam/greptime-proto.git?rev=551c70f2b352b6bc55ab769129344aeab8f22d11#551c70f2b352b6bc55ab769129344aeab8f22d11"
+source = "git+https://github.com/GreptimeTeam/greptime-proto.git?rev=f3103a8c9b8ce162457d0a3e3ca00d53d1a8bd06#f3103a8c9b8ce162457d0a3e3ca00d53d1a8bd06"
 dependencies = [
 "prost 0.13.5",
 "serde",
@@ -5931,7 +5932,7 @@ dependencies = [

 [[package]]
 name = "index"
-version = "0.15.4"
+version = "0.15.5"
 dependencies = [
 "async-trait",
 "asynchronous-codec",
@@ -5943,6 +5944,7 @@ dependencies = [
 "common-runtime",
 "common-telemetry",
 "common-test-util",
+ "criterion 0.4.0",
 "fastbloom",
 "fst",
 "futures",
@@ -5955,6 +5957,7 @@ dependencies = [
 "prost 0.13.5",
 "puffin",
 "rand 0.9.0",
+ "rand_chacha 0.9.0",
 "regex",
 "regex-automata 0.4.8",
 "roaring",
@@ -6711,7 +6714,7 @@ source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "4979f22fdb869068da03c9f7528f8297c6fd2606bc3a4affe42e6a823fdb8da4"
 dependencies = [
 "cfg-if",
- "windows-targets 0.48.5",
+ "windows-targets 0.52.6",
 ]

 [[package]]
@@ -6816,7 +6819,7 @@ checksum = "a7a70ba024b9dc04c27ea2f0c0548feb474ec5c54bba33a7f72f873a39d07b24"

 [[package]]
 name = "log-query"
-version = "0.15.4"
+version = "0.15.5"
 dependencies = [
 "chrono",
 "common-error",
@@ -6828,7 +6831,7 @@ dependencies = [

 [[package]]
 name = "log-store"
-version = "0.15.4"
+version = "0.15.5"
 dependencies = [
 "async-stream",
 "async-trait",
@@ -7126,7 +7129,7 @@ dependencies = [

 [[package]]
 name = "meta-client"
-version = "0.15.4"
+version = "0.15.5"
 dependencies = [
 "api",
 "async-trait",
@@ -7154,7 +7157,7 @@ dependencies = [

 [[package]]
 name = "meta-srv"
-version = "0.15.4"
+version = "0.15.5"
 dependencies = [
 "api",
 "async-trait",
@@ -7187,6 +7190,7 @@ dependencies = [
 "deadpool",
 "deadpool-postgres",
 "derive_builder 0.20.1",
+ "either",
 "etcd-client",
 "futures",
 "h2 0.3.26",
@@ -7250,7 +7254,7 @@ dependencies = [

 [[package]]
 name = "metric-engine"
-version = "0.15.4"
+version = "0.15.5"
 dependencies = [
 "api",
 "aquamarine",
@@ -7342,7 +7346,7 @@ dependencies = [

 [[package]]
 name = "mito-codec"
-version = "0.15.4"
+version = "0.15.5"
 dependencies = [
 "api",
 "bytes",
@@ -7365,7 +7369,7 @@ dependencies = [

 [[package]]
 name = "mito2"
-version = "0.15.4"
+version = "0.15.5"
 dependencies = [
 "api",
 "aquamarine",
@@ -8115,10 +8119,11 @@ dependencies = [

 [[package]]
 name = "object-store"
-version = "0.15.4"
+version = "0.15.5"
 dependencies = [
 "anyhow",
 "bytes",
+ "common-error",
 "common-telemetry",
 "common-test-util",
 "futures",
@@ -8429,7 +8434,7 @@ dependencies = [

 [[package]]
 name = "operator"
-version = "0.15.4"
+version = "0.15.5"
 dependencies = [
 "ahash 0.8.11",
 "api",
@@ -8484,7 +8489,7 @@ dependencies = [
 "sql",
 "sqlparser 0.54.0 (git+https://github.com/GreptimeTeam/sqlparser-rs.git?rev=0cf6c04490d59435ee965edd2078e8855bd8471e)",
 "store-api",
- "substrait 0.15.4",
+ "substrait 0.15.5",
 "table",
 "tokio",
 "tokio-util",
@@ -8751,7 +8756,7 @@ dependencies = [

 [[package]]
 name = "partition"
-version = "0.15.4"
+version = "0.15.5"
 dependencies = [
 "api",
 "async-trait",
@@ -9039,7 +9044,7 @@ checksum = "8b870d8c151b6f2fb93e84a13146138f05d02ed11c7e7c54f8826aaaf7c9f184"

 [[package]]
 name = "pipeline"
-version = "0.15.4"
+version = "0.15.5"
 dependencies = [
 "ahash 0.8.11",
 "api",
@@ -9182,7 +9187,7 @@ dependencies = [

 [[package]]
 name = "plugins"
-version = "0.15.4"
+version = "0.15.5"
 dependencies = [
 "auth",
 "clap 4.5.19",
@@ -9495,7 +9500,7 @@ dependencies = [

 [[package]]
 name = "promql"
-version = "0.15.4"
+version = "0.15.5"
 dependencies = [
 "ahash 0.8.11",
 "async-trait",
@@ -9591,7 +9596,7 @@ source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "be769465445e8c1474e9c5dac2018218498557af32d9ed057325ec9a41ae81bf"
 dependencies = [
 "heck 0.5.0",
- "itertools 0.11.0",
+ "itertools 0.14.0",
 "log",
 "multimap",
 "once_cell",
@@ -9637,7 +9642,7 @@ source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "8a56d757972c98b346a9b766e3f02746cde6dd1cd1d1d563472929fdd74bec4d"
 dependencies = [
 "anyhow",
- "itertools 0.11.0",
+ "itertools 0.14.0",
 "proc-macro2",
 "quote",
 "syn 2.0.100",
@@ -9777,7 +9782,7 @@ dependencies = [

 [[package]]
 name = "puffin"
-version = "0.15.4"
+version = "0.15.5"
 dependencies = [
 "async-compression 0.4.13",
 "async-trait",
@@ -9819,7 +9824,7 @@ dependencies = [

 [[package]]
 name = "query"
-version = "0.15.4"
+version = "0.15.5"
 dependencies = [
 "ahash 0.8.11",
 "api",
@@ -9885,7 +9890,7 @@ dependencies = [
 "sqlparser 0.54.0 (git+https://github.com/GreptimeTeam/sqlparser-rs.git?rev=0cf6c04490d59435ee965edd2078e8855bd8471e)",
 "statrs",
 "store-api",
- "substrait 0.15.4",
+ "substrait 0.15.5",
 "table",
 "tokio",
 "tokio-stream",
@@ -11171,7 +11176,7 @@ dependencies = [

 [[package]]
 name = "servers"
-version = "0.15.4"
+version = "0.15.5"
 dependencies = [
 "ahash 0.8.11",
 "api",
@@ -11292,7 +11297,7 @@ dependencies = [

 [[package]]
 name = "session"
-version = "0.15.4"
+version = "0.15.5"
 dependencies = [
 "api",
 "arc-swap",
@@ -11631,7 +11636,7 @@ dependencies = [

 [[package]]
 name = "sql"
-version = "0.15.4"
+version = "0.15.5"
 dependencies = [
 "api",
 "chrono",
@@ -11686,7 +11691,7 @@ dependencies = [

 [[package]]
 name = "sqlness-runner"
-version = "0.15.4"
+version = "0.15.5"
 dependencies = [
 "async-trait",
 "clap 4.5.19",
@@ -11986,7 +11991,7 @@ dependencies = [

 [[package]]
 name = "stat"
-version = "0.15.4"
+version = "0.15.5"
 dependencies = [
 "nix 0.30.1",
 ]
@@ -12012,7 +12017,7 @@ dependencies = [

 [[package]]
 name = "store-api"
-version = "0.15.4"
+version = "0.15.5"
 dependencies = [
 "api",
 "aquamarine",
@@ -12173,7 +12178,7 @@ dependencies = [

 [[package]]
 name = "substrait"
-version = "0.15.4"
+version = "0.15.5"
 dependencies = [
 "async-trait",
 "bytes",
@@ -12353,7 +12358,7 @@ dependencies = [

 [[package]]
 name = "table"
-version = "0.15.4"
+version = "0.15.5"
 dependencies = [
 "api",
 "async-trait",
@@ -12614,7 +12619,7 @@ checksum = "3369f5ac52d5eb6ab48c6b4ffdc8efbcad6b89c765749064ba298f2c68a16a76"

 [[package]]
 name = "tests-fuzz"
-version = "0.15.4"
+version = "0.15.5"
 dependencies = [
 "arbitrary",
 "async-trait",
@@ -12658,7 +12663,7 @@ dependencies = [

 [[package]]
 name = "tests-integration"
-version = "0.15.4"
+version = "0.15.5"
 dependencies = [
 "api",
 "arrow-flight",
@@ -12725,7 +12730,7 @@ dependencies = [
 "sql",
 "sqlx",
 "store-api",
- "substrait 0.15.4",
+ "substrait 0.15.5",
 "table",
 "tempfile",
 "time",
@@ -12735,6 +12740,7 @@ dependencies = [
 "tonic 0.12.3",
 "tower 0.5.2",
 "url",
+ "urlencoding",
 "uuid",
 "yaml-rust",
 "zstd 0.13.2",
@@ -14207,7 +14213,7 @@ version = "0.1.9"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "cf221c93e13a30d793f7645a0e7762c55d169dbb0a49671918a2319d289b10bb"
 dependencies = [
- "windows-sys 0.48.0",
+ "windows-sys 0.59.0",
 ]

 [[package]]
--- a/Cargo.toml
+++ b/Cargo.toml
@@ -71,7 +71,7 @@ members = [
 resolver = "2"

 [workspace.package]
-version = "0.15.4"
+version = "0.15.5"
 edition = "2021"
 license = "Apache-2.0"

@@ -130,11 +130,12 @@ deadpool = "0.12"
 deadpool-postgres = "0.14"
 derive_builder = "0.20"
 dotenv = "0.15"
+either = "1.15"
 etcd-client = "0.14"
 fst = "0.4.7"
 futures = "0.3"
 futures-util = "0.3"
-greptime-proto = { git = "https://github.com/GreptimeTeam/greptime-proto.git", rev = "551c70f2b352b6bc55ab769129344aeab8f22d11" }
+greptime-proto = { git = "https://github.com/GreptimeTeam/greptime-proto.git", rev = "f3103a8c9b8ce162457d0a3e3ca00d53d1a8bd06" }
 hex = "0.4"
 http = "1"
 humantime = "2.1"
--- a/2
+++ b/2
@@ -8,7 +8,7 @@ CARGO_BUILD_OPTS := --locked
 IMAGE_REGISTRY ?= docker.io
 IMAGE_NAMESPACE ?= greptime
 IMAGE_TAG ?= latest
-DEV_BUILDER_IMAGE_TAG ?= 2025-05-19-b2377d4b-20250520045554
+DEV_BUILDER_IMAGE_TAG ?= 2025-05-19-f55023f3-20250829091211
 BUILDX_MULTI_PLATFORM_BUILD ?= false
 BUILDX_BUILDER_NAME ?= gtbuilder
 BASE_IMAGE ?= ubuntu
--- a/src/catalog/Cargo.toml
+++ b/src/catalog/Cargo.toml
@@ -43,6 +43,8 @@ moka = { workspace = true, features = ["future", "sync"] }
 partition.workspace = true
 paste.workspace = true
 prometheus.workspace = true
+promql-parser.workspace = true
+rand.workspace = true
 rustc-hash.workspace = true
 serde_json.workspace = true
 session.workspace = true
--- a/src/catalog/src/process_manager.rs
+++ b/src/catalog/src/process_manager.rs
@@ -14,17 +14,24 @@

 use std::collections::hash_map::Entry;
 use std::collections::HashMap;
-use std::fmt::{Debug, Formatter};
+use std::fmt::{Debug, Display, Formatter};
 use std::sync::atomic::{AtomicU32, Ordering};
 use std::sync::{Arc, RwLock};
+use std::time::{Duration, Instant, UNIX_EPOCH};

 use api::v1::frontend::{KillProcessRequest, ListProcessRequest, ProcessInfo};
 use common_base::cancellation::CancellationHandle;
 use common_frontend::selector::{FrontendSelector, MetaClientSelector};
-use common_telemetry::{debug, info, warn};
+use common_frontend::slow_query_event::SlowQueryEvent;
+use common_telemetry::{debug, error, info, warn};
 use common_time::util::current_time_millis;
 use meta_client::MetaClientRef;
+use promql_parser::parser::EvalStmt;
+use rand::random;
+use session::context::QueryContextRef;
 use snafu::{ensure, OptionExt, ResultExt};
+use sql::statements::statement::Statement;
+use tokio::sync::mpsc::Sender;

 use crate::error;
 use crate::metrics::{PROCESS_KILL_COUNT, PROCESS_LIST_COUNT};
@@ -44,6 +51,23 @@ pub struct ProcessManager {
    frontend_selector: Option<MetaClientSelector>,
 }

+/// Represents a parsed query statement, functionally equivalent to [query::parser::QueryStatement].
+/// This enum is defined here to avoid cyclic dependencies with the query parser module.
+#[derive(Debug, Clone)]
+pub enum QueryStatement {
+    Sql(Statement),
+    Promql(EvalStmt),
+}
+
+impl Display for QueryStatement {
+    fn fmt(&self, f: &mut Formatter<'_>) -> std::fmt::Result {
+        match self {
+            QueryStatement::Sql(stmt) => write!(f, "{}", stmt),
+            QueryStatement::Promql(eval_stmt) => write!(f, "{}", eval_stmt),
+        }
+    }
+}
+
 impl ProcessManager {
    /// Create a [ProcessManager] instance with server address and kv client.
    pub fn new(server_addr: String, meta_client: Option<MetaClientRef>) -> Self {
@@ -67,6 +91,7 @@ impl ProcessManager {
        query: String,
        client: String,
        query_id: Option<ProcessId>,
+        _slow_query_timer: Option<SlowQueryTimer>,
    ) -> Ticket {
        let id = query_id.unwrap_or_else(|| self.next_id.fetch_add(1, Ordering::Relaxed));
        let process = ProcessInfo {
@@ -93,6 +118,7 @@ impl ProcessManager {
            manager: self.clone(),
            id,
            cancellation_handle,
+            _slow_query_timer,
        }
    }

@@ -223,6 +249,7 @@ pub struct Ticket {
    pub(crate) manager: ProcessManagerRef,
    pub(crate) id: ProcessId,
    pub cancellation_handle: Arc<CancellationHandle>,
+    _slow_query_timer: Option<SlowQueryTimer>,
 }

 impl Drop for Ticket {
@@ -263,6 +290,107 @@ impl Debug for CancellableProcess {
    }
 }

+/// SlowQueryTimer is used to log slow query when it's dropped.
+/// In drop(), it will check if the query is slow and send the slow query event to the handler.
+pub struct SlowQueryTimer {
+    start: Instant,
+    stmt: QueryStatement,
+    query_ctx: QueryContextRef,
+    threshold: Option<Duration>,
+    sample_ratio: Option<f64>,
+    tx: Sender<SlowQueryEvent>,
+}
+
+impl SlowQueryTimer {
+    pub fn new(
+        stmt: QueryStatement,
+        query_ctx: QueryContextRef,
+        threshold: Option<Duration>,
+        sample_ratio: Option<f64>,
+        tx: Sender<SlowQueryEvent>,
+    ) -> Self {
+        Self {
+            start: Instant::now(),
+            stmt,
+            query_ctx,
+            threshold,
+            sample_ratio,
+            tx,
+        }
+    }
+}
+
+impl SlowQueryTimer {
+    fn send_slow_query_event(&self, elapsed: Duration, threshold: Duration) {
+        let mut slow_query_event = SlowQueryEvent {
+            cost: elapsed.as_millis() as u64,
+            threshold: threshold.as_millis() as u64,
+            query: "".to_string(),
+            query_ctx: self.query_ctx.clone(),
+
+            // The following fields are only used for PromQL queries.
+            is_promql: false,
+            promql_range: None,
+            promql_step: None,
+            promql_start: None,
+            promql_end: None,
+        };
+
+        match &self.stmt {
+            QueryStatement::Promql(stmt) => {
+                slow_query_event.is_promql = true;
+                slow_query_event.query = stmt.expr.to_string();
+                slow_query_event.promql_step = Some(stmt.interval.as_millis() as u64);
+
+                let start = stmt
+                    .start
+                    .duration_since(UNIX_EPOCH)
+                    .unwrap_or_default()
+                    .as_millis() as i64;
+
+                let end = stmt
+                    .end
+                    .duration_since(UNIX_EPOCH)
+                    .unwrap_or_default()
+                    .as_millis() as i64;
+
+                slow_query_event.promql_range = Some((end - start) as u64);
+                slow_query_event.promql_start = Some(start);
+                slow_query_event.promql_end = Some(end);
+            }
+            QueryStatement::Sql(stmt) => {
+                slow_query_event.query = stmt.to_string();
+            }
+        }
+
+        // Send SlowQueryEvent to the handler.
+        if let Err(e) = self.tx.try_send(slow_query_event) {
+            error!(e; "Failed to send slow query event");
+        }
+    }
+}
+
+impl Drop for SlowQueryTimer {
+    fn drop(&mut self) {
+        if let Some(threshold) = self.threshold {
+            // Calculate the elaspsed duration since the timer is created.
+            let elapsed = self.start.elapsed();
+            if elapsed > threshold {
+                if let Some(ratio) = self.sample_ratio {
+                    // Only capture a portion of slow queries based on sample_ratio.
+                    // Generate a random number in [0, 1) and compare it with sample_ratio.
+                    if ratio >= 1.0 || random::<f64>() <= ratio {
+                        self.send_slow_query_event(elapsed, threshold);
+                    }
+                } else {
+                    // Captures all slow queries if sample_ratio is not set.
+                    self.send_slow_query_event(elapsed, threshold);
+                }
+            }
+        }
+    }
+}
+
 #[cfg(test)]
 mod tests {
    use std::sync::Arc;
@@ -278,6 +406,7 @@ mod tests {
            "SELECT * FROM table".to_string(),
            "".to_string(),
            None,
+            None,
        );

        let running_processes = process_manager.local_processes(None).unwrap();
@@ -301,6 +430,7 @@ mod tests {
            "SELECT * FROM table".to_string(),
            "client1".to_string(),
            Some(custom_id),
+            None,
        );

        assert_eq!(ticket.id, custom_id);
@@ -321,6 +451,7 @@ mod tests {
            "SELECT * FROM table1".to_string(),
            "client1".to_string(),
            None,
+            None,
        );

        let ticket2 = process_manager.clone().register_query(
@@ -329,6 +460,7 @@ mod tests {
            "SELECT * FROM table2".to_string(),
            "client2".to_string(),
            None,
+            None,
        );

        let running_processes = process_manager.local_processes(Some("public")).unwrap();
@@ -350,6 +482,7 @@ mod tests {
            "SELECT * FROM table1".to_string(),
            "client1".to_string(),
            None,
+            None,
        );

        let _ticket2 = process_manager.clone().register_query(
@@ -358,6 +491,7 @@ mod tests {
            "SELECT * FROM table2".to_string(),
            "client2".to_string(),
            None,
+            None,
        );

        // Test listing processes for specific catalog
@@ -384,6 +518,7 @@ mod tests {
            "SELECT * FROM table".to_string(),
            "client1".to_string(),
            None,
+            None,
        );
        assert_eq!(process_manager.local_processes(None).unwrap().len(), 1);
        process_manager.deregister_query("public".to_string(), ticket.id);
@@ -400,6 +535,7 @@ mod tests {
            "SELECT * FROM table".to_string(),
            "client1".to_string(),
            None,
+            None,
        );

        assert!(!ticket.cancellation_handle.is_cancelled());
@@ -417,6 +553,7 @@ mod tests {
            "SELECT * FROM table".to_string(),
            "client1".to_string(),
            None,
+            None,
        );
        assert!(!ticket.cancellation_handle.is_cancelled());
        let killed = process_manager
@@ -462,6 +599,7 @@ mod tests {
            "SELECT COUNT(*) FROM users WHERE age > 18".to_string(),
            "test_client".to_string(),
            Some(42),
+            None,
        );

        let processes = process_manager.local_processes(None).unwrap();
@@ -488,6 +626,7 @@ mod tests {
                "SELECT * FROM table".to_string(),
                "client1".to_string(),
                None,
+                None,
            );

            // Process should be registered
--- a/src/catalog/src/system_schema/information_schema/region_statistics.rs
+++ b/src/catalog/src/system_schema/information_schema/region_statistics.rs
@@ -40,6 +40,7 @@ const REGION_ID: &str = "region_id";
 const TABLE_ID: &str = "table_id";
 const REGION_NUMBER: &str = "region_number";
 const REGION_ROWS: &str = "region_rows";
+const WRITTEN_BYTES: &str = "written_bytes_since_open";
 const DISK_SIZE: &str = "disk_size";
 const MEMTABLE_SIZE: &str = "memtable_size";
 const MANIFEST_SIZE: &str = "manifest_size";
@@ -56,6 +57,7 @@ const INIT_CAPACITY: usize = 42;
 /// - `table_id`: The table id.
 /// - `region_number`: The region number.
 /// - `region_rows`: The number of rows in region.
+/// - `written_bytes_since_open`: The total bytes written of the region since region opened.
 /// - `memtable_size`: The memtable size in bytes.
 /// - `disk_size`: The approximate disk size in bytes.
 /// - `manifest_size`: The manifest size in bytes.
@@ -83,6 +85,7 @@ impl InformationSchemaRegionStatistics {
            ColumnSchema::new(TABLE_ID, ConcreteDataType::uint32_datatype(), false),
            ColumnSchema::new(REGION_NUMBER, ConcreteDataType::uint32_datatype(), false),
            ColumnSchema::new(REGION_ROWS, ConcreteDataType::uint64_datatype(), true),
+            ColumnSchema::new(WRITTEN_BYTES, ConcreteDataType::uint64_datatype(), true),
            ColumnSchema::new(DISK_SIZE, ConcreteDataType::uint64_datatype(), true),
            ColumnSchema::new(MEMTABLE_SIZE, ConcreteDataType::uint64_datatype(), true),
            ColumnSchema::new(MANIFEST_SIZE, ConcreteDataType::uint64_datatype(), true),
@@ -145,6 +148,7 @@ struct InformationSchemaRegionStatisticsBuilder {
    table_ids: UInt32VectorBuilder,
    region_numbers: UInt32VectorBuilder,
    region_rows: UInt64VectorBuilder,
+    written_bytes: UInt64VectorBuilder,
    disk_sizes: UInt64VectorBuilder,
    memtable_sizes: UInt64VectorBuilder,
    manifest_sizes: UInt64VectorBuilder,
@@ -163,6 +167,7 @@ impl InformationSchemaRegionStatisticsBuilder {
            table_ids: UInt32VectorBuilder::with_capacity(INIT_CAPACITY),
            region_numbers: UInt32VectorBuilder::with_capacity(INIT_CAPACITY),
            region_rows: UInt64VectorBuilder::with_capacity(INIT_CAPACITY),
+            written_bytes: UInt64VectorBuilder::with_capacity(INIT_CAPACITY),
            disk_sizes: UInt64VectorBuilder::with_capacity(INIT_CAPACITY),
            memtable_sizes: UInt64VectorBuilder::with_capacity(INIT_CAPACITY),
            manifest_sizes: UInt64VectorBuilder::with_capacity(INIT_CAPACITY),
@@ -193,6 +198,7 @@ impl InformationSchemaRegionStatisticsBuilder {
            (TABLE_ID, &Value::from(region_stat.id.table_id())),
            (REGION_NUMBER, &Value::from(region_stat.id.region_number())),
            (REGION_ROWS, &Value::from(region_stat.num_rows)),
+            (WRITTEN_BYTES, &Value::from(region_stat.written_bytes)),
            (DISK_SIZE, &Value::from(region_stat.approximate_bytes)),
            (MEMTABLE_SIZE, &Value::from(region_stat.memtable_size)),
            (MANIFEST_SIZE, &Value::from(region_stat.manifest_size)),
@@ -211,6 +217,7 @@ impl InformationSchemaRegionStatisticsBuilder {
        self.region_numbers
            .push(Some(region_stat.id.region_number()));
        self.region_rows.push(Some(region_stat.num_rows));
+        self.written_bytes.push(Some(region_stat.written_bytes));
        self.disk_sizes.push(Some(region_stat.approximate_bytes));
        self.memtable_sizes.push(Some(region_stat.memtable_size));
        self.manifest_sizes.push(Some(region_stat.manifest_size));
@@ -226,6 +233,7 @@ impl InformationSchemaRegionStatisticsBuilder {
            Arc::new(self.table_ids.finish()),
            Arc::new(self.region_numbers.finish()),
            Arc::new(self.region_rows.finish()),
+            Arc::new(self.written_bytes.finish()),
            Arc::new(self.disk_sizes.finish()),
            Arc::new(self.memtable_sizes.finish()),
            Arc::new(self.manifest_sizes.finish()),
--- a/src/cli/Cargo.toml
+++ b/src/cli/Cargo.toml
@@ -43,7 +43,6 @@ common-time.workspace = true
 common-version.workspace = true
 common-wal.workspace = true
 datatypes.workspace = true
-either = "1.8"
 etcd-client.workspace = true
 futures.workspace = true
 humantime.workspace = true
--- a/src/cli/src/metadata/control/get.rs
+++ b/src/cli/src/metadata/control/get.rs
@@ -29,7 +29,7 @@ use futures::TryStreamExt;

 use crate::error::InvalidArgumentsSnafu;
 use crate::metadata::common::StoreConfig;
-use crate::metadata::control::utils::{decode_key_value, get_table_id_by_name, json_fromatter};
+use crate::metadata::control::utils::{decode_key_value, get_table_id_by_name, json_formatter};
 use crate::Tool;

 /// Getting metadata from metadata store.
@@ -206,7 +206,7 @@ impl Tool for GetTableTool {
            println!(
                "{}\n{}",
                TableInfoKey::new(table_id),
-                json_fromatter(self.pretty, &*table_info)
+                json_formatter(self.pretty, &*table_info)
            );
        } else {
            println!("Table info not found");
@@ -221,7 +221,7 @@ impl Tool for GetTableTool {
            println!(
                "{}\n{}",
                TableRouteKey::new(table_id),
-                json_fromatter(self.pretty, &table_route)
+                json_formatter(self.pretty, &table_route)
            );
        } else {
            println!("Table route not found");
--- a/src/cli/src/metadata/control/utils.rs
+++ b/src/cli/src/metadata/control/utils.rs
@@ -27,7 +27,7 @@ pub fn decode_key_value(kv: KeyValue) -> CommonMetaResult<(String, String)> {
 }

 /// Formats a value as a JSON string.
-pub fn json_fromatter<T>(pretty: bool, value: &T) -> String
+pub fn json_formatter<T>(pretty: bool, value: &T) -> String
 where
    T: Serialize,
 {
--- a/src/cmd/Cargo.toml
+++ b/src/cmd/Cargo.toml
@@ -52,7 +52,6 @@ common-version.workspace = true
 common-wal.workspace = true
 datanode.workspace = true
 datatypes.workspace = true
-either = "1.8"
 etcd-client.workspace = true
 file-engine.workspace = true
 flow.workspace = true
--- a/src/cmd/src/lib.rs
+++ b/src/cmd/src/lib.rs
@@ -105,7 +105,7 @@ pub trait App: Send {
    }
 }

-/// Log the versions of the application, and the arguments passed to the cli.
+/// Log the versions of the application.
 ///
 /// `version` should be the same as the output of cli "--version";
 /// and the `short_version` is the short version of the codes, often consist of git branch and commit.
@@ -115,10 +115,7 @@ pub fn log_versions(version: &str, short_version: &str, app: &str) {
        .with_label_values(&[common_version::version(), short_version, app])
        .inc();

-    // Log version and argument flags.
    info!("GreptimeDB version: {}", version);
-
-    log_env_flags();
 }

 pub fn create_resource_limit_metrics(app: &str) {
@@ -138,10 +135,3 @@ pub fn create_resource_limit_metrics(app: &str) {
        MEMORY_LIMIT.with_label_values(&[app]).set(memory_limit);
    }
 }
-
-fn log_env_flags() {
-    info!("command line arguments");
-    for argument in std::env::args() {
-        info!("argument: {}", argument);
-    }
-}
--- a/src/cmd/src/metasrv.rs
+++ b/src/cmd/src/metasrv.rs
@@ -54,6 +54,10 @@ impl Instance {
    pub fn get_inner(&self) -> &MetasrvInstance {
        &self.instance
    }
+
+    pub fn mut_inner(&mut self) -> &mut MetasrvInstance {
+        &mut self.instance
+    }
 }

