feat: support tls for pg backend (#6611)

* load tls

Signed-off-by: Ruihang Xia <waynestxia@gmail.com>

* impl tls

Signed-off-by: Ruihang Xia <waynestxia@gmail.com>

* pass options

Signed-off-by: Ruihang Xia <waynestxia@gmail.com>

* implement require mode

Signed-off-by: Ruihang Xia <waynestxia@gmail.com>

* clean up

Signed-off-by: Ruihang Xia <waynestxia@gmail.com>

* update config

Signed-off-by: Ruihang Xia <waynestxia@gmail.com>

* fix clippy

Signed-off-by: Ruihang Xia <waynestxia@gmail.com>

* default to prefer

Signed-off-by: Ruihang Xia <waynestxia@gmail.com>

* update example config

Signed-off-by: Ruihang Xia <waynestxia@gmail.com>

* adjust example config

Signed-off-by: Ruihang Xia <waynestxia@gmail.com>

* handle client cert and key properly

Signed-off-by: Ruihang Xia <waynestxia@gmail.com>

* implement verify_ca and verify_full

Signed-off-by: Ruihang Xia <waynestxia@gmail.com>

* update integration test for config api

Signed-off-by: Ruihang Xia <waynestxia@gmail.com>

* change config name and default mode

Signed-off-by: Ruihang Xia <waynestxia@gmail.com>

---------

Signed-off-by: Ruihang Xia <waynestxia@gmail.com>

.cargo/config.toml

@@ -12,3 +12,6 @@ fetch = true
 checkout = true
 list_files = true
 internal_use_git2 = false
+
+[env]
+CARGO_WORKSPACE_DIR = { value = "", relative = true }

.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

.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
@@ -332,7 +339,7 @@ jobs:
       build-windows-artifacts,
       release-images-to-dockerhub,
     ]
-    runs-on: ubuntu-latest
+    runs-on: ubuntu-latest-16-cores
     # When we push to ACR, it's easy to fail due to some unknown network issues.
     # However, we don't want to fail the whole workflow because of this.
     # The ACR have daily sync with DockerHub, so don't worry about the image not being updated.
@@ -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

.github/workflows/semantic-pull-request.yml

@@ -11,17 +11,17 @@ concurrency:
   group: ${{ github.workflow }}-${{ github.head_ref || github.run_id }}
   cancel-in-progress: true
 
+permissions:
+  issues: write
+  contents: write
+  pull-requests: write
+
 jobs:
   check:
     runs-on: ubuntu-latest
-    permissions:
-      pull-requests: write  # Add permissions to modify PRs
-      issues: write
     timeout-minutes: 10
     steps:
       - uses: actions/checkout@v4
-        with:
-          persist-credentials: false
       - uses: ./.github/actions/setup-cyborg
       - name: Check Pull Request
         working-directory: cyborg

.gitignore

@@ -60,4 +60,7 @@ tests-fuzz/corpus/
 greptimedb_data
 
 # github
-!/.github
+!/.github
+
+# Claude code
+CLAUDE.md

AUTHOR.md

@@ -10,12 +10,10 @@
 * [NiwakaDev](https://github.com/NiwakaDev)
 * [tisonkun](https://github.com/tisonkun)
 
-
 ## Team Members (in alphabetical order)
 
 * [apdong2022](https://github.com/apdong2022)
 * [beryl678](https://github.com/beryl678)
-* [Breeze-P](https://github.com/Breeze-P)
 * [daviderli614](https://github.com/daviderli614)
 * [discord9](https://github.com/discord9)
 * [evenyag](https://github.com/evenyag)

Cargo.toml

@@ -13,6 +13,7 @@ members = [
     "src/common/datasource",
     "src/common/decimal",
     "src/common/error",
+    "src/common/event-recorder",
     "src/common/frontend",
     "src/common/function",
     "src/common/greptimedb-telemetry",
@@ -30,6 +31,7 @@ members = [
     "src/common/recordbatch",
     "src/common/runtime",
     "src/common/session",
+    "src/common/sql",
     "src/common/stat",
     "src/common/substrait",
     "src/common/telemetry",
@@ -71,11 +73,13 @@ members = [
 resolver = "2"
 
 [workspace.package]
-version = "0.15.0"
+version = "0.16.0"
 edition = "2021"
 license = "Apache-2.0"
 
 [workspace.lints]
+clippy.print_stdout = "warn"
+clippy.print_stderr = "warn"
 clippy.dbg_macro = "warn"
 clippy.implicit_clone = "warn"
 clippy.result_large_err = "allow"
@@ -121,6 +125,7 @@ datafusion = { git = "https://github.com/waynexia/arrow-datafusion.git", rev = "
 datafusion-common = { git = "https://github.com/waynexia/arrow-datafusion.git", rev = "12c0381babd52c681043957e9d6ee083a03f7646" }
 datafusion-expr = { git = "https://github.com/waynexia/arrow-datafusion.git", rev = "12c0381babd52c681043957e9d6ee083a03f7646" }
 datafusion-functions = { git = "https://github.com/waynexia/arrow-datafusion.git", rev = "12c0381babd52c681043957e9d6ee083a03f7646" }
+datafusion-functions-aggregate-common = { git = "https://github.com/waynexia/arrow-datafusion.git", rev = "12c0381babd52c681043957e9d6ee083a03f7646" }
 datafusion-optimizer = { git = "https://github.com/waynexia/arrow-datafusion.git", rev = "12c0381babd52c681043957e9d6ee083a03f7646" }
 datafusion-physical-expr = { git = "https://github.com/waynexia/arrow-datafusion.git", rev = "12c0381babd52c681043957e9d6ee083a03f7646" }
 datafusion-physical-plan = { git = "https://github.com/waynexia/arrow-datafusion.git", rev = "12c0381babd52c681043957e9d6ee083a03f7646" }
@@ -130,11 +135,12 @@ 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 = "82fe5c6282f623c185b86f03e898ee8952e50cf9" }
+greptime-proto = { git = "https://github.com/GreptimeTeam/greptime-proto.git", rev = "fe8c13f5f3c1fbef63f57fbdd29f0490dfeb987b" }
 hex = "0.4"
 http = "1"
 humantime = "2.1"
@@ -162,14 +168,13 @@ opentelemetry-proto = { version = "0.27", features = [
     "with-serde",
     "logs",
 ] }
+ordered-float = { version = "4.3", features = ["serde"] }
 parking_lot = "0.12"
 parquet = { version = "54.2", default-features = false, features = ["arrow", "async", "object_store"] }
 paste = "1.0"
 pin-project = "1.0"
 prometheus = { version = "0.13.3", features = ["process"] }
-promql-parser = { git = "https://github.com/GreptimeTeam/promql-parser.git", rev = "0410e8b459dda7cb222ce9596f8bf3971bd07bd2", features = [
-    "ser",
-] }
+promql-parser = { version = "0.6", features = ["ser"] }
 prost = { version = "0.13", features = ["no-recursion-limit"] }
 raft-engine = { version = "0.4.1", default-features = false }
 rand = "0.9"
@@ -220,10 +225,13 @@ tokio-util = { version = "0.7", features = ["io-util", "compat"] }
 toml = "0.8.8"
 tonic = { version = "0.12", features = ["tls", "gzip", "zstd"] }
 tower = "0.5"
+tower-http = "0.6"
+tracing = "0.1"
 tracing-appender = "0.2"
 tracing-subscriber = { version = "0.3", features = ["env-filter", "json", "fmt"] }
 typetag = "0.2"
 uuid = { version = "1.7", features = ["serde", "v4", "fast-rng"] }
+vrl = "0.25"
 zstd = "0.13"
 # DO_NOT_REMOVE_THIS: END_OF_EXTERNAL_DEPENDENCIES
 
@@ -241,6 +249,7 @@ common-config = { path = "src/common/config" }
 common-datasource = { path = "src/common/datasource" }
 common-decimal = { path = "src/common/decimal" }
 common-error = { path = "src/common/error" }
+common-event-recorder = { path = "src/common/event-recorder" }
 common-frontend = { path = "src/common/frontend" }
 common-function = { path = "src/common/function" }
 common-greptimedb-telemetry = { path = "src/common/greptimedb-telemetry" }
@@ -258,6 +267,7 @@ common-query = { path = "src/common/query" }
 common-recordbatch = { path = "src/common/recordbatch" }
 common-runtime = { path = "src/common/runtime" }
 common-session = { path = "src/common/session" }
+common-sql = { path = "src/common/sql" }
 common-telemetry = { path = "src/common/telemetry" }
 common-test-util = { path = "src/common/test-util" }
 common-time = { path = "src/common/time" }

README.md

@@ -75,9 +75,9 @@
 | --------- | ----------- |
 | [Unified Observability Data](https://docs.greptime.com/user-guide/concepts/why-greptimedb) | Store metrics, logs, and traces as timestamped, contextual wide events. Query via [SQL](https://docs.greptime.com/user-guide/query-data/sql), [PromQL](https://docs.greptime.com/user-guide/query-data/promql), and [streaming](https://docs.greptime.com/user-guide/flow-computation/overview). |
 | [High Performance & Cost Effective](https://docs.greptime.com/user-guide/manage-data/data-index) | Written in Rust, with a distributed query engine, [rich indexing](https://docs.greptime.com/user-guide/manage-data/data-index), and optimized columnar storage, delivering sub-second responses at PB scale. |
-| [Cloud-Native Architecture](https://docs.greptime.com/user-guide/concepts/architecture) | Designed for [Kubernetes](https://docs.greptime.com/user-guide/deployments/deploy-on-kubernetes/greptimedb-operator-management), with compute/storage separation, native object storage (AWS S3, Azure Blob, etc.) and seamless cross-cloud access. |
+| [Cloud-Native Architecture](https://docs.greptime.com/user-guide/concepts/architecture) | Designed for [Kubernetes](https://docs.greptime.com/user-guide/deployments-administration/deploy-on-kubernetes/greptimedb-operator-management), with compute/storage separation, native object storage (AWS S3, Azure Blob, etc.) and seamless cross-cloud access. |
 | [Developer-Friendly](https://docs.greptime.com/user-guide/protocols/overview) | Access via SQL/PromQL interfaces, REST API, MySQL/PostgreSQL protocols, and popular ingestion [protocols](https://docs.greptime.com/user-guide/protocols/overview). |
-| [Flexible Deployment](https://docs.greptime.com/user-guide/deployments/overview) | Deploy anywhere: edge (including ARM/[Android](https://docs.greptime.com/user-guide/deployments/run-on-android)) or cloud, with unified APIs and efficient data sync. |
+| [Flexible Deployment](https://docs.greptime.com/user-guide/deployments-administration/overview) | Deploy anywhere: edge (including ARM/[Android](https://docs.greptime.com/user-guide/deployments-administration/run-on-android)) or cloud, with unified APIs and efficient data sync. |
 
 Learn more in [Why GreptimeDB](https://docs.greptime.com/user-guide/concepts/why-greptimedb) and [Observability 2.0 and the Database for It](https://greptime.com/blogs/2025-04-25-greptimedb-observability2-new-database).
 

config/config.md

@@ -185,10 +185,11 @@
 | `logging.dir` | String | `./greptimedb_data/logs` | The directory to store the log files. If set to empty, logs will not be written to files. |
 | `logging.level` | String | Unset | The log level. Can be `info`/`debug`/`warn`/`error`. |
 | `logging.enable_otlp_tracing` | Bool | `false` | Enable OTLP tracing. |
-| `logging.otlp_endpoint` | String | `http://localhost:4317` | The OTLP tracing endpoint. |
+| `logging.otlp_endpoint` | String | `http://localhost:4318` | The OTLP tracing endpoint. |
 | `logging.append_stdout` | Bool | `true` | Whether to append logs to stdout. |
 | `logging.log_format` | String | `text` | The log format. Can be `text`/`json`. |
 | `logging.max_log_files` | Integer | `720` | The maximum amount of log files. |
+| `logging.otlp_export_protocol` | String | `http` | The OTLP tracing export protocol. Can be `grpc`/`http`. |
 | `logging.tracing_sample_ratio` | -- | -- | The percentage of tracing will be sampled and exported.<br/>Valid range `[0, 1]`, 1 means all traces are sampled, 0 means all traces are not sampled, the default value is 1.<br/>ratio > 1 are treated as 1. Fractions < 0 are treated as 0 |
 | `logging.tracing_sample_ratio.default_ratio` | Float | `1.0` | -- |
 | `slow_query` | -- | -- | The slow query log options. |
@@ -206,6 +207,8 @@
 | `export_metrics.remote_write.headers` | InlineTable | -- | HTTP headers of Prometheus remote-write carry. |
 | `tracing` | -- | -- | The tracing options. Only effect when compiled with `tokio-console` feature. |
 | `tracing.tokio_console_addr` | String | Unset | The tokio console address. |
+| `memory` | -- | -- | The memory options. |
+| `memory.enable_heap_profiling` | Bool | `true` | Whether to enable heap profiling activation during startup.<br/>When enabled, heap profiling will be activated if the `MALLOC_CONF` environment variable<br/>is set to "prof:true,prof_active:false". The official image adds this env variable.<br/>Default is true. |
 
 
 ## Distributed Mode
@@ -280,6 +283,7 @@
 | `meta_client.metadata_cache_tti` | String | `5m` | -- |
 | `query` | -- | -- | The query engine options. |
 | `query.parallelism` | Integer | `0` | Parallelism of the query engine.<br/>Default to 0, which means the number of CPU cores. |
+| `query.allow_query_fallback` | Bool | `false` | Whether to allow query fallback when push down optimize fails.<br/>Default to false, meaning when push down optimize failed, return error msg |
 | `datanode` | -- | -- | Datanode options. |
 | `datanode.client` | -- | -- | Datanode client options. |
 | `datanode.client.connect_timeout` | String | `10s` | -- |
@@ -288,10 +292,11 @@
 | `logging.dir` | String | `./greptimedb_data/logs` | The directory to store the log files. If set to empty, logs will not be written to files. |
 | `logging.level` | String | Unset | The log level. Can be `info`/`debug`/`warn`/`error`. |
 | `logging.enable_otlp_tracing` | Bool | `false` | Enable OTLP tracing. |
-| `logging.otlp_endpoint` | String | `http://localhost:4317` | The OTLP tracing endpoint. |
+| `logging.otlp_endpoint` | String | `http://localhost:4318` | The OTLP tracing endpoint. |
 | `logging.append_stdout` | Bool | `true` | Whether to append logs to stdout. |
 | `logging.log_format` | String | `text` | The log format. Can be `text`/`json`. |
 | `logging.max_log_files` | Integer | `720` | The maximum amount of log files. |
+| `logging.otlp_export_protocol` | String | `http` | The OTLP tracing export protocol. Can be `grpc`/`http`. |
 | `logging.tracing_sample_ratio` | -- | -- | The percentage of tracing will be sampled and exported.<br/>Valid range `[0, 1]`, 1 means all traces are sampled, 0 means all traces are not sampled, the default value is 1.<br/>ratio > 1 are treated as 1. Fractions < 0 are treated as 0 |
 | `logging.tracing_sample_ratio.default_ratio` | Float | `1.0` | -- |
 | `slow_query` | -- | -- | The slow query log options. |
@@ -308,6 +313,8 @@
 | `export_metrics.remote_write.headers` | InlineTable | -- | HTTP headers of Prometheus remote-write carry. |
 | `tracing` | -- | -- | The tracing options. Only effect when compiled with `tokio-console` feature. |
 | `tracing.tokio_console_addr` | String | Unset | The tokio console address. |
+| `memory` | -- | -- | The memory options. |
+| `memory.enable_heap_profiling` | Bool | `true` | Whether to enable heap profiling activation during startup.<br/>When enabled, heap profiling will be activated if the `MALLOC_CONF` environment variable<br/>is set to "prof:true,prof_active:false". The official image adds this env variable.<br/>Default is true. |
 
 
 ### Metasrv
@@ -323,12 +330,19 @@
 | `selector` | String | `round_robin` | Datanode selector type.<br/>- `round_robin` (default value)<br/>- `lease_based`<br/>- `load_based`<br/>For details, please see "https://docs.greptime.com/developer-guide/metasrv/selector". |
 | `use_memory_store` | Bool | `false` | Store data in memory. |
 | `enable_region_failover` | Bool | `false` | Whether to enable region failover.<br/>This feature is only available on GreptimeDB running on cluster mode and<br/>- Using Remote WAL<br/>- Using shared storage (e.g., s3). |
+| `region_failure_detector_initialization_delay` | String | `10m` | The delay before starting region failure detection.<br/>This delay helps prevent Metasrv from triggering unnecessary region failovers before all Datanodes are fully started.<br/>Especially useful when the cluster is not deployed with GreptimeDB Operator and maintenance mode is not enabled. |
 | `allow_region_failover_on_local_wal` | Bool | `false` | Whether to allow region failover on local WAL.<br/>**This option is not recommended to be set to true, because it may lead to data loss during failover.** |
 | `node_max_idle_time` | String | `24hours` | Max allowed idle time before removing node info from metasrv memory. |
 | `enable_telemetry` | Bool | `true` | Whether to enable greptimedb telemetry. Enabled by default. |
 | `runtime` | -- | -- | The runtime options. |
 | `runtime.global_rt_size` | Integer | `8` | The number of threads to execute the runtime for global read operations. |
 | `runtime.compact_rt_size` | Integer | `4` | The number of threads to execute the runtime for global write operations. |
+| `backend_tls` | -- | -- | TLS configuration for kv store backend (only applicable for PostgreSQL/MySQL backends)<br/>When using PostgreSQL or MySQL as metadata store, you can configure TLS here |
+| `backend_tls.mode` | String | `prefer` | TLS mode, refer to https://www.postgresql.org/docs/current/libpq-ssl.html<br/>- "disable" - No TLS<br/>- "prefer" (default) - Try TLS, fallback to plain<br/>- "require" - Require TLS<br/>- "verify_ca" - Require TLS and verify CA<br/>- "verify_full" - Require TLS and verify hostname |
+| `backend_tls.cert_path` | String | `""` | Path to client certificate file (for client authentication)<br/>Like "/path/to/client.crt" |
+| `backend_tls.key_path` | String | `""` | Path to client private key file (for client authentication)<br/>Like "/path/to/client.key" |
+| `backend_tls.ca_cert_path` | String | `""` | Path to CA certificate file (for server certificate verification)<br/>Required when using custom CAs or self-signed certificates<br/>Leave empty to use system root certificates only<br/>Like "/path/to/ca.crt" |
+| `backend_tls.watch` | Bool | `false` | Watch for certificate file changes and auto reload |
 | `grpc` | -- | -- | The gRPC server options. |
 | `grpc.bind_addr` | String | `127.0.0.1:3002` | The address to bind the gRPC server. |
 | `grpc.server_addr` | String | `127.0.0.1:3002` | The communication server address for the frontend and datanode to connect to metasrv.<br/>If left empty or unset, the server will automatically use the IP address of the first network interface<br/>on the host, with the same port number as the one specified in `bind_addr`. |
@@ -370,10 +384,11 @@
 | `logging.dir` | String | `./greptimedb_data/logs` | The directory to store the log files. If set to empty, logs will not be written to files. |
 | `logging.level` | String | Unset | The log level. Can be `info`/`debug`/`warn`/`error`. |
 | `logging.enable_otlp_tracing` | Bool | `false` | Enable OTLP tracing. |
-| `logging.otlp_endpoint` | String | `http://localhost:4317` | The OTLP tracing endpoint. |
+| `logging.otlp_endpoint` | String | `http://localhost:4318` | The OTLP tracing endpoint. |
 | `logging.append_stdout` | Bool | `true` | Whether to append logs to stdout. |
 | `logging.log_format` | String | `text` | The log format. Can be `text`/`json`. |
 | `logging.max_log_files` | Integer | `720` | The maximum amount of log files. |
+| `logging.otlp_export_protocol` | String | `http` | The OTLP tracing export protocol. Can be `grpc`/`http`. |
 | `logging.tracing_sample_ratio` | -- | -- | The percentage of tracing will be sampled and exported.<br/>Valid range `[0, 1]`, 1 means all traces are sampled, 0 means all traces are not sampled, the default value is 1.<br/>ratio > 1 are treated as 1. Fractions < 0 are treated as 0 |
 | `logging.tracing_sample_ratio.default_ratio` | Float | `1.0` | -- |
 | `export_metrics` | -- | -- | The metasrv can export its metrics and send to Prometheus compatible service (e.g. `greptimedb` itself) from remote-write API.<br/>This is only used for `greptimedb` to export its own metrics internally. It's different from prometheus scrape. |
@@ -384,6 +399,8 @@
 | `export_metrics.remote_write.headers` | InlineTable | -- | HTTP headers of Prometheus remote-write carry. |
 | `tracing` | -- | -- | The tracing options. Only effect when compiled with `tokio-console` feature. |
 | `tracing.tokio_console_addr` | String | Unset | The tokio console address. |
+| `memory` | -- | -- | The memory options. |
+| `memory.enable_heap_profiling` | Bool | `true` | Whether to enable heap profiling activation during startup.<br/>When enabled, heap profiling will be activated if the `MALLOC_CONF` environment variable<br/>is set to "prof:true,prof_active:false". The official image adds this env variable.<br/>Default is true. |
 
 
 ### Datanode
@@ -432,8 +449,8 @@
 | `wal.provider` | String | `raft_engine` | The provider of the WAL.<br/>- `raft_engine`: the wal is stored in the local file system by raft-engine.<br/>- `kafka`: it's remote wal that data is stored in Kafka. |
 | `wal.dir` | String | Unset | The directory to store the WAL files.<br/>**It's only used when the provider is `raft_engine`**. |
 | `wal.file_size` | String | `128MB` | The size of the WAL segment file.<br/>**It's only used when the provider is `raft_engine`**. |
-| `wal.purge_threshold` | String | `1GB` | The threshold of the WAL size to trigger a flush.<br/>**It's only used when the provider is `raft_engine`**. |
-| `wal.purge_interval` | String | `1m` | The interval to trigger a flush.<br/>**It's only used when the provider is `raft_engine`**. |
+| `wal.purge_threshold` | String | `1GB` | The threshold of the WAL size to trigger a purge.<br/>**It's only used when the provider is `raft_engine`**. |
+| `wal.purge_interval` | String | `1m` | The interval to trigger a purge.<br/>**It's only used when the provider is `raft_engine`**. |
 | `wal.read_batch_size` | Integer | `128` | The read batch size.<br/>**It's only used when the provider is `raft_engine`**. |
 | `wal.sync_write` | Bool | `false` | Whether to use sync write.<br/>**It's only used when the provider is `raft_engine`**. |
 | `wal.enable_log_recycle` | Bool | `true` | Whether to reuse logically truncated log files.<br/>**It's only used when the provider is `raft_engine`**. |
@@ -534,10 +551,11 @@
 | `logging.dir` | String | `./greptimedb_data/logs` | The directory to store the log files. If set to empty, logs will not be written to files. |
 | `logging.level` | String | Unset | The log level. Can be `info`/`debug`/`warn`/`error`. |
 | `logging.enable_otlp_tracing` | Bool | `false` | Enable OTLP tracing. |
-| `logging.otlp_endpoint` | String | `http://localhost:4317` | The OTLP tracing endpoint. |
+| `logging.otlp_endpoint` | String | `http://localhost:4318` | The OTLP tracing endpoint. |
 | `logging.append_stdout` | Bool | `true` | Whether to append logs to stdout. |
 | `logging.log_format` | String | `text` | The log format. Can be `text`/`json`. |
 | `logging.max_log_files` | Integer | `720` | The maximum amount of log files. |
+| `logging.otlp_export_protocol` | String | `http` | The OTLP tracing export protocol. Can be `grpc`/`http`. |
 | `logging.tracing_sample_ratio` | -- | -- | The percentage of tracing will be sampled and exported.<br/>Valid range `[0, 1]`, 1 means all traces are sampled, 0 means all traces are not sampled, the default value is 1.<br/>ratio > 1 are treated as 1. Fractions < 0 are treated as 0 |
 | `logging.tracing_sample_ratio.default_ratio` | Float | `1.0` | -- |
 | `export_metrics` | -- | -- | The datanode can export its metrics and send to Prometheus compatible service (e.g. `greptimedb` itself) from remote-write API.<br/>This is only used for `greptimedb` to export its own metrics internally. It's different from prometheus scrape. |
@@ -548,6 +566,8 @@
 | `export_metrics.remote_write.headers` | InlineTable | -- | HTTP headers of Prometheus remote-write carry. |
 | `tracing` | -- | -- | The tracing options. Only effect when compiled with `tokio-console` feature. |
 | `tracing.tokio_console_addr` | String | Unset | The tokio console address. |
+| `memory` | -- | -- | The memory options. |
+| `memory.enable_heap_profiling` | Bool | `true` | Whether to enable heap profiling activation during startup.<br/>When enabled, heap profiling will be activated if the `MALLOC_CONF` environment variable<br/>is set to "prof:true,prof_active:false". The official image adds this env variable.<br/>Default is true. |
 
 
 ### Flownode
@@ -557,6 +577,16 @@
 | `node_id` | Integer | Unset | The flownode identifier and should be unique in the cluster. |
 | `flow` | -- | -- | flow engine options. |
 | `flow.num_workers` | Integer | `0` | The number of flow worker in flownode.<br/>Not setting(or set to 0) this value will use the number of CPU cores divided by 2. |
+| `flow.batching_mode` | -- | -- | -- |
+| `flow.batching_mode.query_timeout` | String | `600s` | The default batching engine query timeout is 10 minutes. |
+| `flow.batching_mode.slow_query_threshold` | String | `60s` | will output a warn log for any query that runs for more that this threshold |
+| `flow.batching_mode.experimental_min_refresh_duration` | String | `5s` | The minimum duration between two queries execution by batching mode task |
+| `flow.batching_mode.grpc_conn_timeout` | String | `5s` | The gRPC connection timeout |
+| `flow.batching_mode.experimental_grpc_max_retries` | Integer | `3` | The gRPC max retry number |
+| `flow.batching_mode.experimental_frontend_scan_timeout` | String | `30s` | Flow wait for available frontend timeout,<br/>if failed to find available frontend after frontend_scan_timeout elapsed, return error<br/>which prevent flownode from starting |
+| `flow.batching_mode.experimental_frontend_activity_timeout` | String | `60s` | Frontend activity timeout<br/>if frontend is down(not sending heartbeat) for more than frontend_activity_timeout,<br/>it will be removed from the list that flownode use to connect |
+| `flow.batching_mode.experimental_max_filter_num_per_query` | Integer | `20` | Maximum number of filters allowed in a single query |
+| `flow.batching_mode.experimental_time_window_merge_threshold` | Integer | `3` | Time window merge distance |
 | `grpc` | -- | -- | The gRPC server options. |
 | `grpc.bind_addr` | String | `127.0.0.1:6800` | The address to bind the gRPC server. |
 | `grpc.server_addr` | String | `127.0.0.1:6800` | The address advertised to the metasrv,<br/>and used for connections from outside the host |
@@ -584,11 +614,16 @@
 | `logging.dir` | String | `./greptimedb_data/logs` | The directory to store the log files. If set to empty, logs will not be written to files. |
 | `logging.level` | String | Unset | The log level. Can be `info`/`debug`/`warn`/`error`. |
 | `logging.enable_otlp_tracing` | Bool | `false` | Enable OTLP tracing. |
-| `logging.otlp_endpoint` | String | `http://localhost:4317` | The OTLP tracing endpoint. |
+| `logging.otlp_endpoint` | String | `http://localhost:4318` | The OTLP tracing endpoint. |
 | `logging.append_stdout` | Bool | `true` | Whether to append logs to stdout. |
 | `logging.log_format` | String | `text` | The log format. Can be `text`/`json`. |
 | `logging.max_log_files` | Integer | `720` | The maximum amount of log files. |
+| `logging.otlp_export_protocol` | String | `http` | The OTLP tracing export protocol. Can be `grpc`/`http`. |
 | `logging.tracing_sample_ratio` | -- | -- | The percentage of tracing will be sampled and exported.<br/>Valid range `[0, 1]`, 1 means all traces are sampled, 0 means all traces are not sampled, the default value is 1.<br/>ratio > 1 are treated as 1. Fractions < 0 are treated as 0 |
 | `logging.tracing_sample_ratio.default_ratio` | Float | `1.0` | -- |
 | `tracing` | -- | -- | The tracing options. Only effect when compiled with `tokio-console` feature. |
 | `tracing.tokio_console_addr` | String | Unset | The tokio console address. |
+| `query` | -- | -- | -- |
+| `query.parallelism` | Integer | `1` | Parallelism of the query engine for query sent by flownode.<br/>Default to 1, so it won't use too much cpu or memory |
+| `memory` | -- | -- | The memory options. |
+| `memory.enable_heap_profiling` | Bool | `true` | Whether to enable heap profiling activation during startup.<br/>When enabled, heap profiling will be activated if the `MALLOC_CONF` environment variable<br/>is set to "prof:true,prof_active:false". The official image adds this env variable.<br/>Default is true. |

config/datanode.example.toml

@@ -129,11 +129,11 @@ dir = "./greptimedb_data/wal"
 ## **It's only used when the provider is `raft_engine`**.
 file_size = "128MB"
 
-## The threshold of the WAL size to trigger a flush.
+## The threshold of the WAL size to trigger a purge.
 ## **It's only used when the provider is `raft_engine`**.
 purge_threshold = "1GB"
 
-## The interval to trigger a flush.
+## The interval to trigger a purge.
 ## **It's only used when the provider is `raft_engine`**.
 purge_interval = "1m"
 
@@ -629,7 +629,7 @@ level = "info"
 enable_otlp_tracing = false
 
 ## The OTLP tracing endpoint.
-otlp_endpoint = "http://localhost:4317"
+otlp_endpoint = "http://localhost:4318"
 
 ## Whether to append logs to stdout.
 append_stdout = true
@@ -640,6 +640,9 @@ log_format = "text"
 ## The maximum amount of log files.
 max_log_files = 720
 
+## The OTLP tracing export protocol. Can be `grpc`/`http`.
+otlp_export_protocol = "http"
+
 ## The percentage of tracing will be sampled and exported.
 ## Valid range `[0, 1]`, 1 means all traces are sampled, 0 means all traces are not sampled, the default value is 1.
 ## ratio > 1 are treated as 1. Fractions < 0 are treated as 0
@@ -666,3 +669,11 @@ headers = { }
 ## The tokio console address.
 ## @toml2docs:none-default
 #+ tokio_console_addr = "127.0.0.1"
+
+## The memory options.
+[memory]
+## Whether to enable heap profiling activation during startup.
+## When enabled, heap profiling will be activated if the `MALLOC_CONF` environment variable
+## is set to "prof:true,prof_active:false". The official image adds this env variable.
+## Default is true.
+enable_heap_profiling = true

config/flownode.example.toml

@@ -7,6 +7,29 @@ node_id = 14
 ## The number of flow worker in flownode.
 ## Not setting(or set to 0) this value will use the number of CPU cores divided by 2.
 #+num_workers=0
+[flow.batching_mode]
+## The default batching engine query timeout is 10 minutes.
+#+query_timeout="600s"
+## will output a warn log for any query that runs for more that this threshold
+#+slow_query_threshold="60s"
+## The minimum duration between two queries execution by batching mode task
+#+experimental_min_refresh_duration="5s"
+## The gRPC connection timeout
+#+grpc_conn_timeout="5s"
+## The gRPC max retry number
+#+experimental_grpc_max_retries=3
+## Flow wait for available frontend timeout,
+## if failed to find available frontend after frontend_scan_timeout elapsed, return error
+## which prevent flownode from starting
+#+experimental_frontend_scan_timeout="30s"
+## Frontend activity timeout
+## if frontend is down(not sending heartbeat) for more than frontend_activity_timeout,
+## it will be removed from the list that flownode use to connect
+#+experimental_frontend_activity_timeout="60s"
+## Maximum number of filters allowed in a single query
+#+experimental_max_filter_num_per_query=20
+## Time window merge distance
+#+experimental_time_window_merge_threshold=3
 
 ## The gRPC server options.
 [grpc]
@@ -83,7 +106,7 @@ level = "info"
 enable_otlp_tracing = false
 
 ## The OTLP tracing endpoint.
-otlp_endpoint = "http://localhost:4317"
+otlp_endpoint = "http://localhost:4318"
 
 ## Whether to append logs to stdout.
 append_stdout = true
@@ -94,6 +117,9 @@ log_format = "text"
 ## The maximum amount of log files.
 max_log_files = 720
 
+## The OTLP tracing export protocol. Can be `grpc`/`http`.
+otlp_export_protocol = "http"
+
 ## The percentage of tracing will be sampled and exported.
 ## Valid range `[0, 1]`, 1 means all traces are sampled, 0 means all traces are not sampled, the default value is 1.
 ## ratio > 1 are treated as 1. Fractions < 0 are treated as 0
@@ -105,3 +131,16 @@ default_ratio = 1.0
 ## The tokio console address.
 ## @toml2docs:none-default
 #+ tokio_console_addr = "127.0.0.1"
+
+[query]
+## Parallelism of the query engine for query sent by flownode.
+## Default to 1, so it won't use too much cpu or memory
+parallelism = 1
+
+## The memory options.
+[memory]
+## Whether to enable heap profiling activation during startup.
+## When enabled, heap profiling will be activated if the `MALLOC_CONF` environment variable
+## is set to "prof:true,prof_active:false". The official image adds this env variable.
+## Default is true.
+enable_heap_profiling = true

config/frontend.example.toml

@@ -197,6 +197,9 @@ metadata_cache_tti = "5m"
 ## Parallelism of the query engine.
 ## Default to 0, which means the number of CPU cores.
 parallelism = 0
+## Whether to allow query fallback when push down optimize fails.
+## Default to false, meaning when push down optimize failed, return error msg
+allow_query_fallback = false
 
 ## Datanode options.
 [datanode]
@@ -218,7 +221,7 @@ level = "info"
 enable_otlp_tracing = false
 
 ## The OTLP tracing endpoint.
-otlp_endpoint = "http://localhost:4317"
+otlp_endpoint = "http://localhost:4318"
 
 ## Whether to append logs to stdout.
 append_stdout = true
@@ -229,6 +232,9 @@ log_format = "text"
 ## The maximum amount of log files.
 max_log_files = 720
 
+## The OTLP tracing export protocol. Can be `grpc`/`http`.
+otlp_export_protocol = "http"
+
 ## The percentage of tracing will be sampled and exported.
 ## Valid range `[0, 1]`, 1 means all traces are sampled, 0 means all traces are not sampled, the default value is 1.
 ## ratio > 1 are treated as 1. Fractions < 0 are treated as 0
@@ -274,3 +280,11 @@ headers = { }
 ## The tokio console address.
 ## @toml2docs:none-default
 #+ tokio_console_addr = "127.0.0.1"
+
+## The memory options.
+[memory]
+## Whether to enable heap profiling activation during startup.
+## When enabled, heap profiling will be activated if the `MALLOC_CONF` environment variable
+## is set to "prof:true,prof_active:false". The official image adds this env variable.
+## Default is true.
+enable_heap_profiling = true

config/metasrv.example.toml

@@ -43,6 +43,11 @@ use_memory_store = false
 ## - Using shared storage (e.g., s3).
 enable_region_failover = false
 
+## The delay before starting region failure detection.
+## This delay helps prevent Metasrv from triggering unnecessary region failovers before all Datanodes are fully started.
+## Especially useful when the cluster is not deployed with GreptimeDB Operator and maintenance mode is not enabled.
+region_failure_detector_initialization_delay = '10m'
+
 ## Whether to allow region failover on local WAL.
 ## **This option is not recommended to be set to true, because it may lead to data loss during failover.**
 allow_region_failover_on_local_wal = false
@@ -60,6 +65,34 @@ node_max_idle_time = "24hours"
 ## The number of threads to execute the runtime for global write operations.
 #+ compact_rt_size = 4
 
+## TLS configuration for kv store backend (only applicable for PostgreSQL/MySQL backends)
+## When using PostgreSQL or MySQL as metadata store, you can configure TLS here
+[backend_tls]
+## TLS mode, refer to https://www.postgresql.org/docs/current/libpq-ssl.html
+## - "disable" - No TLS
+## - "prefer" (default) - Try TLS, fallback to plain
+## - "require" - Require TLS
+## - "verify_ca" - Require TLS and verify CA
+## - "verify_full" - Require TLS and verify hostname
+mode = "prefer"
+
+## Path to client certificate file (for client authentication)
+## Like "/path/to/client.crt"
+cert_path = ""
+
+## Path to client private key file (for client authentication)
+## Like "/path/to/client.key"
+key_path = ""
+
+## Path to CA certificate file (for server certificate verification)
+## Required when using custom CAs or self-signed certificates
+## Leave empty to use system root certificates only
+## Like "/path/to/ca.crt"
+ca_cert_path = ""
+
+## Watch for certificate file changes and auto reload
+watch = false
+
 ## The gRPC server options.
 [grpc]
 ## The address to bind the gRPC server.
@@ -220,7 +253,7 @@ level = "info"
 enable_otlp_tracing = false
 
 ## The OTLP tracing endpoint.
-otlp_endpoint = "http://localhost:4317"
+otlp_endpoint = "http://localhost:4318"
 
 ## Whether to append logs to stdout.
 append_stdout = true
@@ -231,6 +264,9 @@ log_format = "text"
 ## The maximum amount of log files.
 max_log_files = 720
 
+## The OTLP tracing export protocol. Can be `grpc`/`http`.
+otlp_export_protocol = "http"
+
 ## The percentage of tracing will be sampled and exported.
 ## Valid range `[0, 1]`, 1 means all traces are sampled, 0 means all traces are not sampled, the default value is 1.
 ## ratio > 1 are treated as 1. Fractions < 0 are treated as 0
@@ -257,3 +293,11 @@ headers = { }
 ## The tokio console address.
 ## @toml2docs:none-default
 #+ tokio_console_addr = "127.0.0.1"
+
+## The memory options.
+[memory]
+## Whether to enable heap profiling activation during startup.
+## When enabled, heap profiling will be activated if the `MALLOC_CONF` environment variable
+## is set to "prof:true,prof_active:false". The official image adds this env variable.
+## Default is true.
+enable_heap_profiling = true

config/standalone.example.toml

@@ -720,7 +720,7 @@ level = "info"
 enable_otlp_tracing = false
 
 ## The OTLP tracing endpoint.
-otlp_endpoint = "http://localhost:4317"
+otlp_endpoint = "http://localhost:4318"
 
 ## Whether to append logs to stdout.
 append_stdout = true
@@ -731,6 +731,9 @@ log_format = "text"
 ## The maximum amount of log files.
 max_log_files = 720
 
+## The OTLP tracing export protocol. Can be `grpc`/`http`.
+otlp_export_protocol = "http"
+
 ## The percentage of tracing will be sampled and exported.
 ## Valid range `[0, 1]`, 1 means all traces are sampled, 0 means all traces are not sampled, the default value is 1.
 ## ratio > 1 are treated as 1. Fractions < 0 are treated as 0
@@ -780,3 +783,11 @@ headers = { }
 ## The tokio console address.
 ## @toml2docs:none-default
 #+ tokio_console_addr = "127.0.0.1"
+
+## The memory options.
+[memory]
+## Whether to enable heap profiling activation during startup.
+## When enabled, heap profiling will be activated if the `MALLOC_CONF` environment variable
+## is set to "prof:true,prof_active:false". The official image adds this env variable.
+## Default is true.
+enable_heap_profiling = true

cyborg/bin/follow-up-docs-issue.ts

@@ -55,12 +55,25 @@ async function main() {
         await client.rest.issues.addLabels({
             owner, repo, issue_number: number, labels: [labelDocsRequired],
         })
+
+        // Get available assignees for the docs repo
+        const assigneesResponse = await docsClient.rest.issues.listAssignees({
+            owner: 'GreptimeTeam',
+            repo: 'docs',
+        })
+        const validAssignees = assigneesResponse.data.map(assignee => assignee.login)
+        core.info(`Available assignees: ${validAssignees.join(', ')}`)
+
+        // Check if the actor is a valid assignee, otherwise fallback to fengjiachun
+        const assignee = validAssignees.includes(actor) ? actor : 'fengjiachun'
+        core.info(`Assigning issue to: ${assignee}`)
+
         await docsClient.rest.issues.create({
             owner: 'GreptimeTeam',
             repo: 'docs',
             title: `Update docs for ${title}`,
             body: `A document change request is generated from ${html_url}`,
-            assignee: actor,
+            assignee: assignee,
         }).then((res) => {
             core.info(`Created issue ${res.data}`)
         })

docker/buildx/centos/Dockerfile

@@ -47,4 +47,6 @@ WORKDIR /greptime
 COPY --from=builder /out/target/${OUTPUT_DIR}/greptime /greptime/bin/
 ENV PATH /greptime/bin/:$PATH
 
+ENV MALLOC_CONF="prof:true,prof_active:false"
+
 ENTRYPOINT ["greptime"]

docker/buildx/ubuntu/Dockerfile

@@ -47,4 +47,6 @@ WORKDIR /greptime
 COPY --from=builder /out/target/${OUTPUT_DIR}/greptime /greptime/bin/
 ENV PATH /greptime/bin/:$PATH
 
+ENV MALLOC_CONF="prof:true,prof_active:false"
+
 ENTRYPOINT ["greptime"]

docker/ci/centos/Dockerfile

@@ -15,4 +15,6 @@ ADD $TARGETARCH/greptime /greptime/bin/
 
 ENV PATH /greptime/bin/:$PATH
 
+ENV MALLOC_CONF="prof:true,prof_active:false"
+
 ENTRYPOINT ["greptime"]

docker/ci/ubuntu/Dockerfile

@@ -18,4 +18,6 @@ ENV PATH /greptime/bin/:$PATH
 
 ENV TARGET_BIN=$TARGET_BIN
 
+ENV MALLOC_CONF="prof:true,prof_active:false"
+
 ENTRYPOINT ["sh", "-c", "exec $TARGET_BIN \"$@\"", "--"]

docs/benchmarks/log/README.md

@@ -48,4 +48,4 @@ Please refer to [SQL query](./query.sql) for GreptimeDB and Clickhouse, and [que
 
 ## Addition
 - You can tune GreptimeDB's configuration to get better performance.
-- You can setup GreptimeDB to use S3 as storage, see [here](https://docs.greptime.com/user-guide/deployments/configuration#storage-options).
+- You can setup GreptimeDB to use S3 as storage, see [here](https://docs.greptime.com/user-guide/deployments-administration/configuration#storage-options).

docs/how-to/how-to-profile-memory.md

@@ -30,6 +30,23 @@ curl https://raw.githubusercontent.com/brendangregg/FlameGraph/master/flamegraph
 
 ## Profiling
 
+### Configuration
+
+You can control heap profiling activation through configuration. Add the following to your configuration file:
+
+```toml
+[memory]
+# Whether to enable heap profiling activation during startup.
+# When enabled, heap profiling will be activated if the `MALLOC_CONF` environment variable
+# is set to "prof:true,prof_active:false". The official image adds this env variable.
+# Default is true.
+enable_heap_profiling = true
+```
+
+By default, if you set `MALLOC_CONF=prof:true,prof_active:false`, the database will enable profiling during startup. You can disable this behavior by setting `enable_heap_profiling = false` in the configuration.
+
+### Starting with environment variables
+
 Start GreptimeDB instance with environment variables:
 
 ```bash
@@ -40,6 +57,23 @@ MALLOC_CONF=prof:true ./target/debug/greptime standalone start
 _RJEM_MALLOC_CONF=prof:true ./target/debug/greptime standalone start
 ```
 
+### Memory profiling control
+
+You can control heap profiling activation using the new HTTP APIs:
+
+```bash
+# Check current profiling status
+curl -X GET localhost:4000/debug/prof/mem/status
+
+# Activate heap profiling (if not already active)
+curl -X POST localhost:4000/debug/prof/mem/activate
+
+# Deactivate heap profiling
+curl -X POST localhost:4000/debug/prof/mem/deactivate
+```
+
+### Dump memory profiling data
+
 Dump memory profiling data through HTTP API:
 
 ```bash

docs/how-to/how-to-write-aggregate-function.md

@@ -1,4 +1,4 @@
-Currently, our query engine is based on DataFusion, so all aggregate function is executed by DataFusion, through its UDAF interface. You can find DataFusion's UDAF example [here](https://github.com/apache/arrow-datafusion/blob/arrow2/datafusion-examples/examples/simple_udaf.rs). Basically, we provide the same way as DataFusion to write aggregate functions: both are centered in a struct called "Accumulator" to accumulates states along the way in aggregation.
+Currently, our query engine is based on DataFusion, so all aggregate function is executed by DataFusion, through its UDAF interface. You can find DataFusion's UDAF example [here](https://github.com/apache/datafusion/tree/main/datafusion-examples/examples/simple_udaf.rs). Basically, we provide the same way as DataFusion to write aggregate functions: both are centered in a struct called "Accumulator" to accumulates states along the way in aggregation.
 
 However, DataFusion's UDAF implementation has a huge restriction, that it requires user to provide a concrete "Accumulator". Take `Median` aggregate function for example, to aggregate a `u32` datatype column, you have to write a `MedianU32`, and use `SELECT MEDIANU32(x)` in SQL. `MedianU32` cannot be used to aggregate a `i32` datatype column. Or, there's another way: you can use a special type that can hold all kinds of data (like our `Value` enum or Arrow's `ScalarValue`), and `match` all the way up to do aggregate calculations. It might work, though rather tedious. (But I think it's DataFusion's preferred way to write UDAF.)
 

docs/rfcs/2023-02-01-table-compaction.md

@@ -76,7 +76,7 @@ pub trait CompactionStrategy {
 ```
 
 The most suitable compaction strategy for time-series scenario would be 
-a hybrid strategy that combines time window compaction with size-tired compaction, just like [Cassandra](https://cassandra.apache.org/doc/latest/cassandra/operating/compaction/twcs.html) and [ScyllaDB](https://docs.scylladb.com/stable/architecture/compaction/compaction-strategies.html#time-window-compaction-strategy-twcs) does.
+a hybrid strategy that combines time window compaction with size-tired compaction, just like [Cassandra](https://cassandra.apache.org/doc/latest/cassandra/managing/operating/compaction/twcs.html) and [ScyllaDB](https://docs.scylladb.com/stable/architecture/compaction/compaction-strategies.html#time-window-compaction-strategy-twcs) does.
 