 #[async_trait]
--- a/src/cmd/src/standalone.rs
+++ b/src/cmd/src/standalone.rs
@@ -792,6 +792,7 @@ impl InformationExtension for StandaloneInformationExtension {
                    region_manifest: region_stat.manifest.into(),
                    data_topic_latest_entry_id: region_stat.data_topic_latest_entry_id,
                    metadata_topic_latest_entry_id: region_stat.metadata_topic_latest_entry_id,
+                    written_bytes: region_stat.written_bytes,
                }
            })
            .collect::<Vec<_>>();
--- a/src/common/error/src/lib.rs
+++ b/src/common/error/src/lib.rs
@@ -45,3 +45,19 @@ pub fn from_err_code_msg_to_header(code: u32, msg: &str) -> HeaderMap {
    header.insert(GREPTIME_DB_HEADER_ERROR_MSG, msg);
    header
 }
+
+/// Returns the external root cause of the source error (exclude the current error).
+pub fn root_source(err: &dyn std::error::Error) -> Option<&dyn std::error::Error> {
+    // There are some divergence about the behavior of the `sources()` API
+    // in https://github.com/rust-lang/rust/issues/58520
+    // So this function iterates the sources manually.
+    let mut root = err.source();
+    while let Some(r) = root {
+        if let Some(s) = r.source() {
+            root = Some(s);
+        } else {
+            break;
+        }
+    }
+    root
+}
--- a/src/common/frontend/Cargo.toml
+++ b/src/common/frontend/Cargo.toml
@@ -12,6 +12,7 @@ common-macro.workspace = true
 common-meta.workspace = true
 greptime-proto.workspace = true
 meta-client.workspace = true
+session.workspace = true
 snafu.workspace = true
 tonic.workspace = true

--- a/src/common/frontend/src/lib.rs
+++ b/src/common/frontend/src/lib.rs
@@ -19,6 +19,7 @@ use snafu::OptionExt;

 pub mod error;
 pub mod selector;
+pub mod slow_query_event;

 #[derive(Debug, Clone, Eq, PartialEq)]
 pub struct DisplayProcessId {
--- a/src/common/frontend/src/slow_query_event.rs
+++ b/src/common/frontend/src/slow_query_event.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 session::context::QueryContextRef;
+
+#[derive(Debug)]
+pub struct SlowQueryEvent {
+    pub cost: u64,
+    pub threshold: u64,
+    pub query: String,
+    pub is_promql: bool,
+    pub query_ctx: QueryContextRef,
+    pub promql_range: Option<u64>,
+    pub promql_step: Option<u64>,
+    pub promql_start: Option<i64>,
+    pub promql_end: Option<i64>,
+}
--- a/src/common/meta/src/datanode.rs
+++ b/src/common/meta/src/datanode.rs
@@ -97,6 +97,8 @@ pub struct RegionStat {
    pub index_size: u64,
    /// The manifest infoof the region.
    pub region_manifest: RegionManifestInfo,
+    /// The total bytes written of the region since region opened.
+    pub written_bytes: u64,
    /// The latest entry id of topic used by data.
    /// **Only used by remote WAL prune.**
    pub data_topic_latest_entry_id: u64,
@@ -277,6 +279,7 @@ impl From<&api::v1::meta::RegionStat> for RegionStat {
            sst_size: region_stat.sst_size,
            index_size: region_stat.index_size,
            region_manifest: region_stat.manifest.into(),
+            written_bytes: region_stat.written_bytes,
            data_topic_latest_entry_id: region_stat.data_topic_latest_entry_id,
            metadata_topic_latest_entry_id: region_stat.metadata_topic_latest_entry_id,
        }
--- a/src/common/meta/src/kv_backend/rds/mysql.rs
+++ b/src/common/meta/src/kv_backend/rds/mysql.rs
@@ -611,7 +611,7 @@ mod tests {
    #[tokio::test]
    async fn test_mysql_put() {
        maybe_skip_mysql_integration_test!();
-        let kv_backend = build_mysql_kv_backend("put_test").await.unwrap();
+        let kv_backend = build_mysql_kv_backend("put-test").await.unwrap();
        let prefix = b"put/";
        prepare_kv_with_prefix(&kv_backend, prefix.to_vec()).await;
        test_kv_put_with_prefix(&kv_backend, prefix.to_vec()).await;
@@ -621,7 +621,7 @@ mod tests {
    #[tokio::test]
    async fn test_mysql_range() {
        maybe_skip_mysql_integration_test!();
-        let kv_backend = build_mysql_kv_backend("range_test").await.unwrap();
+        let kv_backend = build_mysql_kv_backend("range-test").await.unwrap();
        let prefix = b"range/";
        prepare_kv_with_prefix(&kv_backend, prefix.to_vec()).await;
        test_kv_range_with_prefix(&kv_backend, prefix.to_vec()).await;
@@ -631,7 +631,7 @@ mod tests {
    #[tokio::test]
    async fn test_mysql_range_2() {
        maybe_skip_mysql_integration_test!();
-        let kv_backend = build_mysql_kv_backend("range2_test").await.unwrap();
+        let kv_backend = build_mysql_kv_backend("range2-test").await.unwrap();
        let prefix = b"range2/";
        test_kv_range_2_with_prefix(&kv_backend, prefix.to_vec()).await;
        unprepare_kv(&kv_backend, prefix).await;
@@ -640,7 +640,7 @@ mod tests {
    #[tokio::test]
    async fn test_mysql_all_range() {
        maybe_skip_mysql_integration_test!();
-        let kv_backend = build_mysql_kv_backend("simple_range_test").await.unwrap();
+        let kv_backend = build_mysql_kv_backend("simple_range-test").await.unwrap();
        let prefix = b"";
        prepare_kv_with_prefix(&kv_backend, prefix.to_vec()).await;
        test_simple_kv_range(&kv_backend).await;
@@ -650,7 +650,7 @@ mod tests {
    #[tokio::test]
    async fn test_mysql_batch_get() {
        maybe_skip_mysql_integration_test!();
-        let kv_backend = build_mysql_kv_backend("batch_get_test").await.unwrap();
+        let kv_backend = build_mysql_kv_backend("batch_get-test").await.unwrap();
        let prefix = b"batch_get/";
        prepare_kv_with_prefix(&kv_backend, prefix.to_vec()).await;
        test_kv_batch_get_with_prefix(&kv_backend, prefix.to_vec()).await;
@@ -660,7 +660,7 @@ mod tests {
    #[tokio::test]
    async fn test_mysql_batch_delete() {
        maybe_skip_mysql_integration_test!();
-        let kv_backend = build_mysql_kv_backend("batch_delete_test").await.unwrap();
+        let kv_backend = build_mysql_kv_backend("batch_delete-test").await.unwrap();
        let prefix = b"batch_delete/";
        prepare_kv_with_prefix(&kv_backend, prefix.to_vec()).await;
        test_kv_delete_range_with_prefix(&kv_backend, prefix.to_vec()).await;
@@ -670,7 +670,7 @@ mod tests {
    #[tokio::test]
    async fn test_mysql_batch_delete_with_prefix() {
        maybe_skip_mysql_integration_test!();
-        let kv_backend = build_mysql_kv_backend("batch_delete_with_prefix_test")
+        let kv_backend = build_mysql_kv_backend("batch_delete_with_prefix-test")
            .await
            .unwrap();
        let prefix = b"batch_delete/";
@@ -682,7 +682,7 @@ mod tests {
    #[tokio::test]
    async fn test_mysql_delete_range() {
        maybe_skip_mysql_integration_test!();
-        let kv_backend = build_mysql_kv_backend("delete_range_test").await.unwrap();
+        let kv_backend = build_mysql_kv_backend("delete_range-test").await.unwrap();
        let prefix = b"delete_range/";
        prepare_kv_with_prefix(&kv_backend, prefix.to_vec()).await;
        test_kv_delete_range_with_prefix(&kv_backend, prefix.to_vec()).await;
@@ -692,7 +692,7 @@ mod tests {
    #[tokio::test]
    async fn test_mysql_compare_and_put() {
        maybe_skip_mysql_integration_test!();
-        let kv_backend = build_mysql_kv_backend("compare_and_put_test")
+        let kv_backend = build_mysql_kv_backend("compare_and_put-test")
            .await
            .unwrap();
        let prefix = b"compare_and_put/";
@@ -703,7 +703,7 @@ mod tests {
    #[tokio::test]
    async fn test_mysql_txn() {
        maybe_skip_mysql_integration_test!();
-        let kv_backend = build_mysql_kv_backend("txn_test").await.unwrap();
+        let kv_backend = build_mysql_kv_backend("txn-test").await.unwrap();
        test_txn_one_compare_op(&kv_backend).await;
        text_txn_multi_compare_op(&kv_backend).await;
        test_txn_compare_equal(&kv_backend).await;
--- a/src/common/meta/src/kv_backend/rds/postgres.rs
+++ b/src/common/meta/src/kv_backend/rds/postgres.rs
@@ -585,7 +585,7 @@ mod tests {
    #[tokio::test]
    async fn test_pg_put() {
        maybe_skip_postgres_integration_test!();
-        let kv_backend = build_pg_kv_backend("put_test").await.unwrap();
+        let kv_backend = build_pg_kv_backend("put-test").await.unwrap();
        let prefix = b"put/";
        prepare_kv_with_prefix(&kv_backend, prefix.to_vec()).await;
        test_kv_put_with_prefix(&kv_backend, prefix.to_vec()).await;
@@ -595,7 +595,7 @@ mod tests {
    #[tokio::test]
    async fn test_pg_range() {
        maybe_skip_postgres_integration_test!();
-        let kv_backend = build_pg_kv_backend("range_test").await.unwrap();
+        let kv_backend = build_pg_kv_backend("range-test").await.unwrap();
        let prefix = b"range/";
        prepare_kv_with_prefix(&kv_backend, prefix.to_vec()).await;
        test_kv_range_with_prefix(&kv_backend, prefix.to_vec()).await;
@@ -605,7 +605,7 @@ mod tests {
    #[tokio::test]
    async fn test_pg_range_2() {
        maybe_skip_postgres_integration_test!();
-        let kv_backend = build_pg_kv_backend("range2_test").await.unwrap();
+        let kv_backend = build_pg_kv_backend("range2-test").await.unwrap();
        let prefix = b"range2/";
        test_kv_range_2_with_prefix(&kv_backend, prefix.to_vec()).await;
        unprepare_kv(&kv_backend, prefix).await;
@@ -614,7 +614,7 @@ mod tests {
    #[tokio::test]
    async fn test_pg_all_range() {
        maybe_skip_postgres_integration_test!();
-        let kv_backend = build_pg_kv_backend("simple_range_test").await.unwrap();
+        let kv_backend = build_pg_kv_backend("simple_range-test").await.unwrap();
        let prefix = b"";
        prepare_kv_with_prefix(&kv_backend, prefix.to_vec()).await;
        test_simple_kv_range(&kv_backend).await;
@@ -624,7 +624,7 @@ mod tests {
    #[tokio::test]
    async fn test_pg_batch_get() {
        maybe_skip_postgres_integration_test!();
-        let kv_backend = build_pg_kv_backend("batch_get_test").await.unwrap();
+        let kv_backend = build_pg_kv_backend("batch_get-test").await.unwrap();
        let prefix = b"batch_get/";
        prepare_kv_with_prefix(&kv_backend, prefix.to_vec()).await;
        test_kv_batch_get_with_prefix(&kv_backend, prefix.to_vec()).await;
@@ -634,7 +634,7 @@ mod tests {
    #[tokio::test]
    async fn test_pg_batch_delete() {
        maybe_skip_postgres_integration_test!();
-        let kv_backend = build_pg_kv_backend("batch_delete_test").await.unwrap();
+        let kv_backend = build_pg_kv_backend("batch_delete-test").await.unwrap();
        let prefix = b"batch_delete/";
        prepare_kv_with_prefix(&kv_backend, prefix.to_vec()).await;
        test_kv_delete_range_with_prefix(&kv_backend, prefix.to_vec()).await;
@@ -644,7 +644,7 @@ mod tests {
    #[tokio::test]
    async fn test_pg_batch_delete_with_prefix() {
        maybe_skip_postgres_integration_test!();
-        let kv_backend = build_pg_kv_backend("batch_delete_with_prefix_test")
+        let kv_backend = build_pg_kv_backend("batch_delete_with_prefix-test")
            .await
            .unwrap();
        let prefix = b"batch_delete/";
@@ -656,7 +656,7 @@ mod tests {
    #[tokio::test]
    async fn test_pg_delete_range() {
        maybe_skip_postgres_integration_test!();
-        let kv_backend = build_pg_kv_backend("delete_range_test").await.unwrap();
+        let kv_backend = build_pg_kv_backend("delete_range-test").await.unwrap();
        let prefix = b"delete_range/";
        prepare_kv_with_prefix(&kv_backend, prefix.to_vec()).await;
        test_kv_delete_range_with_prefix(&kv_backend, prefix.to_vec()).await;
@@ -666,7 +666,7 @@ mod tests {
    #[tokio::test]
    async fn test_pg_compare_and_put() {
        maybe_skip_postgres_integration_test!();
-        let kv_backend = build_pg_kv_backend("compare_and_put_test").await.unwrap();
+        let kv_backend = build_pg_kv_backend("compare_and_put-test").await.unwrap();
        let prefix = b"compare_and_put/";
        let kv_backend = Arc::new(kv_backend);
        test_kv_compare_and_put_with_prefix(kv_backend.clone(), prefix.to_vec()).await;
@@ -675,7 +675,7 @@ mod tests {
    #[tokio::test]
    async fn test_pg_txn() {
        maybe_skip_postgres_integration_test!();
-        let kv_backend = build_pg_kv_backend("txn_test").await.unwrap();
+        let kv_backend = build_pg_kv_backend("txn-test").await.unwrap();
        test_txn_one_compare_op(&kv_backend).await;
        text_txn_multi_compare_op(&kv_backend).await;
        test_txn_compare_equal(&kv_backend).await;
--- a/src/common/telemetry/src/tracing_context.rs
+++ b/src/common/telemetry/src/tracing_context.rs
@@ -19,7 +19,7 @@ use opentelemetry::propagation::TextMapPropagator;
 use opentelemetry_sdk::propagation::TraceContextPropagator;
 use tracing_opentelemetry::OpenTelemetrySpanExt;

-// An wapper for `Futures` that provides tracing instrument adapters.
+// A wrapper for `Futures` that provides tracing instrument adapters.
 pub trait FutureExt: std::future::Future + Sized {
    fn trace(self, span: tracing::span::Span) -> tracing::instrument::Instrumented<Self>;
 }
--- a/src/flow/src/heartbeat.rs
+++ b/src/flow/src/heartbeat.rs
@@ -218,6 +218,7 @@ impl HeartbeatTask {
                        if let Some(message) = message {
                            Self::new_heartbeat_request(&heartbeat_request, Some(message), &latest_report)
                        } else {
+                            warn!("Sender has been dropped, exiting the heartbeat loop");
                            // Receives None that means Sender was dropped, we need to break the current loop
                            break
                        }
@@ -259,7 +260,11 @@ impl HeartbeatTask {
                            error!(e; "Error while handling heartbeat response");
                        }
                    }
-                    Ok(None) => break,
+                    Ok(None) => {
+                        warn!("Heartbeat response stream closed");
+                        capture_self.start_with_retry(retry_interval).await;
+                        break;
+                    }
                    Err(e) => {
                        error!(e; "Occur error while reading heartbeat response");
                        capture_self.start_with_retry(retry_interval).await;
--- a/src/flow/src/plan/join.rs
+++ b/src/flow/src/plan/join.rs
@@ -71,6 +71,6 @@ pub struct LinearStagePlan {
    /// The key expressions to use for the lookup relation.
    pub lookup_key: Vec<ScalarExpr>,
    /// The closure to apply to the concatenation of the key columns,
-    /// the stream value columns, and the lookup value colunms.
+    /// the stream value columns, and the lookup value columns.
    pub closure: JoinFilter,
 }
--- a/src/frontend/src/heartbeat.rs
+++ b/src/frontend/src/heartbeat.rs
@@ -23,7 +23,7 @@ use common_meta::heartbeat::handler::{
 };
 use common_meta::heartbeat::mailbox::{HeartbeatMailbox, MailboxRef, OutgoingMessage};
 use common_meta::heartbeat::utils::outgoing_message_to_mailbox_message;
-use common_telemetry::{debug, error, info};
+use common_telemetry::{debug, error, info, warn};
 use meta_client::client::{HeartbeatSender, HeartbeatStream, MetaClient};
 use servers::addrs;
 use servers::heartbeat_options::HeartbeatOptions;
@@ -42,8 +42,8 @@ use crate::metrics::{HEARTBEAT_RECV_COUNT, HEARTBEAT_SENT_COUNT};
 pub struct HeartbeatTask {
    peer_addr: String,
    meta_client: Arc<MetaClient>,
-    report_interval: u64,
-    retry_interval: u64,
+    report_interval: Duration,
+    retry_interval: Duration,
    resp_handler_executor: HeartbeatResponseHandlerExecutorRef,
    start_time_ms: u64,
 }
@@ -58,8 +58,8 @@ impl HeartbeatTask {
        HeartbeatTask {
            peer_addr: addrs::resolve_addr(&opts.grpc.bind_addr, Some(&opts.grpc.server_addr)),
            meta_client,
-            report_interval: heartbeat_opts.interval.as_millis() as u64,
-            retry_interval: heartbeat_opts.retry_interval.as_millis() as u64,
+            report_interval: heartbeat_opts.interval,
+            retry_interval: heartbeat_opts.retry_interval,
            resp_handler_executor,
            start_time_ms: common_time::util::current_time_millis() as u64,
        }
@@ -103,13 +103,15 @@ impl HeartbeatTask {
                            HEARTBEAT_RECV_COUNT.with_label_values(&["success"]).inc();
                        }
                    }
-                    Ok(None) => break,
+                    Ok(None) => {
+                        warn!("Heartbeat response stream closed");
+                        capture_self.start_with_retry(retry_interval).await;
+                        break;
+                    }
                    Err(e) => {
                        HEARTBEAT_RECV_COUNT.with_label_values(&["error"]).inc();
                        error!(e; "Occur error while reading heartbeat response");
-                        capture_self
-                            .start_with_retry(Duration::from_millis(retry_interval))
-                            .await;
+                        capture_self.start_with_retry(retry_interval).await;

                        break;
                    }
@@ -177,12 +179,13 @@ impl HeartbeatTask {
                        if let Some(message) = message {
                            Self::new_heartbeat_request(&heartbeat_request, Some(message))
                        } else {
+                            warn!("Sender has been dropped, exiting the heartbeat loop");
                            // Receives None that means Sender was dropped, we need to break the current loop
                            break
                        }
                    }
                    _ = &mut sleep => {
-                        sleep.as_mut().reset(Instant::now() + Duration::from_millis(report_interval));
+                        sleep.as_mut().reset(Instant::now() + report_interval);
                       Self::new_heartbeat_request(&heartbeat_request, None)
                    }
                };
--- a/src/frontend/src/instance.rs
+++ b/src/frontend/src/instance.rs
@@ -32,7 +32,7 @@ use std::time::{Duration, SystemTime};
 use async_stream::stream;
 use async_trait::async_trait;
 use auth::{PermissionChecker, PermissionCheckerRef, PermissionReq};
-use catalog::process_manager::ProcessManagerRef;
+use catalog::process_manager::{ProcessManagerRef, QueryStatement as CatalogQueryStatement};
 use catalog::CatalogManagerRef;
 use client::OutputData;
 use common_base::cancellation::CancellableFuture;
@@ -64,8 +64,9 @@ use query::parser::{PromQuery, QueryLanguageParser, QueryStatement};
 use query::query_engine::options::{validate_catalog_and_schema, QueryOptions};
 use query::query_engine::DescribeResult;
 use query::QueryEngineRef;
-use servers::error as server_error;
-use servers::error::{AuthSnafu, ExecuteQuerySnafu, ParsePromQLSnafu};
+use servers::error::{
+    self as server_error, AuthSnafu, ExecuteQuerySnafu, ParsePromQLSnafu, UnexpectedResultSnafu,
+};
 use servers::interceptor::{
    PromQueryInterceptor, PromQueryInterceptorRef, SqlQueryInterceptor, SqlQueryInterceptorRef,
 };
@@ -183,36 +184,40 @@ impl Instance {
        let query_interceptor = self.plugins.get::<SqlQueryInterceptorRef<Error>>();
        let query_interceptor = query_interceptor.as_ref();

-        let _slow_query_timer = if let Some(recorder) = &self.slow_query_recorder {
-            recorder.start(QueryStatement::Sql(stmt.clone()), query_ctx.clone())
+        if should_capture_statement(Some(&stmt)) {
+            let slow_query_timer = self.slow_query_recorder.as_ref().and_then(|recorder| {
+                recorder.start(CatalogQueryStatement::Sql(stmt.clone()), query_ctx.clone())
+            });
+
+            let ticket = self.process_manager.register_query(
+                query_ctx.current_catalog().to_string(),
+                vec![query_ctx.current_schema()],
+                stmt.to_string(),
+                query_ctx.conn_info().to_string(),
+                Some(query_ctx.process_id()),
+                slow_query_timer,
+            );
+
+            let query_fut = self.exec_statement_with_timeout(stmt, query_ctx, query_interceptor);
+
+            CancellableFuture::new(query_fut, ticket.cancellation_handle.clone())
+                .await
+                .map_err(|_| error::CancelledSnafu.build())?
+                .map(|output| {
+                    let Output { meta, data } = output;
+
+                    let data = match data {
+                        OutputData::Stream(stream) => OutputData::Stream(Box::pin(
+                            CancellableStreamWrapper::new(stream, ticket),
+                        )),
+                        other => other,
+                    };
+                    Output { data, meta }
+                })
        } else {
-            None
-        };
-
-        let ticket = self.process_manager.register_query(
-            query_ctx.current_catalog().to_string(),
-            vec![query_ctx.current_schema()],
-            stmt.to_string(),
-            query_ctx.conn_info().to_string(),
-            Some(query_ctx.process_id()),
-        );
-
-        let query_fut = self.exec_statement_with_timeout(stmt, query_ctx, query_interceptor);
-
-        CancellableFuture::new(query_fut, ticket.cancellation_handle.clone())
-            .await
-            .map_err(|_| error::CancelledSnafu.build())?
-            .map(|output| {
-                let Output { meta, data } = output;
-
-                let data = match data {
-                    OutputData::Stream(stream) => {
-                        OutputData::Stream(Box::pin(CancellableStreamWrapper::new(stream, ticket)))
-                    }
-                    other => other,
-                };
-                Output { data, meta }
-            })
+            self.exec_statement_with_timeout(stmt, query_ctx, query_interceptor)
+                .await
+        }
    }

    async fn exec_statement_with_timeout(
@@ -428,13 +433,54 @@ impl SqlQueryHandler for Instance {
        }
    }

-    async fn do_exec_plan(&self, plan: LogicalPlan, query_ctx: QueryContextRef) -> Result<Output> {
-        // plan should be prepared before exec
-        // we'll do check there
-        self.query_engine
-            .execute(plan.clone(), query_ctx)
-            .await
-            .context(ExecLogicalPlanSnafu)
+    async fn do_exec_plan(
+        &self,
+        stmt: Option<Statement>,
+        plan: LogicalPlan,
+        query_ctx: QueryContextRef,
+    ) -> Result<Output> {
+        if should_capture_statement(stmt.as_ref()) {
+            // It's safe to unwrap here because we've already checked the type.
+            let stmt = stmt.unwrap();
+            let query = stmt.to_string();
+            let slow_query_timer = self.slow_query_recorder.as_ref().and_then(|recorder| {
+                recorder.start(CatalogQueryStatement::Sql(stmt), query_ctx.clone())
+            });
+
+            let ticket = self.process_manager.register_query(
+                query_ctx.current_catalog().to_string(),
+                vec![query_ctx.current_schema()],
+                query,
+                query_ctx.conn_info().to_string(),
+                Some(query_ctx.process_id()),
+                slow_query_timer,
+            );
+
+            let query_fut = self.query_engine.execute(plan.clone(), query_ctx);
+
+            CancellableFuture::new(query_fut, ticket.cancellation_handle.clone())
+                .await
+                .map_err(|_| error::CancelledSnafu.build())?
+                .map(|output| {
+                    let Output { meta, data } = output;
+
+                    let data = match data {
+                        OutputData::Stream(stream) => OutputData::Stream(Box::pin(
+                            CancellableStreamWrapper::new(stream, ticket),
+                        )),
+                        other => other,
+                    };
+                    Output { data, meta }
+                })
+                .context(ExecLogicalPlanSnafu)
+        } else {
+            // plan should be prepared before exec
+            // we'll do check there
+            self.query_engine
+                .execute(plan.clone(), query_ctx)
+                .await
+                .context(ExecLogicalPlanSnafu)
+        }
    }

    #[tracing::instrument(skip_all)]
@@ -528,12 +574,6 @@ impl PrometheusHandler for Instance {
            }
        })?;

-        let _slow_query_timer = if let Some(recorder) = &self.slow_query_recorder {
-            recorder.start(stmt.clone(), query_ctx.clone())
-        } else {
-            None
-        };
-
        let plan = self
            .statement_executor
            .plan(&stmt, query_ctx.clone())
@@ -543,10 +583,47 @@ impl PrometheusHandler for Instance {

        interceptor.pre_execute(query, Some(&plan), query_ctx.clone())?;

-        let output = self
-            .statement_executor
-            .exec_plan(plan, query_ctx.clone())
+        // Take the EvalStmt from the original QueryStatement and use it to create the CatalogQueryStatement.
+        let query_statement = if let QueryStatement::Promql(eval_stmt) = stmt {
+            CatalogQueryStatement::Promql(eval_stmt)
+        } else {
+            // It should not happen since the query is already parsed successfully.
+            return UnexpectedResultSnafu {
+                reason: "The query should always be promql.".to_string(),
+            }
+            .fail();
+        };
+        let query = query_statement.to_string();
+
+        let slow_query_timer = self
+            .slow_query_recorder
+            .as_ref()
+            .and_then(|recorder| recorder.start(query_statement, query_ctx.clone()));
+
+        let ticket = self.process_manager.register_query(
+            query_ctx.current_catalog().to_string(),
+            vec![query_ctx.current_schema()],
+            query,
+            query_ctx.conn_info().to_string(),
+            Some(query_ctx.process_id()),
+            slow_query_timer,
+        );
+
+        let query_fut = self.statement_executor.exec_plan(plan, query_ctx.clone());
+
+        let output = CancellableFuture::new(query_fut, ticket.cancellation_handle.clone())
            .await
+            .map_err(|_| servers::error::CancelledSnafu.build())?
+            .map(|output| {
+                let Output { meta, data } = output;
+                let data = match data {
+                    OutputData::Stream(stream) => {
+                        OutputData::Stream(Box::pin(CancellableStreamWrapper::new(stream, ticket)))
+                    }
+                    other => other,
+                };
+                Output { data, meta }
+            })
            .map_err(BoxedError::new)
            .context(ExecuteQuerySnafu)?;

@@ -764,6 +841,15 @@ fn validate_database(name: &ObjectName, query_ctx: &QueryContextRef) -> Result<(
        .context(SqlExecInterceptedSnafu)
 }

+// Create a query ticket and slow query timer if the statement is a query or readonly statement.
+fn should_capture_statement(stmt: Option<&Statement>) -> bool {
+    if let Some(stmt) = stmt {
+        matches!(stmt, Statement::Query(_)) || stmt.is_readonly()
+    } else {
+        false
+    }
+}
+
 #[cfg(test)]
 mod tests {
    use std::collections::HashMap;
--- a/src/frontend/src/instance/grpc.rs
+++ b/src/frontend/src/instance/grpc.rs
@@ -119,7 +119,8 @@ impl GrpcQueryHandler for Instance {
                            .await
                            .context(SubstraitDecodeLogicalPlanSnafu)?;
                        let output =
-                            SqlQueryHandler::do_exec_plan(self, logical_plan, ctx.clone()).await?;
+                            SqlQueryHandler::do_exec_plan(self, None, logical_plan, ctx.clone())
+                                .await?;

                        attach_timer(output, timer)
                    }
@@ -397,7 +398,7 @@ impl Instance {
            .context(common_query::error::GeneralDataFusionSnafu)
            .context(SubstraitDecodeLogicalPlanSnafu)?;

-        let output = SqlQueryHandler::do_exec_plan(self, optimized_plan, ctx.clone()).await?;
+        let output = SqlQueryHandler::do_exec_plan(self, None, optimized_plan, ctx.clone()).await?;

        Ok(attach_timer(output, timer))
    }
--- a/src/frontend/src/lib.rs
+++ b/src/frontend/src/lib.rs
@@ -22,5 +22,5 @@ pub(crate) mod limiter;
 pub(crate) mod metrics;
 pub mod server;
 pub mod service_config;
-pub(crate) mod slow_query_recorder;
+pub mod slow_query_recorder;
 mod stream_wrapper;
--- a/src/frontend/src/slow_query_recorder.rs
+++ b/src/frontend/src/slow_query_recorder.rs
@@ -14,22 +14,21 @@

 use std::collections::HashMap;
 use std::sync::Arc;
-use std::time::{Duration, Instant, UNIX_EPOCH};

 use api::v1::value::ValueData;
 use api::v1::{
    ColumnDataType, ColumnDef, ColumnSchema, CreateTableExpr, Row, RowInsertRequest,
    RowInsertRequests, Rows, SemanticType,
 };
+use catalog::process_manager::{QueryStatement as CatalogQueryStatement, SlowQueryTimer};
 use catalog::CatalogManagerRef;
 use common_catalog::consts::{default_engine, DEFAULT_PRIVATE_SCHEMA_NAME};
+use common_frontend::slow_query_event::SlowQueryEvent;
 use common_telemetry::logging::{SlowQueriesRecordType, SlowQueryOptions};
 use common_telemetry::{debug, error, info, slow};
 use common_time::timestamp::{TimeUnit, Timestamp};
 use operator::insert::InserterRef;
 use operator::statement::StatementExecutorRef;
-use query::parser::QueryStatement;
-use rand::random;
 use session::context::{QueryContextBuilder, QueryContextRef};
 use snafu::ResultExt;
 use store_api::mito_engine_options::{APPEND_MODE_KEY, TTL_KEY};
@@ -39,16 +38,16 @@ use tokio::task::JoinHandle;

 use crate::error::{CatalogSnafu, Result, TableOperationSnafu};

-const SLOW_QUERY_TABLE_NAME: &str = "slow_queries";
-const SLOW_QUERY_TABLE_COST_COLUMN_NAME: &str = "cost";
-const SLOW_QUERY_TABLE_THRESHOLD_COLUMN_NAME: &str = "threshold";
-const SLOW_QUERY_TABLE_QUERY_COLUMN_NAME: &str = "query";
-const SLOW_QUERY_TABLE_TIMESTAMP_COLUMN_NAME: &str = "timestamp";
-const SLOW_QUERY_TABLE_IS_PROMQL_COLUMN_NAME: &str = "is_promql";
-const SLOW_QUERY_TABLE_PROMQL_START_COLUMN_NAME: &str = "promql_start";
-const SLOW_QUERY_TABLE_PROMQL_END_COLUMN_NAME: &str = "promql_end";
-const SLOW_QUERY_TABLE_PROMQL_RANGE_COLUMN_NAME: &str = "promql_range";
-const SLOW_QUERY_TABLE_PROMQL_STEP_COLUMN_NAME: &str = "promql_step";
+pub const SLOW_QUERY_TABLE_NAME: &str = "slow_queries";
+pub const SLOW_QUERY_TABLE_COST_COLUMN_NAME: &str = "cost";
+pub const SLOW_QUERY_TABLE_THRESHOLD_COLUMN_NAME: &str = "threshold";
+pub const SLOW_QUERY_TABLE_QUERY_COLUMN_NAME: &str = "query";
+pub const SLOW_QUERY_TABLE_TIMESTAMP_COLUMN_NAME: &str = "timestamp";
+pub const SLOW_QUERY_TABLE_IS_PROMQL_COLUMN_NAME: &str = "is_promql";
+pub const SLOW_QUERY_TABLE_PROMQL_START_COLUMN_NAME: &str = "promql_start";
+pub const SLOW_QUERY_TABLE_PROMQL_END_COLUMN_NAME: &str = "promql_end";
+pub const SLOW_QUERY_TABLE_PROMQL_RANGE_COLUMN_NAME: &str = "promql_range";
+pub const SLOW_QUERY_TABLE_PROMQL_STEP_COLUMN_NAME: &str = "promql_step";

 const DEFAULT_SLOW_QUERY_TABLE_TTL: &str = "30d";
 const DEFAULT_SLOW_QUERY_EVENTS_CHANNEL_SIZE: usize = 1024;
@@ -61,19 +60,6 @@ pub struct SlowQueryRecorder {
    _handle: Arc<JoinHandle<()>>,
 }

-#[derive(Debug)]
-struct SlowQueryEvent {
-    cost: u64,
-    threshold: u64,
-    query: String,
-    is_promql: bool,
-    query_ctx: QueryContextRef,
-    promql_range: Option<u64>,
-    promql_step: Option<u64>,
-    promql_start: Option<i64>,
-    promql_end: Option<i64>,
-}
-
 impl SlowQueryRecorder {
    /// Create a new SlowQueryRecorder.
    pub fn new(
@@ -115,18 +101,17 @@ impl SlowQueryRecorder {
    /// The timer sets the start time when created and calculates the elapsed duration when dropped.
    pub fn start(
        &self,
-        stmt: QueryStatement,
+        stmt: CatalogQueryStatement,
        query_ctx: QueryContextRef,
    ) -> Option<SlowQueryTimer> {
        if self.slow_query_opts.enable {
-            Some(SlowQueryTimer {
+            Some(SlowQueryTimer::new(
                stmt,
                query_ctx,
-                start: Instant::now(), // Set the initial start time.
-                threshold: self.slow_query_opts.threshold,
-                sample_ratio: self.slow_query_opts.sample_ratio,
-                tx: self.tx.clone(),
-            })
+                self.slow_query_opts.threshold,
+                self.slow_query_opts.sample_ratio,
+                self.tx.clone(),
+            ))
        } else {
            None
        }
@@ -447,85 +432,3 @@ impl SlowQueryEventHandler {
        ]
    }
 }
-
-/// SlowQueryTimer is used to log slow query when it's dropped.
-/// In drop(), it will check if the query is slow and send the slow query event to the handler.
-pub struct SlowQueryTimer {
-    start: Instant,
-    stmt: QueryStatement,
-    query_ctx: QueryContextRef,
-    threshold: Option<Duration>,
-    sample_ratio: Option<f64>,
-    tx: Sender<SlowQueryEvent>,
-}
-
-impl SlowQueryTimer {
-    fn send_slow_query_event(&self, elapsed: Duration, threshold: Duration) {
-        let mut slow_query_event = SlowQueryEvent {
-            cost: elapsed.as_millis() as u64,
-            threshold: threshold.as_millis() as u64,
-            query: "".to_string(),
-            query_ctx: self.query_ctx.clone(),
-
-            // The following fields are only used for PromQL queries.
-            is_promql: false,
-            promql_range: None,
-            promql_step: None,
-            promql_start: None,
-            promql_end: None,
-        };
-
-        match &self.stmt {
-            QueryStatement::Promql(stmt) => {
-                slow_query_event.is_promql = true;
-                slow_query_event.query = stmt.expr.to_string();
-                slow_query_event.promql_step = Some(stmt.interval.as_millis() as u64);
-
-                let start = stmt
-                    .start
-                    .duration_since(UNIX_EPOCH)
-                    .unwrap_or_default()
-                    .as_millis() as i64;
-
-                let end = stmt
-                    .end
-                    .duration_since(UNIX_EPOCH)
-                    .unwrap_or_default()
-                    .as_millis() as i64;
-
-                slow_query_event.promql_range = Some((end - start) as u64);
-                slow_query_event.promql_start = Some(start);
-                slow_query_event.promql_end = Some(end);
-            }
-            QueryStatement::Sql(stmt) => {
-                slow_query_event.query = stmt.to_string();
-            }
-        }
-
-        // Send SlowQueryEvent to the handler.
-        if let Err(e) = self.tx.try_send(slow_query_event) {
-            error!(e; "Failed to send slow query event");
-        }
-    }
-}
-
-impl Drop for SlowQueryTimer {
-    fn drop(&mut self) {
-        if let Some(threshold) = self.threshold {
-            // Calculate the elaspsed duration since the timer is created.
-            let elapsed = self.start.elapsed();
-            if elapsed > threshold {
-                if let Some(ratio) = self.sample_ratio {
-                    // Only capture a portion of slow queries based on sample_ratio.
-                    // Generate a random number in [0, 1) and compare it with sample_ratio.
-                    if ratio >= 1.0 || random::<f64>() <= ratio {
-                        self.send_slow_query_event(elapsed, threshold);
-                    }
-                } else {
-                    // Captures all slow queries if sample_ratio is not set.
-                    self.send_slow_query_event(elapsed, threshold);
-                }
-            }
-        }
-    }
-}
--- a/src/frontend/src/stream_wrapper.rs
+++ b/src/frontend/src/stream_wrapper.rs
@@ -187,6 +187,7 @@ mod tests {
            "query".to_string(),
            "client".to_string(),
            None,
+            None,
        );

        let mut cancellable_stream = CancellableStreamWrapper::new(Box::pin(mock_stream), ticket);
@@ -210,6 +211,7 @@ mod tests {
            "query".to_string(),
            "client".to_string(),
            None,
+            None,
        );

        // Cancel before creating the wrapper
@@ -234,6 +236,7 @@ mod tests {
            "query".to_string(),
            "client".to_string(),
            None,
+            None,
        );
        let cancellation_handle = ticket.cancellation_handle.clone();

@@ -261,6 +264,7 @@ mod tests {
            "query".to_string(),
            "client".to_string(),
            None,
+            None,
        );
        let cancellation_handle = ticket.cancellation_handle.clone();

@@ -289,6 +293,7 @@ mod tests {
            "query".to_string(),
            "client".to_string(),
            None,
+            None,
        );

        let mut cancellable_stream = CancellableStreamWrapper::new(Box::pin(mock_stream), ticket);
@@ -319,6 +324,7 @@ mod tests {
            "query".to_string(),
            "client".to_string(),
            None,
+            None,
        );

        let cancellable_stream = CancellableStreamWrapper::new(Box::pin(mock_stream), ticket);
@@ -347,6 +353,7 @@ mod tests {
            "query".to_string(),
            "client".to_string(),
            None,
+            None,
        );
        let cancellation_handle = ticket.cancellation_handle.clone();

--- a/src/index/Cargo.toml
+++ b/src/index/Cargo.toml
@@ -42,7 +42,17 @@ uuid.workspace = true

 [dev-dependencies]
 common-test-util.workspace = true
+criterion = "0.4"
 rand.workspace = true
+rand_chacha = "0.9"
 tempfile.workspace = true
 tokio.workspace = true
 tokio-util.workspace = true
+
+[[bench]]
+name = "tokenizer_bench"
+harness = false
+
+[[bench]]
+name = "bytes_to_u64_vec"
+harness = false
--- a/src/index/benches/bytes_to_u64_vec.rs
+++ b/src/index/benches/bytes_to_u64_vec.rs
@@ -0,0 +1,99 @@
+// Copyright 2023 Greptime Team
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+use std::hint::black_box;
+
+use bytes::Bytes;
+use criterion::{criterion_group, criterion_main, BenchmarkId, Criterion, Throughput};
+use index::bloom_filter::reader::bytes_to_u64_vec;
+use rand::{Rng, SeedableRng};
+use rand_chacha::ChaCha8Rng;
+
+/// Generate test data that is guaranteed to be aligned to 8-byte boundary
+fn generate_aligned_data(size: usize) -> Bytes {
+    let mut rng = ChaCha8Rng::seed_from_u64(42);
+    let u64_count = size / 8; // Number of u64 values
+
+    // Generate random u64 values directly - this guarantees alignment
+    let mut u64_data: Vec<u64> = Vec::with_capacity(u64_count);
+    for _ in 0..u64_count {
+        u64_data.push(rng.random::<u64>());
+    }
+
+    // Transmute Vec<u64> to Vec<u8> while preserving alignment
+    let byte_vec = unsafe {
+        let ptr = u64_data.as_mut_ptr() as *mut u8;
+        let len = u64_data.len() * std::mem::size_of::<u64>();
+        let cap = u64_data.capacity() * std::mem::size_of::<u64>();
+        std::mem::forget(u64_data); // Prevent dropping the original Vec
+        Vec::from_raw_parts(ptr, len, cap)
+    };
+
+    Bytes::from(byte_vec)
+}
+
+/// Generate test data that is guaranteed to be unaligned
+fn generate_unaligned_data(size: usize) -> Bytes {
+    let mut rng = ChaCha8Rng::seed_from_u64(42);
+    let u64_count = size / 8; // Number of u64 values
+
+    // Generate random u64 values - start with aligned data
+    let mut u64_data: Vec<u64> = Vec::with_capacity(u64_count);
+    for _ in 0..u64_count {
+        u64_data.push(rng.random::<u64>());
+    }
+
+    // Transmute Vec<u64> to Vec<u8>
+    let byte_vec = unsafe {
+        let ptr = u64_data.as_mut_ptr() as *mut u8;
+        let len = u64_data.len() * std::mem::size_of::<u64>();
+        let cap = u64_data.capacity() * std::mem::size_of::<u64>();
+        std::mem::forget(u64_data); // Prevent dropping the original Vec
+        Vec::from_raw_parts(ptr, len, cap)
+    };
+
+    let unaligned_bytes = Bytes::from(byte_vec);
+    unaligned_bytes.slice(1..)
+}
+
+fn benchmark_convert(c: &mut Criterion) {
+    let sizes = vec![1024, 16384, 262144, 1048576]; // 1KB to 1MB
+
+    let mut group = c.benchmark_group("bytes_to_u64_vec");
+
+    for size in sizes {
+        let data = generate_aligned_data(size);
+        group.throughput(Throughput::Bytes(data.len() as u64));
+        group.bench_with_input(BenchmarkId::new("aligned", size), &data, |b, data| {
+            b.iter(|| {
+                let result = bytes_to_u64_vec(black_box(data));
+                black_box(result);
+            });
+        });
+
+        let data = generate_unaligned_data(size);
+        group.throughput(Throughput::Bytes(data.len() as u64));
+        group.bench_with_input(BenchmarkId::new("unaligned", size), &data, |b, data| {
+            b.iter(|| {
+                let result = bytes_to_u64_vec(black_box(data));
+                black_box(result);
+            });
+        });
+    }
+
+    group.finish();
+}
+
+criterion_group!(benches, benchmark_convert);
+criterion_main!(benches);
--- a/src/index/benches/tokenizer_bench.rs
+++ b/src/index/benches/tokenizer_bench.rs
@@ -0,0 +1,66 @@
+// Copyright 2023 Greptime Team
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+use criterion::{criterion_group, criterion_main, BenchmarkId, Criterion};
+use index::fulltext_index::tokenizer::{EnglishTokenizer, Tokenizer};
+
+fn bench_english_tokenizer(c: &mut Criterion) {
+    let tokenizer = EnglishTokenizer;
+
+    let texts = vec![
+        ("short", "Hello, world! This is a test."),
+        ("medium", "The quick brown fox jumps over the lazy dog. Lorem ipsum dolor sit amet, consectetur adipiscing elit. Sed do eiusmod tempor incididunt ut labore et dolore magna aliqua."),
+        ("long", "Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do eiusmod tempor incididunt ut labore et dolore magna aliqua. Ut enim ad minim veniam, quis nostrud exercitation ullamco laboris nisi ut aliquip ex ea commodo consequat. Duis aute irure dolor in reprehenderit in voluptate velit esse cillum dolore eu fugiat nulla pariatur. Excepteur sint occaecat cupidatat non proident, sunt in culpa qui officia deserunt mollit anim id est laborum. Sed ut perspiciatis unde omnis iste natus error sit voluptatem accusantium doloremque laudantium, totam rem aperiam, eaque ipsa quae ab illo inventore veritatis et quasi architecto beatae vitae dicta sunt explicabo. Nemo enim ipsam voluptatem quia voluptas sit aspernatur aut odit aut fugit, sed quia consequuntur magni dolores eos qui ratione voluptatem sequi nesciunt."),
+        ("mixed_chars", "Hello123 world!!! This-is_a.test@example.com with various: punctuation; and [brackets] {curly} (parentheses) & symbols* + numbers456."),
+        ("numbers_heavy", "test123 456test test789 abc123def 999888777 hello42world 123 456 789 mix1ng l3tt3rs 4nd numb3rs"),
+        ("punctuation_heavy", "Hello!!! World??? This...is...a...test... With lots of!!! punctuation??? marks!!! And... ellipses???"),
+        ("postgres log", "2025-08-01 21:09:28.928 UTC [27] LOG:  checkpoint complete: wrote 0 buffers (0.0%); 0 WAL file(s) added, 0 removed, 0 recycled; write=0.001 s, sync=0.001 s, total=0.003 s; sync files=0, longest=0.000 s, average=0.000 s; distance=0 kB, estimate=5 kB; lsn=0/1992868, redo lsn=0/1992868"),
+        ("many_short_words", "a b c d e f g h i j k l m n o p q r s t u v w x y z"),
+        ("with_unicode", "这是，一个测试。🈶一些 Unicøde 字符比如 café and naïve words."),
+    ];
+
+    let mut group = c.benchmark_group("english_tokenizer");
+
+    for (size, text) in texts {
+        group.bench_with_input(BenchmarkId::new("tokenize", size), &text, |b, text| {
+            b.iter(|| tokenizer.tokenize(text))
+        });
+    }
+
+    group.finish();
+
+    // Benchmark with repeated tokenization to simulate real-world usage
+    let mut repeat_group = c.benchmark_group("english_tokenizer_repeated");
+
+    let sample_text = "The quick brown fox jumps over the lazy dog. This sentence contains most letters of the alphabet.";
+
+    for repeat_count in [10, 100, 1000] {
+        repeat_group.bench_with_input(
+            BenchmarkId::new("repeated_tokenize", repeat_count),
+            &repeat_count,
+            |b, &repeat_count| {
+                b.iter(|| {
+                    for _ in 0..repeat_count {
+                        tokenizer.tokenize(sample_text);
+                    }
+                })
+            },
+        );
+    }
+
+    repeat_group.finish();
+}
+
+criterion_group!(benches, bench_english_tokenizer);
+criterion_main!(benches);
--- a/src/index/src/bloom_filter/reader.rs
+++ b/src/index/src/bloom_filter/reader.rs
@@ -12,9 +12,10 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.

-use std::ops::Range;
+use std::ops::{Range, Rem};

 use async_trait::async_trait;
+use bytemuck::try_cast_slice;
 use bytes::Bytes;
 use common_base::range_read::RangeReader;
 use fastbloom::BloomFilter;
@@ -33,6 +34,47 @@ const BLOOM_META_LEN_SIZE: u64 = 4;
 /// Default prefetch size of bloom filter meta.
 pub const DEFAULT_PREFETCH_SIZE: u64 = 8192; // 8KiB

+/// Safely converts bytes to Vec<u64> using bytemuck for optimal performance.
+/// Faster than chunking and converting each piece individually.
+///
+/// The input bytes are a sequence of little-endian u64s.
+pub fn bytes_to_u64_vec(bytes: &Bytes) -> Vec<u64> {
+    // drop tailing things, this keeps the same behavior with `chunks_exact`.
+    let aligned_length = bytes.len() - bytes.len().rem(std::mem::size_of::<u64>());
+    let byte_slice = &bytes[..aligned_length];
+
+    // Try fast path first: direct cast if aligned
+    let u64_vec = if let Ok(u64_slice) = try_cast_slice::<u8, u64>(byte_slice) {
+        u64_slice.to_vec()
+    } else {
+        // Slow path: create aligned Vec<u64> and copy data
+        let u64_count = byte_slice.len() / std::mem::size_of::<u64>();
+        let mut u64_vec = Vec::<u64>::with_capacity(u64_count);
+
+        // SAFETY: We're creating a properly sized slice from uninitialized but allocated memory
+        // to copy bytes into. The slice has exactly the right size for the byte data.
+        let dest_slice = unsafe {
+            std::slice::from_raw_parts_mut(u64_vec.as_mut_ptr() as *mut u8, byte_slice.len())
+        };
+        dest_slice.copy_from_slice(byte_slice);
+
+        // SAFETY: We've just initialized exactly u64_count elements worth of bytes
+        unsafe { u64_vec.set_len(u64_count) };
+        u64_vec
+    };
+
+    // Convert from platform endianness to little endian if needed
+    // Just in case.
+    #[cfg(target_endian = "little")]
+    {
+        u64_vec
+    }
+    #[cfg(target_endian = "big")]
+    {
+        u64_vec.into_iter().map(|x| x.swap_bytes()).collect()
+    }
+}
+
 /// `BloomFilterReader` reads the bloom filter from the file.
 #[async_trait]
 pub trait BloomFilterReader: Sync {
@@ -56,10 +98,7 @@ pub trait BloomFilterReader: Sync {
    /// Reads a bloom filter with the given location.
    async fn bloom_filter(&self, loc: &BloomFilterLoc) -> Result<BloomFilter> {
        let bytes = self.range_read(loc.offset, loc.size as _).await?;
-        let vec = bytes
-            .chunks_exact(std::mem::size_of::<u64>())
-            .map(|chunk| u64::from_le_bytes(chunk.try_into().unwrap()))
-            .collect();
+        let vec = bytes_to_u64_vec(&bytes);
        let bm = BloomFilter::from_vec(vec)
            .seed(&SEED)
            .expected_items(loc.element_count as _);
@@ -75,10 +114,7 @@ pub trait BloomFilterReader: Sync {

        let mut result = Vec::with_capacity(bss.len());
        for (bs, loc) in bss.into_iter().zip(locs.iter()) {
-            let vec = bs
-                .chunks_exact(std::mem::size_of::<u64>())
-                .map(|chunk| u64::from_le_bytes(chunk.try_into().unwrap()))
-                .collect();
+            let vec = bytes_to_u64_vec(&bs);
            let bm = BloomFilter::from_vec(vec)
                .seed(&SEED)
                .expected_items(loc.element_count as _);
--- a/src/index/src/fulltext_index/create/bloom_filter.rs
+++ b/src/index/src/fulltext_index/create/bloom_filter.rs
@@ -89,8 +89,12 @@ impl FulltextIndexCreator for BloomFilterFulltextIndexCreator {
        &mut self,
        puffin_writer: &mut (impl PuffinWriter + Send),
        blob_key: &str,
-        put_options: PutOptions,
+        mut put_options: PutOptions,
    ) -> Result<u64> {
+        // Compressing the bloom filter doesn't reduce the size but hurts read performance.
+        // Always disable compression here.
+        put_options.compression = None;
+
        let creator = self.inner.as_mut().context(AbortedSnafu)?;

        let (tx, rx) = tokio::io::duplex(PIPE_BUFFER_SIZE_FOR_SENDING_BLOB);
--- a/src/index/src/fulltext_index/tokenizer.rs
+++ b/src/index/src/fulltext_index/tokenizer.rs
@@ -19,6 +19,29 @@ lazy_static::lazy_static! {
    static ref JIEBA: jieba_rs::Jieba = jieba_rs::Jieba::new();
 }

+/// A-Z, a-z, 0-9, and '_' are true
+const VALID_ASCII_TOKEN: [bool; 256] = [
+    false, false, false, false, false, false, false, false, false, false, false, false, false,
+    false, false, false, false, false, false, false, false, false, false, false, false, false,
+    false, false, false, false, false, false, false, false, false, false, false, false, false,
+    false, false, false, false, false, false, false, false, false, true, true, true, true, true,
+    true, true, true, true, true, false, false, false, false, false, false, false, true, true,
+    true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true,
+    true, true, true, true, true, true, true, true, false, false, false, false, true, false, true,
+    true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true,
+    true, true, true, true, true, true, true, true, true, false, false, false, false, false, false,
+    false, false, false, false, false, false, false, false, false, false, false, false, false,
+    false, false, false, false, false, false, false, false, false, false, false, false, false,
+    false, false, false, false, false, false, false, false, false, false, false, false, false,
+    false, false, false, false, false, false, false, false, false, false, false, false, false,
+    false, false, false, false, false, false, false, false, false, false, false, false, false,
+    false, false, false, false, false, false, false, false, false, false, false, false, false,
+    false, false, false, false, false, false, false, false, false, false, false, false, false,
+    false, false, false, false, false, false, false, false, false, false, false, false, false,
+    false, false, false, false, false, false, false, false, false, false, false, false, false,
+    false, false, false, false, false, false, false, false, false, false,
+];
+
 /// `Tokenizer` tokenizes a text into a list of tokens.
 pub trait Tokenizer: Send {
    fn tokenize<'a>(&self, text: &'a str) -> Vec<&'a str>;
@@ -32,9 +55,28 @@ pub struct EnglishTokenizer;

 impl Tokenizer for EnglishTokenizer {
    fn tokenize<'a>(&self, text: &'a str) -> Vec<&'a str> {
-        text.split(|c: char| !c.is_alphanumeric())
-            .filter(|s| !s.is_empty())
-            .collect()
+        if text.is_ascii() {
+            let mut tokens = Vec::new();
+            let mut start = 0;
+            for (i, &byte) in text.as_bytes().iter().enumerate() {
+                if !VALID_ASCII_TOKEN[byte as usize] {
+                    if start < i {
+                        tokens.push(&text[start..i]);
+                    }
+                    start = i + 1;
+                }
+            }
+
+            if start < text.len() {
+                tokens.push(&text[start..]);
+            }
+
+            tokens
+        } else {
+            text.split(|c: char| !c.is_alphanumeric() && c != '_')
+                .filter(|s| !s.is_empty())
+                .collect()
+        }
    }
 }

@@ -96,9 +138,25 @@ mod tests {
    #[test]
    fn test_english_tokenizer() {
        let tokenizer = EnglishTokenizer;
-        let text = "Hello, world! This is a test0.";
+        let text = "Hello, world!!! This is a----++   test012_345+67890";
        let tokens = tokenizer.tokenize(text);
-        assert_eq!(tokens, vec!["Hello", "world", "This", "is", "a", "test0"]);
+        assert_eq!(
+            tokens,
+            vec!["Hello", "world", "This", "is", "a", "test012_345", "67890"]
+        );
+    }
+
+    #[test]
+    fn test_english_tokenizer_with_utf8() {
+        let tokenizer = EnglishTokenizer;
+        let text = "💸unfold the 纸巾😣and gently 清洁表😭面";
+        let tokens = tokenizer.tokenize(text);
+        assert_eq!(
+            tokens,
+            // Don't care what happens to non-ASCII characters.
+            // It's kind of a misconfiguration to use EnglishTokenizer on non-ASCII text.
+            vec!["unfold", "the", "纸巾", "and", "gently", "清洁表", "面"]
+        );
    }

    #[test]
@@ -109,6 +167,29 @@ mod tests {
        assert_eq!(tokens, vec!["我", "喜欢", "苹果"]);
    }

+    #[test]
+    fn test_valid_ascii_token_lookup_table() {
+        // Test all ASCII values in a single loop
+        for c in 0u8..=255u8 {
+            let is_valid = VALID_ASCII_TOKEN[c as usize];
+            let should_be_valid = (c as char).is_ascii_alphanumeric() || c == b'_';
+
+            assert_eq!(
+                is_valid,
+                should_be_valid,
+                "Character '{}' (byte {}) validity mismatch: expected {}, got {}",
+                if c.is_ascii() && !c.is_ascii_control() {
+                    c as char
+                } else {
+                    '?'
+                },
+                c,
+                should_be_valid,
+                is_valid
+            );
+        }
+    }
+
    #[test]
    fn test_analyzer() {
        let tokenizer = EnglishTokenizer;
--- a/src/meta-client/src/client.rs
+++ b/src/meta-client/src/client.rs
@@ -525,7 +525,7 @@ impl MetaClient {
        self.heartbeat_client()?.ask_leader().await
    }

-    /// Returns a heartbeat bidirectional streaming: (sender, recever), the
+    /// Returns a heartbeat bidirectional streaming: (sender, receiver), the
    /// other end is the leader of `metasrv`.
    ///
    /// The `datanode` needs to use the sender to continuously send heartbeat
--- a/src/meta-srv/Cargo.toml
+++ b/src/meta-srv/Cargo.toml
@@ -48,6 +48,7 @@ datatypes.workspace = true
 deadpool = { workspace = true, optional = true }
 deadpool-postgres = { workspace = true, optional = true }
 derive_builder.workspace = true
+either.workspace = true
 etcd-client.workspace = true
 futures.workspace = true
 h2 = "0.3"
--- a/src/meta-srv/src/bootstrap.rs
+++ b/src/meta-srv/src/bootstrap.rs
@@ -34,6 +34,7 @@ use common_meta::kv_backend::{KvBackendRef, ResettableKvBackendRef};
 use common_telemetry::info;
 #[cfg(feature = "pg_kvbackend")]
 use deadpool_postgres::{Config, Runtime};
+use either::Either;
 use etcd_client::Client;
 use servers::configurator::ConfiguratorRef;
 use servers::export_metrics::ExportMetricsTask;
@@ -77,7 +78,7 @@ use crate::{error, Result};
 pub struct MetasrvInstance {
    metasrv: Arc<Metasrv>,