 We can first group SSTs in level n into buckets according to some predefined time window. Within that window, 
 SSTs are compacted in a size-tired manner (find SSTs with similar size and compact them to level n+1). 

docs/rfcs/2024-01-17-dataflow-framework.md

@@ -28,7 +28,7 @@ In order to do those things while maintaining a low memory footprint, you need t
 - Greptime Flow's is built on top of [Hydroflow](https://github.com/hydro-project/hydroflow).
 - We have three choices for the Dataflow/Streaming process framework for our simple continuous aggregation feature:
 1. Based on the timely/differential dataflow crate that [materialize](https://github.com/MaterializeInc/materialize) based on. Later, it's proved too obscure for a simple usage, and is hard to customize memory usage control.
-2. Based on a simple dataflow framework that we write from ground up, like what [arroyo](https://www.arroyo.dev/) or [risingwave](https://www.risingwave.dev/) did, for example the core streaming logic of [arroyo](https://github.com/ArroyoSystems/arroyo/blob/master/arroyo-datastream/src/lib.rs) only takes up to 2000 line of codes. However, it means maintaining another layer of dataflow framework, which might seem easy in the beginning, but I fear it might be too burdensome to maintain once we need more features.
+2. Based on a simple dataflow framework that we write from ground up, like what [arroyo](https://www.arroyo.dev/) or [risingwave](https://www.risingwave.dev/) did, for example the core streaming logic of [arroyo](https://github.com/ArroyoSystems/arroyo/blob/master/crates/arroyo-datastream/src/lib.rs) only takes up to 2000 line of codes. However, it means maintaining another layer of dataflow framework, which might seem easy in the beginning, but I fear it might be too burdensome to maintain once we need more features.
 3. Based on a simple and lower level dataflow framework that someone else write, like [hydroflow](https://github.com/hydro-project/hydroflow), this approach combines the best of both worlds. Firstly, it boasts ease of comprehension and customization. Secondly, the dataflow framework offers precisely the necessary features for crafting uncomplicated single-node dataflow programs while delivering decent performance.
 
 Hence, we choose the third option, and use a simple logical plan that's anagonistic to the underlying dataflow framework, as it only describe how the dataflow graph should be doing, not how it do that. And we built operator in hydroflow to execute the plan. And the result hydroflow graph is wrapped in a engine that only support data in/out and tick event to flush and compute the result. This provide a thin middle layer that's easy to maintain and allow switching to other dataflow framework if necessary.

docs/rfcs/2025-06-20-repartition.md

@@ -0,0 +1,154 @@
+---
+Feature Name: Repartition
+Tracking Issue: https://github.com/GreptimeTeam/greptimedb/issues/6558
+Date: 2025-06-20
+Author: "Ruihang Xia <waynestxia@gmail.com>"
+---
+
+# Summary
+
+This RFC proposes a method for repartitioning a table, to adjust the partition rule and data distribution.
+
+# Motivation
+
+With time passing, the data distribution and skew pattern of a table might change. We need a way to repartition the table to suit the new pattern.
+
+# Details
+
+Here is a rough workflow diagram of the entire repartition process, each step is described in detail below.
+
+```mermaid
+sequenceDiagram
+    participant Frontend
+    participant Metasrv
+    participant Datanodes
+    participant Region0 as Region 0
+    
+    Frontend->>Frontend: Process request, validation etc.
+    Frontend->>Metasrv: Submit procedure
+    Metasrv->>Metasrv: Compute diff and generate migration plan
+    Metasrv->>Metasrv: Allocate necessary region resources (with Paas)
+    Metasrv->>Datanodes: Stop compaction and snapshot
+    rect rgb(255, 225, 225)
+        note over Frontend, Region0: No Ingestion Period
+        Metasrv->>Frontend: Stop processing write requests
+        Metasrv->>Metasrv: Update metadata
+        Metasrv->>Frontend: Start processing read requests
+    end
+    Metasrv->>Datanodes: Update region rule, stage version changes from now on
+    Region0->>Region0: Compute new manifests for all regions
+    Region0->>Datanodes: Submit manifest changes
+    Metasrv->>Datanodes: Recover compaction and snapshot, make staged changes visible
+
+    note over Frontend, Datanodes: Reload Cache
+    Metasrv->>Metasrv: Release resources (with Paas)
+    Metasrv->>Metasrv: Schedule optional compaction (to remote compactor)
+```
+
+## Preprocessing
+
+This phase is for static analysis of the new partition rule. The server can know whether the repartitioning is possible, how to do the repartitioning, and how much resources are needed.
+
+In theory, the input and output partition rules for repartitioning can be completely unrelated. But in practice, to avoid a very large change set, we'll only allow two simple kinds of change. One splits one region into two regions (region split) and another merges two regions into one (region merge).
+
+After validating the new partition rule using the same validation logic as table creation, we compute the difference between the old and new partition rules. The resulting diff may contain several independent groups of changes. During subsequent processing, each group of changes can be handled independently and can succeed or fail without affecting other groups or creating non-idempotently retryable scenarios.
+
+Next, we generate a repartition plan for each group of changes. Each plan contains this information for all regions involved in that particular plan. And one target region will only be referenced by a single plan.
+
+With those plans, we can determine the resource requirements for the repartition operation, where resources here primarily refer to Regions. Metasrv will coordinate with PaaS layer to pre-allocate the necessary regions at this stage. These new regions start completely empty, and their metadata and manifests will be populated during subsequent modification steps.
+
+## Data Processing
+
+This phase is primarily for region's change, including region's metadata (route table and the corresponding rule) and manifest.
+
+Once we start processing one plan through a procedure, we'll first stop the region's compaction and snapshot. This is to avoid any states being removed due to compaction (which may removes old SST files) and snapshot (which may removes old manifest files).
+
+Metasrv will trying to update the metadata of partition, or the region route table (related to `PartitionRuleManager`). This step is in the "no ingestion" scope, so no new data will be ingested. Since this won't take much time, the affection to the cluster is minimized. Metasrv will also update the region rule to corresponding regions on Datanodes. 
+
+Every regions and all the ingestion requests to the region server will have a version of region rule, to identify under which rule the request is processed. The version can be something like `hash(region_rule)`. Once the region rule on region server is updated, all ingestion request with old rule will be rejected, and all requests with new rule will be accepted but not visible. They can still be flushed to persisted storage, but their version change (new manifest) will be staged.
+
+Then region 0 (or let metasrv to pick any operational region) will compute the new manifests for all target regions. This step is done by first reading all old manifests, and remapping the files with new partition rule, to get the content of new manifests. Notice this step only handles the manifests before region rule change on region server, and won't touch those staged manifests, as they are already with the new rule.
+
+Those new manifest will be submitted to the corresponding target regions by region 0 via a `RegionEdit` request. If this request falls after a few retries, region 0 will try to rollback this change by directly overwriting the manifest on object storage. and report this failure to metasrv and let the entire repartition procedure to fail. And we can also optionally compute the new manifest for those staged version changes (like another repartition) and submit them to the target regions to make the also visible even if the repartition fails.
+
+In the other hand, a successful `RegionEdit` request also acknowledges those staged version changes and make them visible.
+
+After this step, the repartition is done in the data plane. We can start to process compaction and snapshot again.
+
+## Postprocessing
+
+After the main processing is done, we can do some extra postprocessing to reduce the performance impact of repartition. Including reloading caches in frontend's route table, metasrv's kv cache and datanode's read/write/page cache etc.
+
+We can also schedule an optional compaction to reorganize all the data file under the new partition rule to reduce potential fragmentation or read amplification.
+
+## Procedure
+
+Here describe the repartition procedure step by step:
+
+- <on frontend> Validating repartition request
+- <on frontend> Initialize the repartition procedure
+- Calculate rule diff and repartition plan group
+- Allocate necessary new regions
+- Lock the table key
+- For each repartition subprocedure
+  - Stop compaction and snapshot
+  - Forbid new ingestion requests, update metadata, allow ingestion requests.
+  - Update region rule to regions
+  - Pick one region to calculate new manifest for all regions in this repartition group
+  - Let that region to apply new manifest to each region via `RegionEdit`
+  - If failed after some retries, revert this manifest change to other succeeded regions and mark this failure.
+  - If all succeeded, acknowledge those staged version changes and make them visible.
+  - Return result
+- Collect results from subprocedure.
+  - For those who failed, we need to restart those regions to force reconstruct their status from manifests
+  - For those who succeeded, collect and merge their rule diff
+- Unlock the table key
+- Report the result to user.
+- <in background> Reload cache
+- <in background> Maybe trigger a special compaction
+
+In addition of sequential step, rollback is also an important part of this procedure. There are three steps can be rolled back when unrecoverable failure occurs.
+
+If the metadata update is not committed, we can overwrite the metadata to previous version. This step is scoped in the "no ingestion" period, so no new data will be ingested and the status of both datanode and metasrv will be consistent.
+
+If the `RegionEdit` to other regions is not acknowledged, or partial acknowledged, we can directly overwrite the manifest on object storage from the central region (who computes the new manifest), and force region server to reload corresponding region to load its state from object storage to recover.
+
+If the staged version changes are not acknowledged, we can re-compute manifest based on old rule for staged data, and apply them directly like above. This is like another smaller repartition for those staged data.
+
+## Region rule validation and diff calculation
+
+In the current codebase, the rule checker is not complete. It can't check uniqueness and completeness of the rule. This RFC also propose a new way to validate the rule.
+
+The proposed validation way is based on a check-point system, which first generates a group of check-points from the rule, and then check if all the point is covered and only covered by one rule.
+
+All the partition rule expressionis limited to be the form of `<column> <operator> <value>`, and the operator is limited to be comparison operators. Those expressions are allowed to be nested with `AND` and `OR` operators. Based on this, we can first extract all the unique values on each column, adding and subtracting a little epsilon to cover its left and right boundary.
+
+Since we accept integer, float and string as the value type, compute on them directly is not convenient. So we'll first normalize them to a common type and only need to preserve the relative partial ordering. This also avoids the problem of "what is next/previous value" of string and "what's a good precision" for float.
+
+After normalization, we get a set of scatter points for each column. Then we can generate a set of check-points by combining all the scatter points like building a cartesian product. This might bring a large number of check-points, so we can do an prune optimization to remove some of them by merging some of the expression zones. Those expressions who have identical N-1 edge sub-expressions with one adjacent edge can be merged together. This prune check is with a time complexity of O(N * M * log(M)), where N is the number of active dimensions and M is the number of expression zones. Diff calculation is also done by finding different expression zones between the old and new rule set, and check if we can transform one to another by merging some of the expression zones.
+
+The step to validate the check-points set against expressions can be treated as a tiny expression of `PhysicalExpr`. This evaluation will give a boolean matrix of K*M shape, where K is the number of check-points. We then check in each row of the matrix, if there is one and only one true value.
+
+## Compute and use new manifest
+
+We can generate a new set of manifest file based on old manifest and two versions of rule. From abvoe rule processing part, we can tell how a new rule & region is from previous one. So a simple way to get the new manifest is also apply the step of change to manifest files. E.g., if region A is from region B and C, we simply combine all file IDs from B and C to generate the content of A.
+
+If necessary, we can do this better by involving some metadata related to data, like min-max statistics of each file, and pre-evaluate over min-max to filter out unneeded files when generating new manifest.
+
+The way to use new manifest needs one more extra step based on the current implementation. We'll need to record either in manifest or in file metadata, of what rule is used when generating (flush or compaction) a SST file. Then in every single read request, we need to append the current region rule as predicate to the read request, to ensure no data belong to other regions will be read. We can use the stored region rule to reduce the number of new predicates to apply, by removing the identical predicate between the current region rule and the stored region rule. So ideally in a table that has not been repartitioned recently, the overhead of checking region rule is minimal.
+
+## Pre-required tasks
+
+In above steps, we assume some functionalities are implemented. Here list them with where they are used and how to implement them.
+
+### Cross-region read
+
+The current data directory structure is `{table_id}/{region_id}/[data/metadata]/{file_id}`, every region can only access files under their own directory. After repartition, data file may be placed in other previous old regions. So we need to support cross-region read. This new access method allows region to access any file under the same table. Related tracking issue is <https://github.com/GreptimeTeam/greptimedb/issues/6409>.
+
+### Global GC worker
+
+This is to simplify state management of data files. As one file may be referenced in multiple manifests, or no manifest at all. After this, every region and the repartition process only need to care about generateing and using new files, without tracking whether a file should be deleted or not. Leaving the deletion to the global GC worker. This worker basically works by counting reference from manifest file, and remove unused one. Related tracking issue is **TBD**.
+
+# Alternatives
+
+In the "Data Processing" section, we can enlarge the "no ingestion" period to include almost all the steps. This can simplify the entire procedure by a lot, but will bring a longer time of ingestion pause which may not be acceptable.

docs/rfcs/2025-07-04-compatibility-test-framework.md

@@ -0,0 +1,151 @@
+---
+Feature Name: Compatibility Test Framework
+Tracking Issue: TBD
+Date: 2025-07-04
+Author: "Ruihang Xia <waynestxia@gmail.com>"
+---
+
+# Summary
+
+This RFC proposes a compatibility test framework for GreptimeDB to ensure backward/forward compatibility for different versions of GreptimeDB.
+
+# Motivation
+
+In current practice, we don't have a systematic way to test and ensure the compatibility of different versions of GreptimeDB. Each time we release a new version, we need to manually test the compatibility with ad-hoc cases. This is not only time-consuming, but also prone to errors and unmaintainable. Highly rely on the release manager to ensure the compatibility of different versions of GreptimeDB.
+
+We don't have a detailed guide on the release SoP of how to test and ensure the compatibility of the new version. And has broken the compatibility of the new version many times (`v0.14.1` and `v0.15.1` are two examples, which are both released right after the major release).
+
+# Details
+
+This RFC proposes a compatibility test framework that is easy to maintain, extend and run. It can tell the compatibility between any given two versions of GreptimeDB, both backward and forward. It's based on the Sqlness library but used in a different way.
+
+Generally speaking, the framework is composed of two parts:
+
+1. Test cases: A set of test cases that are maintained dedicatedly for the compatibility test. Still in the `.sql` and `.result` format.
+2. Test framework: A new sqlness runner that is used to run the test cases. With some new features that is not required by the integration sqlness test.
+
+## Test Cases
+
+### Structure
+
+The case set is organized in three parts:
+
+- `1.feature`: Use a new feature
+- `2.verify`: Verify database behavior
+- `3.cleanup`: Paired with `1.feature`, cleanup the test environment.
+
+These three parts are organized in a tree structure, and should be run in sequence:
+
+```
+compatibility_test/
+├── 1.feature/
+│   ├── feature-a/
+│   ├── feature-b/
+│   └── feature-c/
+├── 2.verify/
+│   ├── verify-metadata/
+│   ├── verify-data/
+│   └── verify-schema/
+└── 3.cleanup/
+    ├── cleanup-a/
+    ├── cleanup-b/
+    └── cleanup-c/
+```
+
+### Example
+
+For example, for a new feature like adding new index option ([#6416](https://github.com/GreptimeTeam/greptimedb/pull/6416)), we (who implement the feature) create a new test case like this:
+
+```sql
+-- path: compatibility_test/1.feature/index-option/granularity_and_false_positive_rate.sql
+
+-- SQLNESS ARG since=0.15.0
+-- SQLNESS IGNORE_RESULT
+CREATE TABLE granularity_and_false_positive_rate (ts timestamp time index, val double) with ("index.granularity" = "8192", "index.false_positive_rate" = "0.01");
+```
+
+And
+
+```sql
+-- path: compatibility_test/3.cleanup/index-option/granularity_and_false_positive_rate.sql
+drop table granularity_and_false_positive_rate;
+```
+
+Since this new feature don't require some special way to verify the database behavior, we can reuse existing test cases in `2.verify/` to verify the database behavior. For example, we can reuse the `verify-metadata` test case to verify the metadata of the table.
+
+```sql
+-- path: compatibility_test/2.verify/verify-metadata/show-create-table.sql
+
+-- SQLNESS TEMPLATE TABLE="SHOW TABLES";
+SHOW CREATE TABLE $TABLE;
+```
+
+In this example, we use some new sqlness features that will be introduced in the next section (`since`, `IGNORE_RESULT`, `TEMPLATE`).
+
+### Maintenance
+
+Each time implement a new feature that should be covered by the compatibility test, we should create a new test case in `1.feature/` and `3.cleanup/` for them. And check if existing cases in `2.verify/` can be reused to verify the database behavior.
+
+This simulates an enthusiastic user who uses all the new features at the first time. All the new Maintenance burden is on the feature implementer to write one more test case for the new feature, to "fixation" the behavior. And once there is a breaking change in the future, it can be detected by the compatibility test framework automatically.
+
+Another topic is about deprecation. If a feature is deprecated, we should also mark it in the test case. Still use above example, assume we deprecate the `index.granularity` and `index.false_positive_rate` index options in `v0.99.0`, we can mark them as:
+```sql
+-- SQLNESS ARG since=0.15.0 till=0.99.0
+...
+```
+
+This tells the framework to ignore this feature in version `v0.99.0` and later. Currently, we have so many experimental features that are scheduled to be broken in the future, this is a good way to mark them.
+
+## Test Framework
+
+This section is about new sqlness features required by this framework.
+
+### Since and Till
+
+Follows the `ARG` interceptor in sqlness, we can mark a feature is available between two given versions. Only the `since` is required:
+
+```sql
+-- SQLNESS ARG since=VERSION_STRING [till=VERSION_STRING]
+```
+
+### IGNORE_RESULT
+
+`IGNORE_RESULT` is a new interceptor, it tells the runner to ignore the result of the query, only check whether the query is executed successfully.
+
+This is useful to reduce the Maintenance burden of the test cases, unlike the integration sqlness test, in most cases we don't care about the result of the query, only need to make sure the query is executed successfully.
+
+### TEMPLATE
+
+`TEMPLATE` is another new interceptor, it can generate queries from a template based on a runtime data.
+
+In above example, we need to run the `SHOW CREATE TABLE` query for all existing tables, so we can use the `TEMPLATE` interceptor to generate the query with a dynamic table list.
+
+### RUNNER
+
+There are also some extra requirement for the runner itself:
+
+- It should run the test cases in sequence, first `1.feature/`, then `2.verify/`, and finally `3.cleanup/`.
+- It should be able to fetch required version automatically to finish the test.
+- It should handle the `since` and `till` properly.
+
+On the `1.feature` phase, the runner needs to identify all features need to be tested by version number. And then restart with a new version (the `to` version) to run `2.verify/` and `3.cleanup/` phase.
+
+## Test Report
+
+Finally, we can run the compatibility test to verify the compatibility between any given two versions of GreptimeDB, for example:
+
+```bash
+# check backward compatibility between v0.15.0 and v0.16.0 when releasing v0.16.0
+./sqlness run --from=0.15.0 --to=0.16.0
+
+# check forward compatibility when downgrading from v0.15.0 to v0.13.0
+./sqlness run --from=0.15.0 --to=0.13.0
+```
+
+We can also use a script to run the compatibility test for all the versions in a given range to give a quick report with all versions we need.
+
+And we always bump the version in `Cargo.toml` to the next major release version, so the next major release version can be used as "latest" unpublished version for scenarios like local testing.
+
+# Alternatives
+
+There was a previous attempt to implement a compatibility test framework that was disabled due to some reasons [#3728](https://github.com/GreptimeTeam/greptimedb/issues/3728).

grafana/README.md

@@ -83,7 +83,7 @@ If you use the [Helm Chart](https://github.com/GreptimeTeam/helm-charts) to depl
 - `monitoring.enabled=true`: Deploys a standalone GreptimeDB instance dedicated to monitoring the cluster;
 - `grafana.enabled=true`: Deploys Grafana and automatically imports the monitoring dashboard;
 
-The standalone GreptimeDB instance will collect metrics from your cluster, and the dashboard will be available in the Grafana UI. For detailed deployment instructions, please refer to our [Kubernetes deployment guide](https://docs.greptime.com/user-guide/deployments-administration/deploy-on-kubernetes/getting-started).
+The standalone GreptimeDB instance will collect metrics from your cluster, and the dashboard will be available in the Grafana UI. For detailed deployment instructions, please refer to our [Kubernetes deployment guide](https://docs.greptime.com/user-guide/deployments-administration/deploy-on-kubernetes/overview).
 
 ### Self-host Prometheus and import dashboards manually
 

grafana/dashboards/metrics/cluster/dashboard.md

@@ -21,14 +21,14 @@
 # Resources
 | Title | Query | Type | Description | Datasource | Unit | Legend Format |
 | --- | --- | --- | --- | --- | --- | --- |
-| Datanode Memory per Instance | `sum(process_resident_memory_bytes{instance=~"$datanode"}) by (instance, pod)` | `timeseries` | Current memory usage by instance | `prometheus` | `decbytes` | `[{{instance}}]-[{{ pod }}]` |
-| Datanode CPU Usage per Instance | `sum(rate(process_cpu_seconds_total{instance=~"$datanode"}[$__rate_interval]) * 1000) by (instance, pod)` | `timeseries` | Current cpu usage by instance | `prometheus` | `none` | `[{{ instance }}]-[{{ pod }}]` |
-| Frontend Memory per Instance | `sum(process_resident_memory_bytes{instance=~"$frontend"}) by (instance, pod)` | `timeseries` | Current memory usage by instance | `prometheus` | `decbytes` | `[{{ instance }}]-[{{ pod }}]` |
-| Frontend CPU Usage per Instance | `sum(rate(process_cpu_seconds_total{instance=~"$frontend"}[$__rate_interval]) * 1000) by (instance, pod)` | `timeseries` | Current cpu usage by instance | `prometheus` | `none` | `[{{ instance }}]-[{{ pod }}]-cpu` |
-| Metasrv Memory per Instance | `sum(process_resident_memory_bytes{instance=~"$metasrv"}) by (instance, pod)` | `timeseries` | Current memory usage by instance | `prometheus` | `decbytes` | `[{{ instance }}]-[{{ pod }}]-resident` |
-| Metasrv CPU Usage per Instance | `sum(rate(process_cpu_seconds_total{instance=~"$metasrv"}[$__rate_interval]) * 1000) by (instance, pod)` | `timeseries` | Current cpu usage by instance | `prometheus` | `none` | `[{{ instance }}]-[{{ pod }}]` |
-| Flownode Memory per Instance | `sum(process_resident_memory_bytes{instance=~"$flownode"}) by (instance, pod)` | `timeseries` | Current memory usage by instance | `prometheus` | `decbytes` | `[{{ instance }}]-[{{ pod }}]` |
-| Flownode CPU Usage per Instance | `sum(rate(process_cpu_seconds_total{instance=~"$flownode"}[$__rate_interval]) * 1000) by (instance, pod)` | `timeseries` | Current cpu usage by instance | `prometheus` | `none` | `[{{ instance }}]-[{{ pod }}]` |
+| Datanode Memory per Instance | `sum(process_resident_memory_bytes{instance=~"$datanode"}) by (instance, pod)`<br/>`max(greptime_memory_limit_in_bytes{app="greptime-datanode"})` | `timeseries` | Current memory usage by instance | `prometheus` | `bytes` | `[{{instance}}]-[{{ pod }}]` |
+| Datanode CPU Usage per Instance | `sum(rate(process_cpu_seconds_total{instance=~"$datanode"}[$__rate_interval]) * 1000) by (instance, pod)`<br/>`max(greptime_cpu_limit_in_millicores{app="greptime-datanode"})` | `timeseries` | Current cpu usage by instance | `prometheus` | `none` | `[{{ instance }}]-[{{ pod }}]` |
+| Frontend Memory per Instance | `sum(process_resident_memory_bytes{instance=~"$frontend"}) by (instance, pod)`<br/>`max(greptime_memory_limit_in_bytes{app="greptime-frontend"})` | `timeseries` | Current memory usage by instance | `prometheus` | `bytes` | `[{{ instance }}]-[{{ pod }}]` |
+| Frontend CPU Usage per Instance | `sum(rate(process_cpu_seconds_total{instance=~"$frontend"}[$__rate_interval]) * 1000) by (instance, pod)`<br/>`max(greptime_cpu_limit_in_millicores{app="greptime-frontend"})` | `timeseries` | Current cpu usage by instance | `prometheus` | `none` | `[{{ instance }}]-[{{ pod }}]-cpu` |
+| Metasrv Memory per Instance | `sum(process_resident_memory_bytes{instance=~"$metasrv"}) by (instance, pod)`<br/>`max(greptime_memory_limit_in_bytes{app="greptime-metasrv"})` | `timeseries` | Current memory usage by instance | `prometheus` | `bytes` | `[{{ instance }}]-[{{ pod }}]-resident` |
+| Metasrv CPU Usage per Instance | `sum(rate(process_cpu_seconds_total{instance=~"$metasrv"}[$__rate_interval]) * 1000) by (instance, pod)`<br/>`max(greptime_cpu_limit_in_millicores{app="greptime-metasrv"})` | `timeseries` | Current cpu usage by instance | `prometheus` | `none` | `[{{ instance }}]-[{{ pod }}]` |
+| Flownode Memory per Instance | `sum(process_resident_memory_bytes{instance=~"$flownode"}) by (instance, pod)`<br/>`max(greptime_memory_limit_in_bytes{app="greptime-flownode"})` | `timeseries` | Current memory usage by instance | `prometheus` | `bytes` | `[{{ instance }}]-[{{ pod }}]` |
+| Flownode CPU Usage per Instance | `sum(rate(process_cpu_seconds_total{instance=~"$flownode"}[$__rate_interval]) * 1000) by (instance, pod)`<br/>`max(greptime_cpu_limit_in_millicores{app="greptime-flownode"})` | `timeseries` | Current cpu usage by instance | `prometheus` | `none` | `[{{ instance }}]-[{{ pod }}]` |
 # Frontend Requests
 | Title | Query | Type | Description | Datasource | Unit | Legend Format |
 | --- | --- | --- | --- | --- | --- | --- |
@@ -72,18 +72,19 @@
 | Region Worker Handle Bulk Insert Requests | `histogram_quantile(0.95, sum by(le,instance, stage, pod) (rate(greptime_region_worker_handle_write_bucket[$__rate_interval])))`<br/>`sum by(instance, stage, pod) (rate(greptime_region_worker_handle_write_sum[$__rate_interval]))/sum by(instance, stage, pod) (rate(greptime_region_worker_handle_write_count[$__rate_interval]))` | `timeseries` | Per-stage elapsed time for region worker to handle bulk insert region requests. | `prometheus` | `s` | `[{{instance}}]-[{{pod}}]-[{{stage}}]-P95` |
 | Active Series and Field Builders Count | `sum by(instance, pod) (greptime_mito_memtable_active_series_count)`<br/>`sum by(instance, pod) (greptime_mito_memtable_field_builder_count)` | `timeseries` | Compaction oinput output bytes | `prometheus` | `none` | `[{{instance}}]-[{{pod}}]-series` |
 | Region Worker Convert Requests | `histogram_quantile(0.95, sum by(le, instance, stage, pod) (rate(greptime_datanode_convert_region_request_bucket[$__rate_interval])))`<br/>`sum by(le,instance, stage, pod) (rate(greptime_datanode_convert_region_request_sum[$__rate_interval]))/sum by(le,instance, stage, pod) (rate(greptime_datanode_convert_region_request_count[$__rate_interval]))` | `timeseries` | Per-stage elapsed time for region worker to decode requests. | `prometheus` | `s` | `[{{instance}}]-[{{pod}}]-[{{stage}}]-P95` |
+| Cache Miss | `sum by (instance,pod, type) (rate(greptime_mito_cache_miss{instance=~"$datanode"}[$__rate_interval]))` | `timeseries` | The local cache miss of the datanode. | `prometheus` | -- | `[{{instance}}]-[{{pod}}]-[{{type}}]` |
 # OpenDAL
 | Title | Query | Type | Description | Datasource | Unit | Legend Format |
 | --- | --- | --- | --- | --- | --- | --- |
 | QPS per Instance | `sum by(instance, pod, scheme, operation) (rate(opendal_operation_duration_seconds_count{instance=~"$datanode"}[$__rate_interval]))` | `timeseries` | QPS per Instance. | `prometheus` | `ops` | `[{{instance}}]-[{{pod}}]-[{{scheme}}]-[{{operation}}]` |
-| Read QPS per Instance | `sum by(instance, pod, scheme) (rate(opendal_operation_duration_seconds_count{instance=~"$datanode", operation="read"}[$__rate_interval]))` | `timeseries` | Read QPS per Instance. | `prometheus` | `ops` | `[{{instance}}]-[{{pod}}]-[{{scheme}}]` |
-| Read P99 per Instance | `histogram_quantile(0.99, sum by(instance, pod, le, scheme) (rate(opendal_operation_duration_seconds_bucket{instance=~"$datanode",operation="read"}[$__rate_interval])))` | `timeseries` | Read P99 per Instance. | `prometheus` | `s` | `[{{instance}}]-[{{pod}}]-{{scheme}}` |
-| Write QPS per Instance | `sum by(instance, pod, scheme) (rate(opendal_operation_duration_seconds_count{instance=~"$datanode", operation="write"}[$__rate_interval]))` | `timeseries` | Write QPS per Instance. | `prometheus` | `ops` | `[{{instance}}]-[{{pod}}]-{{scheme}}` |
-| Write P99 per Instance | `histogram_quantile(0.99, sum by(instance, pod, le, scheme) (rate(opendal_operation_duration_seconds_bucket{instance=~"$datanode", operation="write"}[$__rate_interval])))` | `timeseries` | Write P99 per Instance. | `prometheus` | `s` | `[{{instance}}]-[{{pod}}]-[{{scheme}}]` |
+| Read QPS per Instance | `sum by(instance, pod, scheme, operation) (rate(opendal_operation_duration_seconds_count{instance=~"$datanode", operation=~"read\|Reader::read"}[$__rate_interval]))` | `timeseries` | Read QPS per Instance. | `prometheus` | `ops` | `[{{instance}}]-[{{pod}}]-[{{scheme}}]-[{{operation}}]` |
+| Read P99 per Instance | `histogram_quantile(0.99, sum by(instance, pod, le, scheme, operation) (rate(opendal_operation_duration_seconds_bucket{instance=~"$datanode",operation=~"read\|Reader::read"}[$__rate_interval])))` | `timeseries` | Read P99 per Instance. | `prometheus` | `s` | `[{{instance}}]-[{{pod}}]-[{{scheme}}]-[{{operation}}]` |
+| Write QPS per Instance | `sum by(instance, pod, scheme, operation) (rate(opendal_operation_duration_seconds_count{instance=~"$datanode", operation=~"write\|Writer::write\|Writer::close"}[$__rate_interval]))` | `timeseries` | Write QPS per Instance. | `prometheus` | `ops` | `[{{instance}}]-[{{pod}}]-[{{scheme}}]-[{{operation}}]` |
+| Write P99 per Instance | `histogram_quantile(0.99, sum by(instance, pod, le, scheme, operation) (rate(opendal_operation_duration_seconds_bucket{instance=~"$datanode", operation =~ "Writer::write\|Writer::close\|write"}[$__rate_interval])))` | `timeseries` | Write P99 per Instance. | `prometheus` | `s` | `[{{instance}}]-[{{pod}}]-[{{scheme}}]-[{{operation}}]` |
 | List QPS per Instance | `sum by(instance, pod, scheme) (rate(opendal_operation_duration_seconds_count{instance=~"$datanode", operation="list"}[$__rate_interval]))` | `timeseries` | List QPS per Instance. | `prometheus` | `ops` | `[{{instance}}]-[{{pod}}]-[{{scheme}}]` |
 | List P99 per Instance | `histogram_quantile(0.99, sum by(instance, pod, le, scheme) (rate(opendal_operation_duration_seconds_bucket{instance=~"$datanode", operation="list"}[$__rate_interval])))` | `timeseries` | List P99 per Instance. | `prometheus` | `s` | `[{{instance}}]-[{{pod}}]-[{{scheme}}]` |
 | Other Requests per Instance | `sum by(instance, pod, scheme, operation) (rate(opendal_operation_duration_seconds_count{instance=~"$datanode",operation!~"read\|write\|list\|stat"}[$__rate_interval]))` | `timeseries` | Other Requests per Instance. | `prometheus` | `ops` | `[{{instance}}]-[{{pod}}]-[{{scheme}}]-[{{operation}}]` |
-| Other Request P99 per Instance | `histogram_quantile(0.99, sum by(instance, pod, le, scheme, operation) (rate(opendal_operation_duration_seconds_bucket{instance=~"$datanode", operation!~"read\|write\|list"}[$__rate_interval])))` | `timeseries` | Other Request P99 per Instance. | `prometheus` | `s` | `[{{instance}}]-[{{pod}}]-[{{scheme}}]-[{{operation}}]` |
+| Other Request P99 per Instance | `histogram_quantile(0.99, sum by(instance, pod, le, scheme, operation) (rate(opendal_operation_duration_seconds_bucket{instance=~"$datanode", operation!~"read\|write\|list\|Writer::write\|Writer::close\|Reader::read"}[$__rate_interval])))` | `timeseries` | Other Request P99 per Instance. | `prometheus` | `s` | `[{{instance}}]-[{{pod}}]-[{{scheme}}]-[{{operation}}]` |
 | Opendal traffic | `sum by(instance, pod, scheme, operation) (rate(opendal_operation_bytes_sum{instance=~"$datanode"}[$__rate_interval]))` | `timeseries` | Total traffic as in bytes by instance and operation | `prometheus` | `decbytes` | `[{{instance}}]-[{{pod}}]-[{{scheme}}]-[{{operation}}]` |
 | OpenDAL errors per Instance | `sum by(instance, pod, scheme, operation, error) (rate(opendal_operation_errors_total{instance=~"$datanode", error!="NotFound"}[$__rate_interval]))` | `timeseries` | OpenDAL error counts per Instance. | `prometheus` | -- | `[{{instance}}]-[{{pod}}]-[{{scheme}}]-[{{operation}}]-[{{error}}]` |
 # Metasrv

grafana/dashboards/metrics/cluster/dashboard.yaml

@@ -180,13 +180,18 @@ groups:
         - title: Datanode Memory per Instance
           type: timeseries
           description: Current memory usage by instance
-          unit: decbytes
+          unit: bytes
           queries:
             - expr: sum(process_resident_memory_bytes{instance=~"$datanode"}) by (instance, pod)
               datasource:
                 type: prometheus
                 uid: ${metrics}
               legendFormat: '[{{instance}}]-[{{ pod }}]'
+            - expr: max(greptime_memory_limit_in_bytes{app="greptime-datanode"})
+              datasource:
+                type: prometheus
+                uid: ${metrics}
+              legendFormat: limit
         - title: Datanode CPU Usage per Instance
           type: timeseries
           description: Current cpu usage by instance
@@ -197,16 +202,26 @@ groups:
                 type: prometheus
                 uid: ${metrics}
               legendFormat: '[{{ instance }}]-[{{ pod }}]'
+            - expr: max(greptime_cpu_limit_in_millicores{app="greptime-datanode"})
+              datasource:
+                type: prometheus
+                uid: ${metrics}
+              legendFormat: limit
         - title: Frontend Memory per Instance
           type: timeseries
           description: Current memory usage by instance
-          unit: decbytes
+          unit: bytes
           queries:
             - expr: sum(process_resident_memory_bytes{instance=~"$frontend"}) by (instance, pod)
               datasource:
                 type: prometheus
                 uid: ${metrics}
               legendFormat: '[{{ instance }}]-[{{ pod }}]'
+            - expr: max(greptime_memory_limit_in_bytes{app="greptime-frontend"})
+              datasource:
+                type: prometheus
+                uid: ${metrics}
+              legendFormat: limit
         - title: Frontend CPU Usage per Instance
           type: timeseries
           description: Current cpu usage by instance
@@ -217,16 +232,26 @@ groups:
                 type: prometheus
                 uid: ${metrics}
               legendFormat: '[{{ instance }}]-[{{ pod }}]-cpu'
+            - expr: max(greptime_cpu_limit_in_millicores{app="greptime-frontend"})
+              datasource:
+                type: prometheus
+                uid: ${metrics}
+              legendFormat: limit
         - title: Metasrv Memory per Instance
           type: timeseries
           description: Current memory usage by instance
-          unit: decbytes
+          unit: bytes
           queries:
             - expr: sum(process_resident_memory_bytes{instance=~"$metasrv"}) by (instance, pod)
               datasource:
                 type: prometheus
                 uid: ${metrics}
               legendFormat: '[{{ instance }}]-[{{ pod }}]-resident'
+            - expr: max(greptime_memory_limit_in_bytes{app="greptime-metasrv"})
+              datasource:
+                type: prometheus
+                uid: ${metrics}
+              legendFormat: limit
         - title: Metasrv CPU Usage per Instance
           type: timeseries
           description: Current cpu usage by instance
@@ -237,16 +262,26 @@ groups:
                 type: prometheus
                 uid: ${metrics}
               legendFormat: '[{{ instance }}]-[{{ pod }}]'
+            - expr: max(greptime_cpu_limit_in_millicores{app="greptime-metasrv"})
+              datasource:
+                type: prometheus
+                uid: ${metrics}
+              legendFormat: limit
         - title: Flownode Memory per Instance
           type: timeseries
           description: Current memory usage by instance
-          unit: decbytes
+          unit: bytes
           queries:
             - expr: sum(process_resident_memory_bytes{instance=~"$flownode"}) by (instance, pod)
               datasource:
                 type: prometheus
                 uid: ${metrics}
               legendFormat: '[{{ instance }}]-[{{ pod }}]'
+            - expr: max(greptime_memory_limit_in_bytes{app="greptime-flownode"})
+              datasource:
+                type: prometheus
+                uid: ${metrics}
+              legendFormat: limit
         - title: Flownode CPU Usage per Instance
           type: timeseries
           description: Current cpu usage by instance
@@ -257,6 +292,11 @@ groups:
                 type: prometheus
                 uid: ${metrics}
               legendFormat: '[{{ instance }}]-[{{ pod }}]'
+            - expr: max(greptime_cpu_limit_in_millicores{app="greptime-flownode"})
+              datasource:
+                type: prometheus
+                uid: ${metrics}
+              legendFormat: limit
     - title: Frontend Requests
       panels:
         - title: HTTP QPS per Instance
@@ -642,6 +682,15 @@ groups:
                 type: prometheus
                 uid: ${metrics}
               legendFormat: '[{{instance}}]-[{{pod}}]-[{{stage}}]-AVG'
+        - title: Cache Miss
+          type: timeseries
+          description: The local cache miss of the datanode.
+          queries:
+            - expr: sum by (instance,pod, type) (rate(greptime_mito_cache_miss{instance=~"$datanode"}[$__rate_interval]))
+              datasource:
+                type: prometheus
+                uid: ${metrics}
+              legendFormat: '[{{instance}}]-[{{pod}}]-[{{type}}]'
     - title: OpenDAL
       panels:
         - title: QPS per Instance
@@ -659,41 +708,41 @@ groups:
           description: Read QPS per Instance.
           unit: ops
           queries:
-            - expr: sum by(instance, pod, scheme) (rate(opendal_operation_duration_seconds_count{instance=~"$datanode", operation="read"}[$__rate_interval]))
+            - expr: sum by(instance, pod, scheme, operation) (rate(opendal_operation_duration_seconds_count{instance=~"$datanode", operation=~"read|Reader::read"}[$__rate_interval]))
               datasource:
                 type: prometheus
                 uid: ${metrics}
-              legendFormat: '[{{instance}}]-[{{pod}}]-[{{scheme}}]'
+              legendFormat: '[{{instance}}]-[{{pod}}]-[{{scheme}}]-[{{operation}}]'
         - title: Read P99 per Instance
           type: timeseries
           description: Read P99 per Instance.
           unit: s
           queries:
-            - expr: histogram_quantile(0.99, sum by(instance, pod, le, scheme) (rate(opendal_operation_duration_seconds_bucket{instance=~"$datanode",operation="read"}[$__rate_interval])))
+            - expr: histogram_quantile(0.99, sum by(instance, pod, le, scheme, operation) (rate(opendal_operation_duration_seconds_bucket{instance=~"$datanode",operation=~"read|Reader::read"}[$__rate_interval])))
               datasource:
                 type: prometheus
                 uid: ${metrics}
-              legendFormat: '[{{instance}}]-[{{pod}}]-{{scheme}}'
+              legendFormat: '[{{instance}}]-[{{pod}}]-[{{scheme}}]-[{{operation}}]'
         - title: Write QPS per Instance
           type: timeseries
           description: Write QPS per Instance.
           unit: ops
           queries:
-            - expr: sum by(instance, pod, scheme) (rate(opendal_operation_duration_seconds_count{instance=~"$datanode", operation="write"}[$__rate_interval]))
+            - expr: sum by(instance, pod, scheme, operation) (rate(opendal_operation_duration_seconds_count{instance=~"$datanode", operation=~"write|Writer::write|Writer::close"}[$__rate_interval]))
               datasource:
                 type: prometheus
                 uid: ${metrics}
-              legendFormat: '[{{instance}}]-[{{pod}}]-{{scheme}}'
+              legendFormat: '[{{instance}}]-[{{pod}}]-[{{scheme}}]-[{{operation}}]'
         - title: Write P99 per Instance
           type: timeseries
           description: Write P99 per Instance.
           unit: s
           queries:
-            - expr: histogram_quantile(0.99, sum by(instance, pod, le, scheme) (rate(opendal_operation_duration_seconds_bucket{instance=~"$datanode", operation="write"}[$__rate_interval])))
+            - expr: histogram_quantile(0.99, sum by(instance, pod, le, scheme, operation) (rate(opendal_operation_duration_seconds_bucket{instance=~"$datanode", operation =~ "Writer::write|Writer::close|write"}[$__rate_interval])))
               datasource:
                 type: prometheus
                 uid: ${metrics}
-              legendFormat: '[{{instance}}]-[{{pod}}]-[{{scheme}}]'
+              legendFormat: '[{{instance}}]-[{{pod}}]-[{{scheme}}]-[{{operation}}]'
         - title: List QPS per Instance
           type: timeseries
           description: List QPS per Instance.
@@ -729,7 +778,7 @@ groups:
           description: Other Request P99 per Instance.
           unit: s
           queries:
-            - expr: histogram_quantile(0.99, sum by(instance, pod, le, scheme, operation) (rate(opendal_operation_duration_seconds_bucket{instance=~"$datanode", operation!~"read|write|list"}[$__rate_interval])))
+            - expr: histogram_quantile(0.99, sum by(instance, pod, le, scheme, operation) (rate(opendal_operation_duration_seconds_bucket{instance=~"$datanode", operation!~"read|write|list|Writer::write|Writer::close|Reader::read"}[$__rate_interval])))
               datasource:
                 type: prometheus
                 uid: ${metrics}

grafana/dashboards/metrics/standalone/dashboard.md

@@ -21,14 +21,14 @@
 # Resources
 | Title | Query | Type | Description | Datasource | Unit | Legend Format |
 | --- | --- | --- | --- | --- | --- | --- |
-| Datanode Memory per Instance | `sum(process_resident_memory_bytes{}) by (instance, pod)` | `timeseries` | Current memory usage by instance | `prometheus` | `decbytes` | `[{{instance}}]-[{{ pod }}]` |
-| Datanode CPU Usage per Instance | `sum(rate(process_cpu_seconds_total{}[$__rate_interval]) * 1000) by (instance, pod)` | `timeseries` | Current cpu usage by instance | `prometheus` | `none` | `[{{ instance }}]-[{{ pod }}]` |
-| Frontend Memory per Instance | `sum(process_resident_memory_bytes{}) by (instance, pod)` | `timeseries` | Current memory usage by instance | `prometheus` | `decbytes` | `[{{ instance }}]-[{{ pod }}]` |
-| Frontend CPU Usage per Instance | `sum(rate(process_cpu_seconds_total{}[$__rate_interval]) * 1000) by (instance, pod)` | `timeseries` | Current cpu usage by instance | `prometheus` | `none` | `[{{ instance }}]-[{{ pod }}]-cpu` |
-| Metasrv Memory per Instance | `sum(process_resident_memory_bytes{}) by (instance, pod)` | `timeseries` | Current memory usage by instance | `prometheus` | `decbytes` | `[{{ instance }}]-[{{ pod }}]-resident` |
-| Metasrv CPU Usage per Instance | `sum(rate(process_cpu_seconds_total{}[$__rate_interval]) * 1000) by (instance, pod)` | `timeseries` | Current cpu usage by instance | `prometheus` | `none` | `[{{ instance }}]-[{{ pod }}]` |
-| Flownode Memory per Instance | `sum(process_resident_memory_bytes{}) by (instance, pod)` | `timeseries` | Current memory usage by instance | `prometheus` | `decbytes` | `[{{ instance }}]-[{{ pod }}]` |
-| Flownode CPU Usage per Instance | `sum(rate(process_cpu_seconds_total{}[$__rate_interval]) * 1000) by (instance, pod)` | `timeseries` | Current cpu usage by instance | `prometheus` | `none` | `[{{ instance }}]-[{{ pod }}]` |
+| Datanode Memory per Instance | `sum(process_resident_memory_bytes{}) by (instance, pod)`<br/>`max(greptime_memory_limit_in_bytes{app="greptime-datanode"})` | `timeseries` | Current memory usage by instance | `prometheus` | `bytes` | `[{{instance}}]-[{{ pod }}]` |
+| Datanode CPU Usage per Instance | `sum(rate(process_cpu_seconds_total{}[$__rate_interval]) * 1000) by (instance, pod)`<br/>`max(greptime_cpu_limit_in_millicores{app="greptime-datanode"})` | `timeseries` | Current cpu usage by instance | `prometheus` | `none` | `[{{ instance }}]-[{{ pod }}]` |
+| Frontend Memory per Instance | `sum(process_resident_memory_bytes{}) by (instance, pod)`<br/>`max(greptime_memory_limit_in_bytes{app="greptime-frontend"})` | `timeseries` | Current memory usage by instance | `prometheus` | `bytes` | `[{{ instance }}]-[{{ pod }}]` |
+| Frontend CPU Usage per Instance | `sum(rate(process_cpu_seconds_total{}[$__rate_interval]) * 1000) by (instance, pod)`<br/>`max(greptime_cpu_limit_in_millicores{app="greptime-frontend"})` | `timeseries` | Current cpu usage by instance | `prometheus` | `none` | `[{{ instance }}]-[{{ pod }}]-cpu` |
+| Metasrv Memory per Instance | `sum(process_resident_memory_bytes{}) by (instance, pod)`<br/>`max(greptime_memory_limit_in_bytes{app="greptime-metasrv"})` | `timeseries` | Current memory usage by instance | `prometheus` | `bytes` | `[{{ instance }}]-[{{ pod }}]-resident` |
+| Metasrv CPU Usage per Instance | `sum(rate(process_cpu_seconds_total{}[$__rate_interval]) * 1000) by (instance, pod)`<br/>`max(greptime_cpu_limit_in_millicores{app="greptime-metasrv"})` | `timeseries` | Current cpu usage by instance | `prometheus` | `none` | `[{{ instance }}]-[{{ pod }}]` |
+| Flownode Memory per Instance | `sum(process_resident_memory_bytes{}) by (instance, pod)`<br/>`max(greptime_memory_limit_in_bytes{app="greptime-flownode"})` | `timeseries` | Current memory usage by instance | `prometheus` | `bytes` | `[{{ instance }}]-[{{ pod }}]` |
+| Flownode CPU Usage per Instance | `sum(rate(process_cpu_seconds_total{}[$__rate_interval]) * 1000) by (instance, pod)`<br/>`max(greptime_cpu_limit_in_millicores{app="greptime-flownode"})` | `timeseries` | Current cpu usage by instance | `prometheus` | `none` | `[{{ instance }}]-[{{ pod }}]` |
 # Frontend Requests
 | Title | Query | Type | Description | Datasource | Unit | Legend Format |
 | --- | --- | --- | --- | --- | --- | --- |
@@ -72,18 +72,19 @@
 | Region Worker Handle Bulk Insert Requests | `histogram_quantile(0.95, sum by(le,instance, stage, pod) (rate(greptime_region_worker_handle_write_bucket[$__rate_interval])))`<br/>`sum by(instance, stage, pod) (rate(greptime_region_worker_handle_write_sum[$__rate_interval]))/sum by(instance, stage, pod) (rate(greptime_region_worker_handle_write_count[$__rate_interval]))` | `timeseries` | Per-stage elapsed time for region worker to handle bulk insert region requests. | `prometheus` | `s` | `[{{instance}}]-[{{pod}}]-[{{stage}}]-P95` |
 | Active Series and Field Builders Count | `sum by(instance, pod) (greptime_mito_memtable_active_series_count)`<br/>`sum by(instance, pod) (greptime_mito_memtable_field_builder_count)` | `timeseries` | Compaction oinput output bytes | `prometheus` | `none` | `[{{instance}}]-[{{pod}}]-series` |
 | Region Worker Convert Requests | `histogram_quantile(0.95, sum by(le, instance, stage, pod) (rate(greptime_datanode_convert_region_request_bucket[$__rate_interval])))`<br/>`sum by(le,instance, stage, pod) (rate(greptime_datanode_convert_region_request_sum[$__rate_interval]))/sum by(le,instance, stage, pod) (rate(greptime_datanode_convert_region_request_count[$__rate_interval]))` | `timeseries` | Per-stage elapsed time for region worker to decode requests. | `prometheus` | `s` | `[{{instance}}]-[{{pod}}]-[{{stage}}]-P95` |
+| Cache Miss | `sum by (instance,pod, type) (rate(greptime_mito_cache_miss{}[$__rate_interval]))` | `timeseries` | The local cache miss of the datanode. | `prometheus` | -- | `[{{instance}}]-[{{pod}}]-[{{type}}]` |
 # OpenDAL
 | Title | Query | Type | Description | Datasource | Unit | Legend Format |
 | --- | --- | --- | --- | --- | --- | --- |
 | QPS per Instance | `sum by(instance, pod, scheme, operation) (rate(opendal_operation_duration_seconds_count{}[$__rate_interval]))` | `timeseries` | QPS per Instance. | `prometheus` | `ops` | `[{{instance}}]-[{{pod}}]-[{{scheme}}]-[{{operation}}]` |
-| Read QPS per Instance | `sum by(instance, pod, scheme) (rate(opendal_operation_duration_seconds_count{ operation="read"}[$__rate_interval]))` | `timeseries` | Read QPS per Instance. | `prometheus` | `ops` | `[{{instance}}]-[{{pod}}]-[{{scheme}}]` |
-| Read P99 per Instance | `histogram_quantile(0.99, sum by(instance, pod, le, scheme) (rate(opendal_operation_duration_seconds_bucket{operation="read"}[$__rate_interval])))` | `timeseries` | Read P99 per Instance. | `prometheus` | `s` | `[{{instance}}]-[{{pod}}]-{{scheme}}` |
-| Write QPS per Instance | `sum by(instance, pod, scheme) (rate(opendal_operation_duration_seconds_count{ operation="write"}[$__rate_interval]))` | `timeseries` | Write QPS per Instance. | `prometheus` | `ops` | `[{{instance}}]-[{{pod}}]-{{scheme}}` |
-| Write P99 per Instance | `histogram_quantile(0.99, sum by(instance, pod, le, scheme) (rate(opendal_operation_duration_seconds_bucket{ operation="write"}[$__rate_interval])))` | `timeseries` | Write P99 per Instance. | `prometheus` | `s` | `[{{instance}}]-[{{pod}}]-[{{scheme}}]` |
+| Read QPS per Instance | `sum by(instance, pod, scheme, operation) (rate(opendal_operation_duration_seconds_count{ operation=~"read\|Reader::read"}[$__rate_interval]))` | `timeseries` | Read QPS per Instance. | `prometheus` | `ops` | `[{{instance}}]-[{{pod}}]-[{{scheme}}]-[{{operation}}]` |
+| Read P99 per Instance | `histogram_quantile(0.99, sum by(instance, pod, le, scheme, operation) (rate(opendal_operation_duration_seconds_bucket{operation=~"read\|Reader::read"}[$__rate_interval])))` | `timeseries` | Read P99 per Instance. | `prometheus` | `s` | `[{{instance}}]-[{{pod}}]-[{{scheme}}]-[{{operation}}]` |
+| Write QPS per Instance | `sum by(instance, pod, scheme, operation) (rate(opendal_operation_duration_seconds_count{ operation=~"write\|Writer::write\|Writer::close"}[$__rate_interval]))` | `timeseries` | Write QPS per Instance. | `prometheus` | `ops` | `[{{instance}}]-[{{pod}}]-[{{scheme}}]-[{{operation}}]` |
+| Write P99 per Instance | `histogram_quantile(0.99, sum by(instance, pod, le, scheme, operation) (rate(opendal_operation_duration_seconds_bucket{ operation =~ "Writer::write\|Writer::close\|write"}[$__rate_interval])))` | `timeseries` | Write P99 per Instance. | `prometheus` | `s` | `[{{instance}}]-[{{pod}}]-[{{scheme}}]-[{{operation}}]` |
 | List QPS per Instance | `sum by(instance, pod, scheme) (rate(opendal_operation_duration_seconds_count{ operation="list"}[$__rate_interval]))` | `timeseries` | List QPS per Instance. | `prometheus` | `ops` | `[{{instance}}]-[{{pod}}]-[{{scheme}}]` |
 | List P99 per Instance | `histogram_quantile(0.99, sum by(instance, pod, le, scheme) (rate(opendal_operation_duration_seconds_bucket{ operation="list"}[$__rate_interval])))` | `timeseries` | List P99 per Instance. | `prometheus` | `s` | `[{{instance}}]-[{{pod}}]-[{{scheme}}]` |
 | Other Requests per Instance | `sum by(instance, pod, scheme, operation) (rate(opendal_operation_duration_seconds_count{operation!~"read\|write\|list\|stat"}[$__rate_interval]))` | `timeseries` | Other Requests per Instance. | `prometheus` | `ops` | `[{{instance}}]-[{{pod}}]-[{{scheme}}]-[{{operation}}]` |
-| Other Request P99 per Instance | `histogram_quantile(0.99, sum by(instance, pod, le, scheme, operation) (rate(opendal_operation_duration_seconds_bucket{ operation!~"read\|write\|list"}[$__rate_interval])))` | `timeseries` | Other Request P99 per Instance. | `prometheus` | `s` | `[{{instance}}]-[{{pod}}]-[{{scheme}}]-[{{operation}}]` |
+| Other Request P99 per Instance | `histogram_quantile(0.99, sum by(instance, pod, le, scheme, operation) (rate(opendal_operation_duration_seconds_bucket{ operation!~"read\|write\|list\|Writer::write\|Writer::close\|Reader::read"}[$__rate_interval])))` | `timeseries` | Other Request P99 per Instance. | `prometheus` | `s` | `[{{instance}}]-[{{pod}}]-[{{scheme}}]-[{{operation}}]` |
 | Opendal traffic | `sum by(instance, pod, scheme, operation) (rate(opendal_operation_bytes_sum{}[$__rate_interval]))` | `timeseries` | Total traffic as in bytes by instance and operation | `prometheus` | `decbytes` | `[{{instance}}]-[{{pod}}]-[{{scheme}}]-[{{operation}}]` |
 | OpenDAL errors per Instance | `sum by(instance, pod, scheme, operation, error) (rate(opendal_operation_errors_total{ error!="NotFound"}[$__rate_interval]))` | `timeseries` | OpenDAL error counts per Instance. | `prometheus` | -- | `[{{instance}}]-[{{pod}}]-[{{scheme}}]-[{{operation}}]-[{{error}}]` |
 # Metasrv

grafana/dashboards/metrics/standalone/dashboard.yaml

@@ -180,13 +180,18 @@ groups:
         - title: Datanode Memory per Instance
           type: timeseries
           description: Current memory usage by instance
-          unit: decbytes
+          unit: bytes
           queries:
             - expr: sum(process_resident_memory_bytes{}) by (instance, pod)
               datasource:
                 type: prometheus
                 uid: ${metrics}
               legendFormat: '[{{instance}}]-[{{ pod }}]'
+            - expr: max(greptime_memory_limit_in_bytes{app="greptime-datanode"})
+              datasource:
+                type: prometheus
+                uid: ${metrics}
+              legendFormat: limit
         - title: Datanode CPU Usage per Instance
           type: timeseries
           description: Current cpu usage by instance
@@ -197,16 +202,26 @@ groups:
                 type: prometheus
                 uid: ${metrics}
               legendFormat: '[{{ instance }}]-[{{ pod }}]'
+            - expr: max(greptime_cpu_limit_in_millicores{app="greptime-datanode"})
+              datasource:
+                type: prometheus
+                uid: ${metrics}
+              legendFormat: limit
         - title: Frontend Memory per Instance
           type: timeseries
           description: Current memory usage by instance
-          unit: decbytes
+          unit: bytes
           queries:
             - expr: sum(process_resident_memory_bytes{}) by (instance, pod)
               datasource:
                 type: prometheus
                 uid: ${metrics}
               legendFormat: '[{{ instance }}]-[{{ pod }}]'
+            - expr: max(greptime_memory_limit_in_bytes{app="greptime-frontend"})
+              datasource:
+                type: prometheus
+                uid: ${metrics}
+              legendFormat: limit
         - title: Frontend CPU Usage per Instance
           type: timeseries
           description: Current cpu usage by instance
@@ -217,16 +232,26 @@ groups:
                 type: prometheus
                 uid: ${metrics}
               legendFormat: '[{{ instance }}]-[{{ pod }}]-cpu'
+            - expr: max(greptime_cpu_limit_in_millicores{app="greptime-frontend"})
+              datasource:
+                type: prometheus
+                uid: ${metrics}
+              legendFormat: limit
         - title: Metasrv Memory per Instance
           type: timeseries
           description: Current memory usage by instance
-          unit: decbytes
+          unit: bytes
           queries:
             - expr: sum(process_resident_memory_bytes{}) by (instance, pod)
               datasource:
                 type: prometheus
                 uid: ${metrics}
               legendFormat: '[{{ instance }}]-[{{ pod }}]-resident'
+            - expr: max(greptime_memory_limit_in_bytes{app="greptime-metasrv"})
+              datasource:
+                type: prometheus
+                uid: ${metrics}
+              legendFormat: limit
         - title: Metasrv CPU Usage per Instance
           type: timeseries
           description: Current cpu usage by instance
@@ -237,16 +262,26 @@ groups:
                 type: prometheus
                 uid: ${metrics}
               legendFormat: '[{{ instance }}]-[{{ pod }}]'
+            - expr: max(greptime_cpu_limit_in_millicores{app="greptime-metasrv"})
+              datasource:
+                type: prometheus
+                uid: ${metrics}
+              legendFormat: limit
         - title: Flownode Memory per Instance
           type: timeseries
           description: Current memory usage by instance
-          unit: decbytes
+          unit: bytes
           queries:
             - expr: sum(process_resident_memory_bytes{}) by (instance, pod)
               datasource:
                 type: prometheus
                 uid: ${metrics}
               legendFormat: '[{{ instance }}]-[{{ pod }}]'
+            - expr: max(greptime_memory_limit_in_bytes{app="greptime-flownode"})
+              datasource:
+                type: prometheus
+                uid: ${metrics}
+              legendFormat: limit
         - title: Flownode CPU Usage per Instance
           type: timeseries
           description: Current cpu usage by instance
@@ -257,6 +292,11 @@ groups:
                 type: prometheus
                 uid: ${metrics}
               legendFormat: '[{{ instance }}]-[{{ pod }}]'
+            - expr: max(greptime_cpu_limit_in_millicores{app="greptime-flownode"})
+              datasource:
+                type: prometheus
+                uid: ${metrics}
+              legendFormat: limit
     - title: Frontend Requests
       panels:
         - title: HTTP QPS per Instance
@@ -642,6 +682,15 @@ groups:
                 type: prometheus
                 uid: ${metrics}
               legendFormat: '[{{instance}}]-[{{pod}}]-[{{stage}}]-AVG'
+        - title: Cache Miss
+          type: timeseries
+          description: The local cache miss of the datanode.
+          queries:
+            - expr: sum by (instance,pod, type) (rate(greptime_mito_cache_miss{}[$__rate_interval]))
+              datasource:
+                type: prometheus
+                uid: ${metrics}
+              legendFormat: '[{{instance}}]-[{{pod}}]-[{{type}}]'
     - title: OpenDAL
       panels:
         - title: QPS per Instance
@@ -659,41 +708,41 @@ groups:
           description: Read QPS per Instance.
           unit: ops
           queries:
-            - expr: sum by(instance, pod, scheme) (rate(opendal_operation_duration_seconds_count{ operation="read"}[$__rate_interval]))
+            - expr: sum by(instance, pod, scheme, operation) (rate(opendal_operation_duration_seconds_count{ operation=~"read|Reader::read"}[$__rate_interval]))
               datasource:
                 type: prometheus
                 uid: ${metrics}
-              legendFormat: '[{{instance}}]-[{{pod}}]-[{{scheme}}]'
+              legendFormat: '[{{instance}}]-[{{pod}}]-[{{scheme}}]-[{{operation}}]'
         - title: Read P99 per Instance
           type: timeseries
           description: Read P99 per Instance.
           unit: s
           queries:
-            - expr: histogram_quantile(0.99, sum by(instance, pod, le, scheme) (rate(opendal_operation_duration_seconds_bucket{operation="read"}[$__rate_interval])))
+            - expr: histogram_quantile(0.99, sum by(instance, pod, le, scheme, operation) (rate(opendal_operation_duration_seconds_bucket{operation=~"read|Reader::read"}[$__rate_interval])))
               datasource:
                 type: prometheus
                 uid: ${metrics}
-              legendFormat: '[{{instance}}]-[{{pod}}]-{{scheme}}'
+              legendFormat: '[{{instance}}]-[{{pod}}]-[{{scheme}}]-[{{operation}}]'
         - title: Write QPS per Instance
           type: timeseries
           description: Write QPS per Instance.
           unit: ops
           queries:
-            - expr: sum by(instance, pod, scheme) (rate(opendal_operation_duration_seconds_count{ operation="write"}[$__rate_interval]))
+            - expr: sum by(instance, pod, scheme, operation) (rate(opendal_operation_duration_seconds_count{ operation=~"write|Writer::write|Writer::close"}[$__rate_interval]))
               datasource:
                 type: prometheus
                 uid: ${metrics}
-              legendFormat: '[{{instance}}]-[{{pod}}]-{{scheme}}'
+              legendFormat: '[{{instance}}]-[{{pod}}]-[{{scheme}}]-[{{operation}}]'
         - title: Write P99 per Instance
           type: timeseries
           description: Write P99 per Instance.
           unit: s
           queries:
-            - expr: histogram_quantile(0.99, sum by(instance, pod, le, scheme) (rate(opendal_operation_duration_seconds_bucket{ operation="write"}[$__rate_interval])))
+            - expr: histogram_quantile(0.99, sum by(instance, pod, le, scheme, operation) (rate(opendal_operation_duration_seconds_bucket{ operation =~ "Writer::write|Writer::close|write"}[$__rate_interval])))
               datasource:
                 type: prometheus
                 uid: ${metrics}
-              legendFormat: '[{{instance}}]-[{{pod}}]-[{{scheme}}]'
+              legendFormat: '[{{instance}}]-[{{pod}}]-[{{scheme}}]-[{{operation}}]'
         - title: List QPS per Instance
           type: timeseries
           description: List QPS per Instance.
@@ -729,7 +778,7 @@ groups:
           description: Other Request P99 per Instance.
           unit: s
           queries:
-            - expr: histogram_quantile(0.99, sum by(instance, pod, le, scheme, operation) (rate(opendal_operation_duration_seconds_bucket{ operation!~"read|write|list"}[$__rate_interval])))
+            - expr: histogram_quantile(0.99, sum by(instance, pod, le, scheme, operation) (rate(opendal_operation_duration_seconds_bucket{ operation!~"read|write|list|Writer::write|Writer::close|Reader::read"}[$__rate_interval])))
               datasource:
                 type: prometheus
                 uid: ${metrics}

grafana/scripts/check.sh

@@ -26,7 +26,7 @@ check_dashboards_generation() {
   ./grafana/scripts/gen-dashboards.sh
 
   if [[ -n "$(git diff --name-only grafana/dashboards/metrics)" ]]; then
-    echo "Error: The dashboards are not generated correctly. You should execute the `make dashboards` command."
+    echo "Error: The dashboards are not generated correctly. You should execute the 'make dashboards' command."
     exit 1
   fi
 }

licenserc.toml

@@ -29,10 +29,14 @@ excludes = [
     # enterprise
     "src/common/meta/src/rpc/ddl/trigger.rs",
     "src/operator/src/expr_helper/trigger.rs",
+    "src/sql/src/statements/alter/trigger.rs",
     "src/sql/src/statements/create/trigger.rs",
     "src/sql/src/statements/show/trigger.rs",
+    "src/sql/src/statements/drop/trigger.rs",
+    "src/sql/src/parsers/alter_parser/trigger.rs",
     "src/sql/src/parsers/create_parser/trigger.rs",
     "src/sql/src/parsers/show_parser/trigger.rs",
+    "src/mito2/src/extension.rs",
 ]
 
 [properties]

scripts/check-super-imports.py

@@ -13,8 +13,8 @@
 # limitations under the License.
 
 import os
-import re
 from multiprocessing import Pool
+from pathlib import Path
 
 
 def find_rust_files(directory):
@@ -24,6 +24,10 @@ def find_rust_files(directory):
         if "test" in root.lower():
             continue
 
+        # Skip the target directory
+        if "target" in Path(root).parts:
+            continue
+
         for file in files:
             # Skip files with "test" in the filename
             if "test" in file.lower():

scripts/install.sh

@@ -53,6 +53,54 @@ get_arch_type() {
   esac
 }
 
+# Verify SHA256 checksum
+verify_sha256() {
+  file="$1"
+  expected_sha256="$2"
+
+  if command -v sha256sum >/dev/null 2>&1; then
+    actual_sha256=$(sha256sum "$file" | cut -d' ' -f1)
+  elif command -v shasum >/dev/null 2>&1; then
+    actual_sha256=$(shasum -a 256 "$file" | cut -d' ' -f1)
+  else
+    echo "Warning: No SHA256 verification tool found (sha256sum or shasum). Skipping checksum verification."
+    return 0
+  fi
+
+  if [ "$actual_sha256" = "$expected_sha256" ]; then
+    echo "SHA256 checksum verified successfully."
+    return 0
+  else
+    echo "Error: SHA256 checksum verification failed!"
+    echo "Expected: $expected_sha256"
+    echo "Actual: $actual_sha256"
+    return 1
+  fi
+}
+
+# Prompt for user confirmation (compatible with different shells)
+prompt_confirmation() {
+  message="$1"
+  printf "%s (y/N): " "$message"
+
+  # Try to read user input, fallback if read fails
+  answer=""
+  if read answer </dev/tty 2>/dev/null; then
+    case "$answer" in
+      [Yy]|[Yy][Ee][Ss])
+        return 0
+        ;;
+      *)
+        return 1
+        ;;
+    esac
+  else
+    echo ""
+    echo "Cannot read user input. Defaulting to No."
+    return 1
+  fi
+}
+
 download_artifact() {
   if [ -n "${OS_TYPE}" ] && [ -n "${ARCH_TYPE}" ]; then
     # Use the latest stable released version.
@@ -71,17 +119,104 @@ download_artifact() {
     fi
 
     echo "Downloading ${BIN}, OS: ${OS_TYPE}, Arch: ${ARCH_TYPE}, Version: ${VERSION}"
-    PACKAGE_NAME="${BIN}-${OS_TYPE}-${ARCH_TYPE}-${VERSION}.tar.gz"
+    PKG_NAME="${BIN}-${OS_TYPE}-${ARCH_TYPE}-${VERSION}"
+    PACKAGE_NAME="${PKG_NAME}.tar.gz"
+    SHA256_FILE="${PKG_NAME}.sha256sum"
 
     if [ -n "${PACKAGE_NAME}" ]; then
-      wget "https://github.com/${GITHUB_ORG}/${GITHUB_REPO}/releases/download/${VERSION}/${PACKAGE_NAME}"
+      # Check if files already exist and prompt for override
+      if [ -f "${PACKAGE_NAME}" ]; then
+        echo "File ${PACKAGE_NAME} already exists."
+        if prompt_confirmation "Do you want to override it?"; then
+          echo "Overriding existing file..."
+          rm -f "${PACKAGE_NAME}"
+        else
+          echo "Skipping download. Using existing file."
+        fi
+      fi
+
+      if [ -f "${BIN}" ]; then
+        echo "Binary ${BIN} already exists."
+        if prompt_confirmation "Do you want to override it?"; then
+          echo "Will override existing binary..."
+          rm -f "${BIN}"
+        else
+          echo "Installation cancelled."
+          exit 0
+        fi
+      fi
+
+      # Download package if not exists
+      if [ ! -f "${PACKAGE_NAME}" ]; then
+        echo "Downloading ${PACKAGE_NAME}..."
+        # Use curl instead of wget for better compatibility
+        if command -v curl >/dev/null 2>&1; then
+          if ! curl -L -o "${PACKAGE_NAME}" "https://github.com/${GITHUB_ORG}/${GITHUB_REPO}/releases/download/${VERSION}/${PACKAGE_NAME}"; then
+            echo "Error: Failed to download ${PACKAGE_NAME}"
+            exit 1
+          fi
+        elif command -v wget >/dev/null 2>&1; then
+          if ! wget -O "${PACKAGE_NAME}" "https://github.com/${GITHUB_ORG}/${GITHUB_REPO}/releases/download/${VERSION}/${PACKAGE_NAME}"; then
+            echo "Error: Failed to download ${PACKAGE_NAME}"
+            exit 1
+          fi
+        else
+          echo "Error: Neither curl nor wget is available for downloading."
+          exit 1
+        fi
+      fi
+
+      # Download and verify SHA256 checksum
+      echo "Downloading SHA256 checksum..."
+      sha256_download_success=0
+      if command -v curl >/dev/null 2>&1; then
+        if curl -L -s -o "${SHA256_FILE}" "https://github.com/${GITHUB_ORG}/${GITHUB_REPO}/releases/download/${VERSION}/${SHA256_FILE}" 2>/dev/null; then
+          sha256_download_success=1
+        fi
+      elif command -v wget >/dev/null 2>&1; then
+        if wget -q -O "${SHA256_FILE}" "https://github.com/${GITHUB_ORG}/${GITHUB_REPO}/releases/download/${VERSION}/${SHA256_FILE}" 2>/dev/null; then
+          sha256_download_success=1
+        fi
+      fi
+
+      if [ $sha256_download_success -eq 1 ] && [ -f "${SHA256_FILE}" ]; then
+        expected_sha256=$(cat "${SHA256_FILE}" | cut -d' ' -f1)
+        if [ -n "$expected_sha256" ]; then
+          if ! verify_sha256 "${PACKAGE_NAME}" "${expected_sha256}"; then
+            echo "SHA256 verification failed. Removing downloaded file."
+            rm -f "${PACKAGE_NAME}" "${SHA256_FILE}"
+            exit 1
+          fi
+        else
+          echo "Warning: Could not parse SHA256 checksum from file."
+        fi
+        rm -f "${SHA256_FILE}"
+      else
+        echo "Warning: Could not download SHA256 checksum file. Skipping verification."
+      fi
 
       # Extract the binary and clean the rest.
-      tar xvf "${PACKAGE_NAME}" && \
-      mv "${PACKAGE_NAME%.tar.gz}/${BIN}" "${PWD}" && \
-      rm -r "${PACKAGE_NAME}" && \
-      rm -r "${PACKAGE_NAME%.tar.gz}" && \
-      echo "Run './${BIN} --help' to get started"
+      echo "Extracting ${PACKAGE_NAME}..."
+      if ! tar xf "${PACKAGE_NAME}"; then
+        echo "Error: Failed to extract ${PACKAGE_NAME}"
+        exit 1
+      fi
+
+      # Find the binary in the extracted directory
+      extracted_dir="${PACKAGE_NAME%.tar.gz}"
+      if [ -f "${extracted_dir}/${BIN}" ]; then
+        mv "${extracted_dir}/${BIN}" "${PWD}/"
+        rm -f "${PACKAGE_NAME}"
+        rm -rf "${extracted_dir}"
+        chmod +x "${BIN}"
+        echo "Installation completed successfully!"
+        echo "Run './${BIN} --help' to get started"
+      else
+        echo "Error: Binary ${BIN} not found in extracted archive"
+        rm -f "${PACKAGE_NAME}"
+        rm -rf "${extracted_dir}"
+        exit 1
+      fi
     fi
   fi
 }

src/api/src/helper.rs

@@ -291,6 +291,7 @@ impl TryFrom<ConcreteDataType> for ColumnDataTypeWrapper {
             ConcreteDataType::Vector(_) => ColumnDataType::Vector,
             ConcreteDataType::Null(_)
             | ConcreteDataType::List(_)
+            | ConcreteDataType::Struct(_)
             | ConcreteDataType::Dictionary(_)
             | ConcreteDataType::Duration(_) => {
                 return error::IntoColumnDataTypeSnafu { from: datatype }.fail()
@@ -703,6 +704,7 @@ pub fn pb_values_to_vector_ref(data_type: &ConcreteDataType, values: Values) ->
         ConcreteDataType::Vector(_) => Arc::new(BinaryVector::from_vec(values.binary_values)),
         ConcreteDataType::Null(_)
         | ConcreteDataType::List(_)
+        | ConcreteDataType::Struct(_)
         | ConcreteDataType::Dictionary(_)
         | ConcreteDataType::Duration(_)
         | ConcreteDataType::Json(_) => {
@@ -864,6 +866,7 @@ pub fn pb_values_to_values(data_type: &ConcreteDataType, values: Values) -> Vec<
         ConcreteDataType::Vector(_) => values.binary_values.into_iter().map(|v| v.into()).collect(),
         ConcreteDataType::Null(_)
         | ConcreteDataType::List(_)
+        | ConcreteDataType::Struct(_)
         | ConcreteDataType::Dictionary(_)
         | ConcreteDataType::Duration(_)
         | ConcreteDataType::Json(_) => {

src/api/src/region.rs

@@ -22,6 +22,7 @@ use greptime_proto::v1::region::RegionResponse as RegionResponseV1;
 pub struct RegionResponse {
     pub affected_rows: AffectedRows,
     pub extensions: HashMap<String, Vec<u8>>,
+    pub metadata: Vec<u8>,
 }
 
 impl RegionResponse {
@@ -29,6 +30,7 @@ impl RegionResponse {
         Self {
             affected_rows: region_response.affected_rows as _,
             extensions: region_response.extensions,
+            metadata: region_response.metadata,
         }
     }
 
@@ -37,6 +39,16 @@ impl RegionResponse {
         Self {
             affected_rows,
             extensions: Default::default(),
+            metadata: Vec::new(),
+        }
+    }
+
+    /// Creates one response with metadata.
+    pub fn from_metadata(metadata: Vec<u8>) -> Self {
+        Self {
+            affected_rows: 0,
+            extensions: Default::default(),
+            metadata,
         }
     }
 }

src/api/src/v1/column_def.rs

@@ -24,7 +24,7 @@ use greptime_proto::v1::{
 };
 use snafu::ResultExt;
 
-use crate::error::{self, Result};
+use crate::error::{self, ConvertColumnDefaultConstraintSnafu, Result};
 use crate::helper::ColumnDataTypeWrapper;
 use crate::v1::{ColumnDef, ColumnOptions, SemanticType};
 
@@ -77,6 +77,48 @@ pub fn try_as_column_schema(column_def: &ColumnDef) -> Result<ColumnSchema> {
         })
 }
 