-    http_server: HttpServer,
+    http_server: Either<Option<HttpServerBuilder>, HttpServer>,

    opts: MetasrvOptions,

@@ -103,11 +104,10 @@ impl MetasrvInstance {
        // Wire up the admin_axum_router as an extra router
        let extra_routers = admin_axum_router(metasrv.clone());

-        let http_server = HttpServerBuilder::new(opts.http.clone())
+        let builder = HttpServerBuilder::new(opts.http.clone())
            .with_metrics_handler(MetricsHandler)
            .with_greptime_config_options(opts.to_toml().context(error::TomlFormatSnafu)?)
-            .with_extra_router(extra_routers)
-            .build();
+            .with_extra_router(extra_routers);

        // put metasrv into plugins for later use
        plugins.insert::<Arc<Metasrv>>(metasrv.clone());
@@ -115,7 +115,7 @@ impl MetasrvInstance {
            .context(error::InitExportMetricsTaskSnafu)?;
        Ok(MetasrvInstance {
            metasrv,
-            http_server,
+            http_server: Either::Left(Some(builder)),
            opts,
            signal_sender: None,
            plugins,
@@ -126,6 +126,25 @@ impl MetasrvInstance {
    }

    pub async fn start(&mut self) -> Result<()> {
+        if let Some(builder) = self.http_server.as_mut().left()
+            && let Some(builder) = builder.take()
+        {
+            let mut server = builder.build();
+
+            let addr = self.opts.http.addr.parse().context(error::ParseAddrSnafu {
+                addr: &self.opts.http.addr,
+            })?;
+            info!("starting http server at {}", addr);
+            server.start(addr).await.context(error::StartHttpSnafu)?;
+
+            self.http_server = Either::Right(server);
+        } else {
+            // If the http server builder is not present, the Metasrv has to be called "start"
+            // already, regardless of the startup was successful or not. Return an `Ok` here for
+            // simplicity.
+            return Ok(());
+        };
+
        self.metasrv.try_start().await?;

        if let Some(t) = self.export_metrics_task.as_ref() {
@@ -149,14 +168,6 @@ impl MetasrvInstance {
                .await?;
        self.bind_addr = Some(socket_addr);

-        let addr = self.opts.http.addr.parse().context(error::ParseAddrSnafu {
-            addr: &self.opts.http.addr,
-        })?;
-        self.http_server
-            .start(addr)
-            .await
-            .context(error::StartHttpSnafu)?;
-
        *self.serve_state.lock().await = Some(serve_state_rx);
        Ok(())
    }
@@ -174,12 +185,15 @@ impl MetasrvInstance {
                .context(error::SendShutdownSignalSnafu)?;
        }
        self.metasrv.shutdown().await?;
-        self.http_server
-            .shutdown()
-            .await
-            .context(error::ShutdownServerSnafu {
-                server: self.http_server.name(),
-            })?;
+
+        if let Some(http_server) = self.http_server.as_ref().right() {
+            http_server
+                .shutdown()
+                .await
+                .context(error::ShutdownServerSnafu {
+                    server: http_server.name(),
+                })?;
+        }
        Ok(())
    }

@@ -193,6 +207,14 @@ impl MetasrvInstance {
    pub fn bind_addr(&self) -> &Option<SocketAddr> {
        &self.bind_addr
    }
+
+    pub fn mut_http_server(&mut self) -> &mut Either<Option<HttpServerBuilder>, HttpServer> {
+        &mut self.http_server
+    }
+
+    pub fn http_server(&self) -> Option<&HttpServer> {
+        self.http_server.as_ref().right()
+    }
 }

 pub async fn bootstrap_metasrv_with_router(
--- a/src/meta-srv/src/election/rds/mysql.rs
+++ b/src/meta-srv/src/election/rds/mysql.rs
@@ -17,7 +17,7 @@ use std::sync::Arc;
 use std::time::Duration;

 use common_meta::key::{CANDIDATES_ROOT, ELECTION_KEY};
-use common_telemetry::{error, warn};
+use common_telemetry::{error, info, warn};
 use common_time::Timestamp;
 use snafu::{ensure, OptionExt, ResultExt};
 use sqlx::mysql::{MySqlArguments, MySqlRow};
@@ -149,7 +149,7 @@ impl<'a> ElectionSqlFactory<'a> {
    }

    fn delete_value_sql(&self) -> String {
-        format!("DELETE FROM {} WHERE k = ?;", self.table_name)
+        format!("DELETE FROM `{}` WHERE k = ?;", self.table_name)
    }
 }

@@ -645,6 +645,13 @@ impl Election for MySqlElection {
    }

    async fn reset_campaign(&self) {
+        info!("Resetting campaign");
+        if self.is_leader.load(Ordering::Relaxed) {
+            if let Err(err) = self.step_down_without_lock().await {
+                error!(err; "Failed to step down without lock");
+            }
+            info!("Step down without lock successfully, due to reset campaign");
+        }
        if let Err(err) = self.client.lock().await.reset_client().await {
            error!(err; "Failed to reset client");
        }
@@ -1007,7 +1014,7 @@ mod tests {
            execution_timeout,
            Duration::from_secs(1),
            wait_timeout,
-            table_name.unwrap_or("default_greptime_metakv_election"),
+            table_name.unwrap_or("default_greptime_metakv-election"),
        );
        client.maybe_init_client().await?;
        if table_name.is_some() {
@@ -1018,7 +1025,7 @@ mod tests {

    async fn drop_table(client: &Mutex<ElectionMysqlClient>, table_name: &str) {
        let mut client = client.lock().await;
-        let sql = format!("DROP TABLE IF EXISTS {};", table_name);
+        let sql = format!("DROP TABLE IF EXISTS `{}`;", table_name);
        client.execute(sqlx::query(&sql), &sql).await.unwrap();
    }

@@ -1029,7 +1036,7 @@ mod tests {
        let value = "test_value".to_string();

        let uuid = uuid::Uuid::new_v4().to_string();
-        let table_name = "test_mysql_crud_greptime_metakv";
+        let table_name = "test_mysql_crud_greptime-metakv";
        let candidate_lease_ttl = Duration::from_secs(10);
        let meta_lease_ttl = Duration::from_secs(2);

@@ -1043,7 +1050,7 @@ mod tests {
            let mut a = client.lock().await;
            let txn = a.transaction().await.unwrap();
            let mut executor = Executor::Txn(txn);
-            let raw_query = format!("SELECT * FROM {} FOR UPDATE;", table_name);
+            let raw_query = format!("SELECT * FROM `{}` FOR UPDATE;", table_name);
            let query = sqlx::query(&raw_query);
            let _ = executor.query(query, &raw_query).await.unwrap();
        }
@@ -1174,7 +1181,7 @@ mod tests {
        let meta_lease_ttl = Duration::from_secs(2);
        let idle_session_timeout = Duration::from_secs(0);
        let uuid = uuid::Uuid::new_v4().to_string();
-        let table_name = "test_candidate_registration_greptime_metakv";
+        let table_name = "test_candidate_registration_greptime-metakv";
        let mut handles = vec![];
        let client = create_mysql_client(Some(table_name), execution_timeout, idle_session_timeout)
            .await
@@ -1190,7 +1197,7 @@ mod tests {
            ));
            handles.push(handle);
        }
-        // Wait for candidates to registrate themselves and renew their leases at least once.
+        // Wait for candidates to register themselves and renew their leases at least once.
        tokio::time::sleep(candidate_lease_ttl / 2 + Duration::from_secs(1)).await;

        let (tx, _) = broadcast::channel(100);
@@ -1244,7 +1251,7 @@ mod tests {
        let mut client = election.client.lock().await;
        let txn = client.transaction().await.unwrap();
        let mut executor = Executor::Txn(txn);
-        let raw_query = format!("SELECT * FROM {} FOR UPDATE;", table_name);
+        let raw_query = format!("SELECT * FROM `{}` FOR UPDATE;", table_name);
        let query = sqlx::query(&raw_query);
        let _ = executor.query(query, &raw_query).await.unwrap();
        election.elected(executor, expected_lease).await
@@ -1268,7 +1275,7 @@ mod tests {
        let execution_timeout = Duration::from_secs(10);
        let idle_session_timeout = Duration::from_secs(0);
        let uuid = uuid::Uuid::new_v4().to_string();
-        let table_name = "test_elected_failed_greptime_metakv";
+        let table_name = "test_elected_failed_greptime-metakv";
        let client = create_mysql_client(Some(table_name), execution_timeout, idle_session_timeout)
            .await
            .unwrap();
@@ -1301,7 +1308,7 @@ mod tests {
        maybe_skip_mysql_integration_test!();
        let leader_value = "test_leader".to_string();
        let uuid = uuid::Uuid::new_v4().to_string();
-        let table_name = "test_reelection_greptime_metakv";
+        let table_name = "test_reelection_greptime-metakv";
        let candidate_lease_ttl = Duration::from_secs(5);
        let meta_lease_ttl = Duration::from_secs(5);
        let execution_timeout = Duration::from_secs(10);
@@ -1369,7 +1376,7 @@ mod tests {
        let execution_timeout = Duration::from_secs(10);
        let idle_session_timeout = Duration::from_secs(0);
        let uuid = uuid::Uuid::new_v4().to_string();
-        let table_name = "test_elected_and_step_down_greptime_metakv";
+        let table_name = "test_elected_and_step_down_greptime-metakv";
        let client = create_mysql_client(Some(table_name), execution_timeout, idle_session_timeout)
            .await
            .unwrap();
@@ -1458,7 +1465,7 @@ mod tests {
        maybe_skip_mysql_integration_test!();
        let leader_value = "test_leader".to_string();
        let uuid = uuid::Uuid::new_v4().to_string();
-        let table_name = "test_leader_action_greptime_metakv";
+        let table_name = "test_leader_action_greptime-metakv";
        let candidate_lease_ttl = Duration::from_secs(5);
        let meta_lease_ttl = Duration::from_secs(2);
        let execution_timeout = Duration::from_secs(10);
@@ -1639,6 +1646,41 @@ mod tests {
        drop_table(&leader_mysql_election.client, table_name).await;
    }

+    #[tokio::test]
+    async fn test_reset_campaign() {
+        maybe_skip_mysql_integration_test!();
+        common_telemetry::init_default_ut_logging();
+        let leader_value = "test_leader".to_string();
+        let uuid = uuid::Uuid::new_v4().to_string();
+        let table_name = "test_reset_campaign_greptime-metakv";
+        let candidate_lease_ttl = Duration::from_secs(5);
+        let meta_lease_ttl = Duration::from_secs(2);
+        let execution_timeout = Duration::from_secs(10);
+        let idle_session_timeout = Duration::from_secs(0);
+        let client = create_mysql_client(Some(table_name), execution_timeout, idle_session_timeout)
+            .await
+            .unwrap();
+
+        let (tx, _) = broadcast::channel(100);
+        let leader_mysql_election = MySqlElection {
+            leader_value,
+            client,
+            is_leader: AtomicBool::new(false),
+            leader_infancy: AtomicBool::new(true),
+            leader_watcher: tx,
+            store_key_prefix: uuid,
+            candidate_lease_ttl,
+            meta_lease_ttl,
+            sql_set: ElectionSqlFactory::new(table_name).build(),
+        };
+        leader_mysql_election
+            .is_leader
+            .store(true, Ordering::Relaxed);
+        leader_mysql_election.reset_campaign().await;
+        assert!(!leader_mysql_election.is_leader());
+        drop_table(&leader_mysql_election.client, table_name).await;
+    }
+
    #[tokio::test]
    async fn test_follower_action() {
        maybe_skip_mysql_integration_test!();
@@ -1648,7 +1690,7 @@ mod tests {
        let execution_timeout = Duration::from_secs(10);
        let idle_session_timeout = Duration::from_secs(0);
        let uuid = uuid::Uuid::new_v4().to_string();
-        let table_name = "test_follower_action_greptime_metakv";
+        let table_name = "test_follower_action_greptime-metakv";

        let follower_client =
            create_mysql_client(Some(table_name), execution_timeout, idle_session_timeout)
--- a/src/meta-srv/src/election/rds/postgres.rs
+++ b/src/meta-srv/src/election/rds/postgres.rs
@@ -17,7 +17,7 @@ use std::sync::Arc;
 use std::time::Duration;

 use common_meta::key::{CANDIDATES_ROOT, ELECTION_KEY};
-use common_telemetry::{error, warn};
+use common_telemetry::{error, info, warn};
 use common_time::Timestamp;
 use deadpool_postgres::{Manager, Pool};
 use snafu::{ensure, OptionExt, ResultExt};
@@ -454,6 +454,13 @@ impl Election for PgElection {
    }

    async fn reset_campaign(&self) {
+        info!("Resetting campaign");
+        if self.is_leader.load(Ordering::Relaxed) {
+            if let Err(err) = self.step_down_without_lock().await {
+                error!(err; "Failed to step down without lock");
+            }
+            info!("Step down without lock successfully, due to reset campaign");
+        }
        if let Err(err) = self.pg_client.write().await.reset_client().await {
            error!(err; "Failed to reset client");
        }
@@ -749,18 +756,12 @@ impl PgElection {
            key: key.clone(),
            ..Default::default()
        };
-        if self
-            .is_leader
-            .compare_exchange(true, false, Ordering::AcqRel, Ordering::Acquire)
-            .is_ok()
-        {
-            if let Err(e) = self
-                .leader_watcher
-                .send(LeaderChangeMessage::StepDown(Arc::new(leader_key)))
-            {
-                error!(e; "Failed to send leader change message");
-            }
-        }
+        send_leader_change_and_set_flags(
+            &self.is_leader,
+            &self.leader_infancy,
+            &self.leader_watcher,
+            LeaderChangeMessage::StepDown(Arc::new(leader_key)),
+        );
        Ok(())
    }

@@ -1012,7 +1013,7 @@ mod tests {
            ));
            handles.push(handle);
        }
-        // Wait for candidates to registrate themselves and renew their leases at least once.
+        // Wait for candidates to register themselves and renew their leases at least once.
        tokio::time::sleep(Duration::from_secs(3)).await;

        let (tx, _) = broadcast::channel(100);
@@ -1551,6 +1552,44 @@ mod tests {
        drop_table(&follower_pg_election, table_name).await;
    }

+    #[tokio::test]
+    async fn test_reset_campaign() {
+        maybe_skip_postgres_integration_test!();
+        let leader_value = "test_leader".to_string();
+        let uuid = uuid::Uuid::new_v4().to_string();
+        let table_name = "test_reset_campaign_greptime_metakv";
+        let candidate_lease_ttl = Duration::from_secs(5);
+        let execution_timeout = Duration::from_secs(10);
+        let statement_timeout = Duration::from_secs(10);
+        let meta_lease_ttl = Duration::from_secs(2);
+        let idle_session_timeout = Duration::from_secs(0);
+        let client = create_postgres_client(
+            Some(table_name),
+            execution_timeout,
+            idle_session_timeout,
+            statement_timeout,
+        )
+        .await
+        .unwrap();
+
+        let (tx, _) = broadcast::channel(100);
+        let leader_pg_election = PgElection {
+            leader_value,
+            pg_client: RwLock::new(client),
+            is_leader: AtomicBool::new(false),
+            leader_infancy: AtomicBool::new(true),
+            leader_watcher: tx,
+            store_key_prefix: uuid,
+            candidate_lease_ttl,
+            meta_lease_ttl,
+            sql_set: ElectionSqlFactory::new(28321, table_name).build(),
+        };
+        leader_pg_election.is_leader.store(true, Ordering::Relaxed);
+        leader_pg_election.reset_campaign().await;
+        assert!(!leader_pg_election.is_leader());
+        drop_table(&leader_pg_election, table_name).await;
+    }
+
    #[tokio::test]
    async fn test_idle_session_timeout() {
        maybe_skip_postgres_integration_test!();
--- a/src/meta-srv/src/handler/collect_leader_region_handler.rs
+++ b/src/meta-srv/src/handler/collect_leader_region_handler.rs
@@ -124,6 +124,7 @@ mod tests {
            index_size: 0,
            data_topic_latest_entry_id: 0,
            metadata_topic_latest_entry_id: 0,
+            written_bytes: 0,
        }
    }

--- a/src/meta-srv/src/handler/failure_handler.rs
+++ b/src/meta-srv/src/handler/failure_handler.rs
@@ -104,6 +104,7 @@ mod tests {
                },
                data_topic_latest_entry_id: 0,
                metadata_topic_latest_entry_id: 0,
+                written_bytes: 0,
            }
        }
        acc.stat = Some(Stat {
--- a/src/meta-srv/src/handler/region_lease_handler.rs
+++ b/src/meta-srv/src/handler/region_lease_handler.rs
@@ -19,6 +19,7 @@ use api::v1::meta::{HeartbeatRequest, RegionLease, Role};
 use async_trait::async_trait;
 use common_meta::key::TableMetadataManagerRef;
 use common_meta::region_keeper::MemoryRegionKeeperRef;
+use common_telemetry::error;
 use store_api::region_engine::GrantedRegion;
 use store_api::storage::RegionId;

@@ -83,36 +84,44 @@ impl HeartbeatHandler for RegionLeaseHandler {
        let regions = stat.regions();
        let datanode_id = stat.id;

-        let RenewRegionLeasesResponse {
-            non_exists,
-            renewed,
-        } = self
+        match self
            .region_lease_keeper
            .renew_region_leases(datanode_id, &regions)
-            .await?;
+            .await
+        {
+            Ok(RenewRegionLeasesResponse {
+                non_exists,
+                renewed,
+            }) => {
+                let renewed = if let Some(renewer) = &self.customized_region_lease_renewer {
+                    renewer
+                        .renew(ctx, renewed)
+                        .into_iter()
+                        .map(|region| region.into())
+                        .collect()
+                } else {
+                    renewed
+                        .into_iter()
+                        .map(|(region_id, region_lease_info)| {
+                            GrantedRegion::new(region_id, region_lease_info.role).into()
+                        })
+                        .collect::<Vec<_>>()
+                };

-        let renewed = if let Some(renewer) = &self.customized_region_lease_renewer {
-            renewer
-                .renew(ctx, renewed)
-                .into_iter()
-                .map(|region| region.into())
-                .collect()
-        } else {
-            renewed
-                .into_iter()
-                .map(|(region_id, region_lease_info)| {
-                    GrantedRegion::new(region_id, region_lease_info.role).into()
-                })
-                .collect::<Vec<_>>()
-        };
-
-        acc.region_lease = Some(RegionLease {
-            regions: renewed,
-            duration_since_epoch: req.duration_since_epoch,
-            lease_seconds: self.region_lease_seconds,
-            closeable_region_ids: non_exists.iter().map(|region| region.as_u64()).collect(),
-        });
-        acc.inactive_region_ids = non_exists;
+                acc.region_lease = Some(RegionLease {
+                    regions: renewed,
+                    duration_since_epoch: req.duration_since_epoch,
+                    lease_seconds: self.region_lease_seconds,
+                    closeable_region_ids: non_exists.iter().map(|region| region.as_u64()).collect(),
+                });
+                acc.inactive_region_ids = non_exists;
+            }
+            Err(e) => {
+                error!(e; "Failed to renew region leases for datanode: {datanode_id:?}, regions: {:?}", regions);
+                // If we throw error here, the datanode will be marked as failure by region failure handler.
+                // So we only log the error and continue.
+            }
+        }

        Ok(HandleControl::Continue)
    }
@@ -120,18 +129,27 @@ impl HeartbeatHandler for RegionLeaseHandler {

 #[cfg(test)]
 mod test {
+    use std::any::Any;
    use std::collections::{HashMap, HashSet};
    use std::sync::Arc;

    use common_meta::datanode::{RegionManifestInfo, RegionStat, Stat};
    use common_meta::distributed_time_constants;
+    use common_meta::error::Result as MetaResult;
    use common_meta::key::table_route::TableRouteValue;
    use common_meta::key::test_utils::new_test_table_info;
    use common_meta::key::TableMetadataManager;
    use common_meta::kv_backend::memory::MemoryKvBackend;
+    use common_meta::kv_backend::txn::{Txn, TxnResponse};
+    use common_meta::kv_backend::{KvBackend, TxnService};
    use common_meta::peer::Peer;
    use common_meta::region_keeper::MemoryRegionKeeper;
    use common_meta::rpc::router::{LeaderState, Region, RegionRoute};
+    use common_meta::rpc::store::{
+        BatchDeleteRequest, BatchDeleteResponse, BatchGetRequest, BatchGetResponse,
+        BatchPutRequest, BatchPutResponse, DeleteRangeRequest, DeleteRangeResponse, PutRequest,
+        PutResponse, RangeRequest, RangeResponse,
+    };
    use store_api::region_engine::RegionRole;
    use store_api::storage::RegionId;

@@ -166,6 +184,7 @@ mod test {
            },
            data_topic_latest_entry_id: 0,
            metadata_topic_latest_entry_id: 0,
+            written_bytes: 0,
        }
    }

@@ -403,4 +422,102 @@ mod test {

        assert_eq!(granted, expected);
    }
+
+    struct MockKvBackend;
+
+    #[async_trait::async_trait]
+    impl TxnService for MockKvBackend {
+        type Error = common_meta::error::Error;
+
+        async fn txn(&self, _txn: Txn) -> MetaResult<TxnResponse> {
+            unimplemented!()
+        }
+
+        fn max_txn_ops(&self) -> usize {
+            unimplemented!()
+        }
+    }
+
+    #[async_trait::async_trait]
+    impl KvBackend for MockKvBackend {
+        fn name(&self) -> &str {
+            "mock_kv_backend"
+        }
+
+        fn as_any(&self) -> &dyn Any {
+            self
+        }
+
+        async fn range(&self, _req: RangeRequest) -> MetaResult<RangeResponse> {
+            unimplemented!()
+        }
+
+        async fn put(&self, _req: PutRequest) -> MetaResult<PutResponse> {
+            unimplemented!()
+        }
+
+        async fn batch_put(&self, _req: BatchPutRequest) -> MetaResult<BatchPutResponse> {
+            unimplemented!()
+        }
+
+        async fn batch_get(&self, _req: BatchGetRequest) -> MetaResult<BatchGetResponse> {
+            common_meta::error::UnexpectedSnafu {
+                err_msg: "mock err",
+            }
+            .fail()
+        }
+
+        async fn delete_range(&self, _req: DeleteRangeRequest) -> MetaResult<DeleteRangeResponse> {
+            unimplemented!()
+        }
+
+        async fn batch_delete(&self, _req: BatchDeleteRequest) -> MetaResult<BatchDeleteResponse> {
+            unimplemented!()
+        }
+    }
+
+    #[tokio::test]
+    async fn test_handle_renew_region_lease_failure() {
+        common_telemetry::init_default_ut_logging();
+        let kvbackend = Arc::new(MockKvBackend);
+        let table_metadata_manager = Arc::new(TableMetadataManager::new(kvbackend));
+
+        let datanode_id = 1;
+        let region_number = 1u32;
+        let table_id = 10;
+        let region_id = RegionId::new(table_id, region_number);
+        let another_region_id = RegionId::new(table_id, region_number + 1);
+        let no_exist_region_id = RegionId::new(table_id, region_number + 2);
+        let peer = Peer::empty(datanode_id);
+
+        let builder = MetasrvBuilder::new();
+        let metasrv = builder.build().await.unwrap();
+        let ctx = &mut metasrv.new_ctx();
+
+        let req = HeartbeatRequest {
+            duration_since_epoch: 1234,
+            ..Default::default()
+        };
+
+        let acc = &mut HeartbeatAccumulator::default();
+        acc.stat = Some(Stat {
+            id: peer.id,
+            region_stats: vec![
+                new_empty_region_stat(region_id, RegionRole::Leader),
+                new_empty_region_stat(another_region_id, RegionRole::Leader),
+                new_empty_region_stat(no_exist_region_id, RegionRole::Leader),
+            ],
+            ..Default::default()
+        });
+        let handler = RegionLeaseHandler::new(
+            distributed_time_constants::REGION_LEASE_SECS,
+            table_metadata_manager.clone(),
+            Default::default(),
+            None,
+        );
+        handler.handle(&req, ctx, acc).await.unwrap();
+
+        assert!(acc.region_lease.is_none());
+        assert!(acc.inactive_region_ids.is_empty());
+    }
 }
--- a/src/meta-srv/src/lib.rs
+++ b/src/meta-srv/src/lib.rs
@@ -15,6 +15,7 @@
 #![feature(result_flattening)]
 #![feature(assert_matches)]
 #![feature(hash_set_entry)]
+#![feature(let_chains)]

 pub mod bootstrap;
 pub mod cache_invalidator;
--- a/src/meta-srv/src/metasrv/builder.rs
+++ b/src/meta-srv/src/metasrv/builder.rs
@@ -345,7 +345,8 @@ impl MetasrvBuilder {
                region_migration_manager.clone(),
                runtime_switch_manager.clone(),
                peer_lookup_service.clone(),
-            );
+            )
+            .with_state(state.clone());

            Some(RegionFailureHandler::new(
                region_supervisor,
--- a/src/meta-srv/src/procedure/region_migration.rs
+++ b/src/meta-srv/src/procedure/region_migration.rs
@@ -38,7 +38,7 @@ use common_meta::key::table_info::TableInfoValue;
 use common_meta::key::table_route::TableRouteValue;
 use common_meta::key::{DeserializedValueWithBytes, TableMetadataManagerRef};
 use common_meta::kv_backend::ResettableKvBackendRef;
-use common_meta::lock_key::{CatalogLock, RegionLock, SchemaLock};
+use common_meta::lock_key::{CatalogLock, RegionLock, SchemaLock, TableLock};
 use common_meta::peer::Peer;
 use common_meta::region_keeper::{MemoryRegionKeeperRef, OperatingRegionGuard};
 use common_procedure::error::{
@@ -215,8 +215,6 @@ pub struct VolatileContext {
    /// the corresponding [RegionRoute](common_meta::rpc::router::RegionRoute) of the opening region
    /// was written into [TableRouteValue](common_meta::key::table_route::TableRouteValue).
    opening_region_guard: Option<OperatingRegionGuard>,
-    /// `table_route` is stored via previous steps for future use.
-    table_route: Option<DeserializedValueWithBytes<TableRouteValue>>,
    /// `datanode_table` is stored via previous steps for future use.
    from_peer_datanode_table: Option<DatanodeTableValue>,
    /// `table_info` is stored via previous steps for future use.
@@ -383,29 +381,23 @@ impl Context {
    /// Retry:
    /// - Failed to retrieve the metadata of table.
    pub async fn get_table_route_value(
-        &mut self,
-    ) -> Result<&DeserializedValueWithBytes<TableRouteValue>> {
-        let table_route_value = &mut self.volatile_ctx.table_route;
+        &self,
+    ) -> Result<DeserializedValueWithBytes<TableRouteValue>> {
+        let table_id = self.persistent_ctx.region_id.table_id();
+        let table_route = self
+            .table_metadata_manager
+            .table_route_manager()
+            .table_route_storage()
+            .get_with_raw_bytes(table_id)
+            .await
+            .context(error::TableMetadataManagerSnafu)
+            .map_err(BoxedError::new)
+            .with_context(|_| error::RetryLaterWithSourceSnafu {
+                reason: format!("Failed to get TableRoute: {table_id}"),
+            })?
+            .context(error::TableRouteNotFoundSnafu { table_id })?;

-        if table_route_value.is_none() {
-            let table_id = self.persistent_ctx.region_id.table_id();
-            let table_route = self
-                .table_metadata_manager
-                .table_route_manager()
-                .table_route_storage()
-                .get_with_raw_bytes(table_id)
-                .await
-                .context(error::TableMetadataManagerSnafu)
-                .map_err(BoxedError::new)
-                .with_context(|_| error::RetryLaterWithSourceSnafu {
-                    reason: format!("Failed to get TableRoute: {table_id}"),
-                })?
-                .context(error::TableRouteNotFoundSnafu { table_id })?;
-
-            *table_route_value = Some(table_route);
-        }
-
-        Ok(table_route_value.as_ref().unwrap())
+        Ok(table_route)
    }