+/// Tries to construct a `ColumnDef` from the given `ColumnSchema`.
+///
+/// TODO(weny): Add tests for this function.
+pub fn try_as_column_def(column_schema: &ColumnSchema, is_primary_key: bool) -> Result<ColumnDef> {
+    let column_datatype =
+        ColumnDataTypeWrapper::try_from(column_schema.data_type.clone()).map(|w| w.to_parts())?;
+
+    let semantic_type = if column_schema.is_time_index() {
+        SemanticType::Timestamp
+    } else if is_primary_key {
+        SemanticType::Tag
+    } else {
+        SemanticType::Field
+    } as i32;
+    let comment = column_schema
+        .metadata()
+        .get(COMMENT_KEY)
+        .cloned()
+        .unwrap_or_default();
+
+    let default_constraint = match column_schema.default_constraint() {
+        None => vec![],
+        Some(v) => v
+            .clone()
+            .try_into()
+            .context(ConvertColumnDefaultConstraintSnafu {
+                column: &column_schema.name,
+            })?,
+    };
+    let options = options_from_column_schema(column_schema);
+    Ok(ColumnDef {
+        name: column_schema.name.clone(),
+        data_type: column_datatype.0 as i32,
+        is_nullable: column_schema.is_nullable(),
+        default_constraint,
+        semantic_type,
+        comment,
+        datatype_extension: column_datatype.1,
+        options,
+    })
+}
+
 /// Constructs a `ColumnOptions` from the given `ColumnSchema`.
 pub fn options_from_column_schema(column_schema: &ColumnSchema) -> Option<ColumnOptions> {
     let mut options = ColumnOptions::default();
@@ -226,18 +268,20 @@ mod tests {
         assert!(options.is_none());
 
         let mut schema = ColumnSchema::new("test", ConcreteDataType::string_datatype(), true)
-            .with_fulltext_options(FulltextOptions {
-                enable: true,
-                analyzer: FulltextAnalyzer::English,
-                case_sensitive: false,
-                backend: FulltextBackend::Bloom,
-            })
+            .with_fulltext_options(FulltextOptions::new_unchecked(
+                true,
+                FulltextAnalyzer::English,
+                false,
+                FulltextBackend::Bloom,
+                10240,
+                0.01,
+            ))
             .unwrap();
         schema.set_inverted_index(true);
         let options = options_from_column_schema(&schema).unwrap();
         assert_eq!(
             options.options.get(FULLTEXT_GRPC_KEY).unwrap(),
-            "{\"enable\":true,\"analyzer\":\"English\",\"case-sensitive\":false,\"backend\":\"bloom\"}"
+            "{\"enable\":true,\"analyzer\":\"English\",\"case-sensitive\":false,\"backend\":\"bloom\",\"granularity\":10240,\"false-positive-rate-in-10000\":100}"
         );
         assert_eq!(
             options.options.get(INVERTED_INDEX_GRPC_KEY).unwrap(),
@@ -247,16 +291,18 @@ mod tests {
 
     #[test]
     fn test_options_with_fulltext() {
-        let fulltext = FulltextOptions {
-            enable: true,
-            analyzer: FulltextAnalyzer::English,
-            case_sensitive: false,
-            backend: FulltextBackend::Bloom,
-        };
+        let fulltext = FulltextOptions::new_unchecked(
+            true,
+            FulltextAnalyzer::English,
+            false,
+            FulltextBackend::Bloom,
+            10240,
+            0.01,
+        );
         let options = options_from_fulltext(&fulltext).unwrap().unwrap();
         assert_eq!(
             options.options.get(FULLTEXT_GRPC_KEY).unwrap(),
-            "{\"enable\":true,\"analyzer\":\"English\",\"case-sensitive\":false,\"backend\":\"bloom\"}"
+            "{\"enable\":true,\"analyzer\":\"English\",\"case-sensitive\":false,\"backend\":\"bloom\",\"granularity\":10240,\"false-positive-rate-in-10000\":100}"
         );
     }
 

src/catalog/Cargo.toml

@@ -5,6 +5,7 @@ edition.workspace = true
 license.workspace = true
 
 [features]
+enterprise = []
 testing = []
 
 [lints]

src/catalog/src/information_extension.rs

@@ -16,8 +16,8 @@ use api::v1::meta::ProcedureStatus;
 use common_error::ext::BoxedError;
 use common_meta::cluster::{ClusterInfo, NodeInfo};
 use common_meta::datanode::RegionStat;
-use common_meta::ddl::{ExecutorContext, ProcedureExecutor};
 use common_meta::key::flow::flow_state::FlowStat;
+use common_meta::procedure_executor::{ExecutorContext, ProcedureExecutor};
 use common_meta::rpc::procedure;
 use common_procedure::{ProcedureInfo, ProcedureState};
 use meta_client::MetaClientRef;

src/catalog/src/kvbackend.rs

@@ -14,9 +14,11 @@
 
 pub use client::{CachedKvBackend, CachedKvBackendBuilder, MetaKvBackend};
 
+mod builder;
 mod client;
 mod manager;
 mod table_cache;
 
+pub use builder::KvBackendCatalogManagerBuilder;
 pub use manager::KvBackendCatalogManager;
 pub use table_cache::{new_table_cache, TableCache, TableCacheRef};

src/catalog/src/kvbackend/builder.rs

@@ -0,0 +1,131 @@
+// Copyright 2023 Greptime Team
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+use std::sync::Arc;
+
+use common_catalog::consts::DEFAULT_CATALOG_NAME;
+use common_meta::cache::LayeredCacheRegistryRef;
+use common_meta::key::flow::FlowMetadataManager;
+use common_meta::key::TableMetadataManager;
+use common_meta::kv_backend::KvBackendRef;
+use common_procedure::ProcedureManagerRef;
+use moka::sync::Cache;
+use partition::manager::PartitionRuleManager;
+
+#[cfg(feature = "enterprise")]
+use crate::information_schema::InformationSchemaTableFactoryRef;
+use crate::information_schema::{InformationExtensionRef, InformationSchemaProvider};
+use crate::kvbackend::manager::{SystemCatalog, CATALOG_CACHE_MAX_CAPACITY};
+use crate::kvbackend::KvBackendCatalogManager;
+use crate::process_manager::ProcessManagerRef;
+use crate::system_schema::pg_catalog::PGCatalogProvider;
+
+pub struct KvBackendCatalogManagerBuilder {
+    information_extension: InformationExtensionRef,
+    backend: KvBackendRef,
+    cache_registry: LayeredCacheRegistryRef,
+    procedure_manager: Option<ProcedureManagerRef>,
+    process_manager: Option<ProcessManagerRef>,
+    #[cfg(feature = "enterprise")]
+    extra_information_table_factories:
+        std::collections::HashMap<String, InformationSchemaTableFactoryRef>,
+}
+
+impl KvBackendCatalogManagerBuilder {
+    pub fn new(
+        information_extension: InformationExtensionRef,
+        backend: KvBackendRef,
+        cache_registry: LayeredCacheRegistryRef,
+    ) -> Self {
+        Self {
+            information_extension,
+            backend,
+            cache_registry,
+            procedure_manager: None,
+            process_manager: None,
+            #[cfg(feature = "enterprise")]
+            extra_information_table_factories: std::collections::HashMap::new(),
+        }
+    }
+
+    pub fn with_procedure_manager(mut self, procedure_manager: ProcedureManagerRef) -> Self {
+        self.procedure_manager = Some(procedure_manager);
+        self
+    }
+
+    pub fn with_process_manager(mut self, process_manager: ProcessManagerRef) -> Self {
+        self.process_manager = Some(process_manager);
+        self
+    }
+
+    /// Sets the extra information tables.
+    #[cfg(feature = "enterprise")]
+    pub fn with_extra_information_table_factories(
+        mut self,
+        factories: std::collections::HashMap<String, InformationSchemaTableFactoryRef>,
+    ) -> Self {
+        self.extra_information_table_factories = factories;
+        self
+    }
+
+    pub fn build(self) -> Arc<KvBackendCatalogManager> {
+        let Self {
+            information_extension,
+            backend,
+            cache_registry,
+            procedure_manager,
+            process_manager,
+            #[cfg(feature = "enterprise")]
+            extra_information_table_factories,
+        } = self;
+        Arc::new_cyclic(|me| KvBackendCatalogManager {
+            information_extension,
+            partition_manager: Arc::new(PartitionRuleManager::new(
+                backend.clone(),
+                cache_registry
+                    .get()
+                    .expect("Failed to get table_route_cache"),
+            )),
+            table_metadata_manager: Arc::new(TableMetadataManager::new(backend.clone())),
+            system_catalog: SystemCatalog {
+                catalog_manager: me.clone(),
+                catalog_cache: Cache::new(CATALOG_CACHE_MAX_CAPACITY),
+                pg_catalog_cache: Cache::new(CATALOG_CACHE_MAX_CAPACITY),
+                information_schema_provider: {
+                    let provider = InformationSchemaProvider::new(
+                        DEFAULT_CATALOG_NAME.to_string(),
+                        me.clone(),
+                        Arc::new(FlowMetadataManager::new(backend.clone())),
+                        process_manager.clone(),
+                        backend.clone(),
+                    );
+                    #[cfg(feature = "enterprise")]
+                    let provider = provider
+                        .with_extra_table_factories(extra_information_table_factories.clone());
+                    Arc::new(provider)
+                },
+                pg_catalog_provider: Arc::new(PGCatalogProvider::new(
+                    DEFAULT_CATALOG_NAME.to_string(),
+                    me.clone(),
+                )),
+                backend,
+                process_manager,
+                #[cfg(feature = "enterprise")]
+                extra_information_table_factories,
+            },
+            cache_registry,
+            procedure_manager,
+        })
+    }
+}

src/catalog/src/kvbackend/manager.rs

@@ -28,17 +28,18 @@ use common_meta::cache::{
 use common_meta::key::catalog_name::CatalogNameKey;
 use common_meta::key::flow::FlowMetadataManager;
 use common_meta::key::schema_name::SchemaNameKey;
-use common_meta::key::table_info::TableInfoValue;
+use common_meta::key::table_info::{TableInfoManager, TableInfoValue};
 use common_meta::key::table_name::TableNameKey;
-use common_meta::key::{TableMetadataManager, TableMetadataManagerRef};
+use common_meta::key::TableMetadataManagerRef;
 use common_meta::kv_backend::KvBackendRef;
 use common_procedure::ProcedureManagerRef;
 use futures_util::stream::BoxStream;
 use futures_util::{StreamExt, TryStreamExt};
 use moka::sync::Cache;
-use partition::manager::{PartitionRuleManager, PartitionRuleManagerRef};
+use partition::manager::PartitionRuleManagerRef;
 use session::context::{Channel, QueryContext};
 use snafu::prelude::*;
+use store_api::metric_engine_consts::METRIC_ENGINE_NAME;
 use table::dist_table::DistTable;
 use table::metadata::TableId;
 use table::table::numbers::{NumbersTable, NUMBERS_TABLE_NAME};
@@ -51,6 +52,8 @@ use crate::error::{
     CacheNotFoundSnafu, GetTableCacheSnafu, InvalidTableInfoInCatalogSnafu, ListCatalogsSnafu,
     ListSchemasSnafu, ListTablesSnafu, Result, TableMetadataManagerSnafu,
 };
+#[cfg(feature = "enterprise")]
+use crate::information_schema::InformationSchemaTableFactoryRef;
 use crate::information_schema::{InformationExtensionRef, InformationSchemaProvider};
 use crate::kvbackend::TableCacheRef;
 use crate::process_manager::ProcessManagerRef;
@@ -66,60 +69,22 @@ use crate::CatalogManager;
 #[derive(Clone)]
 pub struct KvBackendCatalogManager {
     /// Provides the extension methods for the `information_schema` tables
-    information_extension: InformationExtensionRef,
+    pub(super) information_extension: InformationExtensionRef,
     /// Manages partition rules.
-    partition_manager: PartitionRuleManagerRef,
+    pub(super) partition_manager: PartitionRuleManagerRef,
     /// Manages table metadata.
-    table_metadata_manager: TableMetadataManagerRef,
+    pub(super) table_metadata_manager: TableMetadataManagerRef,
     /// A sub-CatalogManager that handles system tables
-    system_catalog: SystemCatalog,
+    pub(super) system_catalog: SystemCatalog,
     /// Cache registry for all caches.
-    cache_registry: LayeredCacheRegistryRef,
+    pub(super) cache_registry: LayeredCacheRegistryRef,
     /// Only available in `Standalone` mode.
-    procedure_manager: Option<ProcedureManagerRef>,
+    pub(super) procedure_manager: Option<ProcedureManagerRef>,
 }
 
-const CATALOG_CACHE_MAX_CAPACITY: u64 = 128;
+pub(super) const CATALOG_CACHE_MAX_CAPACITY: u64 = 128;
 
 impl KvBackendCatalogManager {
-    pub fn new(
-        information_extension: InformationExtensionRef,
-        backend: KvBackendRef,
-        cache_registry: LayeredCacheRegistryRef,
-        procedure_manager: Option<ProcedureManagerRef>,
-        process_manager: Option<ProcessManagerRef>,
-    ) -> Arc<Self> {
-        Arc::new_cyclic(|me| Self {
-            information_extension,
-            partition_manager: Arc::new(PartitionRuleManager::new(
-                backend.clone(),
-                cache_registry
-                    .get()
-                    .expect("Failed to get table_route_cache"),
-            )),
-            table_metadata_manager: Arc::new(TableMetadataManager::new(backend.clone())),
-            system_catalog: SystemCatalog {
-                catalog_manager: me.clone(),
-                catalog_cache: Cache::new(CATALOG_CACHE_MAX_CAPACITY),
-                pg_catalog_cache: Cache::new(CATALOG_CACHE_MAX_CAPACITY),
-                information_schema_provider: Arc::new(InformationSchemaProvider::new(
-                    DEFAULT_CATALOG_NAME.to_string(),
-                    me.clone(),
-                    Arc::new(FlowMetadataManager::new(backend.clone())),
-                    process_manager.clone(),
-                )),
-                pg_catalog_provider: Arc::new(PGCatalogProvider::new(
-                    DEFAULT_CATALOG_NAME.to_string(),
-                    me.clone(),
-                )),
-                backend,
-                process_manager,
-            },
-            cache_registry,
-            procedure_manager,
-        })
-    }
-
     pub fn view_info_cache(&self) -> Result<ViewInfoCacheRef> {
         self.cache_registry.get().context(CacheNotFoundSnafu {
             name: "view_info_cache",
@@ -142,6 +107,61 @@ impl KvBackendCatalogManager {
     pub fn procedure_manager(&self) -> Option<ProcedureManagerRef> {
         self.procedure_manager.clone()
     }
+
+    // Override logical table's partition key indices with physical table's.
+    async fn override_logical_table_partition_key_indices(
+        table_route_cache: &TableRouteCacheRef,
+        table_info_manager: &TableInfoManager,
+        table: TableRef,
+    ) -> Result<TableRef> {
+        // If the table is not a metric table, return the table directly.
+        if table.table_info().meta.engine != METRIC_ENGINE_NAME {
+            return Ok(table);
+        }
+
+        if let Some(table_route_value) = table_route_cache
+            .get(table.table_info().table_id())
+            .await
+            .context(TableMetadataManagerSnafu)?
+            && let TableRoute::Logical(logical_route) = &*table_route_value
+            && let Some(physical_table_info_value) = table_info_manager
+                .get(logical_route.physical_table_id())
+                .await
+                .context(TableMetadataManagerSnafu)?
+        {
+            let mut new_table_info = (*table.table_info()).clone();
+
+            // Remap partition key indices from physical table to logical table
+            new_table_info.meta.partition_key_indices = physical_table_info_value
+                .table_info
+                .meta
+                .partition_key_indices
+                .iter()
+                .filter_map(|&physical_index| {
+                    // Get the column name from the physical table using the physical index
+                    physical_table_info_value
+                        .table_info
+                        .meta
+                        .schema
+                        .column_schemas
+                        .get(physical_index)
+                        .and_then(|physical_column| {
+                            // Find the corresponding index in the logical table schema
+                            new_table_info
+                                .meta
+                                .schema
+                                .column_index_by_name(physical_column.name.as_str())
+                        })
+                })
+                .collect();
+
+            let new_table = DistTable::table(Arc::new(new_table_info));
+
+            return Ok(new_table);
+        }
+
+        Ok(table)
+    }
 }
 
 #[async_trait::async_trait]
@@ -268,10 +288,7 @@ impl CatalogManager for KvBackendCatalogManager {
         let table_cache: TableCacheRef = self.cache_registry.get().context(CacheNotFoundSnafu {
             name: "table_cache",
         })?;
-        let table_route_cache: TableRouteCacheRef =
-            self.cache_registry.get().context(CacheNotFoundSnafu {
-                name: "table_route_cache",
-            })?;
+
         let table = table_cache
             .get_by_ref(&TableName {
                 catalog_name: catalog_name.to_string(),
@@ -281,55 +298,18 @@ impl CatalogManager for KvBackendCatalogManager {
             .await
             .context(GetTableCacheSnafu)?;
 
-        // Override logical table's partition key indices with physical table's.
-        if let Some(table) = &table
-            && let Some(table_route_value) = table_route_cache
-                .get(table.table_info().table_id())
-                .await
-                .context(TableMetadataManagerSnafu)?
-            && let TableRoute::Logical(logical_route) = &*table_route_value
-            && let Some(physical_table_info_value) = self
-                .table_metadata_manager
-                .table_info_manager()
-                .get(logical_route.physical_table_id())
-                .await
-                .context(TableMetadataManagerSnafu)?
-        {
-            let mut new_table_info = (*table.table_info()).clone();
-            // Gather all column names from the logical table
-            let logical_column_names: std::collections::HashSet<_> = new_table_info
-                .meta
-                .schema
-                .column_schemas()
-                .iter()
-                .map(|col| &col.name)
-                .collect();
-
-            // Only preserve partition key indices where the corresponding columns exist in logical table
-            new_table_info.meta.partition_key_indices = physical_table_info_value
-                .table_info
-                .meta
-                .partition_key_indices
-                .iter()
-                .filter(|&&index| {
-                    if let Some(physical_column) = physical_table_info_value
-                        .table_info
-                        .meta
-                        .schema
-                        .column_schemas
-                        .get(index)
-                    {
-                        logical_column_names.contains(&physical_column.name)
-                    } else {
-                        false
-                    }
-                })
-                .cloned()
-                .collect();
-
-            let new_table = DistTable::table(Arc::new(new_table_info));
-
-            return Ok(Some(new_table));
+        if let Some(table) = table {
+            let table_route_cache: TableRouteCacheRef =
+                self.cache_registry.get().context(CacheNotFoundSnafu {
+                    name: "table_route_cache",
+                })?;
+            return Self::override_logical_table_partition_key_indices(
+                &table_route_cache,
+                self.table_metadata_manager.table_info_manager(),
+                table,
+            )
+            .await
+            .map(Some);
         }
 
         if channel == Channel::Postgres {
@@ -342,7 +322,7 @@ impl CatalogManager for KvBackendCatalogManager {
             }
         }
 
-        Ok(table)
+        Ok(None)
     }
 
     async fn tables_by_ids(
@@ -394,8 +374,20 @@ impl CatalogManager for KvBackendCatalogManager {
         let catalog = catalog.to_string();
         let schema = schema.to_string();
         let semaphore = Arc::new(Semaphore::new(CONCURRENCY));
+        let table_route_cache: Result<TableRouteCacheRef> =
+            self.cache_registry.get().context(CacheNotFoundSnafu {
+                name: "table_route_cache",
+            });
 
         common_runtime::spawn_global(async move {
+            let table_route_cache = match table_route_cache {
+                Ok(table_route_cache) => table_route_cache,
+                Err(e) => {
+                    let _ = tx.send(Err(e)).await;
+                    return;
+                }
+            };
+
             let table_id_stream = metadata_manager
                 .table_name_manager()
                 .tables(&catalog, &schema)
@@ -422,6 +414,7 @@ impl CatalogManager for KvBackendCatalogManager {
                 let metadata_manager = metadata_manager.clone();
                 let tx = tx.clone();
                 let semaphore = semaphore.clone();
+                let table_route_cache = table_route_cache.clone();
                 common_runtime::spawn_global(async move {
                     // we don't explicitly close the semaphore so just ignore the potential error.
                     let _ = semaphore.acquire().await;
@@ -439,6 +432,16 @@ impl CatalogManager for KvBackendCatalogManager {
                     };
 
                     for table in table_info_values.into_values().map(build_table) {
+                        let table = if let Ok(table) = table {
+                            Self::override_logical_table_partition_key_indices(
+                                &table_route_cache,
+                                metadata_manager.table_info_manager(),
+                                table,
+                            )
+                            .await
+                        } else {
+                            table
+                        };
                         if tx.send(table).await.is_err() {
                             return;
                         }
@@ -468,16 +471,19 @@ fn build_table(table_info_value: TableInfoValue) -> Result<TableRef> {
 /// - information_schema.{tables}
 /// - pg_catalog.{tables}
 #[derive(Clone)]
-struct SystemCatalog {
-    catalog_manager: Weak<KvBackendCatalogManager>,
-    catalog_cache: Cache<String, Arc<InformationSchemaProvider>>,
-    pg_catalog_cache: Cache<String, Arc<PGCatalogProvider>>,
+pub(super) struct SystemCatalog {
+    pub(super) catalog_manager: Weak<KvBackendCatalogManager>,
+    pub(super) catalog_cache: Cache<String, Arc<InformationSchemaProvider>>,
+    pub(super) pg_catalog_cache: Cache<String, Arc<PGCatalogProvider>>,
 
     // system_schema_provider for default catalog
-    information_schema_provider: Arc<InformationSchemaProvider>,
-    pg_catalog_provider: Arc<PGCatalogProvider>,
-    backend: KvBackendRef,
-    process_manager: Option<ProcessManagerRef>,
+    pub(super) information_schema_provider: Arc<InformationSchemaProvider>,
+    pub(super) pg_catalog_provider: Arc<PGCatalogProvider>,
+    pub(super) backend: KvBackendRef,
+    pub(super) process_manager: Option<ProcessManagerRef>,
+    #[cfg(feature = "enterprise")]
+    pub(super) extra_information_table_factories:
+        std::collections::HashMap<String, InformationSchemaTableFactoryRef>,
 }
 
 impl SystemCatalog {
@@ -541,12 +547,17 @@ impl SystemCatalog {
         if schema == INFORMATION_SCHEMA_NAME {
             let information_schema_provider =
                 self.catalog_cache.get_with_by_ref(catalog, move || {
-                    Arc::new(InformationSchemaProvider::new(
+                    let provider = InformationSchemaProvider::new(
                         catalog.to_string(),
                         self.catalog_manager.clone(),
                         Arc::new(FlowMetadataManager::new(self.backend.clone())),
                         self.process_manager.clone(),
-                    ))
+                        self.backend.clone(),
+                    );
+                    #[cfg(feature = "enterprise")]
+                    let provider = provider
+                        .with_extra_table_factories(self.extra_information_table_factories.clone());
+                    Arc::new(provider)
                 });
             information_schema_provider.table(table_name)
         } else if schema == PG_CATALOG_NAME && channel == Channel::Postgres {

src/catalog/src/memory/manager.rs

@@ -352,11 +352,13 @@ impl MemoryCatalogManager {
     }
 
     fn create_catalog_entry(self: &Arc<Self>, catalog: String) -> SchemaEntries {
+        let backend = Arc::new(MemoryKvBackend::new());
         let information_schema_provider = InformationSchemaProvider::new(
             catalog,
             Arc::downgrade(self) as Weak<dyn CatalogManager>,
-            Arc::new(FlowMetadataManager::new(Arc::new(MemoryKvBackend::new()))),
+            Arc::new(FlowMetadataManager::new(backend.clone())),
             None, // we don't need ProcessManager on regions server.
+            backend,
         );
         let information_schema = information_schema_provider.tables().clone();
 

src/catalog/src/process_manager.rs

@@ -21,7 +21,7 @@ use std::sync::{Arc, RwLock};
 use api::v1::frontend::{KillProcessRequest, ListProcessRequest, ProcessInfo};
 use common_base::cancellation::CancellationHandle;
 use common_frontend::selector::{FrontendSelector, MetaClientSelector};
-use common_telemetry::{debug, info};
+use common_telemetry::{debug, info, warn};
 use common_time::util::current_time_millis;
 use meta_client::MetaClientRef;
 use snafu::{ensure, OptionExt, ResultExt};
@@ -141,14 +141,20 @@ impl ProcessManager {
                 .await
                 .context(error::InvokeFrontendSnafu)?;
             for mut f in frontends {
-                processes.extend(
-                    f.list_process(ListProcessRequest {
+                let result = f
+                    .list_process(ListProcessRequest {
                         catalog: catalog.unwrap_or_default().to_string(),
                     })
                     .await
-                    .context(error::InvokeFrontendSnafu)?
-                    .processes,
-                );
+                    .context(error::InvokeFrontendSnafu);
+                match result {
+                    Ok(resp) => {
+                        processes.extend(resp.processes);
+                    }
+                    Err(e) => {
+                        warn!(e; "Skipping failing node: {:?}", f)
+                    }
+                }
             }
         }
         processes.extend(self.local_processes(catalog)?);

src/catalog/src/system_schema.rs

@@ -15,7 +15,7 @@
 pub mod information_schema;
 mod memory_table;
 pub mod pg_catalog;
-mod predicate;
+pub mod predicate;
 mod utils;
 
 use std::collections::HashMap;
@@ -96,7 +96,7 @@ trait SystemSchemaProviderInner {
     }
 }
 
-pub(crate) trait SystemTable {
+pub trait SystemTable {
     fn table_id(&self) -> TableId;
 
     fn table_name(&self) -> &'static str;
@@ -110,7 +110,7 @@ pub(crate) trait SystemTable {
     }
 }
 
-pub(crate) type SystemTableRef = Arc<dyn SystemTable + Send + Sync>;
+pub type SystemTableRef = Arc<dyn SystemTable + Send + Sync>;
 
 struct SystemTableDataSource {
     table: SystemTableRef,

src/catalog/src/system_schema/information_schema.rs

@@ -19,7 +19,7 @@ mod information_memory_table;
 pub mod key_column_usage;
 mod partitions;
 mod procedure_info;
-mod process_list;
+pub mod process_list;
 pub mod region_peers;
 mod region_statistics;
 mod runtime_metrics;
@@ -38,6 +38,7 @@ use common_meta::cluster::NodeInfo;
 use common_meta::datanode::RegionStat;
 use common_meta::key::flow::flow_state::FlowStat;
 use common_meta::key::flow::FlowMetadataManager;
+use common_meta::kv_backend::KvBackendRef;
 use common_procedure::ProcedureInfo;
 use common_recordbatch::SendableRecordBatchStream;
 use datatypes::schema::SchemaRef;
@@ -112,6 +113,25 @@ macro_rules! setup_memory_table {
     };
 }
 
+#[cfg(feature = "enterprise")]
+pub struct MakeInformationTableRequest {
+    pub catalog_name: String,
+    pub catalog_manager: Weak<dyn CatalogManager>,
+    pub kv_backend: KvBackendRef,
+}
+
+/// A factory trait for making information schema tables.
+///
+/// This trait allows for extensibility of the information schema by providing
+/// a way to dynamically create custom information schema tables.
+#[cfg(feature = "enterprise")]
+pub trait InformationSchemaTableFactory {
+    fn make_information_table(&self, req: MakeInformationTableRequest) -> SystemTableRef;
+}
+
+#[cfg(feature = "enterprise")]
+pub type InformationSchemaTableFactoryRef = Arc<dyn InformationSchemaTableFactory + Send + Sync>;
+
 /// The `information_schema` tables info provider.
 pub struct InformationSchemaProvider {
     catalog_name: String,
@@ -119,6 +139,10 @@ pub struct InformationSchemaProvider {
     process_manager: Option<ProcessManagerRef>,
     flow_metadata_manager: Arc<FlowMetadataManager>,
     tables: HashMap<String, TableRef>,
+    #[allow(dead_code)]
+    kv_backend: KvBackendRef,
+    #[cfg(feature = "enterprise")]
+    extra_table_factories: HashMap<String, InformationSchemaTableFactoryRef>,
 }
 
 impl SystemSchemaProvider for InformationSchemaProvider {
@@ -128,6 +152,7 @@ impl SystemSchemaProvider for InformationSchemaProvider {
         &self.tables
     }
 }
+
 impl SystemSchemaProviderInner for InformationSchemaProvider {
     fn catalog_name(&self) -> &str {
         &self.catalog_name
@@ -137,6 +162,16 @@ impl SystemSchemaProviderInner for InformationSchemaProvider {
     }
 
     fn system_table(&self, name: &str) -> Option<SystemTableRef> {
+        #[cfg(feature = "enterprise")]
+        if let Some(factory) = self.extra_table_factories.get(name) {
+            let req = MakeInformationTableRequest {
+                catalog_name: self.catalog_name.clone(),
+                catalog_manager: self.catalog_manager.clone(),
+                kv_backend: self.kv_backend.clone(),
+            };
+            return Some(factory.make_information_table(req));
+        }
+
         match name.to_ascii_lowercase().as_str() {
             TABLES => Some(Arc::new(InformationSchemaTables::new(
                 self.catalog_name.clone(),
@@ -226,6 +261,7 @@ impl InformationSchemaProvider {
         catalog_manager: Weak<dyn CatalogManager>,
         flow_metadata_manager: Arc<FlowMetadataManager>,
         process_manager: Option<ProcessManagerRef>,
+        kv_backend: KvBackendRef,
     ) -> Self {
         let mut provider = Self {
             catalog_name,
@@ -233,6 +269,9 @@ impl InformationSchemaProvider {
             flow_metadata_manager,
             process_manager,
             tables: HashMap::new(),
+            kv_backend,
+            #[cfg(feature = "enterprise")]
+            extra_table_factories: HashMap::new(),
         };
 
         provider.build_tables();
@@ -240,6 +279,16 @@ impl InformationSchemaProvider {
         provider
     }
 
+    #[cfg(feature = "enterprise")]
+    pub(crate) fn with_extra_table_factories(
+        mut self,
+        factories: HashMap<String, InformationSchemaTableFactoryRef>,
+    ) -> Self {
+        self.extra_table_factories = factories;
+        self.build_tables();
+        self
+    }
+
     fn build_tables(&mut self) {
         let mut tables = HashMap::new();
 
@@ -290,16 +339,19 @@ impl InformationSchemaProvider {
         if let Some(process_list) = self.build_table(PROCESS_LIST) {
             tables.insert(PROCESS_LIST.to_string(), process_list);
         }
+        #[cfg(feature = "enterprise")]
+        for name in self.extra_table_factories.keys() {
+            tables.insert(name.to_string(), self.build_table(name).expect(name));
+        }
         // Add memory tables
         for name in MEMORY_TABLES.iter() {
             tables.insert((*name).to_string(), self.build_table(name).expect(name));
         }
-
         self.tables = tables;
     }
 }
 
-trait InformationTable {
+pub trait InformationTable {
     fn table_id(&self) -> TableId;
 
     fn table_name(&self) -> &'static str;

src/catalog/src/system_schema/information_schema/information_memory_table.rs

@@ -15,7 +15,8 @@
 use std::sync::Arc;
 
 use common_catalog::consts::{METRIC_ENGINE, MITO_ENGINE};
-use datatypes::schema::{Schema, SchemaRef};
+use datatypes::data_type::ConcreteDataType;
+use datatypes::schema::{ColumnSchema, Schema, SchemaRef};
 use datatypes::vectors::{Int64Vector, StringVector, VectorRef};
 
 use crate::system_schema::information_schema::table_names::*;
@@ -367,28 +368,18 @@ pub(super) fn get_schema_columns(table_name: &str) -> (SchemaRef, Vec<VectorRef>
 
         TRIGGERS => (
             vec![
-                string_column("TRIGGER_CATALOG"),
-                string_column("TRIGGER_SCHEMA"),
                 string_column("TRIGGER_NAME"),
-                string_column("EVENT_MANIPULATION"),
-                string_column("EVENT_OBJECT_CATALOG"),
-                string_column("EVENT_OBJECT_SCHEMA"),
-                string_column("EVENT_OBJECT_TABLE"),
-                bigint_column("ACTION_ORDER"),
-                string_column("ACTION_CONDITION"),
-                string_column("ACTION_STATEMENT"),
-                string_column("ACTION_ORIENTATION"),
-                string_column("ACTION_TIMING"),
-                string_column("ACTION_REFERENCE_OLD_TABLE"),
-                string_column("ACTION_REFERENCE_NEW_TABLE"),
-                string_column("ACTION_REFERENCE_OLD_ROW"),
-                string_column("ACTION_REFERENCE_NEW_ROW"),
-                timestamp_micro_column("CREATED"),
-                string_column("SQL_MODE"),
-                string_column("DEFINER"),
-                string_column("CHARACTER_SET_CLIENT"),
-                string_column("COLLATION_CONNECTION"),
-                string_column("DATABASE_COLLATION"),
+                ColumnSchema::new(
+                    "trigger_id",
+                    ConcreteDataType::uint64_datatype(),
+                    false,
+                ),
+                string_column("TRIGGER_DEFINITION"),
+                ColumnSchema::new(
+                    "flownode_id",
+                    ConcreteDataType::uint64_datatype(),
+                    true,
+                ),
             ],
             vec![],
         ),

src/catalog/src/system_schema/information_schema/partitions.rs

@@ -329,13 +329,8 @@ impl InformationSchemaPartitionsBuilder {
             self.partition_names.push(Some(&partition_name));
             self.partition_ordinal_positions
                 .push(Some((index + 1) as i64));
-            let expressions = if partition.partition.partition_columns().is_empty() {
-                None
-            } else {
-                Some(partition.partition.to_string())
-            };
-
-            self.partition_expressions.push(expressions.as_deref());
+            let expression = partition.partition_expr.as_ref().map(|e| e.to_string());
+            self.partition_expressions.push(expression.as_deref());
             self.create_times.push(Some(TimestampMicrosecond::from(
                 table_info.meta.created_on.timestamp_millis(),
             )));

src/catalog/src/system_schema/information_schema/process_list.rs

@@ -39,14 +39,14 @@ use crate::process_manager::ProcessManagerRef;
 use crate::system_schema::information_schema::InformationTable;
 
 /// Column names of `information_schema.process_list`
-const ID: &str = "id";
-const CATALOG: &str = "catalog";
-const SCHEMAS: &str = "schemas";
-const QUERY: &str = "query";
-const CLIENT: &str = "client";
-const FRONTEND: &str = "frontend";
-const START_TIMESTAMP: &str = "start_timestamp";
-const ELAPSED_TIME: &str = "elapsed_time";
+pub const ID: &str = "id";
+pub const CATALOG: &str = "catalog";
+pub const SCHEMAS: &str = "schemas";
+pub const QUERY: &str = "query";
+pub const CLIENT: &str = "client";
+pub const FRONTEND: &str = "frontend";
+pub const START_TIMESTAMP: &str = "start_timestamp";
+pub const ELAPSED_TIME: &str = "elapsed_time";
 
 /// `information_schema.process_list` table implementation that tracks running
 /// queries in current cluster.

src/catalog/src/system_schema/information_schema/region_statistics.rs

@@ -44,6 +44,7 @@ const DISK_SIZE: &str = "disk_size";
 const MEMTABLE_SIZE: &str = "memtable_size";
 const MANIFEST_SIZE: &str = "manifest_size";
 const SST_SIZE: &str = "sst_size";
+const SST_NUM: &str = "sst_num";
 const INDEX_SIZE: &str = "index_size";
 const ENGINE: &str = "engine";
 const REGION_ROLE: &str = "region_role";
@@ -87,6 +88,7 @@ impl InformationSchemaRegionStatistics {
             ColumnSchema::new(MEMTABLE_SIZE, ConcreteDataType::uint64_datatype(), true),
             ColumnSchema::new(MANIFEST_SIZE, ConcreteDataType::uint64_datatype(), true),
             ColumnSchema::new(SST_SIZE, ConcreteDataType::uint64_datatype(), true),
+            ColumnSchema::new(SST_NUM, ConcreteDataType::uint64_datatype(), true),
             ColumnSchema::new(INDEX_SIZE, ConcreteDataType::uint64_datatype(), true),
             ColumnSchema::new(ENGINE, ConcreteDataType::string_datatype(), true),
             ColumnSchema::new(REGION_ROLE, ConcreteDataType::string_datatype(), true),
@@ -149,6 +151,7 @@ struct InformationSchemaRegionStatisticsBuilder {
     memtable_sizes: UInt64VectorBuilder,
     manifest_sizes: UInt64VectorBuilder,
     sst_sizes: UInt64VectorBuilder,
+    sst_nums: UInt64VectorBuilder,
     index_sizes: UInt64VectorBuilder,
     engines: StringVectorBuilder,
     region_roles: StringVectorBuilder,
@@ -167,6 +170,7 @@ impl InformationSchemaRegionStatisticsBuilder {
             memtable_sizes: UInt64VectorBuilder::with_capacity(INIT_CAPACITY),
             manifest_sizes: UInt64VectorBuilder::with_capacity(INIT_CAPACITY),
             sst_sizes: UInt64VectorBuilder::with_capacity(INIT_CAPACITY),
+            sst_nums: UInt64VectorBuilder::with_capacity(INIT_CAPACITY),
             index_sizes: UInt64VectorBuilder::with_capacity(INIT_CAPACITY),
             engines: StringVectorBuilder::with_capacity(INIT_CAPACITY),
             region_roles: StringVectorBuilder::with_capacity(INIT_CAPACITY),
@@ -197,6 +201,7 @@ impl InformationSchemaRegionStatisticsBuilder {
             (MEMTABLE_SIZE, &Value::from(region_stat.memtable_size)),
             (MANIFEST_SIZE, &Value::from(region_stat.manifest_size)),
             (SST_SIZE, &Value::from(region_stat.sst_size)),
+            (SST_NUM, &Value::from(region_stat.sst_num)),
             (INDEX_SIZE, &Value::from(region_stat.index_size)),
             (ENGINE, &Value::from(region_stat.engine.as_str())),
             (REGION_ROLE, &Value::from(region_stat.role.to_string())),
@@ -215,6 +220,7 @@ impl InformationSchemaRegionStatisticsBuilder {
         self.memtable_sizes.push(Some(region_stat.memtable_size));
         self.manifest_sizes.push(Some(region_stat.manifest_size));
         self.sst_sizes.push(Some(region_stat.sst_size));
+        self.sst_nums.push(Some(region_stat.sst_num));
         self.index_sizes.push(Some(region_stat.index_size));
         self.engines.push(Some(&region_stat.engine));
         self.region_roles.push(Some(&region_stat.role.to_string()));
@@ -230,6 +236,7 @@ impl InformationSchemaRegionStatisticsBuilder {
             Arc::new(self.memtable_sizes.finish()),
             Arc::new(self.manifest_sizes.finish()),
             Arc::new(self.sst_sizes.finish()),
+            Arc::new(self.sst_nums.finish()),
             Arc::new(self.index_sizes.finish()),
             Arc::new(self.engines.finish()),
             Arc::new(self.region_roles.finish()),

src/catalog/src/system_schema/pg_catalog/pg_class.rs

@@ -169,7 +169,7 @@ impl DfPartitionStream for PGClass {
 }
 
 /// Builds the `pg_catalog.pg_class` table row by row
-/// TODO(J0HN50N133): `relowner` is always the [`DUMMY_OWNER_ID`] cuz we don't have user.
+/// TODO(J0HN50N133): `relowner` is always the [`DUMMY_OWNER_ID`] because we don't have users.
 /// Once we have user system, make it the actual owner of the table.
 struct PGClassBuilder {
     schema: SchemaRef,

src/catalog/src/table_source.rs

@@ -70,6 +70,11 @@ impl DfTableSourceProvider {
         }
     }
 
+    /// Returns the query context.
+    pub fn query_ctx(&self) -> &QueryContextRef {
+        &self.query_ctx
+    }
+
     pub fn resolve_table_ref(&self, table_ref: TableReference) -> Result<ResolvedTableReference> {
         if self.disallow_cross_catalog_query {
             match &table_ref {
@@ -207,6 +212,7 @@ mod tests {
     use session::context::QueryContext;
 
     use super::*;
+    use crate::kvbackend::KvBackendCatalogManagerBuilder;
     use crate::memory::MemoryCatalogManager;
 
     #[test]
@@ -323,13 +329,13 @@ mod tests {
             .build(),
         );
 
-        let catalog_manager = KvBackendCatalogManager::new(
+        let catalog_manager = KvBackendCatalogManagerBuilder::new(
             Arc::new(NoopInformationExtension),
             backend.clone(),
             layered_cache_registry,
-            None,
-            None,
-        );
+        )
+        .build();
+
         let table_metadata_manager = TableMetadataManager::new(backend);
         let mut view_info = common_meta::key::test_utils::new_test_table_info(1024, vec![]);
         view_info.table_type = TableType::View;

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

src/cli/src/bench.rs

@@ -160,6 +160,7 @@ fn create_table_info(table_id: TableId, table_name: TableName) -> RawTableInfo {
         options: Default::default(),
         region_numbers: (1..=100).collect(),
         partition_key_indices: vec![],
+        column_ids: vec![],
     };
 
     RawTableInfo {
@@ -187,6 +188,7 @@ fn create_region_routes(regions: Vec<RegionNumber>) -> Vec<RegionRoute> {
                 name: String::new(),
                 partition: None,
                 attrs: BTreeMap::new(),
+                partition_expr: Default::default(),
             },
             leader_peer: Some(Peer {
                 id: rng.random_range(0..10),

src/cli/src/lib.rs

@@ -12,6 +12,7 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
+#![allow(clippy::print_stdout)]
 mod bench;
 mod data;
 mod database;

src/cli/src/metadata/common.rs

@@ -75,7 +75,7 @@ impl StoreConfig {
                 #[cfg(feature = "pg_kvbackend")]
                 BackendImpl::PostgresStore => {
                     let table_name = &self.meta_table_name;
-                    let pool = meta_srv::bootstrap::create_postgres_pool(store_addrs)
+                    let pool = meta_srv::bootstrap::create_postgres_pool(store_addrs, None)
                         .await
                         .map_err(BoxedError::new)?;
                     Ok(common_meta::kv_backend::rds::PgStore::with_pg_pool(

src/cli/src/metadata/repair.rs

@@ -241,7 +241,6 @@ impl RepairTool {
             let alter_table_request = alter_table::make_alter_region_request_for_peer(
                 logical_table_id,
                 &alter_table_expr,
-                full_table_metadata.table_info.ident.version,
                 peer,
                 physical_region_routes,
             )?;

src/cli/src/metadata/repair/alter_table.rs

@@ -66,7 +66,6 @@ pub fn generate_alter_table_expr_for_all_columns(
 pub fn make_alter_region_request_for_peer(
     logical_table_id: TableId,
     alter_table_expr: &AlterTableExpr,
-    schema_version: u64,
     peer: &Peer,
     region_routes: &[RegionRoute],
 ) -> Result<RegionRequest> {
@@ -74,7 +73,7 @@ pub fn make_alter_region_request_for_peer(
     let mut requests = Vec::with_capacity(regions_on_this_peer.len());
     for region_number in &regions_on_this_peer {
         let region_id = RegionId::new(logical_table_id, *region_number);
-        let request = make_alter_region_request(region_id, alter_table_expr, schema_version);
+        let request = make_alter_region_request(region_id, alter_table_expr);
         requests.push(request);
     }
 

src/cli/src/metadata/snapshot.rs

@@ -301,7 +301,6 @@ struct MetaInfoTool {
 
 #[async_trait]
 impl Tool for MetaInfoTool {
-    #[allow(clippy::print_stdout)]
     async fn do_work(&self) -> std::result::Result<(), BoxedError> {
         let result = MetadataSnapshotManager::info(
             &self.inner,

src/client/src/database.rs

@@ -23,7 +23,7 @@ use api::v1::greptime_request::Request;
 use api::v1::query_request::Query;
 use api::v1::{
     AlterTableExpr, AuthHeader, Basic, CreateTableExpr, DdlRequest, GreptimeRequest,
-    InsertRequests, QueryRequest, RequestHeader,
+    InsertRequests, QueryRequest, RequestHeader, RowInsertRequests,
 };
 use arrow_flight::{FlightData, Ticket};
 use async_stream::stream;
@@ -31,7 +31,7 @@ use base64::prelude::BASE64_STANDARD;
 use base64::Engine;
 use common_catalog::build_db_string;
 use common_catalog::consts::{DEFAULT_CATALOG_NAME, DEFAULT_SCHEMA_NAME};
-use common_error::ext::{BoxedError, ErrorExt};
+use common_error::ext::BoxedError;
 use common_grpc::flight::do_put::DoPutResponse;
 use common_grpc::flight::{FlightDecoder, FlightMessage};
 use common_query::Output;
@@ -42,13 +42,13 @@ use common_telemetry::{error, warn};
 use futures::future;
 use futures_util::{Stream, StreamExt, TryStreamExt};
 use prost::Message;
-use snafu::{ensure, ResultExt};
+use snafu::{ensure, OptionExt, ResultExt};
 use tonic::metadata::{AsciiMetadataKey, AsciiMetadataValue, MetadataMap, MetadataValue};
 use tonic::transport::Channel;
 
 use crate::error::{
     ConvertFlightDataSnafu, Error, FlightGetSnafu, IllegalFlightMessagesSnafu,
-    InvalidTonicMetadataValueSnafu, ServerSnafu,
+    InvalidTonicMetadataValueSnafu,
 };
 use crate::{error, from_grpc_response, Client, Result};
 
@@ -118,6 +118,7 @@ impl Database {
         }
     }
 
+    /// Set the catalog for the database client.
     pub fn set_catalog(&mut self, catalog: impl Into<String>) {
         self.catalog = catalog.into();
     }
@@ -130,6 +131,7 @@ impl Database {
         }
     }
 
+    /// Set the schema for the database client.
     pub fn set_schema(&mut self, schema: impl Into<String>) {
         self.schema = schema.into();
     }
@@ -142,20 +144,24 @@ impl Database {
         }
     }
 
+    /// Set the timezone for the database client.
     pub fn set_timezone(&mut self, timezone: impl Into<String>) {
         self.timezone = timezone.into();
     }
 
+    /// Set the auth scheme for the database client.
     pub fn set_auth(&mut self, auth: AuthScheme) {
         self.ctx.auth_header = Some(AuthHeader {
             auth_scheme: Some(auth),
         });
     }
 
+    /// Make an InsertRequests request to the database.
     pub async fn insert(&self, requests: InsertRequests) -> Result<u32> {
         self.handle(Request::Inserts(requests)).await
     }
 
+    /// Make an InsertRequests request to the database with hints.
     pub async fn insert_with_hints(
         &self,
         requests: InsertRequests,
@@ -172,6 +178,28 @@ impl Database {
         from_grpc_response(response)
     }
 
+    /// Make a RowInsertRequests request to the database.
+    pub async fn row_inserts(&self, requests: RowInsertRequests) -> Result<u32> {
+        self.handle(Request::RowInserts(requests)).await
+    }
+
+    /// Make a RowInsertRequests request to the database with hints.
+    pub async fn row_inserts_with_hints(
+        &self,
+        requests: RowInsertRequests,
+        hints: &[(&str, &str)],
+    ) -> Result<u32> {
+        let mut client = make_database_client(&self.client)?.inner;
+        let request = self.to_rpc_request(Request::RowInserts(requests));
+
+        let mut request = tonic::Request::new(request);
+        let metadata = request.metadata_mut();
+        Self::put_hints(metadata, hints)?;
+
+        let response = client.handle(request).await?.into_inner();
+        from_grpc_response(response)
+    }
+
     fn put_hints(metadata: &mut MetadataMap, hints: &[(&str, &str)]) -> Result<()> {
         let Some(value) = hints
             .iter()
@@ -187,6 +215,7 @@ impl Database {
         Ok(())
     }
 
+    /// Make a request to the database.
     pub async fn handle(&self, request: Request) -> Result<u32> {
         let mut client = make_database_client(&self.client)?.inner;
         let request = self.to_rpc_request(request);
@@ -196,12 +225,22 @@ impl Database {
 
     /// Retry if connection fails, max_retries is the max number of retries, so the total wait time
     /// is `max_retries * GRPC_CONN_TIMEOUT`
-    pub async fn handle_with_retry(&self, request: Request, max_retries: u32) -> Result<u32> {
+    pub async fn handle_with_retry(
+        &self,
+        request: Request,
+        max_retries: u32,
+        hints: &[(&str, &str)],
+    ) -> Result<u32> {
         let mut client = make_database_client(&self.client)?.inner;
         let mut retries = 0;
+
         let request = self.to_rpc_request(request);
+
         loop {
-            let raw_response = client.handle(request.clone()).await;
+            let mut tonic_request = tonic::Request::new(request.clone());
+            let metadata = tonic_request.metadata_mut();
+            Self::put_hints(metadata, hints)?;
+            let raw_response = client.handle(tonic_request).await;
             match (raw_response, retries < max_retries) {
                 (Ok(resp), _) => return from_grpc_response(resp.into_inner()),
                 (Err(err), true) => {
@@ -211,12 +250,18 @@ impl Database {
                         retries += 1;
                         warn!("Retrying {} times with error = {:?}", retries, err);
                         continue;
+                    } else {
+                        error!(
+                            err; "Failed to send request to grpc handle, retries = {}, not retryable error, aborting",
+                            retries
+                        );
+                        return Err(err.into());
                     }
                 }
                 (Err(err), false) => {
                     error!(
-                        "Failed to send request to grpc handle after {} retries, error = {:?}",
-                        retries, err
+                        err; "Failed to send request to grpc handle after {} retries",
+                        retries,
                     );
                     return Err(err.into());
                 }
@@ -240,6 +285,7 @@ impl Database {
         }
     }
 
+    /// Executes a SQL query without any hints.
     pub async fn sql<S>(&self, sql: S) -> Result<Output>
     where
         S: AsRef<str>,
@@ -247,6 +293,7 @@ impl Database {
         self.sql_with_hint(sql, &[]).await
     }
 
+    /// Executes a SQL query with optional hints for query optimization.
     pub async fn sql_with_hint<S>(&self, sql: S, hints: &[(&str, &str)]) -> Result<Output>
     where
         S: AsRef<str>,
@@ -257,6 +304,7 @@ impl Database {
         self.do_get(request, hints).await
     }
 
+    /// Executes a logical plan directly without SQL parsing.
     pub async fn logical_plan(&self, logical_plan: Vec<u8>) -> Result<Output> {
         let request = Request::Query(QueryRequest {
             query: Some(Query::LogicalPlan(logical_plan)),
@@ -264,6 +312,7 @@ impl Database {
         self.do_get(request, &[]).await
     }
 
+    /// Creates a new table using the provided table expression.
     pub async fn create(&self, expr: CreateTableExpr) -> Result<Output> {
         let request = Request::Ddl(DdlRequest {
             expr: Some(DdlExpr::CreateTable(expr)),
@@ -271,6 +320,7 @@ impl Database {
         self.do_get(request, &[]).await
     }
 
+    /// Alters an existing table using the provided alter expression.
     pub async fn alter(&self, expr: AlterTableExpr) -> Result<Output> {
         let request = Request::Ddl(DdlRequest {
             expr: Some(DdlExpr::AlterTable(expr)),
@@ -292,21 +342,16 @@ impl Database {
         let response = client.mut_inner().do_get(request).await.or_else(|e| {
             let tonic_code = e.code();
             let e: Error = e.into();
-            let code = e.status_code();
-            let msg = e.to_string();
-            let error =
-                Err(BoxedError::new(ServerSnafu { code, msg }.build())).with_context(|_| {
-                    FlightGetSnafu {
-                        addr: client.addr().to_string(),
-                        tonic_code,
-                    }
-                });
             error!(
                 "Failed to do Flight get, addr: {}, code: {}, source: {:?}",
                 client.addr(),
                 tonic_code,
-                error
+                e
             );
+            let error = Err(BoxedError::new(e)).with_context(|_| FlightGetSnafu {
+                addr: client.addr().to_string(),
+                tonic_code,
+            });
             error
         })?;
 
@@ -316,7 +361,10 @@ impl Database {
         let mut flight_message_stream = flight_data_stream.map(move |flight_data| {
             flight_data
                 .map_err(Error::from)
-                .and_then(|data| decoder.try_decode(&data).context(ConvertFlightDataSnafu))
+                .and_then(|data| decoder.try_decode(&data).context(ConvertFlightDataSnafu))?
+                .context(IllegalFlightMessagesSnafu {
+                    reason: "none message",
+                })
         });
 
         let Some(first_flight_message) = flight_message_stream.next().await else {
@@ -436,8 +484,11 @@ mod tests {
 
     use api::v1::auth_header::AuthScheme;
     use api::v1::{AuthHeader, Basic};
+    use common_error::status_code::StatusCode;
+    use tonic::{Code, Status};
 
     use super::*;
+    use crate::error::TonicSnafu;
 
     #[test]
     fn test_flight_ctx() {
@@ -460,4 +511,19 @@ mod tests {
             })
         )
     }
+
+    #[test]
+    fn test_from_tonic_status() {
+        let expected = TonicSnafu {
+            code: StatusCode::Internal,
+            msg: "blabla".to_string(),
+            tonic_code: Code::Internal,
+        }
+        .build();
+
+        let status = Status::new(Code::Internal, "blabla");
+        let actual: Error = status.into();
+
+        assert_eq!(expected.to_string(), actual.to_string());
+    }
 }

src/client/src/error.rs

@@ -14,13 +14,13 @@
 
 use std::any::Any;
 
+use common_error::define_from_tonic_status;
 use common_error::ext::{BoxedError, ErrorExt};
-use common_error::status_code::{convert_tonic_code_to_status_code, StatusCode};
-use common_error::{GREPTIME_DB_HEADER_ERROR_CODE, GREPTIME_DB_HEADER_ERROR_MSG};
+use common_error::status_code::StatusCode;
 use common_macro::stack_trace_debug;
 use snafu::{location, Location, Snafu};
 use tonic::metadata::errors::InvalidMetadataValue;
-use tonic::{Code, Status};
+use tonic::Code;
 
 #[derive(Snafu)]
 #[snafu(visibility(pub))]
@@ -124,6 +124,15 @@ pub enum Error {
         location: Location,
         source: datatypes::error::Error,
     },
+
+    #[snafu(display("{}", msg))]
+    Tonic {
+        code: StatusCode,
+        msg: String,
+        tonic_code: Code,
+        #[snafu(implicit)]
+        location: Location,
+    },
 }
 
 pub type Result<T> = std::result::Result<T, Error>;
@@ -135,7 +144,7 @@ impl ErrorExt for Error {
             | Error::MissingField { .. }
             | Error::IllegalDatabaseResponse { .. } => StatusCode::Internal,
 
-            Error::Server { code, .. } => *code,
+            Error::Server { code, .. } | Error::Tonic { code, .. } => *code,
             Error::FlightGet { source, .. }
             | Error::RegionServer { source, .. }
             | Error::FlowServer { source, .. } => source.status_code(),
@@ -153,34 +162,7 @@ impl ErrorExt for Error {
     }
 }
 
-impl From<Status> for Error {
-    fn from(e: Status) -> Self {
-        fn get_metadata_value(e: &Status, key: &str) -> Option<String> {
-            e.metadata()
-                .get(key)
-                .and_then(|v| String::from_utf8(v.as_bytes().to_vec()).ok())
-        }
-
-        let code = get_metadata_value(&e, GREPTIME_DB_HEADER_ERROR_CODE).and_then(|s| {
-            if let Ok(code) = s.parse::<u32>() {
-                StatusCode::from_u32(code)
-            } else {
-                None
-            }
-        });
-        let tonic_code = e.code();
-        let code = code.unwrap_or_else(|| convert_tonic_code_to_status_code(tonic_code));
-
-        let msg = get_metadata_value(&e, GREPTIME_DB_HEADER_ERROR_MSG)
-            .unwrap_or_else(|| e.message().to_string());
-
-        Self::Server {
-            code,
-            msg,
-            location: location!(),
-        }
-    }
-}
+define_from_tonic_status!(Error, Tonic);
 
 impl Error {
     pub fn should_retry(&self) -> bool {

src/client/src/region.rs

@@ -21,7 +21,7 @@ use arc_swap::ArcSwapOption;
 use arrow_flight::Ticket;
 use async_stream::stream;
 use async_trait::async_trait;
-use common_error::ext::{BoxedError, ErrorExt};
+use common_error::ext::BoxedError;
 use common_error::status_code::StatusCode;
 use common_grpc::flight::{FlightDecoder, FlightMessage};
 use common_meta::error::{self as meta_error, Result as MetaResult};
@@ -107,24 +107,18 @@ impl RegionRequester {
             .mut_inner()
             .do_get(ticket)
             .await
-            .map_err(|e| {
+            .or_else(|e| {
                 let tonic_code = e.code();
                 let e: error::Error = e.into();
-                let code = e.status_code();
-                let msg = e.to_string();
-                let error = ServerSnafu { code, msg }
-                    .fail::<()>()
-                    .map_err(BoxedError::new)
-                    .with_context(|_| FlightGetSnafu {
-                        tonic_code,
-                        addr: flight_client.addr().to_string(),
-                    })
-                    .unwrap_err();
                 error!(
                     e; "Failed to do Flight get, addr: {}, code: {}",
                     flight_client.addr(),
                     tonic_code
                 );
+                let error = Err(BoxedError::new(e)).with_context(|_| FlightGetSnafu {
+                    addr: flight_client.addr().to_string(),
+                    tonic_code,
+                });
                 error
             })?;
 
@@ -134,7 +128,10 @@ impl RegionRequester {
         let mut flight_message_stream = flight_data_stream.map(move |flight_data| {
             flight_data
                 .map_err(Error::from)
-                .and_then(|data| decoder.try_decode(&data).context(ConvertFlightDataSnafu))
+                .and_then(|data| decoder.try_decode(&data).context(ConvertFlightDataSnafu))?
+                .context(IllegalFlightMessagesSnafu {
+                    reason: "none message",
+                })
         });
 
         let Some(first_flight_message) = flight_message_stream.next().await else {
@@ -163,19 +160,70 @@ impl RegionRequester {
             let _span = tracing_context.attach(common_telemetry::tracing::info_span!(
                 "poll_flight_data_stream"
             ));
-            while let Some(flight_message) = flight_message_stream.next().await {
-                let flight_message = flight_message
-                    .map_err(BoxedError::new)
-                    .context(ExternalSnafu)?;
+
+            let mut buffered_message: Option<FlightMessage> = None;
+            let mut stream_ended = false;
+
+            while !stream_ended {
+                // get the next message from the buffered message or read from the flight message stream
+                let flight_message_item = if let Some(msg) = buffered_message.take() {
+                    Some(Ok(msg))
+                } else {
+                    flight_message_stream.next().await
+                };
+
+                let flight_message = match flight_message_item {
+                    Some(Ok(message)) => message,
+                    Some(Err(e)) => {
+                        yield Err(BoxedError::new(e)).context(ExternalSnafu);
+                        break;
+                    }
+                    None => break,
+                };
 
                 match flight_message {
                     FlightMessage::RecordBatch(record_batch) => {
-                        yield RecordBatch::try_from_df_record_batch(
+                        let result_to_yield = RecordBatch::try_from_df_record_batch(
                             schema_cloned.clone(),
                             record_batch,
-                        )
+                        );
+
+                        // get the next message from the stream. normally it should be a metrics message.
+                        if let Some(next_flight_message_result) = flight_message_stream.next().await
+                        {
+                            match next_flight_message_result {
+                                Ok(FlightMessage::Metrics(s)) => {
+                                    let m = serde_json::from_str(&s).ok().map(Arc::new);
+                                    metrics_ref.swap(m);
+                                }
+                                Ok(FlightMessage::RecordBatch(rb)) => {
+                                    // for some reason it's not a metrics message, so we need to buffer this record batch
+                                    // and yield it in the next iteration.
+                                    buffered_message = Some(FlightMessage::RecordBatch(rb));
+                                }
+                                Ok(_) => {
+                                    yield IllegalFlightMessagesSnafu {
+                                        reason: "A RecordBatch message can only be succeeded by a Metrics message or another RecordBatch message"
+                                    }
+                                    .fail()
+                                    .map_err(BoxedError::new)
+                                    .context(ExternalSnafu);
+                                    break;
+                                }
+                                Err(e) => {
+                                    yield Err(BoxedError::new(e)).context(ExternalSnafu);
+                                    break;
+                                }
+                            }
+                        } else {
+                            // the stream has ended
+                            stream_ended = true;
+                        }
+
+                        yield result_to_yield;
                     }
                     FlightMessage::Metrics(s) => {
+                        // just a branch in case of some metrics message comes after other things.
                         let m = serde_json::from_str(&s).ok().map(Arc::new);
                         metrics_ref.swap(m);
                         break;

src/cmd/Cargo.toml

@@ -16,7 +16,7 @@ default = [
     "meta-srv/pg_kvbackend",
     "meta-srv/mysql_kvbackend",
 ]
-enterprise = ["common-meta/enterprise", "frontend/enterprise", "meta-srv/enterprise"]
+enterprise = ["common-meta/enterprise", "frontend/enterprise", "meta-srv/enterprise", "catalog/enterprise"]
 tokio-console = ["common-telemetry/tokio-console"]
 
 [lints]
@@ -38,6 +38,7 @@ common-config.workspace = true
 common-error.workspace = true
 common-grpc.workspace = true
 common-macro.workspace = true
+common-mem-prof.workspace = true
 common-meta.workspace = true
 common-options.workspace = true
 common-procedure.workspace = true
@@ -52,7 +53,6 @@ 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
@@ -67,6 +67,7 @@ metric-engine.workspace = true
 mito2.workspace = true
 moka.workspace = true
 nu-ansi-term = "0.46"
+object-store.workspace = true
 plugins.workspace = true
 prometheus.workspace = true
 prost.workspace = true

src/cmd/src/bin/greptime.rs

@@ -20,11 +20,11 @@ use cmd::error::{InitTlsProviderSnafu, Result};
 use cmd::options::GlobalOptions;
 use cmd::{cli, datanode, flownode, frontend, metasrv, standalone, App};
 use common_base::Plugins;
-use common_version::version;
+use common_version::{verbose_version, version};
 use servers::install_ring_crypto_provider;
 
 #[derive(Parser)]
-#[command(name = "greptime", author, version, long_version = version(), about)]
+#[command(name = "greptime", author, version, long_version = verbose_version(), about)]
 #[command(propagate_version = true)]
 pub(crate) struct Command {
     #[clap(subcommand)]
@@ -143,10 +143,8 @@ async fn start(cli: Command) -> Result<()> {
 }
 
 fn setup_human_panic() {
-    human_panic::setup_panic!(
-        human_panic::Metadata::new("GreptimeDB", env!("CARGO_PKG_VERSION"))
-            .homepage("https://github.com/GreptimeTeam/greptimedb/discussions")
-    );
+    human_panic::setup_panic!(human_panic::Metadata::new("GreptimeDB", version())
+        .homepage("https://github.com/GreptimeTeam/greptimedb/discussions"));
 
     common_telemetry::set_panic_hook();
 }

src/cmd/src/datanode.rs

@@ -280,7 +280,7 @@ mod tests {
 
     use common_config::ENV_VAR_SEP;
     use common_test_util::temp_dir::create_named_temp_file;
-    use datanode::config::{FileConfig, GcsConfig, ObjectStoreConfig, S3Config};
+    use object_store::config::{FileConfig, GcsConfig, ObjectStoreConfig, S3Config};
     use servers::heartbeat_options::HeartbeatOptions;
 
     use super::*;

src/cmd/src/datanode/builder.rs

@@ -19,7 +19,7 @@ use catalog::kvbackend::MetaKvBackend;
 use common_base::Plugins;
 use common_meta::cache::LayeredCacheRegistryBuilder;
 use common_telemetry::info;
-use common_version::{short_version, version};
+use common_version::{short_version, verbose_version};
 use datanode::datanode::DatanodeBuilder;
 use datanode::service::DatanodeServiceBuilder;
 use meta_client::MetaClientType;
@@ -28,7 +28,7 @@ use tracing_appender::non_blocking::WorkerGuard;
 
 use crate::datanode::{DatanodeOptions, Instance, APP_NAME};
 use crate::error::{MetaClientInitSnafu, MissingConfigSnafu, Result, StartDatanodeSnafu};
-use crate::{create_resource_limit_metrics, log_versions};
+use crate::{create_resource_limit_metrics, log_versions, maybe_activate_heap_profile};
 
 /// Builder for Datanode instance.
 pub struct InstanceBuilder {
@@ -67,7 +67,8 @@ impl InstanceBuilder {
             None,
         );
 
-        log_versions(version(), short_version(), APP_NAME);
+        log_versions(verbose_version(), short_version(), APP_NAME);
+        maybe_activate_heap_profile(&dn_opts.memory);
         create_resource_limit_metrics(APP_NAME);
 
         plugins::setup_datanode_plugins(plugins, &opts.plugins, dn_opts)

src/cmd/src/flownode.rs

@@ -18,7 +18,7 @@ use std::time::Duration;
 
 use cache::{build_fundamental_cache_registry, with_default_composite_cache_registry};
 use catalog::information_extension::DistributedInformationExtension;
-use catalog::kvbackend::{CachedKvBackendBuilder, KvBackendCatalogManager, MetaKvBackend};
+use catalog::kvbackend::{CachedKvBackendBuilder, KvBackendCatalogManagerBuilder, MetaKvBackend};
 use clap::Parser;
 use client::client_manager::NodeClients;
 use common_base::Plugins;
@@ -32,7 +32,7 @@ use common_meta::key::flow::FlowMetadataManager;
 use common_meta::key::TableMetadataManager;
 use common_telemetry::info;
 use common_telemetry::logging::{TracingOptions, DEFAULT_LOGGING_DIR};
-use common_version::{short_version, version};
+use common_version::{short_version, verbose_version};
 use flow::{
     get_flow_auth_options, FlownodeBuilder, FlownodeInstance, FlownodeServiceBuilder,
     FrontendClient, FrontendInvoker,
@@ -46,7 +46,7 @@ use crate::error::{
     MissingConfigSnafu, Result, ShutdownFlownodeSnafu, StartFlownodeSnafu,
 };
 use crate::options::{GlobalOptions, GreptimeOptions};
-use crate::{create_resource_limit_metrics, log_versions, App};
+use crate::{create_resource_limit_metrics, log_versions, maybe_activate_heap_profile, App};
 
 pub const APP_NAME: &str = "greptime-flownode";
 
@@ -279,7 +279,8 @@ impl StartCommand {
             None,
         );
 
-        log_versions(version(), short_version(), APP_NAME);
+        log_versions(verbose_version(), short_version(), APP_NAME);
+        maybe_activate_heap_profile(&opts.component.memory);
         create_resource_limit_metrics(APP_NAME);
 
         info!("Flownode start command: {:#?}", self);
@@ -342,13 +343,12 @@ impl StartCommand {
 
         let information_extension =
             Arc::new(DistributedInformationExtension::new(meta_client.clone()));
-        let catalog_manager = KvBackendCatalogManager::new(
+        let catalog_manager = KvBackendCatalogManagerBuilder::new(
             information_extension,
             cached_meta_backend.clone(),
             layered_cache_registry.clone(),
-            None,
-            None,
-        );
+        )
+        .build();
 
         let table_metadata_manager =
             Arc::new(TableMetadataManager::new(cached_meta_backend.clone()));
@@ -371,8 +371,12 @@ impl StartCommand {
 
         let flow_metadata_manager = Arc::new(FlowMetadataManager::new(cached_meta_backend.clone()));
         let flow_auth_header = get_flow_auth_options(&opts).context(StartFlownodeSnafu)?;
-        let frontend_client =
-            FrontendClient::from_meta_client(meta_client.clone(), flow_auth_header);
+        let frontend_client = FrontendClient::from_meta_client(
+            meta_client.clone(),
+            flow_auth_header,
+            opts.query.clone(),
+            opts.flow.batching_mode.clone(),
+        );
         let frontend_client = Arc::new(frontend_client);
         let flownode_builder = FlownodeBuilder::new(
             opts.clone(),

src/cmd/src/frontend.rs

@@ -19,7 +19,7 @@ use std::time::Duration;
 use async_trait::async_trait;
 use cache::{build_fundamental_cache_registry, with_default_composite_cache_registry};
 use catalog::information_extension::DistributedInformationExtension;
-use catalog::kvbackend::{CachedKvBackendBuilder, KvBackendCatalogManager, MetaKvBackend};
+use catalog::kvbackend::{CachedKvBackendBuilder, KvBackendCatalogManagerBuilder, MetaKvBackend};
 use catalog::process_manager::ProcessManager;
 use clap::Parser;
 use client::client_manager::NodeClients;
@@ -33,7 +33,7 @@ use common_meta::heartbeat::handler::HandlerGroupExecutor;
 use common_telemetry::info;
 use common_telemetry::logging::{TracingOptions, DEFAULT_LOGGING_DIR};
 use common_time::timezone::set_default_timezone;
-use common_version::{short_version, version};
+use common_version::{short_version, verbose_version};
 use frontend::frontend::Frontend;
 use frontend::heartbeat::HeartbeatTask;
 use frontend::instance::builder::FrontendBuilder;
@@ -47,7 +47,7 @@ use tracing_appender::non_blocking::WorkerGuard;
 
 use crate::error::{self, Result};
 use crate::options::{GlobalOptions, GreptimeOptions};
-use crate::{create_resource_limit_metrics, log_versions, App};
+use crate::{create_resource_limit_metrics, log_versions, maybe_activate_heap_profile, App};
 
 type FrontendOptions = GreptimeOptions<frontend::frontend::FrontendOptions>;
 
@@ -102,7 +102,7 @@ impl App for Instance {
 #[derive(Parser)]
 pub struct Command {
     #[clap(subcommand)]
-    subcmd: SubCommand,
+    pub subcmd: SubCommand,
 }
 
 impl Command {
@@ -116,7 +116,7 @@ impl Command {
 }
 
 #[derive(Parser)]
-enum SubCommand {
+pub enum SubCommand {
     Start(StartCommand),
 }
 
@@ -153,7 +153,7 @@ pub struct StartCommand {
     #[clap(long)]
     postgres_addr: Option<String>,
     #[clap(short, long)]
-    config_file: Option<String>,
+    pub config_file: Option<String>,
     #[clap(short, long)]
     influxdb_enable: Option<bool>,
     #[clap(long, value_delimiter = ',', num_args = 1..)]
@@ -169,7 +169,7 @@ pub struct StartCommand {
     #[clap(long)]
     disable_dashboard: Option<bool>,
     #[clap(long, default_value = "GREPTIMEDB_FRONTEND")]
-    env_prefix: String,
+    pub env_prefix: String,
 }
 
 impl StartCommand {
@@ -282,7 +282,8 @@ impl StartCommand {
             opts.component.slow_query.as_ref(),
         );
 
-        log_versions(version(), short_version(), APP_NAME);
+        log_versions(verbose_version(), short_version(), APP_NAME);
+        maybe_activate_heap_profile(&opts.component.memory);
         create_resource_limit_metrics(APP_NAME);
 
         info!("Frontend start command: {:#?}", self);
@@ -350,13 +351,20 @@ impl StartCommand {
             addrs::resolve_addr(&opts.grpc.bind_addr, Some(&opts.grpc.server_addr)),
             Some(meta_client.clone()),
         ));
-        let catalog_manager = KvBackendCatalogManager::new(
+
+        let builder = KvBackendCatalogManagerBuilder::new(
             information_extension,
             cached_meta_backend.clone(),
             layered_cache_registry.clone(),
-            None,
-            Some(process_manager.clone()),
-        );
+        )
+        .with_process_manager(process_manager.clone());
+        #[cfg(feature = "enterprise")]
+        let builder = if let Some(factories) = plugins.get() {
+            builder.with_extra_information_table_factories(factories)
+        } else {
+            builder
+        };
+        let catalog_manager = builder.build();
 
         let executor = HandlerGroupExecutor::new(vec![
             Arc::new(ParseMailboxMessageHandler),

src/cmd/src/lib.rs

@@ -15,7 +15,10 @@
 #![feature(assert_matches, let_chains)]
 
 use async_trait::async_trait;
-use common_telemetry::{error, info};
+use common_error::ext::ErrorExt;
+use common_error::status_code::StatusCode;
+use common_mem_prof::activate_heap_profile;
+use common_telemetry::{error, info, warn};
 use stat::{get_cpu_limit, get_memory_limit};
 
 use crate::error::Result;
@@ -112,7 +115,7 @@ pub trait App: Send {
 pub fn log_versions(version: &str, short_version: &str, app: &str) {
     // Report app version as gauge.
     APP_VERSION
-        .with_label_values(&[env!("CARGO_PKG_VERSION"), short_version, app])
+        .with_label_values(&[common_version::version(), short_version, app])
         .inc();
 
     // Log version and argument flags.
@@ -145,3 +148,20 @@ fn log_env_flags() {
         info!("argument: {}", argument);
     }
 }
+
+pub fn maybe_activate_heap_profile(memory_options: &common_options::memory::MemoryOptions) {
+    if memory_options.enable_heap_profiling {
+        match activate_heap_profile() {
+            Ok(()) => {
+                info!("Heap profile is active");
+            }
+            Err(err) => {
+                if err.status_code() == StatusCode::Unsupported {
+                    info!("Heap profile is not supported");
+                } else {
+                    warn!(err; "Failed to activate heap profile");
+                }
+            }
+        }
+    }
+}

src/cmd/src/metasrv.rs

@@ -22,7 +22,7 @@ use common_base::Plugins;
 use common_config::Configurable;
 use common_telemetry::info;
 use common_telemetry::logging::{TracingOptions, DEFAULT_LOGGING_DIR};
-use common_version::{short_version, version};
+use common_version::{short_version, verbose_version};
 use meta_srv::bootstrap::MetasrvInstance;
 use meta_srv::metasrv::BackendImpl;
 use snafu::ResultExt;
@@ -30,7 +30,7 @@ use tracing_appender::non_blocking::WorkerGuard;
 
 use crate::error::{self, LoadLayeredConfigSnafu, Result, StartMetaServerSnafu};
 use crate::options::{GlobalOptions, GreptimeOptions};
-use crate::{create_resource_limit_metrics, log_versions, App};
+use crate::{create_resource_limit_metrics, log_versions, maybe_activate_heap_profile, App};
 
 type MetasrvOptions = GreptimeOptions<meta_srv::metasrv::MetasrvOptions>;
 
@@ -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]
@@ -320,7 +324,8 @@ impl StartCommand {
             None,
         );
 
-        log_versions(version(), short_version(), APP_NAME);
+        log_versions(verbose_version(), short_version(), APP_NAME);
+        maybe_activate_heap_profile(&opts.component.memory);
         create_resource_limit_metrics(APP_NAME);
 
         info!("Metasrv start command: {:#?}", self);
@@ -336,12 +341,12 @@ impl StartCommand {
             .await
             .context(StartMetaServerSnafu)?;
 
-        let builder = meta_srv::bootstrap::metasrv_builder(&opts, plugins.clone(), None)
+        let builder = meta_srv::bootstrap::metasrv_builder(&opts, plugins, None)
             .await
             .context(error::BuildMetaServerSnafu)?;
         let metasrv = builder.build().await.context(error::BuildMetaServerSnafu)?;
 
-        let instance = MetasrvInstance::new(opts, plugins, metasrv)
+        let instance = MetasrvInstance::new(metasrv)
             .await
             .context(error::BuildMetaServerSnafu)?;
 

src/cmd/src/standalone.rs

@@ -20,7 +20,7 @@ use std::{fs, path};
 use async_trait::async_trait;
 use cache::{build_fundamental_cache_registry, with_default_composite_cache_registry};
 use catalog::information_schema::InformationExtension;
-use catalog::kvbackend::KvBackendCatalogManager;
+use catalog::kvbackend::KvBackendCatalogManagerBuilder;
 use catalog::process_manager::ProcessManager;
 use clap::Parser;
 use client::api::v1::meta::RegionRole;
@@ -30,32 +30,30 @@ use common_catalog::consts::{MIN_USER_FLOW_ID, MIN_USER_TABLE_ID};
 use common_config::{metadata_store_dir, Configurable, KvBackendConfig};
 use common_error::ext::BoxedError;
 use common_meta::cache::LayeredCacheRegistryBuilder;
-use common_meta::cache_invalidator::CacheInvalidatorRef;
 use common_meta::cluster::{NodeInfo, NodeStatus};
 use common_meta::datanode::RegionStat;
-use common_meta::ddl::flow_meta::{FlowMetadataAllocator, FlowMetadataAllocatorRef};
-use common_meta::ddl::table_meta::{TableMetadataAllocator, TableMetadataAllocatorRef};
-use common_meta::ddl::{DdlContext, NoopRegionFailureDetectorControl, ProcedureExecutorRef};
+use common_meta::ddl::flow_meta::FlowMetadataAllocator;
+use common_meta::ddl::table_meta::TableMetadataAllocator;
+use common_meta::ddl::{DdlContext, NoopRegionFailureDetectorControl};
 use common_meta::ddl_manager::DdlManager;
-#[cfg(feature = "enterprise")]
-use common_meta::ddl_manager::TriggerDdlManagerRef;
 use common_meta::key::flow::flow_state::FlowStat;
-use common_meta::key::flow::{FlowMetadataManager, FlowMetadataManagerRef};
+use common_meta::key::flow::FlowMetadataManager;
 use common_meta::key::{TableMetadataManager, TableMetadataManagerRef};
 use common_meta::kv_backend::KvBackendRef;
-use common_meta::node_manager::NodeManagerRef;
 use common_meta::peer::Peer;
+use common_meta::procedure_executor::LocalProcedureExecutor;
 use common_meta::region_keeper::MemoryRegionKeeper;
 use common_meta::region_registry::LeaderRegionRegistry;
 use common_meta::sequence::SequenceBuilder;
 use common_meta::wal_options_allocator::{build_wal_options_allocator, WalOptionsAllocatorRef};
+use common_options::memory::MemoryOptions;
 use common_procedure::{ProcedureInfo, ProcedureManagerRef};
 use common_telemetry::info;
 use common_telemetry::logging::{
     LoggingOptions, SlowQueryOptions, TracingOptions, DEFAULT_LOGGING_DIR,
 };
 use common_time::timezone::set_default_timezone;
-use common_version::{short_version, version};
+use common_version::{short_version, verbose_version};
 use common_wal::config::DatanodeWalConfig;
 use datanode::config::{DatanodeOptions, ProcedureConfig, RegionEngineConfig, StorageConfig};
 use datanode::datanode::{Datanode, DatanodeBuilder};
@@ -87,7 +85,7 @@ use tracing_appender::non_blocking::WorkerGuard;
 
 use crate::error::{Result, StartFlownodeSnafu};
 use crate::options::{GlobalOptions, GreptimeOptions};
-use crate::{create_resource_limit_metrics, error, log_versions, App};
+use crate::{create_resource_limit_metrics, error, log_versions, maybe_activate_heap_profile, App};
 
 pub const APP_NAME: &str = "greptime-standalone";
 
@@ -161,6 +159,7 @@ pub struct StandaloneOptions {
     pub max_in_flight_write_bytes: Option<ReadableSize>,
     pub slow_query: Option<SlowQueryOptions>,
     pub query: QueryOptions,
+    pub memory: MemoryOptions,
 }
 
 impl Default for StandaloneOptions {
@@ -194,6 +193,7 @@ impl Default for StandaloneOptions {
             max_in_flight_write_bytes: None,
             slow_query: Some(SlowQueryOptions::default()),
             query: QueryOptions::default(),
+            memory: MemoryOptions::default(),
         }
     }
 }
@@ -261,15 +261,34 @@ pub struct Instance {
     flownode: FlownodeInstance,
     procedure_manager: ProcedureManagerRef,
     wal_options_allocator: WalOptionsAllocatorRef,
+
+    // The components of standalone, which make it easier to expand based
+    // on the components.
+    #[cfg(feature = "enterprise")]
+    components: Components,
+
     // Keep the logging guard to prevent the worker from being dropped.
     _guard: Vec<WorkerGuard>,
 }
 
+#[cfg(feature = "enterprise")]
+pub struct Components {
+    pub plugins: Plugins,
+    pub kv_backend: KvBackendRef,
+    pub frontend_client: Arc<FrontendClient>,
+    pub catalog_manager: catalog::CatalogManagerRef,
+}
+
 impl Instance {
     /// Find the socket addr of a server by its `name`.
     pub fn server_addr(&self, name: &str) -> Option<SocketAddr> {
         self.frontend.server_handlers().addr(name)
     }
+
+    #[cfg(feature = "enterprise")]
+    pub fn components(&self) -> &Components {
+        &self.components
+    }
 }
 
 #[async_trait]
@@ -470,7 +489,8 @@ impl StartCommand {
             opts.component.slow_query.as_ref(),
         );
 
-        log_versions(version(), short_version(), APP_NAME);
+        log_versions(verbose_version(), short_version(), APP_NAME);
+        maybe_activate_heap_profile(&opts.component.memory);
         create_resource_limit_metrics(APP_NAME);
 
         info!("Standalone start command: {:#?}", self);
@@ -529,13 +549,20 @@ impl StartCommand {
         ));
 
         let process_manager = Arc::new(ProcessManager::new(opts.grpc.server_addr.clone(), None));
-        let catalog_manager = KvBackendCatalogManager::new(
+        let builder = KvBackendCatalogManagerBuilder::new(
             information_extension.clone(),
             kv_backend.clone(),
             layered_cache_registry.clone(),
-            Some(procedure_manager.clone()),
-            Some(process_manager.clone()),
-        );
+        )
+        .with_procedure_manager(procedure_manager.clone())
+        .with_process_manager(process_manager.clone());
+        #[cfg(feature = "enterprise")]
+        let builder = if let Some(factories) = plugins.get() {
+            builder.with_extra_information_table_factories(factories)
+        } else {
+            builder
+        };
+        let catalog_manager = builder.build();
 
         let table_metadata_manager =
             Self::create_table_metadata_manager(kv_backend.clone()).await?;
@@ -549,14 +576,15 @@ impl StartCommand {
         // for standalone not use grpc, but get a handler to frontend grpc client without
         // actually make a connection
         let (frontend_client, frontend_instance_handler) =
-            FrontendClient::from_empty_grpc_handler();
+            FrontendClient::from_empty_grpc_handler(opts.query.clone());
+        let frontend_client = Arc::new(frontend_client);
         let flow_builder = FlownodeBuilder::new(
             flownode_options,
             plugins.clone(),
             table_metadata_manager.clone(),
             catalog_manager.clone(),
             flow_metadata_manager.clone(),
-            Arc::new(frontend_client.clone()),
+            frontend_client.clone(),
         );
         let flownode = flow_builder
             .build()
@@ -594,28 +622,39 @@ impl StartCommand {
             .await
             .context(error::BuildWalOptionsAllocatorSnafu)?;
         let wal_options_allocator = Arc::new(wal_options_allocator);
-        let table_meta_allocator = Arc::new(TableMetadataAllocator::new(
+        let table_metadata_allocator = Arc::new(TableMetadataAllocator::new(
             table_id_sequence,
             wal_options_allocator.clone(),
         ));
-        let flow_meta_allocator = Arc::new(FlowMetadataAllocator::with_noop_peer_allocator(
+        let flow_metadata_allocator = Arc::new(FlowMetadataAllocator::with_noop_peer_allocator(
             flow_id_sequence,
         ));
 