    /// Notifies the RegionSupervisor to register failure detectors of failed region.
@@ -447,12 +439,6 @@ impl Context {
            .await;
    }

-    /// Removes the `table_route` of [VolatileContext], returns true if any.
-    pub fn remove_table_route_value(&mut self) -> bool {
-        let value = self.volatile_ctx.table_route.take();
-        value.is_some()
-    }
-
    /// Returns the `table_info` of [VolatileContext] if any.
    /// Otherwise, returns the value retrieved from remote.
    ///
@@ -627,14 +613,13 @@ impl RegionMigrationProcedure {
        })
    }

-    async fn rollback_inner(&mut self) -> Result<()> {
+    async fn rollback_inner(&mut self, procedure_ctx: &ProcedureContext) -> Result<()> {
        let _timer = METRIC_META_REGION_MIGRATION_EXECUTE
            .with_label_values(&["rollback"])
            .start_timer();

        let table_id = self.context.region_id().table_id();
        let region_id = self.context.region_id();
-        self.context.remove_table_route_value();
        let table_metadata_manager = self.context.table_metadata_manager.clone();
        let table_route = self.context.get_table_route_value().await?;

@@ -647,9 +632,11 @@ impl RegionMigrationProcedure {
            .any(|route| route.is_leader_downgrading());

        if downgraded {
+            let table_lock = TableLock::Write(region_id.table_id()).into();
+            let _guard = procedure_ctx.provider.acquire_lock(&table_lock).await;
            info!("Rollbacking downgraded region leader table route, region: {region_id}");
            table_metadata_manager
-                    .update_leader_region_status(table_id, table_route, |route| {
+                    .update_leader_region_status(table_id, &table_route, |route| {
                        if route.region.id == region_id {
                            Some(None)
                        } else {
@@ -676,8 +663,8 @@ impl Procedure for RegionMigrationProcedure {
        Self::TYPE_NAME
    }

-    async fn rollback(&mut self, _ctx: &ProcedureContext) -> ProcedureResult<()> {
-        self.rollback_inner()
+    async fn rollback(&mut self, ctx: &ProcedureContext) -> ProcedureResult<()> {
+        self.rollback_inner(ctx)
            .await
            .map_err(ProcedureError::external)
    }
--- a/src/meta-srv/src/procedure/region_migration/update_metadata/downgrade_leader_region.rs
+++ b/src/meta-srv/src/procedure/region_migration/update_metadata/downgrade_leader_region.rs
@@ -46,7 +46,7 @@ impl UpdateMetadata {

        // TODO(weny): ensures the leader region peer is the `from_peer`.
        if let Err(err) = table_metadata_manager
-            .update_leader_region_status(table_id, current_table_route_value, |route| {
+            .update_leader_region_status(table_id, &current_table_route_value, |route| {
                if route.region.id == region_id
                    && route
                        .leader_peer
@@ -61,7 +61,6 @@ impl UpdateMetadata {
            .await
            .context(error::TableMetadataManagerSnafu)
        {
-            ctx.remove_table_route_value();
            return Err(BoxedError::new(err)).context(error::RetryLaterWithSourceSnafu {
                reason: format!(
                    "Failed to update the table route during the downgrading leader region, region_id: {region_id}, from_peer_id: {from_peer_id}"
@@ -69,8 +68,6 @@ impl UpdateMetadata {
            });
        }

-        ctx.remove_table_route_value();
-
        Ok(())
    }
 }
@@ -81,7 +78,7 @@ mod tests {

    use common_meta::key::test_utils::new_test_table_info;
    use common_meta::peer::Peer;
-    use common_meta::rpc::router::{LeaderState, Region, RegionRoute};
+    use common_meta::rpc::router::{Region, RegionRoute};
    use store_api::storage::RegionId;

    use crate::error::Error;
@@ -115,63 +112,6 @@ mod tests {
        assert!(!err.is_retryable());
    }

-    #[tokio::test]
-    async fn test_failed_to_update_table_route_error() {
-        let state = UpdateMetadata::Downgrade;
-        let persistent_context = new_persistent_context();
-        let from_peer = persistent_context.from_peer.clone();
-
-        let env = TestingEnv::new();
-        let mut ctx = env.context_factory().new_context(persistent_context);
-        let table_id = ctx.region_id().table_id();
-
-        let table_info = new_test_table_info(1024, vec![1, 2]).into();
-        let region_routes = vec![
-            RegionRoute {
-                region: Region::new_test(RegionId::new(1024, 1)),
-                leader_peer: Some(from_peer.clone()),
-                ..Default::default()
-            },
-            RegionRoute {
-                region: Region::new_test(RegionId::new(1024, 2)),
-                leader_peer: Some(Peer::empty(4)),
-                ..Default::default()
-            },
-        ];
-
-        env.create_physical_table_metadata(table_info, region_routes)
-            .await;
-
-        let table_metadata_manager = env.table_metadata_manager();
-        let original_table_route = table_metadata_manager
-            .table_route_manager()
-            .table_route_storage()
-            .get_with_raw_bytes(table_id)
-            .await
-            .unwrap()
-            .unwrap();
-
-        // modifies the table route.
-        table_metadata_manager
-            .update_leader_region_status(table_id, &original_table_route, |route| {
-                if route.region.id == RegionId::new(1024, 2) {
-                    Some(Some(LeaderState::Downgrading))
-                } else {
-                    None
-                }
-            })
-            .await
-            .unwrap();
-
-        // sets the old table route.
-        ctx.volatile_ctx.table_route = Some(original_table_route);
-
-        let err = state.downgrade_leader_region(&mut ctx).await.unwrap_err();
-        assert!(ctx.volatile_ctx.table_route.is_none());
-        assert!(err.is_retryable());
-        assert!(format!("{err:?}").contains("Failed to update the table route"));
-    }
-
    #[tokio::test]
    async fn test_only_downgrade_from_peer() {
        let mut state = Box::new(UpdateMetadata::Downgrade);
@@ -212,7 +152,6 @@ mod tests {
        // It should remain unchanged.
        assert_eq!(latest_table_route.version().unwrap(), 0);
        assert!(!latest_table_route.region_routes().unwrap()[0].is_leader_downgrading());
-        assert!(ctx.volatile_ctx.table_route.is_none());
    }

    #[tokio::test]
@@ -254,6 +193,5 @@ mod tests {
            .unwrap();

        assert!(latest_table_route.region_routes().unwrap()[0].is_leader_downgrading());
-        assert!(ctx.volatile_ctx.table_route.is_none());
    }
 }
--- a/src/meta-srv/src/procedure/region_migration/update_metadata/rollback_downgraded_region.rs
+++ b/src/meta-srv/src/procedure/region_migration/update_metadata/rollback_downgraded_region.rs
@@ -35,7 +35,7 @@ impl UpdateMetadata {
        let current_table_route_value = ctx.get_table_route_value().await?;

        if let Err(err) = table_metadata_manager
-            .update_leader_region_status(table_id, current_table_route_value, |route| {
+            .update_leader_region_status(table_id, &current_table_route_value, |route| {
                if route.region.id == region_id {
                    Some(None)
                } else {
@@ -45,14 +45,12 @@ impl UpdateMetadata {
            .await
            .context(error::TableMetadataManagerSnafu)
        {
-            ctx.remove_table_route_value();
            return Err(BoxedError::new(err)).context(error::RetryLaterWithSourceSnafu {
                reason: format!("Failed to update the table route during the rollback downgraded leader region: {region_id}"),
            });
        }

        ctx.register_failure_detectors().await;
-        ctx.remove_table_route_value();

        Ok(())
    }
@@ -61,7 +59,6 @@ impl UpdateMetadata {
 #[cfg(test)]
 mod tests {
    use std::assert_matches::assert_matches;
-    use std::sync::Arc;

    use common_meta::key::test_utils::new_test_table_info;
    use common_meta::peer::Peer;
@@ -73,7 +70,6 @@ mod tests {
    use crate::procedure::region_migration::test_util::{self, new_procedure_context, TestingEnv};
    use crate::procedure::region_migration::update_metadata::UpdateMetadata;
    use crate::procedure::region_migration::{ContextFactory, PersistentContext, State};
-    use crate::region::supervisor::RegionFailureDetectorControl;

    fn new_persistent_context() -> PersistentContext {
        test_util::new_persistent_context(1, 2, RegionId::new(1024, 1))
@@ -93,101 +89,6 @@ mod tests {
        assert!(!err.is_retryable());
    }

-    #[tokio::test]
-    async fn test_update_table_route_with_retry() {
-        let state = UpdateMetadata::Rollback;
-        let persistent_context = new_persistent_context();
-        let from_peer = persistent_context.from_peer.clone();
-
-        let env = TestingEnv::new();
-        let mut ctx = env.context_factory().new_context(persistent_context);
-        let (tx, mut rx) = tokio::sync::mpsc::channel(8);
-        ctx.region_failure_detector_controller = Arc::new(RegionFailureDetectorControl::new(tx));
-        let table_id = ctx.region_id().table_id();
-
-        let table_info = new_test_table_info(1024, vec![1, 2, 3]).into();
-        let region_routes = vec![
-            RegionRoute {
-                region: Region::new_test(RegionId::new(1024, 1)),
-                leader_peer: Some(from_peer.clone()),
-                leader_state: Some(LeaderState::Downgrading),
-                ..Default::default()
-            },
-            RegionRoute {
-                region: Region::new_test(RegionId::new(1024, 2)),
-                leader_peer: Some(Peer::empty(4)),
-                leader_state: Some(LeaderState::Downgrading),
-                ..Default::default()
-            },
-            RegionRoute {
-                region: Region::new_test(RegionId::new(1024, 3)),
-                leader_peer: Some(Peer::empty(5)),
-                ..Default::default()
-            },
-        ];
-
-        let expected_region_routes = {
-            let mut region_routes = region_routes.clone();
-            region_routes[0].leader_state = None;
-            region_routes[1].leader_state = None;
-            region_routes
-        };
-
-        env.create_physical_table_metadata(table_info, region_routes)
-            .await;
-
-        let table_metadata_manager = env.table_metadata_manager();
-        let old_table_route = table_metadata_manager
-            .table_route_manager()
-            .table_route_storage()
-            .get_with_raw_bytes(table_id)
-            .await
-            .unwrap()
-            .unwrap();
-
-        // modifies the table route.
-        table_metadata_manager
-            .update_leader_region_status(table_id, &old_table_route, |route| {
-                if route.region.id == RegionId::new(1024, 2) {
-                    Some(None)
-                } else {
-                    None
-                }
-            })
-            .await
-            .unwrap();
-
-        ctx.volatile_ctx.table_route = Some(old_table_route);
-
-        let err = state
-            .rollback_downgraded_region(&mut ctx)
-            .await
-            .unwrap_err();
-        assert!(ctx.volatile_ctx.table_route.is_none());
-        assert!(err.is_retryable());
-        assert!(format!("{err:?}").contains("Failed to update the table route"));
-        assert_eq!(rx.len(), 0);
-        state.rollback_downgraded_region(&mut ctx).await.unwrap();
-        let event = rx.try_recv().unwrap();
-        let detecting_regions = event.into_region_failure_detectors();
-        assert_eq!(
-            detecting_regions,
-            vec![(from_peer.id, ctx.persistent_ctx.region_id)]
-        );
-
-        let table_route = table_metadata_manager
-            .table_route_manager()
-            .table_route_storage()
-            .get(table_id)
-            .await
-            .unwrap()
-            .unwrap();
-        assert_eq!(
-            &expected_region_routes,
-            table_route.region_routes().unwrap()
-        );
-    }
-
    #[tokio::test]
    async fn test_next_migration_end_state() {
        let mut state = Box::new(UpdateMetadata::Rollback);
@@ -238,8 +139,6 @@ mod tests {
            .downcast_ref::<RegionMigrationAbort>()
            .unwrap();

-        assert!(ctx.volatile_ctx.table_route.is_none());
-
        let table_route = table_metadata_manager
            .table_route_manager()
            .table_route_storage()
--- a/src/meta-srv/src/procedure/region_migration/update_metadata/upgrade_candidate_region.rs
+++ b/src/meta-srv/src/procedure/region_migration/update_metadata/upgrade_candidate_region.rs
@@ -160,7 +160,7 @@ impl UpdateMetadata {
                    region_options: region_options.clone(),
                    region_wal_options: region_wal_options.clone(),
                },
-                table_route_value,
+                &table_route_value,
                region_routes,
                &region_options,
                &region_wal_options,
@@ -168,13 +168,11 @@ impl UpdateMetadata {
            .await
            .context(error::TableMetadataManagerSnafu)
        {
-            ctx.remove_table_route_value();
            return Err(BoxedError::new(err)).context(error::RetryLaterWithSourceSnafu {
                reason: format!("Failed to update the table route during the upgrading candidate region: {region_id}"),
            });
        };

-        ctx.remove_table_route_value();
        ctx.deregister_failure_detectors().await;
        // Consumes the guard.
        ctx.volatile_ctx.opening_region_guard.take();
@@ -304,71 +302,6 @@ mod tests {
        assert_eq!(new_region_routes[0].leader_peer.as_ref().unwrap().id, 2);
    }

-    #[tokio::test]
-    async fn test_failed_to_update_table_route_error() {
-        let state = UpdateMetadata::Upgrade;
-        let env = TestingEnv::new();
-        let persistent_context = new_persistent_context();
-        let mut ctx = env.context_factory().new_context(persistent_context);
-        let opening_keeper = MemoryRegionKeeper::default();
-
-        let table_id = 1024;
-        let table_info = new_test_table_info(table_id, vec![1]).into();
-        let region_routes = vec![
-            RegionRoute {
-                region: Region::new_test(RegionId::new(table_id, 1)),
-                leader_peer: Some(Peer::empty(1)),
-                follower_peers: vec![Peer::empty(5), Peer::empty(3)],
-                leader_state: Some(LeaderState::Downgrading),
-                leader_down_since: Some(current_time_millis()),
-            },
-            RegionRoute {
-                region: Region::new_test(RegionId::new(table_id, 2)),
-                leader_peer: Some(Peer::empty(4)),
-                leader_state: Some(LeaderState::Downgrading),
-                ..Default::default()
-            },
-        ];
-
-        env.create_physical_table_metadata(table_info, region_routes)
-            .await;
-
-        let table_metadata_manager = env.table_metadata_manager();
-        let original_table_route = table_metadata_manager
-            .table_route_manager()
-            .table_route_storage()
-            .get_with_raw_bytes(table_id)
-            .await
-            .unwrap()
-            .unwrap();
-
-        // modifies the table route.
-        table_metadata_manager
-            .update_leader_region_status(table_id, &original_table_route, |route| {
-                if route.region.id == RegionId::new(1024, 2) {
-                    // Removes the status.
-                    Some(None)
-                } else {
-                    None
-                }
-            })
-            .await
-            .unwrap();
-
-        // sets the old table route.
-        ctx.volatile_ctx.table_route = Some(original_table_route);
-        let guard = opening_keeper
-            .register(2, RegionId::new(table_id, 1))
-            .unwrap();
-        ctx.volatile_ctx.opening_region_guard = Some(guard);
-        let err = state.upgrade_candidate_region(&mut ctx).await.unwrap_err();
-
-        assert!(ctx.volatile_ctx.table_route.is_none());
-        assert!(ctx.volatile_ctx.opening_region_guard.is_some());
-        assert!(err.is_retryable());
-        assert!(format!("{err:?}").contains("Failed to update the table route"));
-    }
-
    #[tokio::test]
    async fn test_check_metadata() {
        let state = UpdateMetadata::Upgrade;
@@ -486,7 +419,6 @@ mod tests {
            .unwrap();
        let region_routes = table_route.region_routes().unwrap();

-        assert!(ctx.volatile_ctx.table_route.is_none());
        assert!(ctx.volatile_ctx.opening_region_guard.is_none());
        assert_eq!(region_routes.len(), 1);
        assert!(!region_routes[0].is_leader_downgrading());
--- a/src/meta-srv/src/region/supervisor.rs
+++ b/src/meta-srv/src/region/supervisor.rs
@@ -42,6 +42,7 @@ use crate::procedure::region_migration::{
 };
 use crate::region::failure_detector::RegionFailureDetector;
 use crate::selector::SelectorOptions;
+use crate::state::StateRef;

 /// `DatanodeHeartbeat` represents the heartbeat signal sent from a datanode.
 /// It includes identifiers for the cluster and datanode, a list of regions being monitored,
@@ -86,16 +87,6 @@ pub(crate) enum Event {
    Dump(tokio::sync::oneshot::Sender<RegionFailureDetector>),
 }

-#[cfg(test)]
-impl Event {
-    pub(crate) fn into_region_failure_detectors(self) -> Vec<DetectingRegion> {
-        match self {
-            Self::RegisterFailureDetectors(detecting_regions) => detecting_regions,
-            _ => unreachable!(),
-        }
-    }
-}
-
 impl Debug for Event {
    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
        match self {
@@ -228,6 +219,8 @@ pub struct RegionSupervisor {
    runtime_switch_manager: RuntimeSwitchManagerRef,
    /// Peer lookup service
    peer_lookup: PeerLookupServiceRef,
+    /// The meta state, used to check if the current metasrv is the leader.
+    state: Option<StateRef>,
 }

 /// Controller for managing failure detectors for regions.
@@ -308,12 +301,29 @@ impl RegionSupervisor {
            region_migration_manager,
            runtime_switch_manager,
            peer_lookup,
+            state: None,
        }
    }

+    /// Sets the meta state.
+    pub(crate) fn with_state(mut self, state: StateRef) -> Self {
+        self.state = Some(state);
+        self
+    }
+
    /// Runs the main loop.
    pub(crate) async fn run(&mut self) {
        while let Some(event) = self.receiver.recv().await {
+            if let Some(state) = self.state.as_ref()
+                && !state.read().unwrap().is_leader()
+            {
+                warn!(
+                    "The current metasrv is not the leader, ignore {:?} event",
+                    event
+                );
+                continue;
+            }
+
            match event {
                Event::Tick => {
                    let regions = self.detect_region_failure();
@@ -326,7 +336,10 @@ impl RegionSupervisor {
                    self.deregister_failure_detectors(detecting_regions).await
                }
                Event::HeartbeatArrived(heartbeat) => self.on_heartbeat_arrived(heartbeat),
-                Event::Clear => self.clear(),
+                Event::Clear => {
+                    self.clear();
+                    info!("Region supervisor is initialized.");
+                }
                #[cfg(test)]
                Event::Dump(sender) => {
                    let _ = sender.send(self.failure_detector.dump());
@@ -759,6 +772,7 @@ pub(crate) mod tests {
            while let Ok(event) = rx.try_recv() {
                assert_matches!(event, Event::Tick | Event::Clear);
            }
+            assert!(ticker.tick_handle.lock().unwrap().is_none());
        }
    }

--- a/src/meta-srv/src/selector/weight_compute.rs
+++ b/src/meta-srv/src/selector/weight_compute.rs
@@ -197,6 +197,7 @@ mod tests {
                },
                data_topic_latest_entry_id: 0,
                metadata_topic_latest_entry_id: 0,
+                written_bytes: 0,
            }],
            ..Default::default()
        }
@@ -224,6 +225,7 @@ mod tests {
                },
                data_topic_latest_entry_id: 0,
                metadata_topic_latest_entry_id: 0,
+                written_bytes: 0,
            }],
            ..Default::default()
        }
@@ -251,6 +253,7 @@ mod tests {
                },
                data_topic_latest_entry_id: 0,
                metadata_topic_latest_entry_id: 0,
+                written_bytes: 0,
            }],
            ..Default::default()
        }
--- a/src/meta-srv/src/service/heartbeat.rs
+++ b/src/meta-srv/src/service/heartbeat.rs
@@ -27,10 +27,9 @@ use snafu::OptionExt;
 use tokio::sync::mpsc;
 use tokio::sync::mpsc::Sender;
 use tokio_stream::wrappers::ReceiverStream;
-use tonic::{Request, Response, Streaming};
+use tonic::{Request, Response, Status, Streaming};

-use crate::error;
-use crate::error::Result;
+use crate::error::{self, Result};
 use crate::handler::{HeartbeatHandlerGroup, Pusher, PusherId};
 use crate::metasrv::{Context, Metasrv};
 use crate::metrics::METRIC_META_HEARTBEAT_RECV;
@@ -80,6 +79,7 @@ impl heartbeat_server::Heartbeat for Metasrv {
                        let res = handler_group
                            .handle(req, ctx.clone())
                            .await
+                            .inspect_err(|e| warn!(e; "Failed to handle heartbeat request, pusher: {pusher_id:?}", ))
                            .map_err(|e| e.into());

                        is_not_leader = res.as_ref().is_ok_and(|r| r.is_not_leader());
@@ -109,6 +109,12 @@ impl heartbeat_server::Heartbeat for Metasrv {

                if is_not_leader {
                    warn!("Quit because it is no longer the leader");
+                    let _ = tx
+                        .send(Err(Status::aborted(format!(
+                            "The requested metasrv node is not leader, node addr: {}",
+                            ctx.server_addr
+                        ))))
+                        .await;
                    break;
                }
            }
--- a/src/meta-srv/src/service/store.rs
+++ b/src/meta-srv/src/service/store.rs
@@ -97,7 +97,7 @@ impl store_server::Store for Metasrv {
        let req = req.into_inner();

        let _timer = METRIC_META_KV_REQUEST_ELAPSED
-            .with_label_values(&[self.kv_backend().name(), "batch_pub"])
+            .with_label_values(&[self.kv_backend().name(), "batch_put"])
            .start_timer();

        let req: BatchPutRequest = req.into();
--- a/src/meta-srv/src/state.rs
+++ b/src/meta-srv/src/state.rs
@@ -75,6 +75,12 @@ impl State {
        })
    }

+    /// Returns true if the current state is a leader.
+    pub fn is_leader(&self) -> bool {
+        matches!(self, State::Leader(_))
+    }
+
+    /// Returns true if the leader cache is enabled.
    pub fn enable_leader_cache(&self) -> bool {
        match &self {
            State::Leader(leader) => leader.enable_leader_cache,
--- a/src/metric-engine/src/metadata_region.rs
+++ b/src/metric-engine/src/metadata_region.rs
@@ -66,6 +66,10 @@ const COLUMN_PREFIX: &str = "__column_";
 /// itself.
 pub struct MetadataRegion {
    pub(crate) mito: MitoEngine,
+    /// The cache for contents(key-value pairs) of region metadata.
+    ///
+    /// The cache should be invalidated when any new values are put into the metadata region or any
+    /// values are deleted from the metadata region.
    cache: Cache<RegionId, RegionMetadataCacheEntry>,
    /// Logical lock for operations that need to be serialized. Like update & read region columns.
    ///
@@ -80,14 +84,19 @@ struct RegionMetadataCacheEntry {
    size: usize,
 }

+/// The max size of the region metadata cache.
 const MAX_CACHE_SIZE: u64 = ReadableSize::mb(128).as_bytes();
+/// The TTL of the region metadata cache.
+const CACHE_TTL: Duration = Duration::from_secs(5 * 60);

 impl MetadataRegion {
    pub fn new(mito: MitoEngine) -> Self {
        let cache = Cache::builder()
            .max_capacity(MAX_CACHE_SIZE)
+            // Use the LRU eviction policy to minimize frequent mito scans.
+            // Recently accessed items are retained longer in the cache.
            .eviction_policy(EvictionPolicy::lru())
-            .time_to_live(Duration::from_secs(60))
+            .time_to_live(CACHE_TTL)
            .weigher(|_, v: &RegionMetadataCacheEntry| v.size as u32)
            .build();
        Self {
@@ -382,11 +391,11 @@ impl MetadataRegion {
        }
    }

-    async fn load_all(&self, region_id: RegionId) -> Result<RegionMetadataCacheEntry> {
+    async fn load_all(&self, metadata_region_id: RegionId) -> Result<RegionMetadataCacheEntry> {
        let scan_req = MetadataRegion::build_read_request();
        let record_batch_stream = self
            .mito
-            .scan_to_stream(region_id, scan_req)
+            .scan_to_stream(metadata_region_id, scan_req)
            .await
            .context(MitoReadOperationSnafu)?;

@@ -407,12 +416,12 @@ impl MetadataRegion {

    async fn get_all_with_prefix(
        &self,
-        region_id: RegionId,
+        metadata_region_id: RegionId,
        prefix: &str,
    ) -> Result<HashMap<String, String>> {
        let region_metadata = self
            .cache
-            .try_get_with(region_id, self.load_all(region_id))
+            .try_get_with(metadata_region_id, self.load_all(metadata_region_id))
            .await
            .context(CacheGetSnafu)?;

@@ -446,16 +455,17 @@ impl MetadataRegion {

    /// Delete the given keys. For performance consideration, this method
    /// doesn't check if those keys exist or not.
-    async fn delete(&self, region_id: RegionId, keys: &[String]) -> Result<()> {
+    async fn delete(&self, metadata_region_id: RegionId, keys: &[String]) -> Result<()> {
        let delete_request = Self::build_delete_request(keys);
        self.mito
            .handle_request(
-                region_id,
+                metadata_region_id,
                store_api::region_request::RegionRequest::Delete(delete_request),
            )
            .await
            .context(MitoWriteOperationSnafu)?;
-        self.cache.invalidate(&region_id).await;
+        // Invalidates the region metadata cache if any values are deleted from the metadata region.
+        self.cache.invalidate(&metadata_region_id).await;

        Ok(())
    }
@@ -546,7 +556,7 @@ impl MetadataRegion {
        write_region_id: bool,
        logical_regions: impl Iterator<Item = (RegionId, HashMap<&str, &ColumnMetadata>)>,
    ) -> Result<()> {
-        let region_id = utils::to_metadata_region_id(physical_region_id);
+        let metadata_region_id = utils::to_metadata_region_id(physical_region_id);
        let iter = logical_regions
            .into_iter()
            .flat_map(|(logical_region_id, column_metadatas)| {
@@ -573,12 +583,13 @@ impl MetadataRegion {
        let put_request = MetadataRegion::build_put_request_from_iter(iter.into_iter());
        self.mito
            .handle_request(
-                region_id,
+                metadata_region_id,
                store_api::region_request::RegionRequest::Put(put_request),
            )
            .await
            .context(MitoWriteOperationSnafu)?;
-        self.cache.invalidate(&region_id).await;
+        // Invalidates the region metadata cache if any new values are put into the metadata region.
+        self.cache.invalidate(&metadata_region_id).await;

        Ok(())
    }
--- a/src/metric-engine/src/utils.rs
+++ b/src/metric-engine/src/utils.rs
@@ -61,6 +61,7 @@ pub fn get_region_statistic(mito: &MitoEngine, region_id: RegionId) -> Option<Re
                metadata_flushed_entry_id: metadata_stat.manifest.data_flushed_entry_id(),
                metadata_manifest_version: metadata_stat.manifest.data_manifest_version(),
            },
+            written_bytes: metadata_stat.written_bytes + data_stat.written_bytes,
            data_topic_latest_entry_id: data_stat.data_topic_latest_entry_id,
            metadata_topic_latest_entry_id: metadata_stat.metadata_topic_latest_entry_id,
        }),
--- a/src/mito2/src/access_layer.rs
+++ b/src/mito2/src/access_layer.rs
@@ -13,6 +13,7 @@
 // limitations under the License.