+        let ddl_context = DdlContext {
+            node_manager: node_manager.clone(),
+            cache_invalidator: layered_cache_registry.clone(),
+            memory_region_keeper: Arc::new(MemoryRegionKeeper::default()),
+            leader_region_registry: Arc::new(LeaderRegionRegistry::default()),
+            table_metadata_manager: table_metadata_manager.clone(),
+            table_metadata_allocator: table_metadata_allocator.clone(),
+            flow_metadata_manager: flow_metadata_manager.clone(),
+            flow_metadata_allocator: flow_metadata_allocator.clone(),
+            region_failure_detector_controller: Arc::new(NoopRegionFailureDetectorControl),
+        };
+
+        let ddl_manager = DdlManager::try_new(ddl_context, procedure_manager.clone(), true)
+            .context(error::InitDdlManagerSnafu)?;
         #[cfg(feature = "enterprise")]
-        let trigger_ddl_manager: Option<TriggerDdlManagerRef> = plugins.get();
-        let ddl_task_executor = Self::create_ddl_task_executor(
+        let ddl_manager = {
+            let trigger_ddl_manager: Option<common_meta::ddl_manager::TriggerDdlManagerRef> =
+                plugins.get();
+            ddl_manager.with_trigger_ddl_manager(trigger_ddl_manager)
+        };
+
+        let procedure_executor = Arc::new(LocalProcedureExecutor::new(
+            Arc::new(ddl_manager),
             procedure_manager.clone(),
-            node_manager.clone(),
-            layered_cache_registry.clone(),
-            table_metadata_manager,
-            table_meta_allocator,
-            flow_metadata_manager,
-            flow_meta_allocator,
-            #[cfg(feature = "enterprise")]
-            trigger_ddl_manager,
-        )
-        .await?;
+        ));
 
         let fe_instance = FrontendBuilder::new(
             fe_opts.clone(),
@@ -623,7 +662,7 @@ impl StartCommand {
             layered_cache_registry.clone(),
             catalog_manager.clone(),
             node_manager.clone(),
-            ddl_task_executor.clone(),
+            procedure_executor.clone(),
             process_manager,
         )
         .with_plugin(plugins.clone())
@@ -648,7 +687,7 @@ impl StartCommand {
             catalog_manager.clone(),
             kv_backend.clone(),
             layered_cache_registry.clone(),
-            ddl_task_executor.clone(),
+            procedure_executor,
             node_manager,
         )
         .await
@@ -658,7 +697,7 @@ impl StartCommand {
         let export_metrics_task = ExportMetricsTask::try_new(&opts.export_metrics, Some(&plugins))
             .context(error::ServersSnafu)?;
 
-        let servers = Services::new(opts, fe_instance.clone(), plugins)
+        let servers = Services::new(opts, fe_instance.clone(), plugins.clone())
             .build()
             .context(error::StartFrontendSnafu)?;
 
@@ -669,51 +708,26 @@ impl StartCommand {
             export_metrics_task,
         };
 
+        #[cfg(feature = "enterprise")]
+        let components = Components {
+            plugins,
+            kv_backend,
+            frontend_client,
+            catalog_manager,
+        };
+
         Ok(Instance {
             datanode,
             frontend,
             flownode,
             procedure_manager,
             wal_options_allocator,
+            #[cfg(feature = "enterprise")]
+            components,
             _guard: guard,
         })
     }
 
-    #[allow(clippy::too_many_arguments)]
-    pub async fn create_ddl_task_executor(
-        procedure_manager: ProcedureManagerRef,
-        node_manager: NodeManagerRef,
-        cache_invalidator: CacheInvalidatorRef,
-        table_metadata_manager: TableMetadataManagerRef,
-        table_metadata_allocator: TableMetadataAllocatorRef,
-        flow_metadata_manager: FlowMetadataManagerRef,
-        flow_metadata_allocator: FlowMetadataAllocatorRef,
-        #[cfg(feature = "enterprise")] trigger_ddl_manager: Option<TriggerDdlManagerRef>,
-    ) -> Result<ProcedureExecutorRef> {
-        let procedure_executor: ProcedureExecutorRef = Arc::new(
-            DdlManager::try_new(
-                DdlContext {
-                    node_manager,
-                    cache_invalidator,
-                    memory_region_keeper: Arc::new(MemoryRegionKeeper::default()),
-                    leader_region_registry: Arc::new(LeaderRegionRegistry::default()),
-                    table_metadata_manager,
-                    table_metadata_allocator,
-                    flow_metadata_manager,
-                    flow_metadata_allocator,
-                    region_failure_detector_controller: Arc::new(NoopRegionFailureDetectorControl),
-                },
-                procedure_manager,
-                true,
-                #[cfg(feature = "enterprise")]
-                trigger_ddl_manager,
-            )
-            .context(error::InitDdlManagerSnafu)?,
-        );
-
-        Ok(procedure_executor)
-    }
-
     pub async fn create_table_metadata_manager(
         kv_backend: KvBackendRef,
     ) -> Result<TableMetadataManagerRef> {
@@ -815,6 +829,7 @@ impl InformationExtension for StandaloneInformationExtension {
                     memtable_size: region_stat.memtable_size,
                     manifest_size: region_stat.manifest_size,
                     sst_size: region_stat.sst_size,
+                    sst_num: region_stat.sst_num,
                     index_size: region_stat.index_size,
                     region_manifest: region_stat.manifest.into(),
                     data_topic_latest_entry_id: region_stat.data_topic_latest_entry_id,
@@ -849,7 +864,7 @@ mod tests {
     use common_config::ENV_VAR_SEP;
     use common_test_util::temp_dir::create_named_temp_file;
     use common_wal::config::DatanodeWalConfig;
-    use datanode::config::{FileConfig, GcsConfig};
+    use object_store::config::{FileConfig, GcsConfig};
 
     use super::*;
     use crate::options::GlobalOptions;
@@ -968,15 +983,15 @@ mod tests {
 
         assert!(matches!(
             &dn_opts.storage.store,
-            datanode::config::ObjectStoreConfig::File(FileConfig { .. })
+            object_store::config::ObjectStoreConfig::File(FileConfig { .. })
         ));
         assert_eq!(dn_opts.storage.providers.len(), 2);
         assert!(matches!(
             dn_opts.storage.providers[0],
-            datanode::config::ObjectStoreConfig::Gcs(GcsConfig { .. })
+            object_store::config::ObjectStoreConfig::Gcs(GcsConfig { .. })
         ));
         match &dn_opts.storage.providers[1] {
-            datanode::config::ObjectStoreConfig::S3(s3_config) => {
+            object_store::config::ObjectStoreConfig::S3(s3_config) => {
                 assert_eq!(
                     "SecretBox<alloc::string::String>([REDACTED])".to_string(),
                     format!("{:?}", s3_config.access_key_id)

src/cmd/tests/load_config_test.rs

@@ -18,20 +18,23 @@ use cmd::options::GreptimeOptions;
 use cmd::standalone::StandaloneOptions;
 use common_config::{Configurable, DEFAULT_DATA_HOME};
 use common_options::datanode::{ClientOptions, DatanodeClientOptions};
-use common_telemetry::logging::{LoggingOptions, DEFAULT_LOGGING_DIR, DEFAULT_OTLP_ENDPOINT};
+use common_telemetry::logging::{LoggingOptions, DEFAULT_LOGGING_DIR, DEFAULT_OTLP_HTTP_ENDPOINT};
 use common_wal::config::raft_engine::RaftEngineConfig;
 use common_wal::config::DatanodeWalConfig;
 use datanode::config::{DatanodeOptions, RegionEngineConfig, StorageConfig};
 use file_engine::config::EngineConfig as FileEngineConfig;
+use flow::FlownodeOptions;
 use frontend::frontend::FrontendOptions;
 use meta_client::MetaClientOptions;
 use meta_srv::metasrv::MetasrvOptions;
 use meta_srv::selector::SelectorType;
 use metric_engine::config::EngineConfig as MetricEngineConfig;
 use mito2::config::MitoConfig;
+use query::options::QueryOptions;
 use servers::export_metrics::ExportMetricsOption;
 use servers::grpc::GrpcOptions;
 use servers::http::HttpOptions;
+use servers::tls::{TlsMode, TlsOption};
 use store_api::path_utils::WAL_DIR;
 
 #[allow(deprecated)]
@@ -81,7 +84,7 @@ fn test_load_datanode_example_config() {
             logging: LoggingOptions {
                 level: Some("info".to_string()),
                 dir: format!("{}/{}", DEFAULT_DATA_HOME, DEFAULT_LOGGING_DIR),
-                otlp_endpoint: Some(DEFAULT_OTLP_ENDPOINT.to_string()),
+                otlp_endpoint: Some(DEFAULT_OTLP_HTTP_ENDPOINT.to_string()),
                 tracing_sample_ratio: Some(Default::default()),
                 ..Default::default()
             },
@@ -124,7 +127,7 @@ fn test_load_frontend_example_config() {
             logging: LoggingOptions {
                 level: Some("info".to_string()),
                 dir: format!("{}/{}", DEFAULT_DATA_HOME, DEFAULT_LOGGING_DIR),
-                otlp_endpoint: Some(DEFAULT_OTLP_ENDPOINT.to_string()),
+                otlp_endpoint: Some(DEFAULT_OTLP_HTTP_ENDPOINT.to_string()),
                 tracing_sample_ratio: Some(Default::default()),
                 ..Default::default()
             },
@@ -172,7 +175,7 @@ fn test_load_metasrv_example_config() {
             logging: LoggingOptions {
                 dir: format!("{}/{}", DEFAULT_DATA_HOME, DEFAULT_LOGGING_DIR),
                 level: Some("info".to_string()),
-                otlp_endpoint: Some(DEFAULT_OTLP_ENDPOINT.to_string()),
+                otlp_endpoint: Some(DEFAULT_OTLP_HTTP_ENDPOINT.to_string()),
                 tracing_sample_ratio: Some(Default::default()),
                 ..Default::default()
             },
@@ -188,6 +191,13 @@ fn test_load_metasrv_example_config() {
                 remote_write: Some(Default::default()),
                 ..Default::default()
             },
+            backend_tls: Some(TlsOption {
+                mode: TlsMode::Prefer,
+                cert_path: String::new(),
+                key_path: String::new(),
+                ca_cert_path: String::new(),
+                watch: false,
+            }),
             ..Default::default()
         },
         ..Default::default()
@@ -195,6 +205,61 @@ fn test_load_metasrv_example_config() {
     similar_asserts::assert_eq!(options, expected);
 }
 
+#[test]
+fn test_load_flownode_example_config() {
+    let example_config = common_test_util::find_workspace_path("config/flownode.example.toml");
+    let options =
+        GreptimeOptions::<FlownodeOptions>::load_layered_options(example_config.to_str(), "")
+            .unwrap();
+    let expected = GreptimeOptions::<FlownodeOptions> {
+        component: FlownodeOptions {
+            node_id: Some(14),
+            flow: Default::default(),
+            grpc: GrpcOptions {
+                bind_addr: "127.0.0.1:6800".to_string(),
+                server_addr: "127.0.0.1:6800".to_string(),
+                runtime_size: 2,
+                ..Default::default()
+            },
+            logging: LoggingOptions {
+                dir: format!("{}/{}", DEFAULT_DATA_HOME, DEFAULT_LOGGING_DIR),
+                level: Some("info".to_string()),
+                otlp_endpoint: Some(DEFAULT_OTLP_HTTP_ENDPOINT.to_string()),
+                otlp_export_protocol: Some(common_telemetry::logging::OtlpExportProtocol::Http),
+                tracing_sample_ratio: Some(Default::default()),
+                ..Default::default()
+            },
+            tracing: Default::default(),
+            heartbeat: Default::default(),
+            // flownode deliberately use a slower query parallelism
+            // to avoid overwhelming the frontend with too many queries
+            query: QueryOptions {
+                parallelism: 1,
+                allow_query_fallback: false,
+            },
+            meta_client: Some(MetaClientOptions {
+                metasrv_addrs: vec!["127.0.0.1:3002".to_string()],
+                timeout: Duration::from_secs(3),
+                heartbeat_timeout: Duration::from_millis(500),
+                ddl_timeout: Duration::from_secs(10),
+                connect_timeout: Duration::from_secs(1),
+                tcp_nodelay: true,
+                metadata_cache_max_capacity: 100000,
+                metadata_cache_ttl: Duration::from_secs(600),
+                metadata_cache_tti: Duration::from_secs(300),
+            }),
+            http: HttpOptions {
+                addr: "127.0.0.1:4000".to_string(),
+                ..Default::default()
+            },
+            user_provider: None,
+            memory: Default::default(),
+        },
+        ..Default::default()
+    };
+    similar_asserts::assert_eq!(options, expected);
+}
+
 #[test]
 fn test_load_standalone_example_config() {
     let example_config = common_test_util::find_workspace_path("config/standalone.example.toml");
@@ -229,7 +294,7 @@ fn test_load_standalone_example_config() {
             logging: LoggingOptions {
                 level: Some("info".to_string()),
                 dir: format!("{}/{}", DEFAULT_DATA_HOME, DEFAULT_LOGGING_DIR),
-                otlp_endpoint: Some(DEFAULT_OTLP_ENDPOINT.to_string()),
+                otlp_endpoint: Some(DEFAULT_OTLP_HTTP_ENDPOINT.to_string()),
                 tracing_sample_ratio: Some(Default::default()),
                 ..Default::default()
             },
@@ -242,6 +307,7 @@ fn test_load_standalone_example_config() {
                 cors_allowed_origins: vec!["https://example.com".to_string()],
                 ..Default::default()
             },
+
             ..Default::default()
         },
         ..Default::default()

src/common/catalog/src/consts.rs

@@ -78,7 +78,7 @@ pub const INFORMATION_SCHEMA_ROUTINES_TABLE_ID: u32 = 21;
 pub const INFORMATION_SCHEMA_SCHEMA_PRIVILEGES_TABLE_ID: u32 = 22;
 /// id for information_schema.TABLE_PRIVILEGES
 pub const INFORMATION_SCHEMA_TABLE_PRIVILEGES_TABLE_ID: u32 = 23;
-/// id for information_schema.TRIGGERS
+/// id for information_schema.TRIGGERS (for mysql)
 pub const INFORMATION_SCHEMA_TRIGGERS_TABLE_ID: u32 = 24;
 /// id for information_schema.GLOBAL_STATUS
 pub const INFORMATION_SCHEMA_GLOBAL_STATUS_TABLE_ID: u32 = 25;

src/common/config/Cargo.toml

@@ -14,6 +14,7 @@ common-macro.workspace = true
 config.workspace = true
 humantime-serde.workspace = true
 num_cpus.workspace = true
+object-store.workspace = true
 serde.workspace = true
 serde_json.workspace = true
 serde_with.workspace = true

src/common/config/src/config.rs

@@ -106,7 +106,7 @@ mod tests {
     use common_telemetry::logging::LoggingOptions;
     use common_test_util::temp_dir::create_named_temp_file;
     use common_wal::config::DatanodeWalConfig;
-    use datanode::config::{ObjectStoreConfig, StorageConfig};
+    use datanode::config::StorageConfig;
     use meta_client::MetaClientOptions;
     use serde::{Deserialize, Serialize};
 
@@ -212,7 +212,7 @@ mod tests {
 
                 // Check the configs from environment variables.
                 match &opts.storage.store {
-                    ObjectStoreConfig::S3(s3_config) => {
+                    object_store::config::ObjectStoreConfig::S3(s3_config) => {
                         assert_eq!(s3_config.bucket, "mybucket".to_string());
                     }
                     _ => panic!("unexpected store type"),

src/common/datasource/src/lib.rs

@@ -21,6 +21,7 @@ pub mod error;
 pub mod file_format;
 pub mod lister;
 pub mod object_store;
+pub mod parquet_writer;
 pub mod share_buffer;
 #[cfg(test)]
 pub mod test_util;

src/common/datasource/src/object_store/oss.rs

@@ -77,6 +77,11 @@ pub fn build_oss_backend(
 
     let op = ObjectStore::new(builder)
         .context(error::BuildBackendSnafu)?
+        .layer(
+            object_store::layers::RetryLayer::new()
+                .with_jitter()
+                .with_notify(object_store::util::PrintDetailedError),
+        )
         .layer(object_store::layers::LoggingLayer::default())
         .layer(object_store::layers::TracingLayer)
         .layer(object_store::layers::build_prometheus_metrics_layer(true))

src/common/datasource/src/object_store/s3.rs

@@ -85,6 +85,11 @@ pub fn build_s3_backend(
     // TODO(weny): Consider finding a better way to eliminate duplicate code.
     Ok(ObjectStore::new(builder)
         .context(error::BuildBackendSnafu)?
+        .layer(
+            object_store::layers::RetryLayer::new()
+                .with_jitter()
+                .with_notify(object_store::util::PrintDetailedError),
+        )
         .layer(object_store::layers::LoggingLayer::new(
             DefaultLoggingInterceptor,
         ))

src/common/datasource/src/parquet_writer.rs

@@ -0,0 +1,52 @@
+// 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 bytes::Bytes;
+use futures::future::BoxFuture;
+use object_store::Writer;
+use parquet::arrow::async_writer::AsyncFileWriter;
+use parquet::errors::ParquetError;
+
+/// Bridges opendal [Writer] with parquet [AsyncFileWriter].
+pub struct AsyncWriter {
+    inner: Writer,
+}
+
+impl AsyncWriter {
+    /// Create a [`AsyncWriter`] by given [`Writer`].
+    pub fn new(writer: Writer) -> Self {
+        Self { inner: writer }
+    }
+}
+
+impl AsyncFileWriter for AsyncWriter {
+    fn write(&mut self, bs: Bytes) -> BoxFuture<'_, parquet::errors::Result<()>> {
+        Box::pin(async move {
+            self.inner
+                .write(bs)
+                .await
+                .map_err(|err| ParquetError::External(Box::new(err)))
+        })
+    }
+
+    fn complete(&mut self) -> BoxFuture<'_, parquet::errors::Result<()>> {
+        Box::pin(async move {
+            self.inner
+                .close()
+                .await
+                .map(|_| ())
+                .map_err(|err| ParquetError::External(Box::new(err)))
+        })
+    }
+}

src/common/error/src/status_code.rs

@@ -119,6 +119,11 @@ pub enum StatusCode {
     FlowAlreadyExists = 8000,
     FlowNotFound = 8001,
     // ====== End of flow related status code =====
+
+    // ====== Begin of trigger related status code =====
+    TriggerAlreadyExists = 9000,
+    TriggerNotFound = 9001,
+    // ====== End of trigger related status code =====
 }
 
 impl StatusCode {
@@ -155,6 +160,8 @@ impl StatusCode {
             | StatusCode::RegionNotFound
             | StatusCode::FlowAlreadyExists
             | StatusCode::FlowNotFound
+            | StatusCode::TriggerAlreadyExists
+            | StatusCode::TriggerNotFound
             | StatusCode::RegionReadonly
             | StatusCode::TableColumnNotFound
             | StatusCode::TableColumnExists
@@ -198,6 +205,8 @@ impl StatusCode {
             | StatusCode::PlanQuery
             | StatusCode::FlowAlreadyExists
             | StatusCode::FlowNotFound
+            | StatusCode::TriggerAlreadyExists
+            | StatusCode::TriggerNotFound
             | StatusCode::RegionNotReady
             | StatusCode::RegionBusy
             | StatusCode::RegionReadonly
@@ -230,6 +239,48 @@ impl fmt::Display for StatusCode {
     }
 }
 
+#[macro_export]
+macro_rules! define_from_tonic_status {
+    ($Error: ty, $Variant: ident) => {
+        impl From<tonic::Status> for $Error {
+            fn from(e: tonic::Status) -> Self {
+                use snafu::location;
+
+                fn metadata_value(e: &tonic::Status, key: &str) -> Option<String> {
+                    e.metadata()
+                        .get(key)
+                        .and_then(|v| String::from_utf8(v.as_bytes().to_vec()).ok())
+                }
+
+                let code = metadata_value(&e, $crate::GREPTIME_DB_HEADER_ERROR_CODE)
+                    .and_then(|s| {
+                        if let Ok(code) = s.parse::<u32>() {
+                            StatusCode::from_u32(code)
+                        } else {
+                            None
+                        }
+                    })
+                    .unwrap_or_else(|| match e.code() {
+                        tonic::Code::Cancelled => StatusCode::Cancelled,
+                        tonic::Code::DeadlineExceeded => StatusCode::DeadlineExceeded,
+                        _ => StatusCode::Internal,
+                    });
+
+                let msg = metadata_value(&e, $crate::GREPTIME_DB_HEADER_ERROR_MSG)
+                    .unwrap_or_else(|| e.message().to_string());
+
+                // TODO(LFC): Make the error variant defined automatically.
+                Self::$Variant {
+                    code,
+                    msg,
+                    tonic_code: e.code(),
+                    location: location!(),
+                }
+            }
+        }
+    };
+}
+
 #[macro_export]
 macro_rules! define_into_tonic_status {
     ($Error: ty) => {
@@ -281,12 +332,14 @@ pub fn status_to_tonic_code(status_code: StatusCode) -> Code {
         | StatusCode::TableColumnExists
         | StatusCode::RegionAlreadyExists
         | StatusCode::DatabaseAlreadyExists
+        | StatusCode::TriggerAlreadyExists
         | StatusCode::FlowAlreadyExists => Code::AlreadyExists,
         StatusCode::TableNotFound
         | StatusCode::RegionNotFound
         | StatusCode::TableColumnNotFound
         | StatusCode::DatabaseNotFound
         | StatusCode::UserNotFound
+        | StatusCode::TriggerNotFound
         | StatusCode::FlowNotFound => Code::NotFound,
         StatusCode::TableUnavailable
         | StatusCode::StorageUnavailable
@@ -304,15 +357,6 @@ pub fn status_to_tonic_code(status_code: StatusCode) -> Code {
     }
 }
 
-/// Converts tonic [Code] to [StatusCode].
-pub fn convert_tonic_code_to_status_code(code: Code) -> StatusCode {
-    match code {
-        Code::Cancelled => StatusCode::Cancelled,
-        Code::DeadlineExceeded => StatusCode::DeadlineExceeded,
-        _ => StatusCode::Internal,
-    }
-}
-
 #[cfg(test)]
 mod tests {
     use strum::IntoEnumIterator;

src/common/error/tests/ext.rs

@@ -84,24 +84,33 @@ fn test_to_string() {
     assert_eq!(result.unwrap_err().to_string(), "<root cause>");
 }
 
+fn normalize_path(s: &str) -> String {
+    s.replace('\\', "/")
+}
+
 #[test]
 fn test_debug_format() {
     let result = normal_error();
     let debug_output = format!("{:?}", result.unwrap_err());
-    let normalized_output = debug_output.replace('\\', "/");
+
     assert_eq!(
-        normalized_output,
-        r#"0: A normal error with "display" attribute, message "blabla", at src/common/error/tests/ext.rs:55:22
-1: PlainError { msg: "<root cause>", status_code: Unexpected }"#
+        normalize_path(&debug_output),
+        format!(
+            r#"0: A normal error with "display" attribute, message "blabla", at {}:55:22
+1: PlainError {{ msg: "<root cause>", status_code: Unexpected }}"#,
+            normalize_path(file!())
+        )
     );
 
     let result = transparent_error();
     let debug_output = format!("{:?}", result.unwrap_err());
-    let normalized_output = debug_output.replace('\\', "/");
     assert_eq!(
-        normalized_output,
-        r#"0: <transparent>, at src/common/error/tests/ext.rs:60:5
-1: PlainError { msg: "<root cause>", status_code: Unexpected }"#
+        normalize_path(&debug_output),
+        format!(
+            r#"0: <transparent>, at {}:60:5
+1: PlainError {{ msg: "<root cause>", status_code: Unexpected }}"#,
+            normalize_path(file!())
+        )
     );
 }
 

src/common/event-recorder/Cargo.toml

@@ -0,0 +1,25 @@
+[package]
+name = "common-event-recorder"
+version.workspace = true
+edition.workspace = true
+license.workspace = true
+
+[dependencies]
+api.workspace = true
+async-trait.workspace = true
+backon.workspace = true
+client.workspace = true
+common-error.workspace = true
+common-macro.workspace = true
+common-meta.workspace = true
+common-telemetry.workspace = true
+common-time.workspace = true
+serde.workspace = true
+serde_json.workspace = true
+snafu.workspace = true
+store-api.workspace = true
+tokio.workspace = true
+tokio-util.workspace = true
+
+[lints]
+workspace = true

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

@@ -0,0 +1,53 @@
+// 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 api::v1::ColumnSchema;
+use common_error::ext::ErrorExt;
+use common_error::status_code::StatusCode;
+use common_macro::stack_trace_debug;
+use snafu::{Location, Snafu};
+
+#[derive(Snafu)]
+#[snafu(visibility(pub))]
+#[stack_trace_debug]
+pub enum Error {
+    #[snafu(display("No available frontend"))]
+    NoAvailableFrontend {
+        #[snafu(implicit)]
+        location: Location,
+    },
+
+    #[snafu(display("Mismatched schema, expected: {:?}, actual: {:?}", expected, actual))]
+    MismatchedSchema {
+        #[snafu(implicit)]
+        location: Location,
+        expected: Vec<ColumnSchema>,
+        actual: Vec<ColumnSchema>,
+    },
+}
+
+pub type Result<T> = std::result::Result<T, Error>;
+
+impl ErrorExt for Error {
+    fn status_code(&self) -> StatusCode {
+        match self {
+            Error::MismatchedSchema { .. } => StatusCode::InvalidArguments,
+            Error::NoAvailableFrontend { .. } => StatusCode::Internal,
+        }
+    }
+
+    fn as_any(&self) -> &dyn std::any::Any {
+        self
+    }
+}

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

@@ -0,0 +1,18 @@
+// Copyright 2023 Greptime Team
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+pub mod error;
+pub mod recorder;
+
+pub use recorder::*;

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

@@ -0,0 +1,527 @@
+// Copyright 2023 Greptime Team
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+use std::any::Any;
+use std::collections::HashMap;
+use std::fmt::Debug;
+use std::sync::{Arc, OnceLock};
+use std::time::Duration;
+
+use api::v1::column_data_type_extension::TypeExt;
+use api::v1::value::ValueData;
+use api::v1::{
+    ColumnDataType, ColumnDataTypeExtension, ColumnSchema, JsonTypeExtension, Row,
+    RowInsertRequest, RowInsertRequests, Rows, SemanticType,
+};
+use async_trait::async_trait;
+use backon::{BackoffBuilder, ExponentialBuilder};
+use common_telemetry::{debug, error, info, warn};
+use common_time::timestamp::{TimeUnit, Timestamp};
+use serde::{Deserialize, Serialize};
+use store_api::mito_engine_options::{APPEND_MODE_KEY, TTL_KEY};
+use tokio::sync::mpsc::{channel, Receiver, Sender};
+use tokio::task::JoinHandle;
+use tokio::time::sleep;
+use tokio_util::sync::CancellationToken;
+
+use crate::error::{MismatchedSchemaSnafu, Result};
+
+/// The default table name for storing the events.
+pub const DEFAULT_EVENTS_TABLE_NAME: &str = "events";
+
+/// The column name for the event type.
+pub const EVENTS_TABLE_TYPE_COLUMN_NAME: &str = "type";
+/// The column name for the event payload.
+pub const EVENTS_TABLE_PAYLOAD_COLUMN_NAME: &str = "payload";
+/// The column name for the event timestamp.
+pub const EVENTS_TABLE_TIMESTAMP_COLUMN_NAME: &str = "timestamp";
+
+/// EventRecorderRef is the reference to the event recorder.
+pub type EventRecorderRef = Arc<dyn EventRecorder>;
+
+static EVENTS_TABLE_TTL: OnceLock<String> = OnceLock::new();
+
+/// The time interval for flushing batched events to the event handler.
+pub const DEFAULT_FLUSH_INTERVAL_SECONDS: Duration = Duration::from_secs(5);
+// The default TTL for the events table.
+const DEFAULT_EVENTS_TABLE_TTL: &str = "30d";
+// The capacity of the tokio channel for transmitting events to background processor.
+const DEFAULT_CHANNEL_SIZE: usize = 2048;
+// The size of the buffer for batching events before flushing to event handler.
+const DEFAULT_BUFFER_SIZE: usize = 100;
+// The maximum number of retry attempts when event handler processing fails.
+const DEFAULT_MAX_RETRY_TIMES: u64 = 3;
+
+/// Event trait defines the interface for events that can be recorded and persisted as the system table.
+/// By default, the event will be persisted as the system table with the following schema:
+///
+/// - `type`: the type of the event.
+/// - `payload`: the JSON bytes of the event.
+/// - `timestamp`: the timestamp of the event.
+///
+/// The event can also add the extra schema and row to the event by overriding the `extra_schema` and `extra_row` methods.
+pub trait Event: Send + Sync + Debug {
+    /// Returns the type of the event.
+    fn event_type(&self) -> &str;
+
+    /// Returns the timestamp of the event. Default to the current time.
+    fn timestamp(&self) -> Timestamp {
+        Timestamp::current_time(TimeUnit::Nanosecond)
+    }
+
+    /// Returns the JSON bytes of the event as the payload. It will use JSON type to store the payload.
+    fn json_payload(&self) -> Result<String>;
+
+    /// Add the extra schema to the event with the default schema.
+    fn extra_schema(&self) -> Vec<ColumnSchema> {
+        vec![]
+    }
+
+    /// Add the extra row to the event with the default row.
+    fn extra_row(&self) -> Result<Row> {
+        Ok(Row { values: vec![] })
+    }
+
+    /// Returns the event as any type.
+    fn as_any(&self) -> &dyn Any;
+}
+
+/// Returns the hints for the insert operation.
+pub fn insert_hints() -> Vec<(&'static str, &'static str)> {
+    vec![
+        (
+            TTL_KEY,
+            EVENTS_TABLE_TTL
+                .get()
+                .map(|s| s.as_str())
+                .unwrap_or(DEFAULT_EVENTS_TABLE_TTL),
+        ),
+        (APPEND_MODE_KEY, "true"),
+    ]
+}
+
+/// Builds the row inserts request for the events that will be persisted to the events table.
+pub fn build_row_inserts_request(events: &[Box<dyn Event>]) -> Result<RowInsertRequests> {
+    // Aggregate the events by the event type.
+    let mut event_groups: HashMap<&str, Vec<&Box<dyn Event>>> = HashMap::new();
+
+    for event in events {
+        event_groups
+            .entry(event.event_type())
+            .or_default()
+            .push(event);
+    }
+
+    let mut row_insert_requests = RowInsertRequests {
+        inserts: Vec::with_capacity(event_groups.len()),
+    };
+
+    for (_, events) in event_groups {
+        validate_events(&events)?;
+
+        // We already validated the events, so it's safe to get the first event to build the schema for the RowInsertRequest.
+        let event = &events[0];
+        let mut schema = vec![
+            ColumnSchema {
+                column_name: EVENTS_TABLE_TYPE_COLUMN_NAME.to_string(),
+                datatype: ColumnDataType::String.into(),
+                semantic_type: SemanticType::Tag.into(),
+                ..Default::default()
+            },
+            ColumnSchema {
+                column_name: EVENTS_TABLE_PAYLOAD_COLUMN_NAME.to_string(),
+                datatype: ColumnDataType::Binary as i32,
+                semantic_type: SemanticType::Field as i32,
+                datatype_extension: Some(ColumnDataTypeExtension {
+                    type_ext: Some(TypeExt::JsonType(JsonTypeExtension::JsonBinary.into())),
+                }),
+                ..Default::default()
+            },
+            ColumnSchema {
+                column_name: EVENTS_TABLE_TIMESTAMP_COLUMN_NAME.to_string(),
+                datatype: ColumnDataType::TimestampNanosecond.into(),
+                semantic_type: SemanticType::Timestamp.into(),
+                ..Default::default()
+            },
+        ];
+        schema.extend(event.extra_schema());
+
+        let rows = events
+            .iter()
+            .map(|event| {
+                let mut row = Row {
+                    values: vec![
+                        ValueData::StringValue(event.event_type().to_string()).into(),
+                        ValueData::BinaryValue(event.json_payload()?.as_bytes().to_vec()).into(),
+                        ValueData::TimestampNanosecondValue(event.timestamp().value()).into(),
+                    ],
+                };
+                row.values.extend(event.extra_row()?.values);
+                Ok(row)
+            })
+            .collect::<Result<Vec<_>>>()?;
+
+        row_insert_requests.inserts.push(RowInsertRequest {
+            table_name: DEFAULT_EVENTS_TABLE_NAME.to_string(),
+            rows: Some(Rows { schema, rows }),
+        });
+    }
+
+    Ok(row_insert_requests)
+}
+
+// Ensure the events with the same event type have the same extra schema.
+#[allow(clippy::borrowed_box)]
+fn validate_events(events: &[&Box<dyn Event>]) -> Result<()> {
+    // It's safe to get the first event because the events are already grouped by the event type.
+    let extra_schema = events[0].extra_schema();
+    for event in events {
+        if event.extra_schema() != extra_schema {
+            MismatchedSchemaSnafu {
+                expected: extra_schema.clone(),
+                actual: event.extra_schema(),
+            }
+            .fail()?;
+        }
+    }
+    Ok(())
+}
+
+/// EventRecorder trait defines the interface for recording events.
+pub trait EventRecorder: Send + Sync + 'static {
+    /// Records an event for persistence and processing by [EventHandler].
+    fn record(&self, event: Box<dyn Event>);
+
+    /// Cancels the event recorder.
+    fn close(&self);
+}
+
+/// EventHandler trait defines the interface for how to handle the event.
+#[async_trait]
+pub trait EventHandler: Send + Sync + 'static {
+    /// Processes and handles incoming events. The [DefaultEventHandlerImpl] implementation forwards events to frontend instances for persistence.
+    /// We use `&[Box<dyn Event>]` to avoid consuming the events, so the caller can buffer the events and retry if the handler fails.
+    async fn handle(&self, events: &[Box<dyn Event>]) -> Result<()>;
+}
+
+/// Configuration options for the event recorder.
+#[derive(Debug, Clone, Serialize, Deserialize, PartialEq, Eq)]
+pub struct EventRecorderOptions {
+    /// TTL for the events table that will be used to store the events.
+    pub ttl: String,
+}
+
+impl Default for EventRecorderOptions {
+    fn default() -> Self {
+        Self {
+            ttl: DEFAULT_EVENTS_TABLE_TTL.to_string(),
+        }
+    }
+}
+
+/// Implementation of [EventRecorder] that records the events and processes them in the background by the [EventHandler].
+pub struct EventRecorderImpl {
+    // The channel to send the events to the background processor.
+    tx: Sender<Box<dyn Event>>,
+    // The cancel token to cancel the background processor.
+    cancel_token: CancellationToken,
+    // The background processor to process the events.
+    handle: Option<JoinHandle<()>>,
+}
+
+impl EventRecorderImpl {
+    pub fn new(event_handler: Box<dyn EventHandler>, opts: EventRecorderOptions) -> Self {
+        info!("Creating event recorder with options: {:?}", opts);
+
+        let (tx, rx) = channel(DEFAULT_CHANNEL_SIZE);
+        let cancel_token = CancellationToken::new();
+
+        let mut recorder = Self {
+            tx,
+            handle: None,
+            cancel_token: cancel_token.clone(),
+        };
+
+        let processor = EventProcessor::new(
+            rx,
+            event_handler,
+            DEFAULT_FLUSH_INTERVAL_SECONDS,
+            DEFAULT_MAX_RETRY_TIMES,
+        )
+        .with_cancel_token(cancel_token);
+
+        // Spawn a background task to process the events.
+        let handle = tokio::spawn(async move {
+            processor.process(DEFAULT_BUFFER_SIZE).await;
+        });
+
+        recorder.handle = Some(handle);
+
+        // It only sets the ttl once, so it's safe to skip the error.
+        if EVENTS_TABLE_TTL.set(opts.ttl.clone()).is_err() {
+            info!(
+                "Events table ttl already set to {}, skip setting it",
+                opts.ttl
+            );
+        }
+
+        recorder
+    }
+}
+
+impl EventRecorder for EventRecorderImpl {
+    // Accepts an event and send it to the background handler.
+    fn record(&self, event: Box<dyn Event>) {
+        if let Err(e) = self.tx.try_send(event) {
+            error!("Failed to send event to the background processor: {}", e);
+        }
+    }
+
+    // Closes the event recorder. It will stop the background processor and flush the buffer.
+    fn close(&self) {
+        self.cancel_token.cancel();
+    }
+}
+
+impl Drop for EventRecorderImpl {
+    fn drop(&mut self) {
+        if let Some(handle) = self.handle.take() {
+            handle.abort();
+            info!("Aborted the background processor in event recorder");
+        }
+    }
+}
+
+struct EventProcessor {
+    rx: Receiver<Box<dyn Event>>,
+    event_handler: Box<dyn EventHandler>,
+    max_retry_times: u64,
+    process_interval: Duration,
+    cancel_token: CancellationToken,
+}
+
+impl EventProcessor {
+    fn new(
+        rx: Receiver<Box<dyn Event>>,
+        event_handler: Box<dyn EventHandler>,
+        process_interval: Duration,
+        max_retry_times: u64,
+    ) -> Self {
+        Self {
+            rx,
+            event_handler,
+            max_retry_times,
+            process_interval,
+            cancel_token: CancellationToken::new(),
+        }
+    }
+
+    fn with_cancel_token(mut self, cancel_token: CancellationToken) -> Self {
+        self.cancel_token = cancel_token;
+        self
+    }
+
+    async fn process(mut self, buffer_size: usize) {
+        info!("Start the background processor in event recorder to handle the received events.");
+
+        let mut buffer = Vec::with_capacity(buffer_size);
+        let mut interval = tokio::time::interval(self.process_interval);
+
+        loop {
+            tokio::select! {
+                maybe_event = self.rx.recv() => {
+                    if let Some(maybe_event) = maybe_event {
+                        debug!("Received event: {:?}", maybe_event);
+
+                        if buffer.len() >= buffer_size {
+                            debug!(
+                                "Flushing events to the event handler because the buffer is full with {} events",
+                                buffer.len()
+                            );
+                            self.flush_events_to_handler(&mut buffer).await;
+                        }
+
+                        // Push the event to the buffer, the buffer will be flushed when the interval is triggered or received a closed signal.
+                        buffer.push(maybe_event);
+                    } else {
+                        // When received a closed signal, flush the buffer and exit the loop.
+                        self.flush_events_to_handler(&mut buffer).await;
+                        break;
+                    }
+                }
+                // Cancel the processor through the cancel token.
+                _ = self.cancel_token.cancelled() => {
+                    warn!("Received a cancel signal, flushing the buffer and exiting the loop");
+                    self.flush_events_to_handler(&mut buffer).await;
+                    break;
+                }
+                // When the interval is triggered, flush the buffer and send the events to the event handler.
+                _ = interval.tick() => {
+                    self.flush_events_to_handler(&mut buffer).await;
+                }
+            }
+        }
+    }
+
+    // NOTE: While we implement a retry mechanism for failed event handling, there is no guarantee that all events will be processed successfully.
+    async fn flush_events_to_handler(&self, buffer: &mut Vec<Box<dyn Event>>) {
+        if !buffer.is_empty() {
+            debug!("Flushing {} events to the event handler", buffer.len());
+
+            let mut backoff = ExponentialBuilder::default()
+                .with_min_delay(Duration::from_millis(
+                    DEFAULT_FLUSH_INTERVAL_SECONDS.as_millis() as u64 / self.max_retry_times.max(1),
+                ))
+                .with_max_delay(Duration::from_millis(
+                    DEFAULT_FLUSH_INTERVAL_SECONDS.as_millis() as u64,
+                ))
+                .with_max_times(self.max_retry_times as usize)
+                .build();
+
+            loop {
+                match self.event_handler.handle(buffer).await {
+                    Ok(()) => {
+                        debug!("Successfully handled {} events", buffer.len());
+                        break;
+                    }
+                    Err(e) => {
+                        if let Some(d) = backoff.next() {
+                            warn!(e; "Failed to handle events, retrying...");
+                            sleep(d).await;
+                            continue;
+                        } else {
+                            warn!(
+                                e; "Failed to handle events after {} retries",
+                                self.max_retry_times
+                            );
+                            break;
+                        }
+                    }
+                }
+            }
+        }
+
+        // Clear the buffer to prevent unbounded memory growth, regardless of whether event processing succeeded or failed.
+        buffer.clear();
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[derive(Debug)]
+    struct TestEvent {}
+
+    impl Event for TestEvent {
+        fn event_type(&self) -> &str {
+            "test_event"
+        }
+
+        fn json_payload(&self) -> Result<String> {
+            Ok("{\"procedure_id\": \"1234567890\"}".to_string())
+        }
+
+        fn as_any(&self) -> &dyn Any {
+            self
+        }
+    }
+
+    struct TestEventHandlerImpl {}
+
+    #[async_trait]
+    impl EventHandler for TestEventHandlerImpl {
+        async fn handle(&self, events: &[Box<dyn Event>]) -> Result<()> {
+            let event = events
+                .first()
+                .unwrap()
+                .as_any()
+                .downcast_ref::<TestEvent>()
+                .unwrap();
+            assert_eq!(
+                event.json_payload().unwrap(),
+                "{\"procedure_id\": \"1234567890\"}"
+            );
+            assert_eq!(event.event_type(), "test_event");
+            Ok(())
+        }
+    }
+
+    #[tokio::test]
+    async fn test_event_recorder() {
+        let mut event_recorder = EventRecorderImpl::new(
+            Box::new(TestEventHandlerImpl {}),
+            EventRecorderOptions::default(),
+        );
+        event_recorder.record(Box::new(TestEvent {}));
+
+        // Sleep for a while to let the event be sent to the event handler.
+        sleep(Duration::from_millis(500)).await;
+
+        // Close the event recorder to flush the buffer.
+        event_recorder.close();
+
+        // Sleep for a while to let the background task process the event.
+        sleep(Duration::from_millis(500)).await;
+
+        if let Some(handle) = event_recorder.handle.take() {
+            assert!(handle.await.is_ok());
+        }
+    }
+
+    struct TestEventHandlerImplShouldPanic {}
+
+    #[async_trait]
+    impl EventHandler for TestEventHandlerImplShouldPanic {
+        async fn handle(&self, events: &[Box<dyn Event>]) -> Result<()> {
+            let event = events
+                .first()
+                .unwrap()
+                .as_any()
+                .downcast_ref::<TestEvent>()
+                .unwrap();
+
+            // Set the incorrect payload and event type to trigger the panic.
+            assert_eq!(
+                event.json_payload().unwrap(),
+                "{\"procedure_id\": \"should_panic\"}"
+            );
+            assert_eq!(event.event_type(), "should_panic");
+            Ok(())
+        }
+    }
+
+    #[tokio::test]
+    async fn test_event_recorder_should_panic() {
+        let mut event_recorder = EventRecorderImpl::new(
+            Box::new(TestEventHandlerImplShouldPanic {}),
+            EventRecorderOptions::default(),
+        );
+
+        event_recorder.record(Box::new(TestEvent {}));
+
+        // Sleep for a while to let the event be sent to the event handler.
+        sleep(Duration::from_millis(500)).await;
+
+        // Close the event recorder to flush the buffer.
+        event_recorder.close();
+
+        // Sleep for a while to let the background task process the event.
+        sleep(Duration::from_millis(500)).await;
+
+        if let Some(handle) = event_recorder.handle.take() {
+            assert!(handle.await.unwrap_err().is_panic());
+        }
+    }
+}

src/common/frontend/src/selector.rs

@@ -12,6 +12,7 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
+use std::fmt::Debug;
 use std::time::Duration;
 
 use common_grpc::channel_manager::{ChannelConfig, ChannelManager};
@@ -30,7 +31,7 @@ use crate::error::{MetaSnafu, Result};
 pub type FrontendClientPtr = Box<dyn FrontendClient>;
 