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

 use object_store::services::Fs;
 use object_store::util::{join_dir, with_instrument_layers};
@@ -27,6 +28,7 @@ use crate::cache::write_cache::SstUploadRequest;
 use crate::cache::CacheManagerRef;
 use crate::config::{BloomFilterConfig, FulltextIndexConfig, InvertedIndexConfig};
 use crate::error::{CleanDirSnafu, DeleteIndexSnafu, DeleteSstSnafu, OpenDalSnafu, Result};
+use crate::metrics::{COMPACTION_STAGE_ELAPSED, FLUSH_ELAPSED};
 use crate::read::Source;
 use crate::region::options::IndexOptions;
 use crate::sst::file::{FileHandle, FileId, FileMeta};
@@ -46,6 +48,87 @@ pub const ATOMIC_WRITE_DIR: &str = "tmp/";
 /// For compatibility. Remove this after a major version release.
 pub const OLD_ATOMIC_WRITE_DIR: &str = ".tmp/";

+/// Write operation type.
+#[derive(Eq, PartialEq, Debug)]
+pub enum WriteType {
+    /// Writes from flush
+    Flush,
+    /// Writes from compaction.
+    Compaction,
+}
+
+#[derive(Debug)]
+pub struct Metrics {
+    pub(crate) write_type: WriteType,
+    pub(crate) iter_source: Duration,
+    pub(crate) write_batch: Duration,
+    pub(crate) update_index: Duration,
+    pub(crate) upload_parquet: Duration,
+    pub(crate) upload_puffin: Duration,
+}
+
+impl Metrics {
+    pub(crate) fn new(write_type: WriteType) -> Self {
+        Self {
+            write_type,
+            iter_source: Default::default(),
+            write_batch: Default::default(),
+            update_index: Default::default(),
+            upload_parquet: Default::default(),
+            upload_puffin: Default::default(),
+        }
+    }
+
+    pub(crate) fn merge(mut self, other: Self) -> Self {
+        assert_eq!(self.write_type, other.write_type);
+        self.iter_source += other.iter_source;
+        self.write_batch += other.write_batch;
+        self.update_index += other.update_index;
+        self.upload_parquet += other.upload_parquet;
+        self.upload_puffin += other.upload_puffin;
+        self
+    }
+
+    pub(crate) fn observe(self) {
+        match self.write_type {
+            WriteType::Flush => {
+                FLUSH_ELAPSED
+                    .with_label_values(&["iter_source"])
+                    .observe(self.iter_source.as_secs_f64());
+                FLUSH_ELAPSED
+                    .with_label_values(&["write_batch"])
+                    .observe(self.write_batch.as_secs_f64());
+                FLUSH_ELAPSED
+                    .with_label_values(&["update_index"])
+                    .observe(self.update_index.as_secs_f64());
+                FLUSH_ELAPSED
+                    .with_label_values(&["upload_parquet"])
+                    .observe(self.upload_parquet.as_secs_f64());
+                FLUSH_ELAPSED
+                    .with_label_values(&["upload_puffin"])
+                    .observe(self.upload_puffin.as_secs_f64());
+            }
+            WriteType::Compaction => {
+                COMPACTION_STAGE_ELAPSED
+                    .with_label_values(&["iter_source"])
+                    .observe(self.iter_source.as_secs_f64());
+                COMPACTION_STAGE_ELAPSED
+                    .with_label_values(&["write_batch"])
+                    .observe(self.write_batch.as_secs_f64());
+                COMPACTION_STAGE_ELAPSED
+                    .with_label_values(&["update_index"])
+                    .observe(self.update_index.as_secs_f64());
+                COMPACTION_STAGE_ELAPSED
+                    .with_label_values(&["upload_parquet"])
+                    .observe(self.upload_parquet.as_secs_f64());
+                COMPACTION_STAGE_ELAPSED
+                    .with_label_values(&["upload_puffin"])
+                    .observe(self.upload_puffin.as_secs_f64());
+            }
+        };
+    }
+}
+
 /// A layer to access SST files under the same directory.
 pub struct AccessLayer {
    region_dir: String,
@@ -96,6 +179,11 @@ impl AccessLayer {
        &self.puffin_manager_factory
    }

+    /// Returns the intermediate manager.
+    pub fn intermediate_manager(&self) -> &IntermediateManager {
+        &self.intermediate_manager
+    }
+
    /// Deletes a SST file (and its index file if it has one) with given file id.
    pub(crate) async fn delete_sst(&self, file_meta: &FileMeta) -> Result<()> {
        let path = location::sst_file_path(&self.region_dir, file_meta.file_id);
@@ -129,11 +217,12 @@ impl AccessLayer {
        &self,
        request: SstWriteRequest,
        write_opts: &WriteOptions,
-    ) -> Result<SstInfoArray> {
+        write_type: WriteType,
+    ) -> Result<(SstInfoArray, Metrics)> {
        let region_id = request.metadata.region_id;
        let cache_manager = request.cache_manager.clone();

-        let sst_info = if let Some(write_cache) = cache_manager.write_cache() {
+        let (sst_info, metrics) = if let Some(write_cache) = cache_manager.write_cache() {
            // Write to the write cache.
            write_cache
                .write_and_upload_sst(
@@ -145,6 +234,7 @@ impl AccessLayer {
                        remote_store: self.object_store.clone(),
                    },
                    write_opts,
+                    write_type,
                )
                .await?
        } else {
@@ -173,12 +263,15 @@ impl AccessLayer {
                request.metadata,
                indexer_builder,
                path_provider,
+                Metrics::new(write_type),
            )
            .await
            .with_file_cleaner(cleaner);
-            writer
+            let ssts = writer
                .write_all(request.source, request.max_sequence, write_opts)
-                .await?
+                .await?;
+            let metrics = writer.into_metrics();
+            (ssts, metrics)
        };

        // Put parquet metadata to cache manager.
@@ -194,7 +287,7 @@ impl AccessLayer {
            }
        }

-        Ok(sst_info)
+        Ok((sst_info, metrics))
    }
 }

--- a/src/mito2/src/cache.rs
+++ b/src/mito2/src/cache.rs
@@ -23,6 +23,7 @@ pub(crate) mod test_util;
 pub(crate) mod write_cache;

 use std::mem;
+use std::ops::Range;
 use std::sync::Arc;

 use bytes::Bytes;
@@ -32,7 +33,6 @@ use index::bloom_filter_index::{BloomFilterIndexCache, BloomFilterIndexCacheRef}
 use index::result_cache::IndexResultCache;
 use moka::notification::RemovalCause;
 use moka::sync::Cache;
-use parquet::column::page::Page;
 use parquet::file::metadata::ParquetMetaData;
 use puffin::puffin_manager::cache::{PuffinMetadataCache, PuffinMetadataCacheRef};
 use store_api::storage::{ConcreteDataType, RegionId, TimeSeriesRowSelector};
@@ -674,49 +674,33 @@ pub struct ColumnPagePath {
    column_idx: usize,
 }

-/// Cache key for pages of a SST row group.
+/// Cache key to pages in a row group (after projection).
+///
+/// Different projections will have different cache keys.
+/// We cache all ranges together because they may refer to the same `Bytes`.
 #[derive(Debug, Clone, PartialEq, Eq, Hash)]
-pub enum PageKey {
-    /// Cache key for a compressed page in a row group.
-    Compressed(ColumnPagePath),
-    /// Cache key for all uncompressed pages in a row group.
-    Uncompressed(ColumnPagePath),
+pub struct PageKey {
+    /// Id of the SST file to cache.
+    file_id: FileId,
+    /// Index of the row group.
+    row_group_idx: usize,
+    /// Byte ranges of the pages to cache.
+    ranges: Vec<Range<u64>>,
 }

 impl PageKey {
-    /// Creates a key for a compressed page.
-    pub fn new_compressed(
-        region_id: RegionId,
-        file_id: FileId,
-        row_group_idx: usize,
-        column_idx: usize,
-    ) -> PageKey {
-        PageKey::Compressed(ColumnPagePath {
-            region_id,
+    /// Creates a key for a list of pages.
+    pub fn new(file_id: FileId, row_group_idx: usize, ranges: Vec<Range<u64>>) -> PageKey {
+        PageKey {
            file_id,
            row_group_idx,
-            column_idx,
-        })
-    }
-
-    /// Creates a key for all uncompressed pages in a row group.
-    pub fn new_uncompressed(
-        region_id: RegionId,
-        file_id: FileId,
-        row_group_idx: usize,
-        column_idx: usize,
-    ) -> PageKey {
-        PageKey::Uncompressed(ColumnPagePath {
-            region_id,
-            file_id,
-            row_group_idx,
-            column_idx,
-        })
+            ranges,
+        }
    }

    /// Returns memory used by the key (estimated).
    fn estimated_size(&self) -> usize {
-        mem::size_of::<Self>()
+        mem::size_of::<Self>() + mem::size_of_val(self.ranges.as_slice())
    }
 }

@@ -724,38 +708,26 @@ impl PageKey {
 // We don't use enum here to make it easier to mock and use the struct.
 #[derive(Default)]
 pub struct PageValue {
-    /// Compressed page of the column in the row group.
-    pub compressed: Bytes,
-    /// All pages of the column in the row group.
-    pub row_group: Vec<Page>,
+    /// Compressed page in the row group.
+    pub compressed: Vec<Bytes>,
+    /// Total size of the pages (may be larger than sum of compressed bytes due to gaps).
+    pub page_size: u64,
 }

 impl PageValue {
-    /// Creates a new value from a compressed page.
-    pub fn new_compressed(bytes: Bytes) -> PageValue {
+    /// Creates a new value from a range of compressed pages.
+    pub fn new(bytes: Vec<Bytes>, page_size: u64) -> PageValue {
        PageValue {
            compressed: bytes,
-            row_group: vec![],
-        }
-    }
-
-    /// Creates a new value from all pages in a row group.
-    pub fn new_row_group(pages: Vec<Page>) -> PageValue {
-        PageValue {
-            compressed: Bytes::new(),
-            row_group: pages,
+            page_size,
        }
    }

    /// Returns memory used by the value (estimated).
    fn estimated_size(&self) -> usize {
        mem::size_of::<Self>()
-            + self.compressed.len()
-            + self
-                .row_group
-                .iter()
-                .map(|page| page.buffer().len())
-                .sum::<usize>()
+            + self.page_size as usize
+            + self.compressed.iter().map(mem::size_of_val).sum::<usize>()
    }
 }

@@ -834,7 +806,7 @@ mod tests {
            .get_repeated_vector(&ConcreteDataType::int64_datatype(), &value)
            .is_none());

-        let key = PageKey::new_uncompressed(region_id, file_id, 0, 0);
+        let key = PageKey::new(file_id, 1, vec![Range { start: 0, end: 5 }]);
        let pages = Arc::new(PageValue::default());
        cache.put_pages(key.clone(), pages);
        assert!(cache.get_pages(&key).is_none());
@@ -882,9 +854,8 @@ mod tests {
    #[test]
    fn test_page_cache() {
        let cache = CacheManager::builder().page_cache_size(1000).build();
-        let region_id = RegionId::new(1, 1);
        let file_id = FileId::random();
-        let key = PageKey::new_compressed(region_id, file_id, 0, 0);
+        let key = PageKey::new(file_id, 0, vec![(0..10), (10..20)]);
        assert!(cache.get_pages(&key).is_none());
        let pages = Arc::new(PageValue::default());
        cache.put_pages(key.clone(), pages);
--- a/src/mito2/src/cache/index.rs
+++ b/src/mito2/src/cache/index.rs
@@ -212,6 +212,8 @@ where
    }

    fn put_page(&self, key: K, page_key: PageKey, value: Bytes) {
+        // Clones the value to ensure it doesn't reference a larger buffer.
+        let value = Bytes::from(value.to_vec());
        CACHE_BYTES
            .with_label_values(&[INDEX_CONTENT_TYPE])
            .add((self.weight_of_content)(&(key, page_key), &value).into());
--- a/src/mito2/src/cache/index/bloom_filter_index.rs
+++ b/src/mito2/src/cache/index/bloom_filter_index.rs
@@ -15,10 +15,9 @@
 use std::ops::Range;
 use std::sync::Arc;

-use api::v1::index::BloomFilterMeta;
+use api::v1::index::{BloomFilterLoc, BloomFilterMeta};
 use async_trait::async_trait;
 use bytes::Bytes;
-use futures::future::try_join_all;
 use index::bloom_filter::error::Result;
 use index::bloom_filter::reader::BloomFilterReader;
 use store_api::storage::ColumnId;
@@ -57,11 +56,17 @@ impl BloomFilterIndexCache {
 /// Calculates weight for bloom filter index metadata.
 fn bloom_filter_index_metadata_weight(
    k: &(FileId, ColumnId, Tag),
-    _: &Arc<BloomFilterMeta>,
+    meta: &Arc<BloomFilterMeta>,
 ) -> u32 {
-    (k.0.as_bytes().len()
+    let base = k.0.as_bytes().len()
        + std::mem::size_of::<ColumnId>()
-        + std::mem::size_of::<BloomFilterMeta>()) as u32
+        + std::mem::size_of::<Tag>()
+        + std::mem::size_of::<BloomFilterMeta>();
+
+    let vec_estimated = meta.segment_loc_indices.len() * std::mem::size_of::<u64>()
+        + meta.bloom_filter_locs.len() * std::mem::size_of::<BloomFilterLoc>();
+
+    (base + vec_estimated) as u32
 }

 /// Calculates weight for bloom filter index content.
@@ -120,21 +125,24 @@ impl<R: BloomFilterReader + Send> BloomFilterReader for CachedBloomFilterIndexBl
    }

    async fn read_vec(&self, ranges: &[Range<u64>]) -> Result<Vec<Bytes>> {
-        let fetch = ranges.iter().map(|range| {
+        let mut pages = Vec::with_capacity(ranges.len());
+        for range in ranges {
            let inner = &self.inner;
-            self.cache.get_or_load(
-                (self.file_id, self.column_id, self.tag),
-                self.blob_size,
-                range.start,
-                (range.end - range.start) as u32,
-                move |ranges| async move { inner.read_vec(&ranges).await },
-            )
-        });
-        Ok(try_join_all(fetch)
-            .await?
-            .into_iter()
-            .map(Bytes::from)
-            .collect::<Vec<_>>())
+            let page = self
+                .cache
+                .get_or_load(
+                    (self.file_id, self.column_id, self.tag),
+                    self.blob_size,
+                    range.start,
+                    (range.end - range.start) as u32,
+                    move |ranges| async move { inner.read_vec(&ranges).await },
+                )
+                .await?;
+
+            pages.push(Bytes::from(page));
+        }
+
+        Ok(pages)
    }

    /// Reads the meta information of the bloom filter.
@@ -165,6 +173,45 @@ mod test {

    const FUZZ_REPEAT_TIMES: usize = 100;

+    #[test]
+    fn bloom_filter_metadata_weight_counts_vec_contents() {
+        let file_id = FileId::parse_str("00000000-0000-0000-0000-000000000001").unwrap();
+        let column_id: ColumnId = 42;
+        let tag = Tag::Skipping;
+
+        let meta = BloomFilterMeta {
+            rows_per_segment: 128,
+            segment_count: 2,
+            row_count: 256,
+            bloom_filter_size: 1024,
+            segment_loc_indices: vec![0, 64, 128, 192],
+            bloom_filter_locs: vec![
+                BloomFilterLoc {
+                    offset: 0,
+                    size: 512,
+                    element_count: 1000,
+                },
+                BloomFilterLoc {
+                    offset: 512,
+                    size: 512,
+                    element_count: 1000,
+                },
+            ],
+        };
+
+        let weight =
+            bloom_filter_index_metadata_weight(&(file_id, column_id, tag), &Arc::new(meta.clone()));
+
+        let base = file_id.as_bytes().len()
+            + std::mem::size_of::<ColumnId>()
+            + std::mem::size_of::<Tag>()
+            + std::mem::size_of::<BloomFilterMeta>();
+        let expected_dynamic = meta.segment_loc_indices.len() * std::mem::size_of::<u64>()
+            + meta.bloom_filter_locs.len() * std::mem::size_of::<BloomFilterLoc>();
+
+        assert_eq!(weight as usize, base + expected_dynamic);
+    }
+
    #[test]
    fn fuzz_index_calculation() {
        let mut rng = rand::rng();
--- a/src/mito2/src/cache/index/inverted_index.rs
+++ b/src/mito2/src/cache/index/inverted_index.rs
@@ -18,7 +18,6 @@ use std::sync::Arc;
 use api::v1::index::InvertedIndexMetas;
 use async_trait::async_trait;
 use bytes::Bytes;
-use futures::future::try_join_all;
 use index::inverted_index::error::Result;
 use index::inverted_index::format::reader::InvertedIndexReader;
 use prost::Message;
@@ -93,21 +92,24 @@ impl<R: InvertedIndexReader> InvertedIndexReader for CachedInvertedIndexBlobRead
    }

    async fn read_vec(&self, ranges: &[Range<u64>]) -> Result<Vec<Bytes>> {
-        let fetch = ranges.iter().map(|range| {
+        let mut pages = Vec::with_capacity(ranges.len());
+        for range in ranges {
            let inner = &self.inner;
-            self.cache.get_or_load(
-                self.file_id,
-                self.blob_size,
-                range.start,
-                (range.end - range.start) as u32,
-                move |ranges| async move { inner.read_vec(&ranges).await },
-            )
-        });
-        Ok(try_join_all(fetch)
-            .await?
-            .into_iter()
-            .map(Bytes::from)
-            .collect::<Vec<_>>())
+            let page = self
+                .cache
+                .get_or_load(
+                    self.file_id,
+                    self.blob_size,
+                    range.start,
+                    (range.end - range.start) as u32,
+                    move |ranges| async move { inner.read_vec(&ranges).await },
+                )
+                .await?;
+
+            pages.push(Bytes::from(page));
+        }
+
+        Ok(pages)
    }

    async fn metadata(&self) -> Result<Arc<InvertedIndexMetas>> {
--- a/src/mito2/src/cache/write_cache.rs
+++ b/src/mito2/src/cache/write_cache.rs
@@ -15,7 +15,7 @@
 //! A write-through cache for remote object stores.

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

 use common_base::readable_size::ReadableSize;
 use common_telemetry::{debug, info};
@@ -25,14 +25,13 @@ use snafu::ResultExt;
 use store_api::storage::RegionId;

 use crate::access_layer::{
-    new_fs_cache_store, FilePathProvider, RegionFilePathFactory, SstInfoArray, SstWriteRequest,
-    TempFileCleaner, WriteCachePathProvider,
+    new_fs_cache_store, FilePathProvider, Metrics, RegionFilePathFactory, SstInfoArray,
+    SstWriteRequest, TempFileCleaner, WriteCachePathProvider, WriteType,
 };
 use crate::cache::file_cache::{FileCache, FileCacheRef, FileType, IndexKey, IndexValue};
 use crate::error::{self, Result};
 use crate::metrics::{
-    FLUSH_ELAPSED, UPLOAD_BYTES_TOTAL, WRITE_CACHE_DOWNLOAD_BYTES_TOTAL,
-    WRITE_CACHE_DOWNLOAD_ELAPSED,
+    UPLOAD_BYTES_TOTAL, WRITE_CACHE_DOWNLOAD_BYTES_TOTAL, WRITE_CACHE_DOWNLOAD_ELAPSED,
 };
 use crate::sst::index::intermediate::IntermediateManager;
 use crate::sst::index::puffin_manager::PuffinManagerFactory;
@@ -107,11 +106,8 @@ impl WriteCache {
        write_request: SstWriteRequest,
        upload_request: SstUploadRequest,
        write_opts: &WriteOptions,
-    ) -> Result<SstInfoArray> {
-        let timer = FLUSH_ELAPSED
-            .with_label_values(&["write_sst"])
-            .start_timer();
-
+        write_type: WriteType,
+    ) -> Result<(SstInfoArray, Metrics)> {
        let region_id = write_request.metadata.region_id;

        let store = self.file_cache.local_store();
@@ -137,6 +133,7 @@ impl WriteCache {
            write_request.metadata,
            indexer,
            path_provider.clone(),
+            Metrics::new(write_type),
        )
        .await
        .with_file_cleaner(cleaner);
@@ -144,12 +141,11 @@ impl WriteCache {
        let sst_info = writer
            .write_all(write_request.source, write_request.max_sequence, write_opts)
            .await?;
-
-        timer.stop_and_record();
+        let mut metrics = writer.into_metrics();

        // Upload sst file to remote object store.
        if sst_info.is_empty() {
-            return Ok(sst_info);
+            return Ok((sst_info, metrics));
        }

        let mut upload_tracker = UploadTracker::new(region_id);
@@ -160,10 +156,12 @@ impl WriteCache {
            let parquet_path = upload_request
                .dest_path_provider
                .build_sst_file_path(sst.file_id);
+            let start = Instant::now();
            if let Err(e) = self.upload(parquet_key, &parquet_path, remote_store).await {
                err = Some(e);
                break;
            }
+            metrics.upload_parquet += start.elapsed();
            upload_tracker.push_uploaded_file(parquet_path);

            if sst.index_metadata.file_size > 0 {
@@ -171,10 +169,12 @@ impl WriteCache {
                let puffin_path = upload_request
                    .dest_path_provider
                    .build_index_file_path(sst.file_id);
+                let start = Instant::now();
                if let Err(e) = self.upload(puffin_key, &puffin_path, remote_store).await {
                    err = Some(e);
                    break;
                }
+                metrics.upload_puffin += start.elapsed();
                upload_tracker.push_uploaded_file(puffin_path);
            }
        }
@@ -187,7 +187,7 @@ impl WriteCache {
            return Err(err);
        }

-        Ok(sst_info)
+        Ok((sst_info, metrics))
    }

    /// Removes a file from the cache by `index_key`.
@@ -297,13 +297,7 @@ impl WriteCache {
        let file_type = index_key.file_type;
        let cache_path = self.file_cache.cache_file_path(index_key);

-        let timer = FLUSH_ELAPSED
-            .with_label_values(&[match file_type {
-                FileType::Parquet => "upload_parquet",
-                FileType::Puffin => "upload_puffin",
-            }])
-            .start_timer();
-
+        let start = Instant::now();
        let cached_value = self
            .file_cache
            .local_store()
@@ -347,11 +341,11 @@ impl WriteCache {
        UPLOAD_BYTES_TOTAL.inc_by(bytes_written);

        debug!(
-            "Successfully upload file to remote, region: {}, file: {}, upload_path: {}, cost: {:?}s",
+            "Successfully upload file to remote, region: {}, file: {}, upload_path: {}, cost: {:?}",
            region_id,
            file_id,
            upload_path,
-            timer.stop_and_record()
+            start.elapsed(),
        );

        let index_value = IndexValue {
@@ -493,11 +487,11 @@ mod tests {
        };

        // Write to cache and upload sst to mock remote store
-        let sst_info = write_cache
-            .write_and_upload_sst(write_request, upload_request, &write_opts)
+        let (mut sst_infos, _) = write_cache
+            .write_and_upload_sst(write_request, upload_request, &write_opts, WriteType::Flush)
            .await
-            .unwrap()
-            .remove(0); //todo(hl): we assume it only creates one file.
+            .unwrap();
+        let sst_info = sst_infos.remove(0);

        let file_id = sst_info.file_id;
        let sst_upload_path = path_provider.build_sst_file_path(file_id);
@@ -586,11 +580,11 @@ mod tests {
            remote_store: mock_store.clone(),
        };

-        let sst_info = write_cache
-            .write_and_upload_sst(write_request, upload_request, &write_opts)
+        let (mut sst_infos, _) = write_cache
+            .write_and_upload_sst(write_request, upload_request, &write_opts, WriteType::Flush)
            .await
-            .unwrap()
-            .remove(0);
+            .unwrap();
+        let sst_info = sst_infos.remove(0);
        let write_parquet_metadata = sst_info.file_metadata.unwrap();

        // Read metadata from write cache
@@ -661,7 +655,7 @@ mod tests {
        };

        write_cache
-            .write_and_upload_sst(write_request, upload_request, &write_opts)
+            .write_and_upload_sst(write_request, upload_request, &write_opts, WriteType::Flush)
            .await
            .unwrap_err();
        let atomic_write_dir = write_cache_dir.path().join(ATOMIC_WRITE_DIR);
--- a/src/mito2/src/compaction.rs
+++ b/src/mito2/src/compaction.rs
@@ -294,6 +294,7 @@ impl CompactionScheduler {
            &options,
            &request.current_version.options.compaction,
            request.current_version.options.append_mode,
+            Some(self.engine_config.max_background_compactions),
        );
        let region_id = request.region_id();
        let CompactionRequest {
--- a/src/mito2/src/compaction/compactor.rs
+++ b/src/mito2/src/compaction/compactor.rs
@@ -27,7 +27,7 @@ use snafu::{OptionExt, ResultExt};
 use store_api::metadata::RegionMetadataRef;
 use store_api::storage::RegionId;

-use crate::access_layer::{AccessLayer, AccessLayerRef, OperationType, SstWriteRequest};
+use crate::access_layer::{AccessLayer, AccessLayerRef, OperationType, SstWriteRequest, WriteType};
 use crate::cache::{CacheManager, CacheManagerRef};
 use crate::compaction::picker::{new_picker, PickerOutput};
 use crate::compaction::{find_ttl, CompactionSstReaderBuilder};
@@ -339,7 +339,7 @@ impl Compactor for DefaultCompactor {
                }
                .build_sst_reader()
                .await?;
-                let output_files = sst_layer
+                let (sst_infos, metrics) = sst_layer
                    .write_sst(
                        SstWriteRequest {
                            op_type: OperationType::Compact,
@@ -354,8 +354,10 @@ impl Compactor for DefaultCompactor {
                            bloom_filter_index_config,
                        },
                        &write_opts,
+                        WriteType::Compaction,
                    )
-                    .await?
+                    .await?;
+                let output_files = sst_infos
                    .into_iter()
                    .map(|sst_info| FileMeta {
                        region_id,
@@ -373,9 +375,10 @@ impl Compactor for DefaultCompactor {
                let output_file_names =
                    output_files.iter().map(|f| f.file_id.to_string()).join(",");
                info!(
-                    "Region {} compaction inputs: [{}], outputs: [{}]",
-                    region_id, input_file_names, output_file_names
+                    "Region {} compaction inputs: [{}], outputs: [{}], metrics: {:?}",
+                    region_id, input_file_names, output_file_names, metrics
                );
+                metrics.observe();
                Ok(output_files)
            });
        }
@@ -448,6 +451,7 @@ impl Compactor for DefaultCompactor {
                &compact_request_options,
                &compaction_region.region_options.compaction,
                compaction_region.region_options.append_mode,
+                None,
            )
            .pick(compaction_region);

--- a/src/mito2/src/compaction/picker.rs
+++ b/src/mito2/src/compaction/picker.rs
@@ -125,6 +125,7 @@ pub fn new_picker(
    compact_request_options: &compact_request::Options,
    compaction_options: &CompactionOptions,
    append_mode: bool,
+    max_background_tasks: Option<usize>,
 ) -> Arc<dyn Picker> {
    if let compact_request::Options::StrictWindow(window) = compact_request_options {
        let window = if window.window_seconds == 0 {
@@ -140,6 +141,7 @@ pub fn new_picker(
                time_window_seconds: twcs_opts.time_window_seconds(),
                max_output_file_size: twcs_opts.max_output_file_size.map(|r| r.as_bytes()),
                append_mode,
+                max_background_tasks,
            }) as Arc<_>,
        }
    }
--- a/src/mito2/src/compaction/run.rs
+++ b/src/mito2/src/compaction/run.rs
@@ -163,6 +163,10 @@ impl FileGroup {
        self.files.push(file);
    }

+    pub(crate) fn num_files(&self) -> usize {
+        self.files.len()
+    }
+
    #[cfg(test)]
    pub(crate) fn files(&self) -> &[FileHandle] {
        &self.files[..]
--- a/src/mito2/src/compaction/test_util.rs
+++ b/src/mito2/src/compaction/test_util.rs
@@ -42,6 +42,25 @@ pub fn new_file_handle_with_sequence(
    end_ts_millis: i64,
    level: Level,
    sequence: u64,
+) -> FileHandle {
+    new_file_handle_with_size_and_sequence(
+        file_id,
+        start_ts_millis,
+        end_ts_millis,
+        level,
+        sequence,
+        0,
+    )
+}
+
+/// Test util to create file handles with custom size.
+pub fn new_file_handle_with_size_and_sequence(
+    file_id: FileId,
+    start_ts_millis: i64,
+    end_ts_millis: i64,
+    level: Level,
+    sequence: u64,
+    file_size: u64,
 ) -> FileHandle {
    let file_purger = new_noop_file_purger();
    FileHandle::new(
@@ -53,7 +72,7 @@ pub fn new_file_handle_with_sequence(
                Timestamp::new_millisecond(end_ts_millis),
            ),
            level,
-            file_size: 0,
+            file_size,
            available_indexes: Default::default(),
            index_file_size: 0,
            num_rows: 0,
--- a/src/mito2/src/compaction/twcs.rs
+++ b/src/mito2/src/compaction/twcs.rs
@@ -16,9 +16,11 @@ use std::collections::hash_map::Entry;
 use std::collections::{BTreeMap, HashMap};
 use std::fmt::Debug;
 use std::num::NonZeroU64;
+use std::str::FromStr;

 use common_base::readable_size::ReadableSize;
-use common_telemetry::info;
+use common_telemetry::tracing::warn;
+use common_telemetry::{debug, info};
 use common_time::timestamp::TimeUnit;
 use common_time::timestamp_millis::BucketAligned;
 use common_time::Timestamp;
@@ -36,6 +38,9 @@ use crate::sst::version::LevelMeta;

 const LEVEL_COMPACTED: Level = 1;

+/// Default value for max compaction input file num.
+const DEFAULT_MAX_INPUT_FILE_NUM: usize = 32;
+
 /// `TwcsPicker` picks files of which the max timestamp are in the same time window as compaction
 /// candidates.
 #[derive(Debug)]
@@ -48,6 +53,8 @@ pub struct TwcsPicker {
    pub max_output_file_size: Option<u64>,
    /// Whether the target region is in append mode.
    pub append_mode: bool,
+    /// Max background compaction tasks.
+    pub max_background_tasks: Option<usize>,
 }

 impl TwcsPicker {
@@ -58,19 +65,52 @@ impl TwcsPicker {
        time_windows: &mut BTreeMap<i64, Window>,
        active_window: Option<i64>,
    ) -> Vec<CompactionOutput> {
+        let max_input_file_num = std::env::var("TWCS_MAX_INPUT_FILE_NUM")
+            .ok()
+            .and_then(|v| {
+                usize::from_str(&v)
+                    .inspect_err(|_| {
+                        warn!("Cannot recognize value for TWCS_MAX_INPUT_FILE_NUM: `{v}`");
+                    })
+                    .ok()
+            })
+            .unwrap_or(DEFAULT_MAX_INPUT_FILE_NUM);
+        debug!("Max compaction file num from env: {:?}", max_input_file_num);
        let mut output = vec![];
        for (window, files) in time_windows {
            if files.files.is_empty() {
                continue;
            }
            let mut files_to_merge: Vec<_> = files.files().cloned().collect();
+
+            // Filter out large files in append mode - they won't benefit from compaction
+            if self.append_mode
+                && let Some(max_size) = self.max_output_file_size
+            {
+                let (kept_files, ignored_files) = files_to_merge
+                    .into_iter()
+                    .partition(|fg| fg.size() <= max_size as usize);
+                files_to_merge = kept_files;
+                info!(
+                    "Compaction for {} skipped {} large files in append mode for region {}, window {}, max_size: {}",
+                    region_id,
+                    ignored_files.len(),
+                    region_id,
+                    window,
+                    max_size
+                );
+            }
+
            let sorted_runs = find_sorted_runs(&mut files_to_merge);
            let found_runs = sorted_runs.len();
            // We only remove deletion markers if we found less than 2 runs and not in append mode.
            // because after compaction there will be no overlapping files.
            let filter_deleted = !files.overlapping && found_runs <= 2 && !self.append_mode;
+            if found_runs == 0 {
+                continue;
+            }

-            let inputs = if found_runs > 1 {
+            let mut inputs = if found_runs > 1 {
                reduce_runs(sorted_runs)
            } else {
                let run = sorted_runs.last().unwrap();
@@ -80,8 +120,30 @@ impl TwcsPicker {
                // no overlapping files, try merge small files
                merge_seq_files(run.items(), self.max_output_file_size)
            };
-
-            if !inputs.is_empty() {
+            let total_input_files: usize = inputs.iter().map(|g| g.num_files()).sum();
+            if total_input_files > max_input_file_num {
+                // Sorts file groups by size first.
+                inputs.sort_unstable_by_key(|fg| fg.size());
+                let mut num_picked_files = 0;
+                inputs = inputs
+                    .into_iter()
+                    .take_while(|fg| {
+                        let current_group_file_num = fg.num_files();
+                        if current_group_file_num + num_picked_files <= max_input_file_num {
+                            num_picked_files += current_group_file_num;
+                            true
+                        } else {
+                            false
+                        }
+                    })
+                    .collect::<Vec<_>>();
+                info!(
+                    "Compaction for region {} enforces max input file num limit: {}, current total: {}, input: {:?}",
+                    region_id, max_input_file_num, total_input_files, inputs
+                );
+            }
+            if inputs.len() > 1 {
+                // If we have more than one group to compact.
                log_pick_result(
                    region_id,
                    *window,
@@ -98,6 +160,16 @@ impl TwcsPicker {
                    filter_deleted,
                    output_time_range: None, // we do not enforce output time range in twcs compactions.
                });
+
+                if let Some(max_background_tasks) = self.max_background_tasks
+                    && output.len() >= max_background_tasks
+                {
+                    debug!(
+                        "Region ({:?}) compaction task size larger than max background tasks({}), remaining tasks discarded",
+                        region_id, max_background_tasks
+                    );
+                    break;
+                }
            }
        }
        output
@@ -330,7 +402,9 @@ mod tests {
    use std::collections::HashSet;

    use super::*;
-    use crate::compaction::test_util::{new_file_handle, new_file_handle_with_sequence};
+    use crate::compaction::test_util::{
+        new_file_handle, new_file_handle_with_sequence, new_file_handle_with_size_and_sequence,
+    };
    use crate::sst::file::{FileId, Level};

    #[test]
@@ -645,6 +719,7 @@ mod tests {
                time_window_seconds: None,
                max_output_file_size: None,
                append_mode: false,
+                max_background_tasks: None,
            }
            .build_output(RegionId::from_u64(0), &mut windows, active_window);

@@ -756,5 +831,305 @@ mod tests {
        .check();
    }

+    #[test]
+    fn test_append_mode_filter_large_files() {
+        let file_ids = (0..4).map(|_| FileId::random()).collect::<Vec<_>>();
+        let max_output_file_size = 1000u64;
+
+        // Create files with different sizes
+        let small_file_1 = new_file_handle_with_size_and_sequence(file_ids[0], 0, 999, 0, 1, 500);
+        let large_file_1 = new_file_handle_with_size_and_sequence(file_ids[1], 0, 999, 0, 2, 1500);
+        let small_file_2 = new_file_handle_with_size_and_sequence(file_ids[2], 0, 999, 0, 3, 800);
+        let large_file_2 = new_file_handle_with_size_and_sequence(file_ids[3], 0, 999, 0, 4, 2000);
+
+        // Create file groups (each file is in its own group due to different sequences)
+        let mut files_to_merge = vec![
+            FileGroup::new_with_file(small_file_1),
+            FileGroup::new_with_file(large_file_1),
+            FileGroup::new_with_file(small_file_2),
+            FileGroup::new_with_file(large_file_2),
+        ];
+
+        // Test filtering logic directly
+        let original_count = files_to_merge.len();
+
+        // Apply append mode filtering
+        files_to_merge.retain(|fg| fg.size() <= max_output_file_size as usize);
+
+        // Should have filtered out 2 large files, leaving 2 small files
+        assert_eq!(files_to_merge.len(), 2);
+        assert_eq!(original_count, 4);
+
+        // Verify the remaining files are the small ones
+        for fg in &files_to_merge {
+            assert!(
+                fg.size() <= max_output_file_size as usize,
+                "File size {} should be <= {}",
+                fg.size(),
+                max_output_file_size
+            );
+        }
+    }
+
+    #[test]
+    fn test_build_output_multiple_windows_with_zero_runs() {
+        let file_ids = (0..6).map(|_| FileId::random()).collect::<Vec<_>>();
+
+        let files = [
+            // Window 0: Contains 3 files but not forming any runs (not enough files in sequence to reach trigger_file_num)
+            new_file_handle_with_sequence(file_ids[0], 0, 999, 0, 1),
+            new_file_handle_with_sequence(file_ids[1], 0, 999, 0, 2),
+            new_file_handle_with_sequence(file_ids[2], 0, 999, 0, 3),
+            // Window 3: Contains files that will form 2 runs
+            new_file_handle_with_sequence(file_ids[3], 3000, 3999, 0, 4),
+            new_file_handle_with_sequence(file_ids[4], 3000, 3999, 0, 5),
+            new_file_handle_with_sequence(file_ids[5], 3000, 3999, 0, 6),
+        ];
+
+        let mut windows = assign_to_windows(files.iter(), 3);
+
+        // Create picker with trigger_file_num of 4 so single files won't form runs in first window
+        let picker = TwcsPicker {
+            trigger_file_num: 4, // High enough to prevent runs in first window
+            time_window_seconds: Some(3),
+            max_output_file_size: None,
+            append_mode: false,
+            max_background_tasks: None,
+        };
+
+        let active_window = find_latest_window_in_seconds(files.iter(), 3);
+        let output = picker.build_output(RegionId::from_u64(123), &mut windows, active_window);
+
+        assert!(
+            !output.is_empty(),
+            "Should have output from windows with runs, even when one window has 0 runs"
+        );
+
+        let all_output_files: Vec<_> = output
+            .iter()
+            .flat_map(|o| o.inputs.iter())
+            .map(|f| f.file_id())
+            .collect();
+
+        assert!(
+            all_output_files.contains(&file_ids[3])
+                || all_output_files.contains(&file_ids[4])
+                || all_output_files.contains(&file_ids[5]),
+            "Output should contain files from the window with runs"
+        );
+    }
+
+    #[test]
+    fn test_build_output_single_window_zero_runs() {
+        let file_ids = (0..2).map(|_| FileId::random()).collect::<Vec<_>>();
+
+        let large_file_1 = new_file_handle_with_size_and_sequence(file_ids[0], 0, 999, 0, 1, 2000); // 2000 bytes
+        let large_file_2 = new_file_handle_with_size_and_sequence(file_ids[1], 0, 999, 0, 2, 2500); // 2500 bytes
+
+        let files = [large_file_1, large_file_2];
+
+        let mut windows = assign_to_windows(files.iter(), 3);
+
+        let picker = TwcsPicker {
+            trigger_file_num: 2,
+            time_window_seconds: Some(3),
+            max_output_file_size: Some(1000),
+            append_mode: true,
+            max_background_tasks: None,
+        };
+
+        let active_window = find_latest_window_in_seconds(files.iter(), 3);
+        let output = picker.build_output(RegionId::from_u64(456), &mut windows, active_window);
+
+        // Should return empty output (no compaction needed)
+        assert!(
+            output.is_empty(),
+            "Should return empty output when no runs are found after filtering"
+        );
+    }
+
+    #[test]
+    fn test_max_background_tasks_truncation() {
+        let file_ids = (0..10).map(|_| FileId::random()).collect::<Vec<_>>();
+        let max_background_tasks = 3;
+
+        // Create files across multiple windows that will generate multiple compaction outputs
+        let files = [
+            // Window 0: 4 files that will form a run
+            new_file_handle_with_sequence(file_ids[0], 0, 999, 0, 1),
+            new_file_handle_with_sequence(file_ids[1], 0, 999, 0, 2),
+            new_file_handle_with_sequence(file_ids[2], 0, 999, 0, 3),
+            new_file_handle_with_sequence(file_ids[3], 0, 999, 0, 4),
+            // Window 3: 4 files that will form another run
+            new_file_handle_with_sequence(file_ids[4], 3000, 3999, 0, 5),
+            new_file_handle_with_sequence(file_ids[5], 3000, 3999, 0, 6),
+            new_file_handle_with_sequence(file_ids[6], 3000, 3999, 0, 7),
+            new_file_handle_with_sequence(file_ids[7], 3000, 3999, 0, 8),
+            // Window 6: 4 files that will form another run
+            new_file_handle_with_sequence(file_ids[8], 6000, 6999, 0, 9),
+            new_file_handle_with_sequence(file_ids[9], 6000, 6999, 0, 10),
+        ];
+
+        let mut windows = assign_to_windows(files.iter(), 3);
+
+        let picker = TwcsPicker {
+            trigger_file_num: 4,
+            time_window_seconds: Some(3),
+            max_output_file_size: None,
+            append_mode: false,
+            max_background_tasks: Some(max_background_tasks),
+        };
+
+        let active_window = find_latest_window_in_seconds(files.iter(), 3);
+        let output = picker.build_output(RegionId::from_u64(123), &mut windows, active_window);
+
+        // Should have at most max_background_tasks outputs
+        assert!(
+            output.len() <= max_background_tasks,
+            "Output should be truncated to max_background_tasks: expected <= {}, got {}",
+            max_background_tasks,
+            output.len()
+        );
+
+        // Without max_background_tasks, should have more outputs
+        let picker_no_limit = TwcsPicker {
+            trigger_file_num: 4,
+            time_window_seconds: Some(3),
+            max_output_file_size: None,
+            append_mode: false,
+            max_background_tasks: None,
+        };
+
+        let mut windows_no_limit = assign_to_windows(files.iter(), 3);
+        let output_no_limit = picker_no_limit.build_output(
+            RegionId::from_u64(123),
+            &mut windows_no_limit,
+            active_window,
+        );
+
+        // Without limit, should have more outputs (if there are enough windows)
+        if output_no_limit.len() > max_background_tasks {
+            assert!(
+                output_no_limit.len() > output.len(),
+                "Without limit should have more outputs than with limit"
+            );
+        }
+    }
+
+    #[test]
+    fn test_max_background_tasks_no_truncation_when_under_limit() {
+        let file_ids = (0..4).map(|_| FileId::random()).collect::<Vec<_>>();
+        let max_background_tasks = 10; // Larger than expected outputs
+
+        // Create files in one window that will generate one compaction output
+        let files = [
+            new_file_handle_with_sequence(file_ids[0], 0, 999, 0, 1),
+            new_file_handle_with_sequence(file_ids[1], 0, 999, 0, 2),
+            new_file_handle_with_sequence(file_ids[2], 0, 999, 0, 3),
+            new_file_handle_with_sequence(file_ids[3], 0, 999, 0, 4),
+        ];
+
+        let mut windows = assign_to_windows(files.iter(), 3);
+
+        let picker = TwcsPicker {
+            trigger_file_num: 4,
+            time_window_seconds: Some(3),
+            max_output_file_size: None,
+            append_mode: false,
+            max_background_tasks: Some(max_background_tasks),
+        };
+
+        let active_window = find_latest_window_in_seconds(files.iter(), 3);
+        let output = picker.build_output(RegionId::from_u64(123), &mut windows, active_window);
+
+        // Should have all outputs since we're under the limit
+        assert!(
+            output.len() <= max_background_tasks,
+            "Output should be within limit"
+        );
+        // Should have at least one output
+        assert!(!output.is_empty(), "Should have at least one output");
+    }
+
+    #[test]
+    fn test_pick_multiple_runs() {
+        common_telemetry::init_default_ut_logging();
+
+        let num_files = 8;
+        let file_ids = (0..num_files).map(|_| FileId::random()).collect::<Vec<_>>();
+
+        // Create files with different sequences so they form multiple runs
+        let files: Vec<_> = file_ids
+            .iter()
+            .enumerate()
+            .map(|(idx, file_id)| {
+                new_file_handle_with_size_and_sequence(
+                    *file_id,
+                    0,
+                    999,
+                    0,
+                    (idx + 1) as u64,
+                    1024 * 1024,
+                )
+            })
+            .collect();
+
+        let mut windows = assign_to_windows(files.iter(), 3);
+
+        let picker = TwcsPicker {
+            trigger_file_num: 4,
+            time_window_seconds: Some(3),
+            max_output_file_size: None,
+            append_mode: false,
+            max_background_tasks: None,
+        };
+
+        let active_window = find_latest_window_in_seconds(files.iter(), 3);
+        let output = picker.build_output(RegionId::from_u64(123), &mut windows, active_window);
+
+        assert_eq!(1, output.len());
+        assert_eq!(output[0].inputs.len(), 2);
+    }
+
+    #[test]
+    fn test_limit_max_input_files() {
+        common_telemetry::init_default_ut_logging();
+
+        let num_files = 50;
+        let file_ids = (0..num_files).map(|_| FileId::random()).collect::<Vec<_>>();
+
+        // Create files with different sequences so they form 2 runs
+        let files: Vec<_> = file_ids
+            .iter()
+            .enumerate()
+            .map(|(idx, file_id)| {
+                new_file_handle_with_size_and_sequence(
+                    *file_id,
+                    (idx / 2 * 10) as i64,
+                    (idx / 2 * 10 + 5) as i64,
+                    0,
+                    (idx + 1) as u64,
+                    1024 * 1024,
+                )
+            })
+            .collect();
+
+        let mut windows = assign_to_windows(files.iter(), 3);
+
+        let picker = TwcsPicker {
+            trigger_file_num: 4,
+            time_window_seconds: Some(3),
+            max_output_file_size: None,
+            append_mode: false,
+            max_background_tasks: None,
+        };
+
+        let active_window = find_latest_window_in_seconds(files.iter(), 3);
+        let output = picker.build_output(RegionId::from_u64(123), &mut windows, active_window);
+
+        assert_eq!(1, output.len());
+        assert_eq!(output[0].inputs.len(), 32);
+    }
+
    // TODO(hl): TTL tester that checks if get_expired_ssts function works as expected.
 }
--- a/src/mito2/src/engine/basic_test.rs
+++ b/src/mito2/src/engine/basic_test.rs
@@ -15,6 +15,7 @@
 //! Basic tests for mito engine.

 use std::collections::HashMap;
+use std::sync::atomic::Ordering;

 use api::v1::value::ValueData;
 use api::v1::{Rows, SemanticType};
@@ -86,11 +87,15 @@ async fn test_write_to_region() {
        rows: build_rows(0, 42),
    };
    put_rows(&engine, region_id, rows).await;
+    let region = engine.get_region(region_id).unwrap();
+    assert!(region.written_bytes.load(Ordering::Relaxed) > 0);
 }

 #[apply(multiple_log_store_factories)]

 async fn test_region_replay(factory: Option<LogStoreFactory>) {
+    use std::sync::atomic::Ordering;
+
    use common_wal::options::{KafkaWalOptions, WalOptions};

    common_telemetry::init_default_ut_logging();
@@ -155,6 +160,10 @@ async fn test_region_replay(factory: Option<LogStoreFactory>) {
        .unwrap();
    assert_eq!(0, result.affected_rows);

+    // The replay won't update the write bytes rate meter.
+    let region = engine.get_region(region_id).unwrap();
+    assert_eq!(region.written_bytes.load(Ordering::Relaxed), 0);
+
    let request = ScanRequest::default();
    let stream = engine.scan_to_stream(region_id, request).await.unwrap();
    let batches = RecordBatches::try_collect(stream).await.unwrap();
--- a/src/mito2/src/flush.rs
+++ b/src/mito2/src/flush.rs
@@ -25,7 +25,7 @@ use store_api::storage::RegionId;
 use strum::IntoStaticStr;
 use tokio::sync::{mpsc, watch};

-use crate::access_layer::{AccessLayerRef, OperationType, SstWriteRequest};
+use crate::access_layer::{AccessLayerRef, Metrics, OperationType, SstWriteRequest, WriteType};
 use crate::cache::CacheManagerRef;
 use crate::config::MitoConfig;
 use crate::error::{
@@ -341,6 +341,7 @@ impl RegionFlushTask {
        let mut file_metas = Vec::with_capacity(memtables.len());
        let mut flushed_bytes = 0;
        let mut series_count = 0;
+        let mut flush_metrics = Metrics::new(WriteType::Flush);
        for mem in memtables {
            if mem.is_empty() {
                // Skip empty memtables.
@@ -367,14 +368,15 @@ impl RegionFlushTask {
                bloom_filter_index_config: self.engine_config.bloom_filter_index.clone(),
            };

-            let ssts_written = self
+            let (ssts_written, metrics) = self
                .access_layer
-                .write_sst(write_request, &write_opts)
+                .write_sst(write_request, &write_opts, WriteType::Flush)
                .await?;
            if ssts_written.is_empty() {
                // No data written.
                continue;
            }
+            flush_metrics = flush_metrics.merge(metrics);

            file_metas.extend(ssts_written.into_iter().map(|sst_info| {
                flushed_bytes += sst_info.file_size;
@@ -399,13 +401,15 @@ impl RegionFlushTask {

        let file_ids: Vec<_> = file_metas.iter().map(|f| f.file_id).collect();
        info!(
-            "Successfully flush memtables, region: {}, reason: {}, files: {:?}, series count: {}, cost: {:?}s",
+            "Successfully flush memtables, region: {}, reason: {}, files: {:?}, series count: {}, cost: {:?}, metrics: {:?}",
            self.region_id,
            self.reason.as_str(),
            file_ids,
            series_count,
            timer.stop_and_record(),
+            flush_metrics,
        );
+        flush_metrics.observe();

        let edit = RegionEdit {
            files_to_add: file_metas,
--- a/src/mito2/src/manifest/manager.rs
+++ b/src/mito2/src/manifest/manager.rs
@@ -76,7 +76,7 @@ pub struct RegionManifestOptions {
 ///     -RegionMetadataRef metadata
 /// }
 /// class RegionEdit {
-///     -VersionNumber regoin_version
+///     -VersionNumber region_version
 ///     -Vec~FileMeta~ files_to_add
 ///     -Vec~FileMeta~ files_to_remove
 ///     -SequenceNumber flushed_sequence
--- a/src/mito2/src/metrics.rs
+++ b/src/mito2/src/metrics.rs
@@ -119,6 +119,14 @@ lazy_static! {

    // Compaction metrics
    /// Timer of different stages in compaction.
+    /// - pick
+    /// - merge (in parallel)
+    ///   - iter_source
+    ///   - write_batch
+    ///   - update_index
+    ///   - upload_parquet
+    ///   - upload puffin
+    /// - write_manifest
    pub static ref COMPACTION_STAGE_ELAPSED: HistogramVec = register_histogram_vec!(
        "greptime_mito_compaction_stage_elapsed",
        "mito compaction stage elapsed",
--- a/src/mito2/src/region.rs
+++ b/src/mito2/src/region.rs
@@ -129,6 +129,8 @@ pub struct MitoRegion {
    /// There are no WAL entries in range [flushed_entry_id, topic_latest_entry_id] for current region,
    /// which means these WAL entries maybe able to be pruned up to `topic_latest_entry_id`.
    pub(crate) topic_latest_entry_id: AtomicU64,
+    /// The total bytes written to the region.
+    pub(crate) written_bytes: Arc<AtomicU64>,
    /// Memtable builder for the region.
    pub(crate) memtable_builder: MemtableBuilderRef,
    /// manifest stats
@@ -313,6 +315,7 @@ impl MitoRegion {
        let manifest_version = self.stats.manifest_version();

        let topic_latest_entry_id = self.topic_latest_entry_id.load(Ordering::Relaxed);
+        let written_bytes = self.written_bytes.load(Ordering::Relaxed);

        RegionStatistic {
            num_rows,
@@ -327,6 +330,7 @@ impl MitoRegion {
            },
            data_topic_latest_entry_id: topic_latest_entry_id,
            metadata_topic_latest_entry_id: topic_latest_entry_id,
+            written_bytes,
        }
    }

--- a/src/mito2/src/region/opener.rs
+++ b/src/mito2/src/region/opener.rs
@@ -277,6 +277,7 @@ impl RegionOpener {
            time_provider: self.time_provider.clone(),
            topic_latest_entry_id: AtomicU64::new(0),
            memtable_builder,
+            written_bytes: Arc::new(AtomicU64::new(0)),
            stats: self.stats,
        })
    }
@@ -455,6 +456,7 @@ impl RegionOpener {
            last_compaction_millis: AtomicI64::new(now),
            time_provider: self.time_provider.clone(),
            topic_latest_entry_id: AtomicU64::new(0),
+            written_bytes: Arc::new(AtomicU64::new(0)),
            memtable_builder,
            stats: self.stats.clone(),
        };
@@ -634,7 +636,7 @@ where
        last_entry_id = last_entry_id.max(entry_id);

        let mut region_write_ctx =
-            RegionWriteCtx::new(region_id, version_control, provider.clone());
+            RegionWriteCtx::new(region_id, version_control, provider.clone(), None);
        for mutation in entry.mutations {
            rows_replayed += mutation
                .rows
--- a/src/mito2/src/region/version.rs
+++ b/src/mito2/src/region/version.rs
@@ -354,7 +354,7 @@ impl VersionBuilder {
        self
    }

-    /// Sets truncated entty id.
+    /// Sets truncated entry id.
    pub(crate) fn truncated_entry_id(mut self, entry_id: Option<EntryId>) -> Self {
        self.truncated_entry_id = entry_id;
        self
--- a/src/mito2/src/region_write_ctx.rs
+++ b/src/mito2/src/region_write_ctx.rs
@@ -13,6 +13,7 @@
 // limitations under the License.

 use std::mem;
+use std::sync::atomic::{AtomicU64, Ordering};
 use std::sync::Arc;

 use api::v1::{BulkWalEntry, Mutation, OpType, Rows, WalEntry, WriteHint};
@@ -106,6 +107,8 @@ pub(crate) struct RegionWriteCtx {
    pub(crate) put_num: usize,
    /// Rows to delete.
    pub(crate) delete_num: usize,
+    /// The total bytes written to the region.
+    pub(crate) written_bytes: Option<Arc<AtomicU64>>,
 }

 impl RegionWriteCtx {
@@ -114,6 +117,7 @@ impl RegionWriteCtx {
        region_id: RegionId,
        version_control: &VersionControlRef,
        provider: Provider,
+        written_bytes: Option<Arc<AtomicU64>>,
    ) -> RegionWriteCtx {
        let VersionControlData {
            version,
@@ -136,6 +140,7 @@ impl RegionWriteCtx {
            put_num: 0,
            delete_num: 0,
            bulk_parts: vec![],
+            written_bytes,
        }
    }

@@ -214,6 +219,12 @@ impl RegionWriteCtx {
        }

        let mutable = self.version.memtables.mutable.clone();
+        let prev_memory_usage = if self.written_bytes.is_some() {
+            Some(mutable.memory_usage())
+        } else {
+            None
+        };
+
        let mutations = mem::take(&mut self.wal_entry.mutations)
            .into_iter()
            .enumerate()
@@ -246,6 +257,11 @@ impl RegionWriteCtx {
            }
        }

+        if let Some(written_bytes) = &self.written_bytes {
+            let new_memory_usage = mutable.memory_usage();
+            let bytes = new_memory_usage.saturating_sub(prev_memory_usage.unwrap_or_default());
+            written_bytes.fetch_add(bytes as u64, Ordering::Relaxed);
+        }
        // Updates region sequence and entry id. Since we stores last sequence and entry id in region, we need
        // to decrease `next_sequence` and `next_entry_id` by 1.
        self.version_control
@@ -271,6 +287,13 @@ impl RegionWriteCtx {
            .with_label_values(&["write_bulk"])
            .start_timer();

+        let mutable_memtable = &self.version.memtables.mutable;
+        let prev_memory_usage = if self.written_bytes.is_some() {
+            Some(mutable_memtable.memory_usage())
+        } else {
+            None
+        };
+
        if self.bulk_parts.len() == 1 {
            let part = self.bulk_parts.swap_remove(0);
            let num_rows = part.num_rows();
@@ -300,6 +323,11 @@ impl RegionWriteCtx {
            }
        }

+        if let Some(written_bytes) = &self.written_bytes {
+            let new_memory_usage = mutable_memtable.memory_usage();
+            let bytes = new_memory_usage.saturating_sub(prev_memory_usage.unwrap_or_default());
+            written_bytes.fetch_add(bytes as u64, Ordering::Relaxed);
+        }
        self.version_control
            .set_sequence_and_entry_id(self.next_sequence - 1, self.next_entry_id - 1);
    }
--- a/src/mito2/src/sst/file.rs
+++ b/src/mito2/src/sst/file.rs
@@ -289,6 +289,10 @@ impl FileHandle {
    pub fn num_rows(&self) -> usize {
        self.inner.meta.num_rows as usize
    }
+
+    pub fn level(&self) -> Level {
+        self.inner.meta.level
+    }
 }