 #[async_trait::async_trait]
-pub trait FrontendClient: Send {
+pub trait FrontendClient: Send + Debug {
     async fn list_process(&mut self, req: ListProcessRequest) -> Result<ListProcessResponse>;
 
     async fn kill_process(&mut self, req: KillProcessRequest) -> Result<KillProcessResponse>;

src/common/function/Cargo.toml

@@ -16,6 +16,8 @@ geo = ["geohash", "h3o", "s2", "wkt", "geo-types", "dep:geo"]
 ahash.workspace = true
 api.workspace = true
 arc-swap = "1.0"
+arrow.workspace = true
+arrow-schema.workspace = true
 async-trait.workspace = true
 bincode = "1.3"
 catalog.workspace = true
@@ -33,6 +35,8 @@ common-version.workspace = true
 datafusion.workspace = true
 datafusion-common.workspace = true
 datafusion-expr.workspace = true
+datafusion-functions-aggregate-common.workspace = true
+datafusion-physical-expr.workspace = true
 datatypes.workspace = true
 derive_more = { version = "1", default-features = false, features = ["display"] }
 geo = { version = "0.29", optional = true }
@@ -61,5 +65,7 @@ wkt = { version = "0.11", optional = true }
 
 [dev-dependencies]
 approx = "0.5"
+futures.workspace = true
+pretty_assertions = "1.4.0"
 serde = { version = "1.0", features = ["derive"] }
 tokio.workspace = true

src/common/function/src/aggrs.rs

@@ -13,6 +13,9 @@
 // limitations under the License.
 
 pub mod approximate;
+pub mod count_hash;
 #[cfg(feature = "geo")]
 pub mod geo;
 pub mod vector;
+
+pub mod aggr_wrapper;

src/common/function/src/aggrs/aggr_wrapper.rs

@@ -0,0 +1,538 @@
+// 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.
+
+//! Wrapper for making aggregate functions out of state/merge functions of original aggregate functions.
+//!
+//! i.e. for a aggregate function `foo`, we will have a state function `foo_state` and a merge function `foo_merge`.
+//!
+//! `foo_state`'s input args is the same as `foo`'s, and its output is a state object.
+//! Note that `foo_state` might have multiple output columns, so it's a struct array
+//! that each output column is a struct field.
+//! `foo_merge`'s input arg is the same as `foo_state`'s output, and its output is the same as `foo`'s input.
+//!
+
+use std::sync::Arc;
+
+use arrow::array::StructArray;
+use arrow_schema::Fields;
+use datafusion::optimizer::analyzer::type_coercion::TypeCoercion;
+use datafusion::optimizer::AnalyzerRule;
+use datafusion::physical_planner::create_aggregate_expr_and_maybe_filter;
+use datafusion_common::{Column, ScalarValue};
+use datafusion_expr::expr::AggregateFunction;
+use datafusion_expr::function::StateFieldsArgs;
+use datafusion_expr::{
+    Accumulator, Aggregate, AggregateUDF, AggregateUDFImpl, Expr, ExprSchemable, LogicalPlan,
+    Signature,
+};
+use datafusion_physical_expr::aggregate::AggregateFunctionExpr;
+use datatypes::arrow::datatypes::{DataType, Field};
+
+/// Returns the name of the state function for the given aggregate function name.
+/// The state function is used to compute the state of the aggregate function.
+/// The state function's name is in the format `__<aggr_name>_state
+pub fn aggr_state_func_name(aggr_name: &str) -> String {
+    format!("__{}_state", aggr_name)
+}
+
+/// Returns the name of the merge function for the given aggregate function name.
+/// The merge function is used to merge the states of the state functions.
+/// The merge function's name is in the format `__<aggr_name>_merge
+pub fn aggr_merge_func_name(aggr_name: &str) -> String {
+    format!("__{}_merge", aggr_name)
+}
+
+/// A wrapper to make an aggregate function out of the state and merge functions of the original aggregate function.
+/// It contains the original aggregate function, the state functions, and the merge function.
+///
+/// Notice state functions may have multiple output columns, so it's return type is always a struct array, and the merge function is used to merge the states of the state functions.
+#[derive(Debug, Clone)]
+pub struct StateMergeHelper;
+
+/// A struct to hold the two aggregate plans, one for the state function(lower) and one for the merge function(upper).
+#[allow(unused)]
+#[derive(Debug, Clone)]
+pub struct StepAggrPlan {
+    /// Upper merge plan, which is the aggregate plan that merges the states of the state function.
+    pub upper_merge: Arc<LogicalPlan>,
+    /// Lower state plan, which is the aggregate plan that computes the state of the aggregate function.
+    pub lower_state: Arc<LogicalPlan>,
+}
+
+pub fn get_aggr_func(expr: &Expr) -> Option<&datafusion_expr::expr::AggregateFunction> {
+    let mut expr_ref = expr;
+    while let Expr::Alias(alias) = expr_ref {
+        expr_ref = &alias.expr;
+    }
+    if let Expr::AggregateFunction(aggr_func) = expr_ref {
+        Some(aggr_func)
+    } else {
+        None
+    }
+}
+
+impl StateMergeHelper {
+    /// Split an aggregate plan into two aggregate plans, one for the state function and one for the merge function.
+    pub fn split_aggr_node(aggr_plan: Aggregate) -> datafusion_common::Result<StepAggrPlan> {
+        let aggr = {
+            // certain aggr func need type coercion to work correctly, so we need to analyze the plan first.
+            let aggr_plan = TypeCoercion::new().analyze(
+                LogicalPlan::Aggregate(aggr_plan).clone(),
+                &Default::default(),
+            )?;
+            if let LogicalPlan::Aggregate(aggr) = aggr_plan {
+                aggr
+            } else {
+                return Err(datafusion_common::DataFusionError::Internal(format!(
+                    "Failed to coerce expressions in aggregate plan, expected Aggregate, got: {:?}",
+                    aggr_plan
+                )));
+            }
+        };
+        let mut lower_aggr_exprs = vec![];
+        let mut upper_aggr_exprs = vec![];
+
+        for aggr_expr in aggr.aggr_expr.iter() {
+            let Some(aggr_func) = get_aggr_func(aggr_expr) else {
+                return Err(datafusion_common::DataFusionError::NotImplemented(format!(
+                    "Unsupported aggregate expression for step aggr optimize: {:?}",
+                    aggr_expr
+                )));
+            };
+
+            let original_input_types = aggr_func
+                .args
+                .iter()
+                .map(|e| e.get_type(&aggr.input.schema()))
+                .collect::<Result<Vec<_>, _>>()?;
+
+            // first create the state function from the original aggregate function.
+            let state_func = StateWrapper::new((*aggr_func.func).clone())?;
+
+            let expr = AggregateFunction {
+                func: Arc::new(state_func.into()),
+                args: aggr_func.args.clone(),
+                distinct: aggr_func.distinct,
+                filter: aggr_func.filter.clone(),
+                order_by: aggr_func.order_by.clone(),
+                null_treatment: aggr_func.null_treatment,
+            };
+            let expr = Expr::AggregateFunction(expr);
+            let lower_state_output_col_name = expr.schema_name().to_string();
+
+            lower_aggr_exprs.push(expr);
+
+            let (original_phy_expr, _filter, _ordering) = create_aggregate_expr_and_maybe_filter(
+                aggr_expr,
+                aggr.input.schema(),
+                aggr.input.schema().as_arrow(),
+                &Default::default(),
+            )?;
+
+            let merge_func = MergeWrapper::new(
+                (*aggr_func.func).clone(),
+                original_phy_expr,
+                original_input_types,
+            )?;
+            let arg = Expr::Column(Column::new_unqualified(lower_state_output_col_name));
+            let expr = AggregateFunction {
+                func: Arc::new(merge_func.into()),
+                args: vec![arg],
+                distinct: aggr_func.distinct,
+                filter: aggr_func.filter.clone(),
+                order_by: aggr_func.order_by.clone(),
+                null_treatment: aggr_func.null_treatment,
+            };
+
+            // alias to the original aggregate expr's schema name, so parent plan can refer to it
+            // correctly.
+            let expr = Expr::AggregateFunction(expr).alias(aggr_expr.schema_name().to_string());
+            upper_aggr_exprs.push(expr);
+        }
+
+        let mut lower = aggr.clone();
+        lower.aggr_expr = lower_aggr_exprs;
+        let lower_plan = LogicalPlan::Aggregate(lower);
+
+        // update aggregate's output schema
+        let lower_plan = Arc::new(lower_plan.recompute_schema()?);
+
+        let mut upper = aggr.clone();
+        let aggr_plan = LogicalPlan::Aggregate(aggr);
+        upper.aggr_expr = upper_aggr_exprs;
+        upper.input = lower_plan.clone();
+        // upper schema's output schema should be the same as the original aggregate plan's output schema
+        let upper_check = upper.clone();
+        let upper_plan = Arc::new(LogicalPlan::Aggregate(upper_check).recompute_schema()?);
+        if *upper_plan.schema() != *aggr_plan.schema() {
+            return Err(datafusion_common::DataFusionError::Internal(format!(
+                 "Upper aggregate plan's schema is not the same as the original aggregate plan's schema: \n[transformed]:{}\n[   original]{}",
+                upper_plan.schema(), aggr_plan.schema()
+            )));
+        }
+
+        Ok(StepAggrPlan {
+            lower_state: lower_plan,
+            upper_merge: upper_plan,
+        })
+    }
+}
+
+/// Wrapper to make an aggregate function out of a state function.
+#[derive(Debug, Clone, PartialEq, Eq)]
+pub struct StateWrapper {
+    inner: AggregateUDF,
+    name: String,
+}
+
+impl StateWrapper {
+    /// `state_index`: The index of the state in the output of the state function.
+    pub fn new(inner: AggregateUDF) -> datafusion_common::Result<Self> {
+        let name = aggr_state_func_name(inner.name());
+        Ok(Self { inner, name })
+    }
+
+    pub fn inner(&self) -> &AggregateUDF {
+        &self.inner
+    }
+
+    /// Deduce the return type of the original aggregate function
+    /// based on the accumulator arguments.
+    ///
+    pub fn deduce_aggr_return_type(
+        &self,
+        acc_args: &datafusion_expr::function::AccumulatorArgs,
+    ) -> datafusion_common::Result<DataType> {
+        let input_exprs = acc_args.exprs;
+        let input_schema = acc_args.schema;
+        let input_types = input_exprs
+            .iter()
+            .map(|e| e.data_type(input_schema))
+            .collect::<Result<Vec<_>, _>>()?;
+        let return_type = self.inner.return_type(&input_types)?;
+        Ok(return_type)
+    }
+}
+
+impl AggregateUDFImpl for StateWrapper {
+    fn accumulator<'a, 'b>(
+        &'a self,
+        acc_args: datafusion_expr::function::AccumulatorArgs<'b>,
+    ) -> datafusion_common::Result<Box<dyn Accumulator>> {
+        // fix and recover proper acc args for the original aggregate function.
+        let state_type = acc_args.return_type.clone();
+        let inner = {
+            let old_return_type = self.deduce_aggr_return_type(&acc_args)?;
+            let acc_args = datafusion_expr::function::AccumulatorArgs {
+                return_type: &old_return_type,
+                schema: acc_args.schema,
+                ignore_nulls: acc_args.ignore_nulls,
+                ordering_req: acc_args.ordering_req,
+                is_reversed: acc_args.is_reversed,
+                name: acc_args.name,
+                is_distinct: acc_args.is_distinct,
+                exprs: acc_args.exprs,
+            };
+            self.inner.accumulator(acc_args)?
+        };
+        Ok(Box::new(StateAccum::new(inner, state_type)?))
+    }
+
+    fn as_any(&self) -> &dyn std::any::Any {
+        self
+    }
+    fn name(&self) -> &str {
+        self.name.as_str()
+    }
+
+    fn is_nullable(&self) -> bool {
+        self.inner.is_nullable()
+    }
+
+    /// Return state_fields as the output struct type.
+    ///
+    fn return_type(&self, arg_types: &[DataType]) -> datafusion_common::Result<DataType> {
+        let old_return_type = self.inner.return_type(arg_types)?;
+        let state_fields_args = StateFieldsArgs {
+            name: self.inner().name(),
+            input_types: arg_types,
+            return_type: &old_return_type,
+            // TODO(discord9): how to get this?, probably ok?
+            ordering_fields: &[],
+            is_distinct: false,
+        };
+        let state_fields = self.inner.state_fields(state_fields_args)?;
+        let struct_field = DataType::Struct(state_fields.into());
+        Ok(struct_field)
+    }
+
+    /// The state function's output fields are the same as the original aggregate function's state fields.
+    fn state_fields(
+        &self,
+        args: datafusion_expr::function::StateFieldsArgs,
+    ) -> datafusion_common::Result<Vec<Field>> {
+        let old_return_type = self.inner.return_type(args.input_types)?;
+        let state_fields_args = StateFieldsArgs {
+            name: args.name,
+            input_types: args.input_types,
+            return_type: &old_return_type,
+            ordering_fields: args.ordering_fields,
+            is_distinct: args.is_distinct,
+        };
+        self.inner.state_fields(state_fields_args)
+    }
+
+    /// The state function's signature is the same as the original aggregate function's signature,
+    fn signature(&self) -> &Signature {
+        self.inner.signature()
+    }
+
+    /// Coerce types also do nothing, as optimzer should be able to already make struct types
+    fn coerce_types(&self, arg_types: &[DataType]) -> datafusion_common::Result<Vec<DataType>> {
+        self.inner.coerce_types(arg_types)
+    }
+}
+
+/// The wrapper's input is the same as the original aggregate function's input,
+/// and the output is the state function's output.
+#[derive(Debug)]
+pub struct StateAccum {
+    inner: Box<dyn Accumulator>,
+    state_fields: Fields,
+}
+
+impl StateAccum {
+    pub fn new(
+        inner: Box<dyn Accumulator>,
+        state_type: DataType,
+    ) -> datafusion_common::Result<Self> {
+        let DataType::Struct(fields) = state_type else {
+            return Err(datafusion_common::DataFusionError::Internal(format!(
+                "Expected a struct type for state, got: {:?}",
+                state_type
+            )));
+        };
+        Ok(Self {
+            inner,
+            state_fields: fields,
+        })
+    }
+}
+
+impl Accumulator for StateAccum {
+    fn evaluate(&mut self) -> datafusion_common::Result<ScalarValue> {
+        let state = self.inner.state()?;
+
+        let array = state
+            .iter()
+            .map(|s| s.to_array())
+            .collect::<Result<Vec<_>, _>>()?;
+        let struct_array = StructArray::try_new(self.state_fields.clone(), array, None)?;
+        Ok(ScalarValue::Struct(Arc::new(struct_array)))
+    }
+
+    fn merge_batch(
+        &mut self,
+        states: &[datatypes::arrow::array::ArrayRef],
+    ) -> datafusion_common::Result<()> {
+        self.inner.merge_batch(states)
+    }
+
+    fn update_batch(
+        &mut self,
+        values: &[datatypes::arrow::array::ArrayRef],
+    ) -> datafusion_common::Result<()> {
+        self.inner.update_batch(values)
+    }
+
+    fn size(&self) -> usize {
+        self.inner.size()
+    }
+
+    fn state(&mut self) -> datafusion_common::Result<Vec<ScalarValue>> {
+        self.inner.state()
+    }
+}
+
+/// TODO(discord9): mark this function as non-ser/de able
+///
+/// This wrapper shouldn't be register as a udaf, as it contain extra data that is not serializable.
+/// and changes for different logical plans.
+#[derive(Debug, Clone)]
+pub struct MergeWrapper {
+    inner: AggregateUDF,
+    name: String,
+    merge_signature: Signature,
+    /// The original physical expression of the aggregate function, can't store the original aggregate function directly, as PhysicalExpr didn't implement Any
+    original_phy_expr: Arc<AggregateFunctionExpr>,
+    original_input_types: Vec<DataType>,
+}
+impl MergeWrapper {
+    pub fn new(
+        inner: AggregateUDF,
+        original_phy_expr: Arc<AggregateFunctionExpr>,
+        original_input_types: Vec<DataType>,
+    ) -> datafusion_common::Result<Self> {
+        let name = aggr_merge_func_name(inner.name());
+        // the input type is actually struct type, which is the state fields of the original aggregate function.
+        let merge_signature = Signature::user_defined(datafusion_expr::Volatility::Immutable);
+
+        Ok(Self {
+            inner,
+            name,
+            merge_signature,
+            original_phy_expr,
+            original_input_types,
+        })
+    }
+
+    pub fn inner(&self) -> &AggregateUDF {
+        &self.inner
+    }
+}
+
+impl AggregateUDFImpl for MergeWrapper {
+    fn accumulator<'a, 'b>(
+        &'a self,
+        acc_args: datafusion_expr::function::AccumulatorArgs<'b>,
+    ) -> datafusion_common::Result<Box<dyn Accumulator>> {
+        if acc_args.schema.fields().len() != 1
+            || !matches!(acc_args.schema.field(0).data_type(), DataType::Struct(_))
+        {
+            return Err(datafusion_common::DataFusionError::Internal(format!(
+                "Expected one struct type as input, got: {:?}",
+                acc_args.schema
+            )));
+        }
+        let input_type = acc_args.schema.field(0).data_type();
+        let DataType::Struct(fields) = input_type else {
+            return Err(datafusion_common::DataFusionError::Internal(format!(
+                "Expected a struct type for input, got: {:?}",
+                input_type
+            )));
+        };
+
+        let inner_accum = self.original_phy_expr.create_accumulator()?;
+        Ok(Box::new(MergeAccum::new(inner_accum, fields)))
+    }
+
+    fn as_any(&self) -> &dyn std::any::Any {
+        self
+    }
+    fn name(&self) -> &str {
+        self.name.as_str()
+    }
+
+    fn is_nullable(&self) -> bool {
+        self.inner.is_nullable()
+    }
+
+    /// Notice here the `arg_types` is actually the `state_fields`'s data types,
+    /// so return fixed return type instead of using `arg_types` to determine the return type.
+    fn return_type(&self, _arg_types: &[DataType]) -> datafusion_common::Result<DataType> {
+        // The return type is the same as the original aggregate function's return type.
+        let ret_type = self.inner.return_type(&self.original_input_types)?;
+        Ok(ret_type)
+    }
+    fn signature(&self) -> &Signature {
+        &self.merge_signature
+    }
+
+    /// Coerce types also do nothing, as optimzer should be able to already make struct types
+    fn coerce_types(&self, arg_types: &[DataType]) -> datafusion_common::Result<Vec<DataType>> {
+        // just check if the arg_types are only one and is struct array
+        if arg_types.len() != 1 || !matches!(arg_types.first(), Some(DataType::Struct(_))) {
+            return Err(datafusion_common::DataFusionError::Internal(format!(
+                "Expected one struct type as input, got: {:?}",
+                arg_types
+            )));
+        }
+        Ok(arg_types.to_vec())
+    }
+
+    /// Just return the original aggregate function's state fields.
+    fn state_fields(
+        &self,
+        _args: datafusion_expr::function::StateFieldsArgs,
+    ) -> datafusion_common::Result<Vec<Field>> {
+        self.original_phy_expr.state_fields()
+    }
+}
+
+/// The merge accumulator, which modify `update_batch`'s behavior to accept one struct array which
+/// include the state fields of original aggregate function, and merge said states into original accumulator
+/// the output is the same as original aggregate function
+#[derive(Debug)]
+pub struct MergeAccum {
+    inner: Box<dyn Accumulator>,
+    state_fields: Fields,
+}
+
+impl MergeAccum {
+    pub fn new(inner: Box<dyn Accumulator>, state_fields: &Fields) -> Self {
+        Self {
+            inner,
+            state_fields: state_fields.clone(),
+        }
+    }
+}
+
+impl Accumulator for MergeAccum {
+    fn evaluate(&mut self) -> datafusion_common::Result<ScalarValue> {
+        self.inner.evaluate()
+    }
+
+    fn merge_batch(&mut self, states: &[arrow::array::ArrayRef]) -> datafusion_common::Result<()> {
+        self.inner.merge_batch(states)
+    }
+
+    fn update_batch(&mut self, values: &[arrow::array::ArrayRef]) -> datafusion_common::Result<()> {
+        let value = values.first().ok_or_else(|| {
+            datafusion_common::DataFusionError::Internal("No values provided for merge".to_string())
+        })?;
+        // The input values are states from other accumulators, so we merge them.
+        let struct_arr = value
+            .as_any()
+            .downcast_ref::<StructArray>()
+            .ok_or_else(|| {
+                datafusion_common::DataFusionError::Internal(format!(
+                    "Expected StructArray, got: {:?}",
+                    value.data_type()
+                ))
+            })?;
+        let fields = struct_arr.fields();
+        if fields != &self.state_fields {
+            return Err(datafusion_common::DataFusionError::Internal(format!(
+                "Expected state fields: {:?}, got: {:?}",
+                self.state_fields, fields
+            )));
+        }
+
+        // now fields should be the same, so we can merge the batch
+        // by pass the columns as order should be the same
+        let state_columns = struct_arr.columns();
+        self.inner.merge_batch(state_columns)
+    }
+
+    fn size(&self) -> usize {
+        self.inner.size()
+    }
+
+    fn state(&mut self) -> datafusion_common::Result<Vec<ScalarValue>> {
+        self.inner.state()
+    }
+}
+
+#[cfg(test)]
+mod tests;

src/common/function/src/aggrs/aggr_wrapper/tests.rs

@@ -0,0 +1,804 @@
+// Copyright 2023 Greptime Team
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+use std::any::Any;
+use std::pin::Pin;
+use std::sync::{Arc, Mutex};
+use std::task::{Context, Poll};
+
+use arrow::array::{ArrayRef, Float64Array, Int64Array, UInt64Array};
+use arrow::record_batch::RecordBatch;
+use arrow_schema::SchemaRef;
+use datafusion::catalog::{Session, TableProvider};
+use datafusion::datasource::DefaultTableSource;
+use datafusion::execution::{RecordBatchStream, SendableRecordBatchStream, TaskContext};
+use datafusion::functions_aggregate::average::avg_udaf;
+use datafusion::functions_aggregate::sum::sum_udaf;
+use datafusion::optimizer::analyzer::type_coercion::TypeCoercion;
+use datafusion::optimizer::AnalyzerRule;
+use datafusion::physical_plan::aggregates::AggregateExec;
+use datafusion::physical_plan::execution_plan::{Boundedness, EmissionType};
+use datafusion::physical_plan::{DisplayAs, DisplayFormatType, ExecutionPlan, PlanProperties};
+use datafusion::physical_planner::{DefaultPhysicalPlanner, PhysicalPlanner};
+use datafusion::prelude::SessionContext;
+use datafusion_common::{Column, TableReference};
+use datafusion_expr::expr::AggregateFunction;
+use datafusion_expr::sqlparser::ast::NullTreatment;
+use datafusion_expr::{Aggregate, Expr, LogicalPlan, SortExpr, TableScan};
+use datafusion_physical_expr::aggregate::AggregateExprBuilder;
+use datafusion_physical_expr::{EquivalenceProperties, Partitioning};
+use datatypes::arrow_array::StringArray;
+use futures::{Stream, StreamExt as _};
+use pretty_assertions::assert_eq;
+
+use super::*;
+use crate::aggrs::approximate::hll::HllState;
+use crate::aggrs::approximate::uddsketch::UddSketchState;
+use crate::aggrs::count_hash::CountHash;
+use crate::function::Function as _;
+use crate::scalars::hll_count::HllCalcFunction;
+use crate::scalars::uddsketch_calc::UddSketchCalcFunction;
+
+#[derive(Debug)]
+pub struct MockInputExec {
+    input: Vec<RecordBatch>,
+    schema: SchemaRef,
+    properties: PlanProperties,
+}
+
+impl MockInputExec {
+    pub fn new(input: Vec<RecordBatch>, schema: SchemaRef) -> Self {
+        Self {
+            properties: PlanProperties::new(
+                EquivalenceProperties::new(schema.clone()),
+                Partitioning::UnknownPartitioning(1),
+                EmissionType::Incremental,
+                Boundedness::Bounded,
+            ),
+            input,
+            schema,
+        }
+    }
+}
+
+impl DisplayAs for MockInputExec {
+    fn fmt_as(&self, _t: DisplayFormatType, _f: &mut std::fmt::Formatter) -> std::fmt::Result {
+        unimplemented!()
+    }
+}
+
+impl ExecutionPlan for MockInputExec {
+    fn name(&self) -> &str {
+        "MockInputExec"
+    }
+
+    fn as_any(&self) -> &dyn Any {
+        self
+    }
+
+    fn properties(&self) -> &PlanProperties {
+        &self.properties
+    }
+
+    fn children(&self) -> Vec<&Arc<dyn ExecutionPlan>> {
+        vec![]
+    }
+
+    fn with_new_children(
+        self: Arc<Self>,
+        _children: Vec<Arc<dyn ExecutionPlan>>,
+    ) -> datafusion_common::Result<Arc<dyn ExecutionPlan>> {
+        Ok(self)
+    }
+
+    fn execute(
+        &self,
+        _partition: usize,
+        _context: Arc<TaskContext>,
+    ) -> datafusion_common::Result<SendableRecordBatchStream> {
+        let stream = MockStream {
+            stream: self.input.clone(),
+            schema: self.schema.clone(),
+            idx: 0,
+        };
+        Ok(Box::pin(stream))
+    }
+}
+
+struct MockStream {
+    stream: Vec<RecordBatch>,
+    schema: SchemaRef,
+    idx: usize,
+}
+
+impl Stream for MockStream {
+    type Item = datafusion_common::Result<RecordBatch>;
+    fn poll_next(
+        mut self: Pin<&mut Self>,
+        _cx: &mut Context<'_>,
+    ) -> Poll<Option<datafusion_common::Result<RecordBatch>>> {
+        if self.idx < self.stream.len() {
+            let ret = self.stream[self.idx].clone();
+            self.idx += 1;
+            Poll::Ready(Some(Ok(ret)))
+        } else {
+            Poll::Ready(None)
+        }
+    }
+}
+
+impl RecordBatchStream for MockStream {
+    fn schema(&self) -> SchemaRef {
+        self.schema.clone()
+    }
+}
+
+#[derive(Debug)]
+struct DummyTableProvider {
+    schema: Arc<arrow_schema::Schema>,
+    record_batch: Mutex<Option<RecordBatch>>,
+}
+
+impl DummyTableProvider {
+    #[allow(unused)]
+    pub fn new(schema: Arc<arrow_schema::Schema>, record_batch: Option<RecordBatch>) -> Self {
+        Self {
+            schema,
+            record_batch: Mutex::new(record_batch),
+        }
+    }
+}
+
+impl Default for DummyTableProvider {
+    fn default() -> Self {
+        Self {
+            schema: Arc::new(arrow_schema::Schema::new(vec![Field::new(
+                "number",
+                DataType::Int64,
+                true,
+            )])),
+            record_batch: Mutex::new(None),
+        }
+    }
+}
+
+#[async_trait::async_trait]
+impl TableProvider for DummyTableProvider {
+    fn as_any(&self) -> &dyn std::any::Any {
+        self
+    }
+
+    fn schema(&self) -> Arc<arrow_schema::Schema> {
+        self.schema.clone()
+    }
+
+    fn table_type(&self) -> datafusion_expr::TableType {
+        datafusion_expr::TableType::Base
+    }
+
+    async fn scan(
+        &self,
+        _state: &dyn Session,
+        _projection: Option<&Vec<usize>>,
+        _filters: &[Expr],
+        _limit: Option<usize>,
+    ) -> datafusion::error::Result<Arc<dyn ExecutionPlan>> {
+        let input: Vec<RecordBatch> = self
+            .record_batch
+            .lock()
+            .unwrap()
+            .clone()
+            .map(|r| vec![r])
+            .unwrap_or_default();
+        Ok(Arc::new(MockInputExec::new(input, self.schema.clone())))
+    }
+}
+
+fn dummy_table_scan() -> LogicalPlan {
+    let table_provider = Arc::new(DummyTableProvider::default());
+    let table_source = DefaultTableSource::new(table_provider);
+    LogicalPlan::TableScan(
+        TableScan::try_new(
+            TableReference::bare("Number"),
+            Arc::new(table_source),
+            None,
+            vec![],
+            None,
+        )
+        .unwrap(),
+    )
+}
+
+#[tokio::test]
+async fn test_sum_udaf() {
+    let ctx = SessionContext::new();
+
+    let sum = datafusion::functions_aggregate::sum::sum_udaf();
+    let sum = (*sum).clone();
+    let original_aggr = Aggregate::try_new(
+        Arc::new(dummy_table_scan()),
+        vec![],
+        vec![Expr::AggregateFunction(AggregateFunction::new_udf(
+            Arc::new(sum.clone()),
+            vec![Expr::Column(Column::new_unqualified("number"))],
+            false,
+            None,
+            None,
+            None,
+        ))],
+    )
+    .unwrap();
+    let res = StateMergeHelper::split_aggr_node(original_aggr).unwrap();
+
+    let expected_lower_plan = LogicalPlan::Aggregate(
+        Aggregate::try_new(
+            Arc::new(dummy_table_scan()),
+            vec![],
+            vec![Expr::AggregateFunction(AggregateFunction::new_udf(
+                Arc::new(StateWrapper::new(sum.clone()).unwrap().into()),
+                vec![Expr::Column(Column::new_unqualified("number"))],
+                false,
+                None,
+                None,
+                None,
+            ))],
+        )
+        .unwrap(),
+    )
+    .recompute_schema()
+    .unwrap();
+    assert_eq!(res.lower_state.as_ref(), &expected_lower_plan);
+
+    let expected_merge_plan = LogicalPlan::Aggregate(
+        Aggregate::try_new(
+            Arc::new(expected_lower_plan),
+            vec![],
+            vec![Expr::AggregateFunction(AggregateFunction::new_udf(
+                Arc::new(
+                    MergeWrapper::new(
+                        sum.clone(),
+                        Arc::new(
+                            AggregateExprBuilder::new(
+                                Arc::new(sum.clone()),
+                                vec![Arc::new(
+                                    datafusion::physical_expr::expressions::Column::new(
+                                        "number", 0,
+                                    ),
+                                )],
+                            )
+                            .schema(Arc::new(dummy_table_scan().schema().as_arrow().clone()))
+                            .alias("sum(number)")
+                            .build()
+                            .unwrap(),
+                        ),
+                        vec![DataType::Int64],
+                    )
+                    .unwrap()
+                    .into(),
+                ),
+                vec![Expr::Column(Column::new_unqualified("__sum_state(number)"))],
+                false,
+                None,
+                None,
+                None,
+            ))
+            .alias("sum(number)")],
+        )
+        .unwrap(),
+    );
+    assert_eq!(res.upper_merge.as_ref(), &expected_merge_plan);
+
+    let phy_aggr_state_plan = DefaultPhysicalPlanner::default()
+        .create_physical_plan(&res.lower_state, &ctx.state())
+        .await
+        .unwrap();
+    let aggr_exec = phy_aggr_state_plan
+        .as_any()
+        .downcast_ref::<AggregateExec>()
+        .unwrap();
+    let aggr_func_expr = &aggr_exec.aggr_expr()[0];
+    let mut state_accum = aggr_func_expr.create_accumulator().unwrap();
+
+    // evaluate the state function
+    let input = Int64Array::from(vec![Some(1), Some(2), None, Some(3)]);
+    let values = vec![Arc::new(input) as arrow::array::ArrayRef];
+
+    state_accum.update_batch(&values).unwrap();
+    let state = state_accum.state().unwrap();
+    assert_eq!(state.len(), 1);
+    assert_eq!(state[0], ScalarValue::Int64(Some(6)));
+
+    let eval_res = state_accum.evaluate().unwrap();
+    assert_eq!(
+        eval_res,
+        ScalarValue::Struct(Arc::new(
+            StructArray::try_new(
+                vec![Field::new("sum[sum]", DataType::Int64, true)].into(),
+                vec![Arc::new(Int64Array::from(vec![Some(6)]))],
+                None,
+            )
+            .unwrap(),
+        ))
+    );
+
+    let phy_aggr_merge_plan = DefaultPhysicalPlanner::default()
+        .create_physical_plan(&res.upper_merge, &ctx.state())
+        .await
+        .unwrap();
+    let aggr_exec = phy_aggr_merge_plan
+        .as_any()
+        .downcast_ref::<AggregateExec>()
+        .unwrap();
+    let aggr_func_expr = &aggr_exec.aggr_expr()[0];
+    let mut merge_accum = aggr_func_expr.create_accumulator().unwrap();
+
+    let merge_input =
+        vec![Arc::new(Int64Array::from(vec![Some(6), Some(42), None])) as arrow::array::ArrayRef];
+    let merge_input_struct_arr = StructArray::try_new(
+        vec![Field::new("sum[sum]", DataType::Int64, true)].into(),
+        merge_input,
+        None,
+    )
+    .unwrap();
+
+    merge_accum
+        .update_batch(&[Arc::new(merge_input_struct_arr)])
+        .unwrap();
+    let merge_state = merge_accum.state().unwrap();
+    assert_eq!(merge_state.len(), 1);
+    assert_eq!(merge_state[0], ScalarValue::Int64(Some(48)));
+
+    let merge_eval_res = merge_accum.evaluate().unwrap();
+    assert_eq!(merge_eval_res, ScalarValue::Int64(Some(48)));
+}
+
+#[tokio::test]
+async fn test_avg_udaf() {
+    let ctx = SessionContext::new();
+
+    let avg = datafusion::functions_aggregate::average::avg_udaf();
+    let avg = (*avg).clone();
+
+    let original_aggr = Aggregate::try_new(
+        Arc::new(dummy_table_scan()),
+        vec![],
+        vec![Expr::AggregateFunction(AggregateFunction::new_udf(
+            Arc::new(avg.clone()),
+            vec![Expr::Column(Column::new_unqualified("number"))],
+            false,
+            None,
+            None,
+            None,
+        ))],
+    )
+    .unwrap();
+    let res = StateMergeHelper::split_aggr_node(original_aggr).unwrap();
+
+    let state_func: Arc<AggregateUDF> = Arc::new(StateWrapper::new(avg.clone()).unwrap().into());
+    let expected_aggr_state_plan = LogicalPlan::Aggregate(
+        Aggregate::try_new(
+            Arc::new(dummy_table_scan()),
+            vec![],
+            vec![Expr::AggregateFunction(AggregateFunction::new_udf(
+                state_func,
+                vec![Expr::Column(Column::new_unqualified("number"))],
+                false,
+                None,
+                None,
+                None,
+            ))],
+        )
+        .unwrap(),
+    );
+    // type coerced so avg aggr function can function correctly
+    let coerced_aggr_state_plan = TypeCoercion::new()
+        .analyze(expected_aggr_state_plan.clone(), &Default::default())
+        .unwrap();
+    assert_eq!(res.lower_state.as_ref(), &coerced_aggr_state_plan);
+    assert_eq!(
+        res.lower_state.schema().as_arrow(),
+        &arrow_schema::Schema::new(vec![Field::new(
+            "__avg_state(number)",
+            DataType::Struct(
+                vec![
+                    Field::new("avg[count]", DataType::UInt64, true),
+                    Field::new("avg[sum]", DataType::Float64, true)
+                ]
+                .into()
+            ),
+            true,
+        )])
+    );
+
+    let expected_merge_fn = MergeWrapper::new(
+        avg.clone(),
+        Arc::new(
+            AggregateExprBuilder::new(
+                Arc::new(avg.clone()),
+                vec![Arc::new(
+                    datafusion::physical_expr::expressions::Column::new("number", 0),
+                )],
+            )
+            .schema(Arc::new(dummy_table_scan().schema().as_arrow().clone()))
+            .alias("avg(number)")
+            .build()
+            .unwrap(),
+        ),
+        // coerced to float64
+        vec![DataType::Float64],
+    )
+    .unwrap();
+
+    let expected_merge_plan = LogicalPlan::Aggregate(
+        Aggregate::try_new(
+            Arc::new(coerced_aggr_state_plan.clone()),
+            vec![],
+            vec![Expr::AggregateFunction(AggregateFunction::new_udf(
+                Arc::new(expected_merge_fn.into()),
+                vec![Expr::Column(Column::new_unqualified("__avg_state(number)"))],
+                false,
+                None,
+                None,
+                None,
+            ))
+            .alias("avg(number)")],
+        )
+        .unwrap(),
+    );
+    assert_eq!(res.upper_merge.as_ref(), &expected_merge_plan);
+
+    let phy_aggr_state_plan = DefaultPhysicalPlanner::default()
+        .create_physical_plan(&coerced_aggr_state_plan, &ctx.state())
+        .await
+        .unwrap();
+    let aggr_exec = phy_aggr_state_plan
+        .as_any()
+        .downcast_ref::<AggregateExec>()
+        .unwrap();
+    let aggr_func_expr = &aggr_exec.aggr_expr()[0];
+    let mut state_accum = aggr_func_expr.create_accumulator().unwrap();
+
+    // evaluate the state function
+    let input = Float64Array::from(vec![Some(1.), Some(2.), None, Some(3.)]);
+    let values = vec![Arc::new(input) as arrow::array::ArrayRef];
+
+    state_accum.update_batch(&values).unwrap();
+    let state = state_accum.state().unwrap();
+    assert_eq!(state.len(), 2);
+    assert_eq!(state[0], ScalarValue::UInt64(Some(3)));
+    assert_eq!(state[1], ScalarValue::Float64(Some(6.)));
+
+    let eval_res = state_accum.evaluate().unwrap();
+    let expected = Arc::new(
+        StructArray::try_new(
+            vec![
+                Field::new("avg[count]", DataType::UInt64, true),
+                Field::new("avg[sum]", DataType::Float64, true),
+            ]
+            .into(),
+            vec![
+                Arc::new(UInt64Array::from(vec![Some(3)])),
+                Arc::new(Float64Array::from(vec![Some(6.)])),
+            ],
+            None,
+        )
+        .unwrap(),
+    );
+    assert_eq!(eval_res, ScalarValue::Struct(expected));
+
+    let phy_aggr_merge_plan = DefaultPhysicalPlanner::default()
+        .create_physical_plan(&res.upper_merge, &ctx.state())
+        .await
+        .unwrap();
+    let aggr_exec = phy_aggr_merge_plan
+        .as_any()
+        .downcast_ref::<AggregateExec>()
+        .unwrap();
+    let aggr_func_expr = &aggr_exec.aggr_expr()[0];
+
+    let mut merge_accum = aggr_func_expr.create_accumulator().unwrap();
+
+    let merge_input = vec![
+        Arc::new(UInt64Array::from(vec![Some(3), Some(42), None])) as arrow::array::ArrayRef,
+        Arc::new(Float64Array::from(vec![Some(48.), Some(84.), None])),
+    ];
+    let merge_input_struct_arr = StructArray::try_new(
+        vec![
+            Field::new("avg[count]", DataType::UInt64, true),
+            Field::new("avg[sum]", DataType::Float64, true),
+        ]
+        .into(),
+        merge_input,
+        None,
+    )
+    .unwrap();
+
+    merge_accum
+        .update_batch(&[Arc::new(merge_input_struct_arr)])
+        .unwrap();
+    let merge_state = merge_accum.state().unwrap();
+    assert_eq!(merge_state.len(), 2);
+    assert_eq!(merge_state[0], ScalarValue::UInt64(Some(45)));
+    assert_eq!(merge_state[1], ScalarValue::Float64(Some(132.)));
+
+    let merge_eval_res = merge_accum.evaluate().unwrap();
+    // the merge function returns the average, which is 132 / 45
+    assert_eq!(merge_eval_res, ScalarValue::Float64(Some(132. / 45_f64)));
+}
+
+/// For testing whether the UDAF state fields are correctly implemented.
+/// esp. for our own custom UDAF's state fields.
+/// By compare eval results before and after split to state/merge functions.
+#[tokio::test]
+async fn test_udaf_correct_eval_result() {
+    struct TestCase {
+        func: Arc<AggregateUDF>,
+        args: Vec<Expr>,
+        input_schema: SchemaRef,
+        input: Vec<ArrayRef>,
+        expected_output: Option<ScalarValue>,
+        expected_fn: Option<ExpectedFn>,
+        distinct: bool,
+        filter: Option<Box<Expr>>,
+        order_by: Option<Vec<SortExpr>>,
+        null_treatment: Option<NullTreatment>,
+    }
+    type ExpectedFn = fn(ArrayRef) -> bool;
+
+    let test_cases = vec![
+        TestCase {
+            func: sum_udaf(),
+            input_schema: Arc::new(arrow_schema::Schema::new(vec![Field::new(
+                "number",
+                DataType::Int64,
+                true,
+            )])),
+            args: vec![Expr::Column(Column::new_unqualified("number"))],
+            input: vec![Arc::new(Int64Array::from(vec![
+                Some(1),
+                Some(2),
+                None,
+                Some(3),
+            ]))],
+            expected_output: Some(ScalarValue::Int64(Some(6))),
+            expected_fn: None,
+            distinct: false,
+            filter: None,
+            order_by: None,
+            null_treatment: None,
+        },
+        TestCase {
+            func: avg_udaf(),
+            input_schema: Arc::new(arrow_schema::Schema::new(vec![Field::new(
+                "number",
+                DataType::Int64,
+                true,
+            )])),
+            args: vec![Expr::Column(Column::new_unqualified("number"))],
+            input: vec![Arc::new(Int64Array::from(vec![
+                Some(1),
+                Some(2),
+                None,
+                Some(3),
+            ]))],
+            expected_output: Some(ScalarValue::Float64(Some(2.0))),
+            expected_fn: None,
+            distinct: false,
+            filter: None,
+            order_by: None,
+            null_treatment: None,
+        },
+        TestCase {
+            func: Arc::new(CountHash::udf_impl()),
+            input_schema: Arc::new(arrow_schema::Schema::new(vec![Field::new(
+                "number",
+                DataType::Int64,
+                true,
+            )])),
+            args: vec![Expr::Column(Column::new_unqualified("number"))],
+            input: vec![Arc::new(Int64Array::from(vec![
+                Some(1),
+                Some(2),
+                None,
+                Some(3),
+                Some(3),
+                Some(3),
+            ]))],
+            expected_output: Some(ScalarValue::Int64(Some(4))),
+            expected_fn: None,
+            distinct: false,
+            filter: None,
+            order_by: None,
+            null_treatment: None,
+        },
+        TestCase {
+            func: Arc::new(UddSketchState::state_udf_impl()),
+            input_schema: Arc::new(arrow_schema::Schema::new(vec![Field::new(
+                "number",
+                DataType::Float64,
+                true,
+            )])),
+            args: vec![
+                Expr::Literal(ScalarValue::Int64(Some(128))),
+                Expr::Literal(ScalarValue::Float64(Some(0.05))),
+                Expr::Column(Column::new_unqualified("number")),
+            ],
+            input: vec![Arc::new(Float64Array::from(vec![
+                Some(1.),
+                Some(2.),
+                None,
+                Some(3.),
+                Some(3.),
+                Some(3.),
+            ]))],
+            expected_output: None,
+            expected_fn: Some(|arr| {
+                let percent = ScalarValue::Float64(Some(0.5)).to_array().unwrap();
+                let percent = datatypes::vectors::Helper::try_into_vector(percent).unwrap();
+                let state = datatypes::vectors::Helper::try_into_vector(arr).unwrap();
+                let udd_calc = UddSketchCalcFunction;
+                let res = udd_calc
+                    .eval(&Default::default(), &[percent, state])
+                    .unwrap();
+                let binding = res.to_arrow_array();
+                let res_arr = binding.as_any().downcast_ref::<Float64Array>().unwrap();
+                assert!(res_arr.len() == 1);
+                assert!((res_arr.value(0) - 2.856578984907706f64).abs() <= f64::EPSILON);
+                true
+            }),
+            distinct: false,
+            filter: None,
+            order_by: None,
+            null_treatment: None,
+        },
+        TestCase {
+            func: Arc::new(HllState::state_udf_impl()),
+            input_schema: Arc::new(arrow_schema::Schema::new(vec![Field::new(
+                "word",
+                DataType::Utf8,
+                true,
+            )])),
+            args: vec![Expr::Column(Column::new_unqualified("word"))],
+            input: vec![Arc::new(StringArray::from(vec![
+                Some("foo"),
+                Some("bar"),
+                None,
+                Some("baz"),
+                Some("baz"),
+            ]))],
+            expected_output: None,
+            expected_fn: Some(|arr| {
+                let state = datatypes::vectors::Helper::try_into_vector(arr).unwrap();
+                let hll_calc = HllCalcFunction;
+                let res = hll_calc.eval(&Default::default(), &[state]).unwrap();
+                let binding = res.to_arrow_array();
+                let res_arr = binding.as_any().downcast_ref::<UInt64Array>().unwrap();
+                assert!(res_arr.len() == 1);
+                assert_eq!(res_arr.value(0), 3);
+                true
+            }),
+            distinct: false,
+            filter: None,
+            order_by: None,
+            null_treatment: None,
+        },
+        // TODO(discord9): udd_merge/hll_merge/geo_path/quantile_aggr tests
+    ];
+    let test_table_ref = TableReference::bare("TestTable");
+
+    for case in test_cases {
+        let ctx = SessionContext::new();
+        let table_provider = DummyTableProvider::new(
+            case.input_schema.clone(),
+            Some(RecordBatch::try_new(case.input_schema.clone(), case.input.clone()).unwrap()),
+        );
+        let table_source = DefaultTableSource::new(Arc::new(table_provider));
+        let logical_plan = LogicalPlan::TableScan(
+            TableScan::try_new(
+                test_table_ref.clone(),
+                Arc::new(table_source),
+                None,
+                vec![],
+                None,
+            )
+            .unwrap(),
+        );
+
+        let args = case.args;
+
+        let aggr_expr = Expr::AggregateFunction(AggregateFunction::new_udf(
+            case.func.clone(),
+            args,
+            case.distinct,
+            case.filter,
+            case.order_by,
+            case.null_treatment,
+        ));
+
+        let aggr_plan = LogicalPlan::Aggregate(
+            Aggregate::try_new(Arc::new(logical_plan), vec![], vec![aggr_expr]).unwrap(),
+        );
+
+        // make sure the aggr_plan is type coerced
+        let aggr_plan = TypeCoercion::new()
+            .analyze(aggr_plan, &Default::default())
+            .unwrap();
+
+        // first eval the original aggregate function
+        let phy_full_aggr_plan = DefaultPhysicalPlanner::default()
+            .create_physical_plan(&aggr_plan, &ctx.state())
+            .await
+            .unwrap();
+
+        {
+            let unsplit_result = execute_phy_plan(&phy_full_aggr_plan).await.unwrap();
+            assert_eq!(unsplit_result.len(), 1);
+            let unsplit_batch = &unsplit_result[0];
+            assert_eq!(unsplit_batch.num_columns(), 1);
+            assert_eq!(unsplit_batch.num_rows(), 1);
+            let unsplit_col = unsplit_batch.column(0);
+            if let Some(expected_output) = &case.expected_output {
+                assert_eq!(unsplit_col.data_type(), &expected_output.data_type());
+                assert_eq!(unsplit_col.len(), 1);
+                assert_eq!(unsplit_col, &expected_output.to_array().unwrap());
+            }
+
+            if let Some(expected_fn) = &case.expected_fn {
+                assert!(expected_fn(unsplit_col.clone()));
+            }
+        }
+        let LogicalPlan::Aggregate(aggr_plan) = aggr_plan else {
+            panic!("Expected Aggregate plan");
+        };
+        let split_plan = StateMergeHelper::split_aggr_node(aggr_plan).unwrap();
+
+        let phy_upper_plan = DefaultPhysicalPlanner::default()
+            .create_physical_plan(&split_plan.upper_merge, &ctx.state())
+            .await
+            .unwrap();
+
+        // since upper plan use lower plan as input, execute upper plan should also execute lower plan
+        // which should give the same result as the original aggregate function
+        {
+            let split_res = execute_phy_plan(&phy_upper_plan).await.unwrap();
+
+            assert_eq!(split_res.len(), 1);
+            let split_batch = &split_res[0];
+            assert_eq!(split_batch.num_columns(), 1);
+            assert_eq!(split_batch.num_rows(), 1);
+            let split_col = split_batch.column(0);
+            if let Some(expected_output) = &case.expected_output {
+                assert_eq!(split_col.data_type(), &expected_output.data_type());
+                assert_eq!(split_col.len(), 1);
+                assert_eq!(split_col, &expected_output.to_array().unwrap());
+            }
+
+            if let Some(expected_fn) = &case.expected_fn {
+                assert!(expected_fn(split_col.clone()));
+            }
+        }
+    }
+}
+
+async fn execute_phy_plan(
+    phy_plan: &Arc<dyn ExecutionPlan>,
+) -> datafusion_common::Result<Vec<RecordBatch>> {
+    let task_ctx = Arc::new(TaskContext::default());
+    let mut stream = phy_plan.execute(0, task_ctx)?;
+    let mut batches = Vec::new();
+    while let Some(batch) = stream.next().await {
+        batches.push(batch?);
+    }
+    Ok(batches)
+}