 /// Inner data of [FileHandle].
--- a/src/mito2/src/sst/file_purger.rs
+++ b/src/mito2/src/sst/file_purger.rs
@@ -131,6 +131,14 @@ impl FilePurger for LocalFilePurger {
                error!(e; "Failed to purge stager with index file, file_id: {}, region: {}",
                    file_meta.file_id, file_meta.region_id);
            }
+            let file_id = file_meta.file_id;
+            if let Err(e) = sst_layer
+                .intermediate_manager()
+                .prune_sst_dir(&file_meta.region_id, &file_id)
+                .await
+            {
+                error!(e; "Failed to prune intermediate sst directory, region_id: {}, file_id: {}", file_meta.region_id, file_id);
+            }
        })) {
            error!(e; "Failed to schedule the file purge request");
        }
--- a/src/mito2/src/sst/index.rs
+++ b/src/mito2/src/sst/index.rs
@@ -110,6 +110,7 @@ pub struct Indexer {
    last_mem_fulltext_index: usize,
    bloom_filter_indexer: Option<BloomFilterIndexer>,
    last_mem_bloom_filter: usize,
+    intermediate_manager: Option<IntermediateManager>,
 }

 impl Indexer {
@@ -196,6 +197,7 @@ impl IndexerBuilder for IndexerBuilderImpl {
        indexer.inverted_indexer = self.build_inverted_indexer(file_id);
        indexer.fulltext_indexer = self.build_fulltext_indexer(file_id).await;
        indexer.bloom_filter_indexer = self.build_bloom_filter_indexer(file_id);
+        indexer.intermediate_manager = Some(self.intermediate_manager.clone());
        if indexer.inverted_indexer.is_none()
            && indexer.fulltext_indexer.is_none()
            && indexer.bloom_filter_indexer.is_none()
--- a/src/mito2/src/sst/index/fulltext_index/applier.rs
+++ b/src/mito2/src/sst/index/fulltext_index/applier.rs
@@ -22,7 +22,8 @@ use common_telemetry::warn;
 use index::bloom_filter::applier::{BloomFilterApplier, InListPredicate};
 use index::bloom_filter::reader::BloomFilterReaderImpl;
 use index::fulltext_index::search::{FulltextIndexSearcher, RowId, TantivyFulltextIndexSearcher};
-use index::fulltext_index::Config;
+use index::fulltext_index::tokenizer::{ChineseTokenizer, EnglishTokenizer, Tokenizer};
+use index::fulltext_index::{Analyzer, Config};
 use object_store::ObjectStore;
 use puffin::puffin_manager::cache::PuffinMetadataCacheRef;
 use puffin::puffin_manager::{GuardWithMetadata, PuffinManager, PuffinReader};
@@ -393,21 +394,7 @@ impl FulltextIndexApplier {
                // lowercased terms are not indexed
                continue;
            }
-
-            let ts = term
-                .term
-                .split(|c: char| !c.is_alphanumeric())
-                .filter(|&t| !t.is_empty())
-                .map(|t| {
-                    if !config.case_sensitive {
-                        t.to_lowercase()
-                    } else {
-                        t.to_string()
-                    }
-                    .into_bytes()
-                });
-
-            probes.extend(ts);
+            probes.extend(Self::term_to_probes(&term.term, config));
        }

        probes
@@ -417,6 +404,22 @@ impl FulltextIndexApplier {
            })
            .collect::<Vec<_>>()
    }
+
+    fn term_to_probes<'a>(term: &'a str, config: &'a Config) -> impl Iterator<Item = Vec<u8>> + 'a {
+        let tokens = match config.analyzer {
+            Analyzer::English => EnglishTokenizer {}.tokenize(term),
+            Analyzer::Chinese => ChineseTokenizer {}.tokenize(term),
+        };
+
+        tokens.into_iter().map(|t| {
+            if !config.case_sensitive {
+                t.to_lowercase()
+            } else {
+                t.to_string()
+            }
+            .into_bytes()
+        })
+    }
 }

 /// The source of the index.
--- a/src/mito2/src/sst/index/indexer/abort.rs
+++ b/src/mito2/src/sst/index/indexer/abort.rs
@@ -21,6 +21,7 @@ impl Indexer {
        self.do_abort_inverted_index().await;
        self.do_abort_fulltext_index().await;
        self.do_abort_bloom_filter().await;
+        self.do_prune_intm_sst_dir().await;
        self.puffin_manager = None;
    }

--- a/src/mito2/src/sst/index/indexer/finish.rs
+++ b/src/mito2/src/sst/index/indexer/finish.rs
@@ -53,6 +53,7 @@ impl Indexer {
            return IndexOutput::default();
        }

+        self.do_prune_intm_sst_dir().await;
        output.file_size = self.do_finish_puffin_writer(writer).await;
        output
    }
@@ -266,4 +267,12 @@ impl Indexer {
        output.row_count = row_count;
        output.columns = column_ids;
    }
+
+    pub(crate) async fn do_prune_intm_sst_dir(&mut self) {
+        if let Some(manager) = self.intermediate_manager.take() {
+            if let Err(e) = manager.prune_sst_dir(&self.region_id, &self.file_id).await {
+                warn!(e; "Failed to prune intermediate SST directory, region_id: {}, file_id: {}", self.region_id, self.file_id);
+            }
+        }
+    }
 }
--- a/src/mito2/src/sst/index/intermediate.rs
+++ b/src/mito2/src/sst/index/intermediate.rs
@@ -54,14 +54,22 @@ impl IntermediateManager {
            aux_path.as_ref()
        );

+        // Remove the intermediate directory on bankground
+        let aux_pb = PathBuf::from(aux_path.as_ref());
+        let intm_dir = aux_pb.join(INTERMEDIATE_DIR);
+        let deleted_dir = intm_dir.with_extension(format!("deleted-{}", Uuid::new_v4()));
+        if let Err(err) = tokio::fs::rename(&intm_dir, &deleted_dir).await {
+            warn!(err; "Failed to rename intermediate directory");
+        }
+        tokio::spawn(async move {
+            if let Err(err) = tokio::fs::remove_dir_all(deleted_dir).await {
+                warn!(err; "Failed to remove intermediate directory");
+            }
+        });
+
        let store = new_fs_cache_store(&normalize_dir(aux_path.as_ref())).await?;
        let store = InstrumentedStore::new(store);

-        // Remove all garbage intermediate files from previous runs.
-        if let Err(err) = store.remove_all(INTERMEDIATE_DIR).await {
-            warn!(err; "Failed to remove garbage intermediate files");
-        }
-
        Ok(Self {
            base_dir: PathBuf::from(aux_path.as_ref()),
            store,
@@ -94,6 +102,24 @@ impl IntermediateManager {
            .join(sst_file_id.to_string())
            .join(format!("fulltext-{column_id}-{uuid}"))
    }
+
+    /// Prunes the intermediate directory for SST files.
+    pub(crate) async fn prune_sst_dir(
+        &self,
+        region_id: &RegionId,
+        sst_file_id: &FileId,
+    ) -> Result<()> {
+        let region_id = region_id.as_u64();
+        let sst_dir = format!("{INTERMEDIATE_DIR}/{region_id}/{sst_file_id}/");
+        self.store.remove_all(&sst_dir).await
+    }
+
+    /// Prunes the intermediate directory for region files.
+    pub(crate) async fn prune_region_dir(&self, region_id: &RegionId) -> Result<()> {
+        let region_id = region_id.as_u64();
+        let region_dir = format!("{INTERMEDIATE_DIR}/{region_id}/");
+        self.store.remove_all(&region_dir).await
+    }
 }

 /// `IntermediateLocation` produces paths for intermediate files
@@ -268,6 +294,60 @@ mod tests {
            .unwrap());
    }

+    #[tokio::test]
+    async fn test_cleanup_dir() {
+        let temp_dir = temp_dir::create_temp_dir("test_cleanup_dir_");
+
+        let region_id = RegionId::new(0, 0);
+        let sst_file_id = FileId::random();
+        let region_dir = temp_dir
+            .path()
+            .join(INTERMEDIATE_DIR)
+            .join(region_id.as_u64().to_string());
+        let sst_dir = region_dir.join(sst_file_id.to_string());
+
+        let path = temp_dir.path().to_str().unwrap();
+        let manager = IntermediateManager::init_fs(path).await.unwrap();
+
+        let location = IntermediateLocation::new(&region_id, &sst_file_id);
+        let temp_file_provider = TempFileProvider::new(location, manager.clone());
+
+        let mut f1 = temp_file_provider
+            .create("sky", "000000000000")
+            .await
+            .unwrap();
+        f1.write_all(b"hello").await.unwrap();
+        f1.flush().await.unwrap();
+        f1.close().await.unwrap();
+
+        let mut f2 = temp_file_provider
+            .create("sky", "000000000001")
+            .await
+            .unwrap();
+        f2.write_all(b"world").await.unwrap();
+        f2.flush().await.unwrap();
+        f2.close().await.unwrap();
+
+        temp_file_provider.cleanup().await.unwrap();
+
+        // sst_dir and region_dir still exists
+        assert!(tokio::fs::try_exists(&sst_dir).await.unwrap());
+        assert!(tokio::fs::try_exists(&region_dir).await.unwrap());
+
+        // sst_dir should be deleted, region_dir still exists
+        manager
+            .prune_sst_dir(&region_id, &sst_file_id)
+            .await
+            .unwrap();
+        assert!(tokio::fs::try_exists(&region_dir).await.unwrap());
+        assert!(!tokio::fs::try_exists(&sst_dir).await.unwrap());
+
+        // sst_dir, region_dir should be deleted
+        manager.prune_region_dir(&region_id).await.unwrap();
+        assert!(!tokio::fs::try_exists(&sst_dir).await.unwrap());
+        assert!(!tokio::fs::try_exists(&region_dir).await.unwrap());
+    }
+
    #[test]
    fn test_intermediate_location() {
        let sst_file_id = FileId::random();
--- a/src/mito2/src/sst/index/inverted_index/applier/builder.rs
+++ b/src/mito2/src/sst/index/inverted_index/applier/builder.rs
@@ -338,7 +338,7 @@ mod tests {

    #[test]
    fn test_collect_and_basic() {
-        let (_d, facotry) = PuffinManagerFactory::new_for_test_block("test_collect_and_basic_");
+        let (_d, factory) = PuffinManagerFactory::new_for_test_block("test_collect_and_basic_");

        let metadata = test_region_metadata();
        let mut builder = InvertedIndexApplierBuilder::new(
@@ -346,7 +346,7 @@ mod tests {
            test_object_store(),
            &metadata,
            HashSet::from_iter([1, 2, 3]),
-            facotry,
+            factory,
        );

        let expr = Expr::BinaryExpr(BinaryExpr {
--- a/src/mito2/src/sst/index/inverted_index/applier/builder/between.rs
+++ b/src/mito2/src/sst/index/inverted_index/applier/builder/between.rs
@@ -70,14 +70,14 @@ mod tests {

    #[test]
    fn test_collect_between_basic() {
-        let (_d, facotry) = PuffinManagerFactory::new_for_test_block("test_collect_between_basic_");
+        let (_d, factory) = PuffinManagerFactory::new_for_test_block("test_collect_between_basic_");
        let metadata = test_region_metadata();
        let mut builder = InvertedIndexApplierBuilder::new(
            "test".to_string(),
            test_object_store(),
            &metadata,
            HashSet::from_iter([1, 2, 3]),
-            facotry,
+            factory,
        );

        let between = Between {
@@ -110,7 +110,7 @@ mod tests {

    #[test]
    fn test_collect_between_negated() {
-        let (_d, facotry) =
+        let (_d, factory) =
            PuffinManagerFactory::new_for_test_block("test_collect_between_negated_");
        let metadata = test_region_metadata();
        let mut builder = InvertedIndexApplierBuilder::new(
@@ -118,7 +118,7 @@ mod tests {
            test_object_store(),
            &metadata,
            HashSet::from_iter([1, 2, 3]),
-            facotry,
+            factory,
        );

        let between = Between {
@@ -134,7 +134,7 @@ mod tests {

    #[test]
    fn test_collect_between_field_column() {
-        let (_d, facotry) =
+        let (_d, factory) =
            PuffinManagerFactory::new_for_test_block("test_collect_between_field_column_");
        let metadata = test_region_metadata();
        let mut builder = InvertedIndexApplierBuilder::new(
@@ -142,7 +142,7 @@ mod tests {
            test_object_store(),
            &metadata,
            HashSet::from_iter([1, 2, 3]),
-            facotry,
+            factory,
        );

        let between = Between {
@@ -175,7 +175,7 @@ mod tests {

    #[test]
    fn test_collect_between_type_mismatch() {
-        let (_d, facotry) =
+        let (_d, factory) =
            PuffinManagerFactory::new_for_test_block("test_collect_between_type_mismatch_");
        let metadata = test_region_metadata();
        let mut builder = InvertedIndexApplierBuilder::new(
@@ -183,7 +183,7 @@ mod tests {
            test_object_store(),
            &metadata,
            HashSet::from_iter([1, 2, 3]),
-            facotry,
+            factory,
        );

        let between = Between {
@@ -200,7 +200,7 @@ mod tests {

    #[test]
    fn test_collect_between_nonexistent_column() {
-        let (_d, facotry) =
+        let (_d, factory) =
            PuffinManagerFactory::new_for_test_block("test_collect_between_nonexistent_column_");
        let metadata = test_region_metadata();
        let mut builder = InvertedIndexApplierBuilder::new(
@@ -208,7 +208,7 @@ mod tests {
            test_object_store(),
            &metadata,
            HashSet::from_iter([1, 2, 3]),
-            facotry,
+            factory,
        );

        let between = Between {
--- a/src/mito2/src/sst/index/inverted_index/applier/builder/comparison.rs
+++ b/src/mito2/src/sst/index/inverted_index/applier/builder/comparison.rs
@@ -225,7 +225,7 @@ mod tests {
            ),
        ];

-        let (_d, facotry) =
+        let (_d, factory) =
            PuffinManagerFactory::new_for_test_block("test_collect_comparison_basic_");
        let metadata = test_region_metadata();
        let mut builder = InvertedIndexApplierBuilder::new(
@@ -233,7 +233,7 @@ mod tests {
            test_object_store(),
            &metadata,
            HashSet::from_iter([1, 2, 3]),
-            facotry,
+            factory,
        );

        for ((left, op, right), _) in &cases {
@@ -252,7 +252,7 @@ mod tests {

    #[test]
    fn test_collect_comparison_type_mismatch() {
-        let (_d, facotry) =
+        let (_d, factory) =
            PuffinManagerFactory::new_for_test_block("test_collect_comparison_type_mismatch_");
        let metadata = test_region_metadata();
        let mut builder = InvertedIndexApplierBuilder::new(
@@ -260,7 +260,7 @@ mod tests {
            test_object_store(),
            &metadata,
            HashSet::from_iter([1, 2, 3]),
-            facotry,
+            factory,
        );

        let res = builder.collect_comparison_expr(&tag_column(), &Operator::Lt, &int64_lit(10));
@@ -270,7 +270,7 @@ mod tests {

    #[test]
    fn test_collect_comparison_field_column() {
-        let (_d, facotry) =
+        let (_d, factory) =
            PuffinManagerFactory::new_for_test_block("test_collect_comparison_field_column_");
        let metadata = test_region_metadata();
        let mut builder = InvertedIndexApplierBuilder::new(
@@ -278,7 +278,7 @@ mod tests {
            test_object_store(),
            &metadata,
            HashSet::from_iter([1, 2, 3]),
-            facotry,
+            factory,
        );

        builder
@@ -303,7 +303,7 @@ mod tests {

    #[test]
    fn test_collect_comparison_nonexistent_column() {
-        let (_d, facotry) =
+        let (_d, factory) =
            PuffinManagerFactory::new_for_test_block("test_collect_comparison_nonexistent_column_");
        let metadata = test_region_metadata();
        let mut builder = InvertedIndexApplierBuilder::new(
@@ -311,7 +311,7 @@ mod tests {
            test_object_store(),
            &metadata,
            HashSet::from_iter([1, 2, 3]),
-            facotry,
+            factory,
        );

        let res = builder.collect_comparison_expr(
--- a/src/mito2/src/sst/index/inverted_index/applier/builder/eq_list.rs
+++ b/src/mito2/src/sst/index/inverted_index/applier/builder/eq_list.rs
@@ -134,14 +134,14 @@ mod tests {

    #[test]
    fn test_collect_eq_basic() {
-        let (_d, facotry) = PuffinManagerFactory::new_for_test_block("test_collect_eq_basic_");
+        let (_d, factory) = PuffinManagerFactory::new_for_test_block("test_collect_eq_basic_");
        let metadata = test_region_metadata();
        let mut builder = InvertedIndexApplierBuilder::new(
            "test".to_string(),
            test_object_store(),
            &metadata,
            HashSet::from_iter([1, 2, 3]),
-            facotry,
+            factory,
        );

        builder
@@ -169,7 +169,7 @@ mod tests {

    #[test]
    fn test_collect_eq_field_column() {
-        let (_d, facotry) =
+        let (_d, factory) =
            PuffinManagerFactory::new_for_test_block("test_collect_eq_field_column_");
        let metadata = test_region_metadata();
        let mut builder = InvertedIndexApplierBuilder::new(
@@ -177,7 +177,7 @@ mod tests {
            test_object_store(),
            &metadata,
            HashSet::from_iter([1, 2, 3]),
-            facotry,
+            factory,
        );

        builder
@@ -196,7 +196,7 @@ mod tests {

    #[test]
    fn test_collect_eq_nonexistent_column() {
-        let (_d, facotry) =
+        let (_d, factory) =
            PuffinManagerFactory::new_for_test_block("test_collect_eq_nonexistent_column_");
        let metadata = test_region_metadata();
        let mut builder = InvertedIndexApplierBuilder::new(
@@ -204,7 +204,7 @@ mod tests {
            test_object_store(),
            &metadata,
            HashSet::from_iter([1, 2, 3]),
-            facotry,
+            factory,
        );

        let res = builder.collect_eq(&nonexistent_column(), &string_lit("abc"));
@@ -214,7 +214,7 @@ mod tests {

    #[test]
    fn test_collect_eq_type_mismatch() {
-        let (_d, facotry) =
+        let (_d, factory) =
            PuffinManagerFactory::new_for_test_block("test_collect_eq_type_mismatch_");
        let metadata = test_region_metadata();
        let mut builder = InvertedIndexApplierBuilder::new(
@@ -222,7 +222,7 @@ mod tests {
            test_object_store(),
            &metadata,
            HashSet::from_iter([1, 2, 3]),
-            facotry,
+            factory,
        );

        let res = builder.collect_eq(&tag_column(), &int64_lit(1));
@@ -232,7 +232,7 @@ mod tests {

    #[test]
    fn test_collect_or_eq_list_basic() {
-        let (_d, facotry) =
+        let (_d, factory) =
            PuffinManagerFactory::new_for_test_block("test_collect_or_eq_list_basic_");
        let metadata = test_region_metadata();
        let mut builder = InvertedIndexApplierBuilder::new(
@@ -240,7 +240,7 @@ mod tests {
            test_object_store(),
            &metadata,
            HashSet::from_iter([1, 2, 3]),
-            facotry,
+            factory,
        );

        let eq_expr = DfExpr::BinaryExpr(BinaryExpr {
@@ -289,7 +289,7 @@ mod tests {

    #[test]
    fn test_collect_or_eq_list_invalid_op() {
-        let (_d, facotry) =
+        let (_d, factory) =
            PuffinManagerFactory::new_for_test_block("test_collect_or_eq_list_invalid_op_");
        let metadata = test_region_metadata();
        let mut builder = InvertedIndexApplierBuilder::new(
@@ -297,7 +297,7 @@ mod tests {
            test_object_store(),
            &metadata,
            HashSet::from_iter([1, 2, 3]),
-            facotry,
+            factory,
        );

        let eq_expr = DfExpr::BinaryExpr(BinaryExpr {
@@ -325,7 +325,7 @@ mod tests {

    #[test]
    fn test_collect_or_eq_list_multiple_columns() {
-        let (_d, facotry) =
+        let (_d, factory) =
            PuffinManagerFactory::new_for_test_block("test_collect_or_eq_list_multiple_columns_");
        let metadata = test_region_metadata();
        let mut builder = InvertedIndexApplierBuilder::new(
@@ -333,7 +333,7 @@ mod tests {
            test_object_store(),
            &metadata,
            HashSet::from_iter([1, 2, 3]),
-            facotry,
+            factory,
        );

        let eq_expr = DfExpr::BinaryExpr(BinaryExpr {
--- a/src/mito2/src/sst/index/inverted_index/applier/builder/in_list.rs
+++ b/src/mito2/src/sst/index/inverted_index/applier/builder/in_list.rs
@@ -65,14 +65,14 @@ mod tests {

    #[test]
    fn test_collect_in_list_basic() {
-        let (_d, facotry) = PuffinManagerFactory::new_for_test_block("test_collect_in_list_basic_");
+        let (_d, factory) = PuffinManagerFactory::new_for_test_block("test_collect_in_list_basic_");
        let metadata = test_region_metadata();
        let mut builder = InvertedIndexApplierBuilder::new(
            "test".to_string(),
            test_object_store(),
            &metadata,
            HashSet::from_iter([1, 2, 3]),
-            facotry,
+            factory,
        );

        let in_list = InList {
@@ -95,7 +95,7 @@ mod tests {

    #[test]
    fn test_collect_in_list_negated() {
-        let (_d, facotry) =
+        let (_d, factory) =
            PuffinManagerFactory::new_for_test_block("test_collect_in_list_negated_");
        let metadata = test_region_metadata();
        let mut builder = InvertedIndexApplierBuilder::new(
@@ -103,7 +103,7 @@ mod tests {
            test_object_store(),
            &metadata,
            HashSet::from_iter([1, 2, 3]),
-            facotry,
+            factory,
        );

        let in_list = InList {
@@ -118,7 +118,7 @@ mod tests {

    #[test]
    fn test_collect_in_list_field_column() {
-        let (_d, facotry) =
+        let (_d, factory) =
            PuffinManagerFactory::new_for_test_block("test_collect_in_list_field_column_");
        let metadata = test_region_metadata();
        let mut builder = InvertedIndexApplierBuilder::new(
@@ -126,7 +126,7 @@ mod tests {
            test_object_store(),
            &metadata,
            HashSet::from_iter([1, 2, 3]),
-            facotry,
+            factory,
        );

        let in_list = InList {
@@ -149,7 +149,7 @@ mod tests {

    #[test]
    fn test_collect_in_list_type_mismatch() {
-        let (_d, facotry) =
+        let (_d, factory) =
            PuffinManagerFactory::new_for_test_block("test_collect_in_list_type_mismatch_");
        let metadata = test_region_metadata();
        let mut builder = InvertedIndexApplierBuilder::new(
@@ -157,7 +157,7 @@ mod tests {
            test_object_store(),
            &metadata,
            HashSet::from_iter([1, 2, 3]),
-            facotry,
+            factory,
        );

        let in_list = InList {
@@ -173,7 +173,7 @@ mod tests {

    #[test]
    fn test_collect_in_list_nonexistent_column() {
-        let (_d, facotry) =
+        let (_d, factory) =
            PuffinManagerFactory::new_for_test_block("test_collect_in_list_nonexistent_column_");

        let metadata = test_region_metadata();
@@ -182,7 +182,7 @@ mod tests {
            test_object_store(),
            &metadata,
            HashSet::from_iter([1, 2, 3]),
-            facotry,
+            factory,
        );

        let in_list = InList {
--- a/src/mito2/src/sst/index/inverted_index/applier/builder/regex_match.rs
+++ b/src/mito2/src/sst/index/inverted_index/applier/builder/regex_match.rs
@@ -57,14 +57,14 @@ mod tests {

    #[test]
    fn test_regex_match_basic() {
-        let (_d, facotry) = PuffinManagerFactory::new_for_test_block("test_regex_match_basic_");
+        let (_d, factory) = PuffinManagerFactory::new_for_test_block("test_regex_match_basic_");
        let metadata = test_region_metadata();
        let mut builder = InvertedIndexApplierBuilder::new(
            "test".to_string(),
            test_object_store(),
            &metadata,
            HashSet::from_iter([1, 2, 3]),
-            facotry,
+            factory,
        );

        builder
@@ -83,7 +83,7 @@ mod tests {

    #[test]
    fn test_regex_match_field_column() {
-        let (_d, facotry) =
+        let (_d, factory) =
            PuffinManagerFactory::new_for_test_block("test_regex_match_field_column_");
        let metadata = test_region_metadata();
        let mut builder = InvertedIndexApplierBuilder::new(
@@ -91,7 +91,7 @@ mod tests {
            test_object_store(),
            &metadata,
            HashSet::from_iter([1, 2, 3]),
-            facotry,
+            factory,
        );

        builder
@@ -110,7 +110,7 @@ mod tests {

    #[test]
    fn test_regex_match_type_mismatch() {
-        let (_d, facotry) =
+        let (_d, factory) =
            PuffinManagerFactory::new_for_test_block("test_regex_match_type_mismatch_");
        let metadata = test_region_metadata();
        let mut builder = InvertedIndexApplierBuilder::new(
@@ -118,7 +118,7 @@ mod tests {
            test_object_store(),
            &metadata,
            HashSet::from_iter([1, 2, 3]),
-            facotry,
+            factory,
        );

        builder
@@ -130,7 +130,7 @@ mod tests {

    #[test]
    fn test_regex_match_type_nonexist_column() {
-        let (_d, facotry) =
+        let (_d, factory) =
            PuffinManagerFactory::new_for_test_block("test_regex_match_type_nonexist_column_");
        let metadata = test_region_metadata();
        let mut builder = InvertedIndexApplierBuilder::new(
@@ -138,7 +138,7 @@ mod tests {
            test_object_store(),
            &metadata,
            HashSet::from_iter([1, 2, 3]),
-            facotry,
+            factory,
        );

        let res = builder.collect_regex_match(&nonexistent_column(), &string_lit("abc"));
--- a/src/mito2/src/sst/parquet.rs
+++ b/src/mito2/src/sst/parquet.rs
@@ -27,7 +27,6 @@ pub(crate) mod file_range;
 pub mod format;
 pub(crate) mod helper;
 pub(crate) mod metadata;
-pub(crate) mod page_reader;
 pub mod plain_format;
 pub mod reader;
 pub mod row_group;
@@ -105,7 +104,9 @@ mod tests {
    use tokio_util::compat::FuturesAsyncWriteCompatExt;

    use super::*;
-    use crate::access_layer::{FilePathProvider, OperationType, RegionFilePathFactory};
+    use crate::access_layer::{
+        FilePathProvider, Metrics, OperationType, RegionFilePathFactory, WriteType,
+    };
    use crate::cache::{CacheManager, CacheStrategy, PageKey};
    use crate::read::BatchReader;
    use crate::region::options::{IndexOptions, InvertedIndexOptions};
@@ -176,6 +177,7 @@ mod tests {
            metadata.clone(),
            NoopIndexBuilder,
            file_path,
+            Metrics::new(WriteType::Flush),
        )
        .await;

@@ -233,10 +235,11 @@ mod tests {
            FixedPathProvider {
                file_id: handle.file_id(),
            },
+            Metrics::new(WriteType::Flush),
        )
        .await;

-        writer
+        let sst_info = writer
            .write_all(source, None, &write_opts)
            .await
            .unwrap()
@@ -265,16 +268,24 @@ mod tests {
            .await;
        }

-        // Doesn't have compressed page cached.
-        let page_key = PageKey::new_compressed(metadata.region_id, handle.file_id(), 0, 0);
-        assert!(cache.get_pages(&page_key).is_none());
+        let parquet_meta = sst_info.file_metadata.unwrap();
+        let get_ranges = |row_group_idx: usize| {
+            let row_group = parquet_meta.row_group(row_group_idx);
+            let mut ranges = Vec::with_capacity(row_group.num_columns());
+            for i in 0..row_group.num_columns() {
+                let (start, length) = row_group.column(i).byte_range();
+                ranges.push(start..start + length);
+            }
+
+            ranges
+        };

        // Cache 4 row groups.
        for i in 0..4 {
-            let page_key = PageKey::new_uncompressed(metadata.region_id, handle.file_id(), i, 0);
+            let page_key = PageKey::new(handle.file_id(), i, get_ranges(i));
            assert!(cache.get_pages(&page_key).is_some());
        }
-        let page_key = PageKey::new_uncompressed(metadata.region_id, handle.file_id(), 5, 0);
+        let page_key = PageKey::new(handle.file_id(), 5, vec![]);
        assert!(cache.get_pages(&page_key).is_none());
    }

@@ -304,6 +315,7 @@ mod tests {
            FixedPathProvider {
                file_id: handle.file_id(),
            },
+            Metrics::new(WriteType::Flush),
        )
        .await;

@@ -346,6 +358,7 @@ mod tests {
            FixedPathProvider {
                file_id: handle.file_id(),
            },
+            Metrics::new(WriteType::Flush),
        )
        .await;
        writer
@@ -398,6 +411,7 @@ mod tests {
            FixedPathProvider {
                file_id: handle.file_id(),
            },
+            Metrics::new(WriteType::Flush),
        )
        .await;
        writer
@@ -435,6 +449,7 @@ mod tests {
            FixedPathProvider {
                file_id: handle.file_id(),
            },
+            Metrics::new(WriteType::Flush),
        )
        .await;

@@ -577,6 +592,7 @@ mod tests {
            metadata.clone(),
            NoopIndexBuilder,
            path_provider,
+            Metrics::new(WriteType::Flush),
        )
        .await;

@@ -648,6 +664,7 @@ mod tests {
            metadata.clone(),
            indexer_builder,
            file_path.clone(),
+            Metrics::new(WriteType::Flush),
        )
        .await;

--- a/src/mito2/src/sst/parquet/helper.rs
+++ b/src/mito2/src/sst/parquet/helper.rs
@@ -89,7 +89,7 @@ fn parse_column_orders(
 }

 const FETCH_PARALLELISM: usize = 8;
-const MERGE_GAP: usize = 512 * 1024;
+pub(crate) const MERGE_GAP: usize = 512 * 1024;

 /// Asynchronously fetches byte ranges from an object store.
 ///
--- a/Show More
+++ b/Show More