src/common/function/src/aggrs/count_hash.rs

@@ -0,0 +1,647 @@
+// 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.
+
+//! `CountHash` / `count_hash` is a hash-based approximate distinct count function.
+//!
+//! It is a variant of `CountDistinct` that uses a hash function to approximate the
+//! distinct count.
+//! It is designed to be more efficient than `CountDistinct` for large datasets,
+//! but it is not as accurate, as the hash value may be collision.
+
+use std::collections::HashSet;
+use std::fmt::Debug;
+use std::sync::Arc;
+
+use ahash::RandomState;
+use datafusion_common::cast::as_list_array;
+use datafusion_common::error::Result;
+use datafusion_common::hash_utils::create_hashes;
+use datafusion_common::utils::SingleRowListArrayBuilder;
+use datafusion_common::{internal_err, not_impl_err, ScalarValue};
+use datafusion_expr::function::{AccumulatorArgs, StateFieldsArgs};
+use datafusion_expr::utils::{format_state_name, AggregateOrderSensitivity};
+use datafusion_expr::{
+    Accumulator, AggregateUDF, AggregateUDFImpl, EmitTo, GroupsAccumulator, ReversedUDAF,
+    SetMonotonicity, Signature, TypeSignature, Volatility,
+};
+use datafusion_functions_aggregate_common::aggregate::groups_accumulator::nulls::filtered_null_mask;
+use datatypes::arrow;
+use datatypes::arrow::array::{
+    Array, ArrayRef, AsArray, BooleanArray, Int64Array, ListArray, UInt64Array,
+};
+use datatypes::arrow::buffer::{OffsetBuffer, ScalarBuffer};
+use datatypes::arrow::datatypes::{DataType, Field};
+
+use crate::function_registry::FunctionRegistry;
+
+type HashValueType = u64;
+
+// read from /dev/urandom 4047821dc6144e4b2abddf23ad4171126a52eeecd26eff2191cf673b965a7875
+const RANDOM_SEED_0: u64 = 0x4047821dc6144e4b;
+const RANDOM_SEED_1: u64 = 0x2abddf23ad417112;
+const RANDOM_SEED_2: u64 = 0x6a52eeecd26eff21;
+const RANDOM_SEED_3: u64 = 0x91cf673b965a7875;
+
+impl CountHash {
+    pub fn register(registry: &FunctionRegistry) {
+        registry.register_aggr(CountHash::udf_impl());
+    }
+
+    pub fn udf_impl() -> AggregateUDF {
+        AggregateUDF::new_from_impl(CountHash {
+            signature: Signature::one_of(
+                vec![TypeSignature::VariadicAny, TypeSignature::Nullary],
+                Volatility::Immutable,
+            ),
+        })
+    }
+}
+
+#[derive(Debug, Clone)]
+pub struct CountHash {
+    signature: Signature,
+}
+
+impl AggregateUDFImpl for CountHash {
+    fn as_any(&self) -> &dyn std::any::Any {
+        self
+    }
+
+    fn name(&self) -> &str {
+        "count_hash"
+    }
+
+    fn signature(&self) -> &Signature {
+        &self.signature
+    }
+
+    fn return_type(&self, _arg_types: &[DataType]) -> Result<DataType> {
+        Ok(DataType::Int64)
+    }
+
+    fn is_nullable(&self) -> bool {
+        false
+    }
+
+    fn state_fields(&self, args: StateFieldsArgs) -> Result<Vec<Field>> {
+        Ok(vec![Field::new_list(
+            format_state_name(args.name, "count_hash"),
+            Field::new_list_field(DataType::UInt64, true),
+            // For count_hash accumulator, null list item stands for an
+            // empty value set (i.e., all NULL value so far for that group).
+            true,
+        )])
+    }
+
+    fn accumulator(&self, acc_args: AccumulatorArgs) -> Result<Box<dyn Accumulator>> {
+        if acc_args.exprs.len() > 1 {
+            return not_impl_err!("count_hash with multiple arguments");
+        }
+
+        Ok(Box::new(CountHashAccumulator {
+            values: HashSet::default(),
+            random_state: RandomState::with_seeds(
+                RANDOM_SEED_0,
+                RANDOM_SEED_1,
+                RANDOM_SEED_2,
+                RANDOM_SEED_3,
+            ),
+            batch_hashes: vec![],
+        }))
+    }
+
+    fn aliases(&self) -> &[String] {
+        &[]
+    }
+
+    fn groups_accumulator_supported(&self, _args: AccumulatorArgs) -> bool {
+        true
+    }
+
+    fn create_groups_accumulator(
+        &self,
+        args: AccumulatorArgs,
+    ) -> Result<Box<dyn GroupsAccumulator>> {
+        if args.exprs.len() > 1 {
+            return not_impl_err!("count_hash with multiple arguments");
+        }
+
+        Ok(Box::new(CountHashGroupAccumulator::new()))
+    }
+
+    fn reverse_expr(&self) -> ReversedUDAF {
+        ReversedUDAF::Identical
+    }
+
+    fn order_sensitivity(&self) -> AggregateOrderSensitivity {
+        AggregateOrderSensitivity::Insensitive
+    }
+
+    fn default_value(&self, _data_type: &DataType) -> Result<ScalarValue> {
+        Ok(ScalarValue::Int64(Some(0)))
+    }
+
+    fn set_monotonicity(&self, _data_type: &DataType) -> SetMonotonicity {
+        SetMonotonicity::Increasing
+    }
+}
+
+/// GroupsAccumulator for `count_hash` aggregate function
+#[derive(Debug)]
+pub struct CountHashGroupAccumulator {
+    /// One HashSet per group to track distinct values
+    distinct_sets: Vec<HashSet<HashValueType, RandomState>>,
+    random_state: RandomState,
+    batch_hashes: Vec<HashValueType>,
+}
+
+impl Default for CountHashGroupAccumulator {
+    fn default() -> Self {
+        Self::new()
+    }
+}
+
+impl CountHashGroupAccumulator {
+    pub fn new() -> Self {
+        Self {
+            distinct_sets: vec![],
+            random_state: RandomState::with_seeds(
+                RANDOM_SEED_0,
+                RANDOM_SEED_1,
+                RANDOM_SEED_2,
+                RANDOM_SEED_3,
+            ),
+            batch_hashes: vec![],
+        }
+    }
+
+    fn ensure_sets(&mut self, total_num_groups: usize) {
+        if self.distinct_sets.len() < total_num_groups {
+            self.distinct_sets
+                .resize_with(total_num_groups, HashSet::default);
+        }
+    }
+}
+
+impl GroupsAccumulator for CountHashGroupAccumulator {
+    fn update_batch(
+        &mut self,
+        values: &[ArrayRef],
+        group_indices: &[usize],
+        opt_filter: Option<&BooleanArray>,
+        total_num_groups: usize,
+    ) -> Result<()> {
+        assert_eq!(values.len(), 1, "count_hash expects a single argument");
+        self.ensure_sets(total_num_groups);
+
+        let array = &values[0];
+        self.batch_hashes.clear();
+        self.batch_hashes.resize(array.len(), 0);
+        let hashes = create_hashes(
+            &[ArrayRef::clone(array)],
+            &self.random_state,
+            &mut self.batch_hashes,
+        )?;
+
+        // Use a pattern similar to accumulate_indices to process rows
+        // that are not null and pass the filter
+        let nulls = array.logical_nulls();
+
+        match (nulls.as_ref(), opt_filter) {
+            (None, None) => {
+                // No nulls, no filter - process all rows
+                for (row_idx, &group_idx) in group_indices.iter().enumerate() {
+                    self.distinct_sets[group_idx].insert(hashes[row_idx]);
+                }
+            }
+            (Some(nulls), None) => {
+                // Has nulls, no filter
+                for (row_idx, (&group_idx, is_valid)) in
+                    group_indices.iter().zip(nulls.iter()).enumerate()
+                {
+                    if is_valid {
+                        self.distinct_sets[group_idx].insert(hashes[row_idx]);
+                    }
+                }
+            }
+            (None, Some(filter)) => {
+                // No nulls, has filter
+                for (row_idx, (&group_idx, filter_value)) in
+                    group_indices.iter().zip(filter.iter()).enumerate()
+                {
+                    if let Some(true) = filter_value {
+                        self.distinct_sets[group_idx].insert(hashes[row_idx]);
+                    }
+                }
+            }
+            (Some(nulls), Some(filter)) => {
+                // Has nulls and filter
+                let iter = filter
+                    .iter()
+                    .zip(group_indices.iter())
+                    .zip(nulls.iter())
+                    .enumerate();
+
+                for (row_idx, ((filter_value, &group_idx), is_valid)) in iter {
+                    if is_valid && filter_value == Some(true) {
+                        self.distinct_sets[group_idx].insert(hashes[row_idx]);
+                    }
+                }
+            }
+        }
+
+        Ok(())
+    }
+
+    fn evaluate(&mut self, emit_to: EmitTo) -> Result<ArrayRef> {
+        let distinct_sets: Vec<HashSet<u64, RandomState>> =
+            emit_to.take_needed(&mut self.distinct_sets);
+
+        let counts = distinct_sets
+            .iter()
+            .map(|set| set.len() as i64)
+            .collect::<Vec<_>>();
+        Ok(Arc::new(Int64Array::from(counts)))
+    }
+
+    fn merge_batch(
+        &mut self,
+        values: &[ArrayRef],
+        group_indices: &[usize],
+        _opt_filter: Option<&BooleanArray>,
+        total_num_groups: usize,
+    ) -> Result<()> {
+        assert_eq!(
+            values.len(),
+            1,
+            "count_hash merge expects a single state array"
+        );
+        self.ensure_sets(total_num_groups);
+
+        let list_array = as_list_array(&values[0])?;
+
+        // For each group in the incoming batch
+        for (i, &group_idx) in group_indices.iter().enumerate() {
+            if i < list_array.len() {
+                let inner_array = list_array.value(i);
+                let inner_array = inner_array.as_any().downcast_ref::<UInt64Array>().unwrap();
+                // Add each value to our set for this group
+                for j in 0..inner_array.len() {
+                    if !inner_array.is_null(j) {
+                        self.distinct_sets[group_idx].insert(inner_array.value(j));
+                    }
+                }
+            }
+        }
+
+        Ok(())
+    }
+
+    fn state(&mut self, emit_to: EmitTo) -> Result<Vec<ArrayRef>> {
+        let distinct_sets: Vec<HashSet<u64, RandomState>> =
+            emit_to.take_needed(&mut self.distinct_sets);
+
+        let mut offsets = Vec::with_capacity(distinct_sets.len() + 1);
+        offsets.push(0);
+        let mut curr_len = 0i32;
+
+        let mut value_iter = distinct_sets
+            .into_iter()
+            .flat_map(|set| {
+                // build offset
+                curr_len += set.len() as i32;
+                offsets.push(curr_len);
+                // convert into iter
+                set.into_iter()
+            })
+            .peekable();
+        let data_array: ArrayRef = if value_iter.peek().is_none() {
+            arrow::array::new_empty_array(&DataType::UInt64) as _
+        } else {
+            Arc::new(UInt64Array::from_iter_values(value_iter))
+        };
+        let offset_buffer = OffsetBuffer::new(ScalarBuffer::from(offsets));
+
+        let list_array = ListArray::new(
+            Arc::new(Field::new_list_field(DataType::UInt64, true)),
+            offset_buffer,
+            data_array,
+            None,
+        );
+
+        Ok(vec![Arc::new(list_array) as _])
+    }
+
+    fn convert_to_state(
+        &self,
+        values: &[ArrayRef],
+        opt_filter: Option<&BooleanArray>,
+    ) -> Result<Vec<ArrayRef>> {
+        // For a single hash value per row, create a list array with that value
+        assert_eq!(values.len(), 1, "count_hash expects a single argument");
+        let values = ArrayRef::clone(&values[0]);
+
+        let offsets = OffsetBuffer::new(ScalarBuffer::from_iter(0..values.len() as i32 + 1));
+        let nulls = filtered_null_mask(opt_filter, &values);
+        let list_array = ListArray::new(
+            Arc::new(Field::new_list_field(DataType::UInt64, true)),
+            offsets,
+            values,
+            nulls,
+        );
+
+        Ok(vec![Arc::new(list_array)])
+    }
+
+    fn supports_convert_to_state(&self) -> bool {
+        true
+    }
+
+    fn size(&self) -> usize {
+        // Base size of the struct
+        let mut size = size_of::<Self>();
+
+        // Size of the vector holding the HashSets
+        size += size_of::<Vec<HashSet<HashValueType, RandomState>>>()
+            + self.distinct_sets.capacity() * size_of::<HashSet<HashValueType, RandomState>>();
+
+        // Estimate HashSet contents size more efficiently
+        // Instead of iterating through all values which is expensive, use an approximation
+        for set in &self.distinct_sets {
+            // Base size of the HashSet
+            size += set.capacity() * size_of::<HashValueType>();
+        }
+
+        size
+    }
+}
+
+#[derive(Debug)]
+struct CountHashAccumulator {
+    values: HashSet<HashValueType, RandomState>,
+    random_state: RandomState,
+    batch_hashes: Vec<HashValueType>,
+}
+
+impl CountHashAccumulator {
+    // calculating the size for fixed length values, taking first batch size *
+    // number of batches.
+    fn fixed_size(&self) -> usize {
+        size_of_val(self) + (size_of::<HashValueType>() * self.values.capacity())
+    }
+}
+
+impl Accumulator for CountHashAccumulator {
+    /// Returns the distinct values seen so far as (one element) ListArray.
+    fn state(&mut self) -> Result<Vec<ScalarValue>> {
+        let values = self.values.iter().cloned().collect::<Vec<_>>();
+        let arr = Arc::new(UInt64Array::from(values)) as _;
+        let list_scalar = SingleRowListArrayBuilder::new(arr).build_list_scalar();
+        Ok(vec![list_scalar])
+    }
+
+    fn update_batch(&mut self, values: &[ArrayRef]) -> Result<()> {
+        if values.is_empty() {
+            return Ok(());
+        }
+
+        let arr = &values[0];
+        if arr.data_type() == &DataType::Null {
+            return Ok(());
+        }
+
+        self.batch_hashes.clear();
+        self.batch_hashes.resize(arr.len(), 0);
+        let hashes = create_hashes(
+            &[ArrayRef::clone(arr)],
+            &self.random_state,
+            &mut self.batch_hashes,
+        )?;
+        for hash in hashes.as_slice() {
+            self.values.insert(*hash);
+        }
+        Ok(())
+    }
+
+    /// Merges multiple sets of distinct values into the current set.
+    ///
+    /// The input to this function is a `ListArray` with **multiple** rows,
+    /// where each row contains the values from a partial aggregate's phase (e.g.
+    /// the result of calling `Self::state` on multiple accumulators).
+    fn merge_batch(&mut self, states: &[ArrayRef]) -> Result<()> {
+        if states.is_empty() {
+            return Ok(());
+        }
+        assert_eq!(states.len(), 1, "array_agg states must be singleton!");
+        let array = &states[0];
+        let list_array = array.as_list::<i32>();
+        for inner_array in list_array.iter() {
+            let Some(inner_array) = inner_array else {
+                return internal_err!(
+                    "Intermediate results of count_hash should always be non null"
+                );
+            };
+            let hash_array = inner_array.as_any().downcast_ref::<UInt64Array>().unwrap();
+            for i in 0..hash_array.len() {
+                self.values.insert(hash_array.value(i));
+            }
+        }
+        Ok(())
+    }
+
+    fn evaluate(&mut self) -> Result<ScalarValue> {
+        Ok(ScalarValue::Int64(Some(self.values.len() as i64)))
+    }
+
+    fn size(&self) -> usize {
+        self.fixed_size()
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use datatypes::arrow::array::{Array, BooleanArray, Int32Array, Int64Array};
+
+    use super::*;
+
+    fn create_test_accumulator() -> CountHashAccumulator {
+        CountHashAccumulator {
+            values: HashSet::default(),
+            random_state: RandomState::with_seeds(
+                RANDOM_SEED_0,
+                RANDOM_SEED_1,
+                RANDOM_SEED_2,
+                RANDOM_SEED_3,
+            ),
+            batch_hashes: vec![],
+        }
+    }
+
+    #[test]
+    fn test_count_hash_accumulator() -> Result<()> {
+        let mut acc = create_test_accumulator();
+
+        // Test with some data
+        let array = Arc::new(Int32Array::from(vec![
+            Some(1),
+            Some(2),
+            Some(3),
+            Some(1),
+            Some(2),
+            None,
+        ])) as ArrayRef;
+        acc.update_batch(&[array])?;
+        let result = acc.evaluate()?;
+        assert_eq!(result, ScalarValue::Int64(Some(4)));
+
+        // Test with empty data
+        let mut acc = create_test_accumulator();
+        let array = Arc::new(Int32Array::from(vec![] as Vec<Option<i32>>)) as ArrayRef;
+        acc.update_batch(&[array])?;
+        let result = acc.evaluate()?;
+        assert_eq!(result, ScalarValue::Int64(Some(0)));
+
+        // Test with only nulls
+        let mut acc = create_test_accumulator();
+        let array = Arc::new(Int32Array::from(vec![None, None, None])) as ArrayRef;
+        acc.update_batch(&[array])?;
+        let result = acc.evaluate()?;
+        assert_eq!(result, ScalarValue::Int64(Some(1)));
+
+        Ok(())
+    }
+
+    #[test]
+    fn test_count_hash_accumulator_merge() -> Result<()> {
+        // Accumulator 1
+        let mut acc1 = create_test_accumulator();
+        let array1 = Arc::new(Int32Array::from(vec![Some(1), Some(2), Some(3)])) as ArrayRef;
+        acc1.update_batch(&[array1])?;
+        let state1 = acc1.state()?;
+
+        // Accumulator 2
+        let mut acc2 = create_test_accumulator();
+        let array2 = Arc::new(Int32Array::from(vec![Some(3), Some(4), Some(5)])) as ArrayRef;
+        acc2.update_batch(&[array2])?;
+        let state2 = acc2.state()?;
+
+        // Merge state1 and state2 into a new accumulator
+        let mut acc_merged = create_test_accumulator();
+        let state_array1 = state1[0].to_array()?;
+        let state_array2 = state2[0].to_array()?;
+
+        acc_merged.merge_batch(&[state_array1])?;
+        acc_merged.merge_batch(&[state_array2])?;
+
+        let result = acc_merged.evaluate()?;
+        // Distinct values are {1, 2, 3, 4, 5}, so count is 5
+        assert_eq!(result, ScalarValue::Int64(Some(5)));
+
+        Ok(())
+    }
+
+    fn create_test_group_accumulator() -> CountHashGroupAccumulator {
+        CountHashGroupAccumulator::new()
+    }
+
+    #[test]
+    fn test_count_hash_group_accumulator() -> Result<()> {
+        let mut acc = create_test_group_accumulator();
+        let values = Arc::new(Int32Array::from(vec![1, 2, 1, 3, 2, 4, 5])) as ArrayRef;
+        let group_indices = vec![0, 1, 0, 0, 1, 2, 0];
+        let total_num_groups = 3;
+
+        acc.update_batch(&[values], &group_indices, None, total_num_groups)?;
+
+        let result_array = acc.evaluate(EmitTo::All)?;
+        let result = result_array.as_any().downcast_ref::<Int64Array>().unwrap();
+
+        // Group 0: {1, 3, 5} -> 3
+        // Group 1: {2} -> 1
+        // Group 2: {4} -> 1
+        assert_eq!(result.value(0), 3);
+        assert_eq!(result.value(1), 1);
+        assert_eq!(result.value(2), 1);
+
+        Ok(())
+    }
+
+    #[test]
+    fn test_count_hash_group_accumulator_with_filter() -> Result<()> {
+        let mut acc = create_test_group_accumulator();
+        let values = Arc::new(Int32Array::from(vec![1, 2, 3, 4, 5, 6])) as ArrayRef;
+        let group_indices = vec![0, 0, 1, 1, 2, 2];
+        let filter = BooleanArray::from(vec![true, false, true, true, false, true]);
+        let total_num_groups = 3;
+
+        acc.update_batch(&[values], &group_indices, Some(&filter), total_num_groups)?;
+
+        let result_array = acc.evaluate(EmitTo::All)?;
+        let result = result_array.as_any().downcast_ref::<Int64Array>().unwrap();
+
+        // Group 0: {1} (2 is filtered out) -> 1
+        // Group 1: {3, 4} -> 2
+        // Group 2: {6} (5 is filtered out) -> 1
+        assert_eq!(result.value(0), 1);
+        assert_eq!(result.value(1), 2);
+        assert_eq!(result.value(2), 1);
+
+        Ok(())
+    }
+
+    #[test]
+    fn test_count_hash_group_accumulator_merge() -> Result<()> {
+        // Accumulator 1
+        let mut acc1 = create_test_group_accumulator();
+        let values1 = Arc::new(Int32Array::from(vec![1, 2, 3, 4])) as ArrayRef;
+        let group_indices1 = vec![0, 0, 1, 1];
+        acc1.update_batch(&[values1], &group_indices1, None, 2)?;
+        // acc1 state: group 0 -> {1, 2}, group 1 -> {3, 4}
+        let state1 = acc1.state(EmitTo::All)?;
+
+        // Accumulator 2
+        let mut acc2 = create_test_group_accumulator();
+        let values2 = Arc::new(Int32Array::from(vec![5, 6, 1, 3])) as ArrayRef;
+        // Merge into different group indices
+        let group_indices2 = vec![2, 2, 0, 1];
+        acc2.update_batch(&[values2], &group_indices2, None, 3)?;
+        // acc2 state: group 0 -> {1}, group 1 -> {3}, group 2 -> {5, 6}
+
+        // Merge state from acc1 into acc2
+        // We will merge acc1's group 0 into acc2's group 0
+        // and acc1's group 1 into acc2's group 2
+        let merge_group_indices = vec![0, 2];
+        acc2.merge_batch(&state1, &merge_group_indices, None, 3)?;
+
+        let result_array = acc2.evaluate(EmitTo::All)?;
+        let result = result_array.as_any().downcast_ref::<Int64Array>().unwrap();
+
+        // Final state of acc2:
+        // Group 0: {1} U {1, 2} -> {1, 2}, count = 2
+        // Group 1: {3}, count = 1
+        // Group 2: {5, 6} U {3, 4} -> {3, 4, 5, 6}, count = 4
+        assert_eq!(result.value(0), 2);
+        assert_eq!(result.value(1), 1);
+        assert_eq!(result.value(2), 4);
+
+        Ok(())
+    }
+
+    #[test]
+    fn test_size() {
+        let acc = create_test_group_accumulator();
+        // Just test it doesn't crash and returns a value.
+        assert!(acc.size() > 0);
+    }
+}

src/common/function/src/function_registry.rs

@@ -21,6 +21,7 @@ use once_cell::sync::Lazy;
 
 use crate::admin::AdminFunction;
 use crate::aggrs::approximate::ApproximateFunction;
+use crate::aggrs::count_hash::CountHash;
 use crate::aggrs::vector::VectorFunction as VectorAggrFunction;
 use crate::function::{AsyncFunctionRef, Function, FunctionRef};
 use crate::function_factory::ScalarFunctionFactory;
@@ -144,6 +145,9 @@ pub static FUNCTION_REGISTRY: Lazy<Arc<FunctionRegistry>> = Lazy::new(|| {
     // Approximate functions
     ApproximateFunction::register(&function_registry);
 
+    // CountHash function
+    CountHash::register(&function_registry);
+
     Arc::new(function_registry)
 });
 

src/common/function/src/scalars/math.rs

@@ -14,7 +14,6 @@
 
 pub mod clamp;
 mod modulo;
-mod pow;
 mod rate;
 
 use std::fmt;
@@ -26,7 +25,6 @@ use datafusion::error::DataFusionError;
 use datafusion::logical_expr::Volatility;
 use datatypes::prelude::ConcreteDataType;
 use datatypes::vectors::VectorRef;
-pub use pow::PowFunction;
 pub use rate::RateFunction;
 use snafu::ResultExt;
 
@@ -39,7 +37,6 @@ pub(crate) struct MathFunction;
 impl MathFunction {
     pub fn register(registry: &FunctionRegistry) {
         registry.register_scalar(ModuloFunction);
-        registry.register_scalar(PowFunction);
         registry.register_scalar(RateFunction);
         registry.register_scalar(RangeFunction);
         registry.register_scalar(ClampFunction);

src/common/function/src/scalars/math/clamp.rs

@@ -81,7 +81,8 @@ impl Function for ClampFunction {
             }
         );
         ensure!(
-            columns[1].len() == 1 && columns[2].len() == 1,
+            (columns[1].len() == 1 || columns[1].is_const())
+                && (columns[2].len() == 1 || columns[2].is_const()),
             InvalidFuncArgsSnafu {
                 err_msg: format!(
                     "The second and third args should be scalar, have: {:?}, {:?}",
@@ -204,7 +205,7 @@ impl Function for ClampMinFunction {
             }
         );
         ensure!(
-            columns[1].len() == 1,
+            columns[1].len() == 1 || columns[1].is_const(),
             InvalidFuncArgsSnafu {
                 err_msg: format!(
                     "The second arg (min) should be scalar, have: {:?}",
@@ -292,7 +293,7 @@ impl Function for ClampMaxFunction {
             }
         );
         ensure!(
-            columns[1].len() == 1,
+            columns[1].len() == 1 || columns[1].is_const(),
             InvalidFuncArgsSnafu {
                 err_msg: format!(
                     "The second arg (max) should be scalar, have: {:?}",

src/common/function/src/scalars/math/pow.rs

@@ -1,120 +0,0 @@
-// Copyright 2023 Greptime Team
-//
-// Licensed under the Apache License, Version 2.0 (the "License");
-// you may not use this file except in compliance with the License.
-// You may obtain a copy of the License at
-//
-//     http://www.apache.org/licenses/LICENSE-2.0
-//
-// Unless required by applicable law or agreed to in writing, software
-// distributed under the License is distributed on an "AS IS" BASIS,
-// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-// See the License for the specific language governing permissions and
-// limitations under the License.
-
-use std::fmt;
-use std::sync::Arc;
-
-use common_query::error::Result;
-use common_query::prelude::{Signature, Volatility};
-use datatypes::data_type::DataType;
-use datatypes::prelude::ConcreteDataType;
-use datatypes::types::LogicalPrimitiveType;
-use datatypes::vectors::VectorRef;
-use datatypes::with_match_primitive_type_id;
-use num::traits::Pow;
-use num_traits::AsPrimitive;
-
-use crate::function::{Function, FunctionContext};
-use crate::scalars::expression::{scalar_binary_op, EvalContext};
-
-#[derive(Clone, Debug, Default)]
-pub struct PowFunction;
-
-impl Function for PowFunction {
-    fn name(&self) -> &str {
-        "pow"
-    }
-
-    fn return_type(&self, _input_types: &[ConcreteDataType]) -> Result<ConcreteDataType> {
-        Ok(ConcreteDataType::float64_datatype())
-    }
-
-    fn signature(&self) -> Signature {
-        Signature::uniform(2, ConcreteDataType::numerics(), Volatility::Immutable)
-    }
-
-    fn eval(&self, _func_ctx: &FunctionContext, columns: &[VectorRef]) -> Result<VectorRef> {
-        with_match_primitive_type_id!(columns[0].data_type().logical_type_id(), |$S| {
-            with_match_primitive_type_id!(columns[1].data_type().logical_type_id(), |$T| {
-                let col = scalar_binary_op::<<$S as LogicalPrimitiveType>::Native, <$T as LogicalPrimitiveType>::Native, f64, _>(&columns[0], &columns[1], scalar_pow, &mut EvalContext::default())?;
-                Ok(Arc::new(col))
-            },{
-                unreachable!()
-            })
-        },{
-            unreachable!()
-        })
-    }
-}
-
-#[inline]
-fn scalar_pow<S, T>(value: Option<S>, base: Option<T>, _ctx: &mut EvalContext) -> Option<f64>
-where
-    S: AsPrimitive<f64>,
-    T: AsPrimitive<f64>,
-{
-    match (value, base) {
-        (Some(value), Some(base)) => Some(value.as_().pow(base.as_())),
-        _ => None,
-    }
-}
-
-impl fmt::Display for PowFunction {
-    fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
-        write!(f, "POW")
-    }
-}
-
-#[cfg(test)]
-mod tests {
-    use common_query::prelude::TypeSignature;
-    use datatypes::value::Value;
-    use datatypes::vectors::{Float32Vector, Int8Vector};
-
-    use super::*;
-    use crate::function::FunctionContext;
-    #[test]
-    fn test_pow_function() {
-        let pow = PowFunction;
-
-        assert_eq!("pow", pow.name());
-        assert_eq!(
-            ConcreteDataType::float64_datatype(),
-            pow.return_type(&[]).unwrap()
-        );
-
-        assert!(matches!(pow.signature(),
-                         Signature {
-                             type_signature: TypeSignature::Uniform(2, valid_types),
-                             volatility: Volatility::Immutable
-                         } if  valid_types == ConcreteDataType::numerics()
-        ));
-
-        let values = vec![1.0, 2.0, 3.0];
-        let bases = vec![0i8, -1i8, 3i8];
-
-        let args: Vec<VectorRef> = vec![
-            Arc::new(Float32Vector::from_vec(values.clone())),
-            Arc::new(Int8Vector::from_vec(bases.clone())),
-        ];
-
-        let vector = pow.eval(&FunctionContext::default(), &args).unwrap();
-        assert_eq!(3, vector.len());
-
-        for i in 0..3 {
-            let p: f64 = (values[i] as f64).pow(bases[i] as f64);
-            assert!(matches!(vector.get(i), Value::Float64(v) if v == p));
-        }
-    }
-}

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

@@ -12,8 +12,8 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
+use std::fmt;
 use std::sync::Arc;
-use std::{env, fmt};
 
 use common_query::error::Result;
 use common_query::prelude::{Signature, Volatility};
@@ -47,7 +47,7 @@ impl Function for PGVersionFunction {
     fn eval(&self, _func_ctx: &FunctionContext, _columns: &[VectorRef]) -> Result<VectorRef> {
         let result = StringVector::from(vec![format!(
             "PostgreSQL 16.3 GreptimeDB {}",
-            env!("CARGO_PKG_VERSION")
+            common_version::version()
         )]);
         Ok(Arc::new(result))
     }