chore: update all arm64 builders (#5215)

Revert "feat: bump opendal and switch prometheus layer to the upstream impl (#5179)"

This reverts commit 422d18da8b.

.github/actions/setup-kafka-cluster/action.yml

@@ -18,6 +18,8 @@ runs:
         --set controller.replicaCount=${{ inputs.controller-replicas }} \
         --set controller.resources.requests.cpu=50m \
         --set controller.resources.requests.memory=128Mi \
+        --set controller.resources.limits.cpu=2000m \
+        --set controller.resources.limits.memory=2Gi \
         --set listeners.controller.protocol=PLAINTEXT \
         --set listeners.client.protocol=PLAINTEXT \
         --create-namespace \

.github/cargo-blacklist.txt

@@ -1,2 +1,3 @@
 native-tls
 openssl
+aws-lc-sys

.github/pull_request_template.md

@@ -4,7 +4,8 @@ I hereby agree to the terms of the [GreptimeDB CLA](https://github.com/GreptimeT
 
 ## What's changed and what's your intention?
 
-__!!! DO NOT LEAVE THIS BLOCK EMPTY !!!__
+<!--    
+ __!!! DO NOT LEAVE THIS BLOCK EMPTY !!!__
 
 Please explain IN DETAIL what the changes are in this PR and why they are needed:
 
@@ -12,9 +13,14 @@ Please explain IN DETAIL what the changes are in this PR and why they are needed
 - How does this PR work? Need a brief introduction for the changed logic (optional)
 - Describe clearly one logical change and avoid lazy messages (optional)
 - Describe any limitations of the current code (optional)
+- Describe if this PR will break **API or data compatibility**  (optional)
+-->
 
-## Checklist
+## PR Checklist
+Please convert it to a draft if some of the following conditions are not met.
 
 - [ ] I have written the necessary rustdoc comments.
 - [ ] I have added the necessary unit tests and integration tests.
 - [ ] This PR requires documentation updates.
+- [ ] API changes are backward compatible.
+- [ ] Schema or data changes are backward compatible.

.github/workflows/dev-build.yml

@@ -29,7 +29,7 @@ on:
       linux_arm64_runner:
         type: choice
         description: The runner uses to build linux-arm64 artifacts
-        default: ec2-c6g.4xlarge-arm64
+        default: ec2-c6g.8xlarge-arm64
         options:
           - ec2-c6g.xlarge-arm64 # 4C8G
           - ec2-c6g.2xlarge-arm64 # 8C16G

.github/workflows/develop.yml

@@ -269,13 +269,6 @@ jobs:
       - name: Install cargo-gc-bin
         shell: bash
         run: cargo install cargo-gc-bin
-      - name: Check aws-lc-sys will not build
-        shell: bash
-        run: |
-             if cargo tree -i aws-lc-sys -e features | grep -q aws-lc-sys; then
-               echo "Found aws-lc-sys, which has compilation problems on older gcc versions. Please replace it with ring until its building experience improves."
-               exit 1
-             fi
       - name: Build greptime bianry
         shell: bash
         # `cargo gc` will invoke `cargo build` with specified args
@@ -330,8 +323,6 @@ jobs:
         uses: ./.github/actions/setup-kafka-cluster
       - name: Setup Etcd cluser
         uses: ./.github/actions/setup-etcd-cluster
-      - name: Setup Postgres cluser
-        uses: ./.github/actions/setup-postgres-cluster
       # Prepares for fuzz tests
       - uses: arduino/setup-protoc@v3
         with:
@@ -481,8 +472,6 @@ jobs:
         uses: ./.github/actions/setup-kafka-cluster
       - name: Setup Etcd cluser
         uses: ./.github/actions/setup-etcd-cluster
-      - name: Setup Postgres cluser
-        uses: ./.github/actions/setup-postgres-cluster
       # Prepares for fuzz tests
       - uses: arduino/setup-protoc@v3
         with:

.github/workflows/nightly-build.yml

@@ -12,7 +12,7 @@ on:
       linux_amd64_runner:
         type: choice
         description: The runner uses to build linux-amd64 artifacts
-        default: ec2-c6i.2xlarge-amd64
+        default: ec2-c6i.4xlarge-amd64
         options:
           - ubuntu-20.04
           - ubuntu-20.04-8-cores
@@ -27,7 +27,7 @@ on:
       linux_arm64_runner:
         type: choice
         description: The runner uses to build linux-arm64 artifacts
-        default: ec2-c6g.2xlarge-arm64
+        default: ec2-c6g.8xlarge-arm64
         options:
           - ec2-c6g.xlarge-arm64 # 4C8G
           - ec2-c6g.2xlarge-arm64 # 8C16G

.github/workflows/nightly-ci.yml

@@ -114,6 +114,17 @@ jobs:
           GT_S3_REGION: ${{ vars.AWS_CI_TEST_BUCKET_REGION }}
           UNITTEST_LOG_DIR: "__unittest_logs"
 
+  cleanbuild-linux-nix:
+    runs-on: ubuntu-latest-8-cores
+    timeout-minutes: 60
+    needs: [coverage, fmt, clippy, check]
+    steps:
+      - uses: actions/checkout@v4
+      - uses: cachix/install-nix-action@v27
+        with:
+          nix_path: nixpkgs=channel:nixos-unstable
+      - run: nix-shell --pure --run "cargo build"
+
   check-status:
     name: Check status
     needs: [sqlness-test, sqlness-windows, test-on-windows]

.github/workflows/release.yml

@@ -31,7 +31,7 @@ on:
       linux_arm64_runner:
         type: choice
         description: The runner uses to build linux-arm64 artifacts
-        default: ec2-c6g.4xlarge-arm64
+        default: ec2-c6g.8xlarge-arm64
         options:
           - ubuntu-2204-32-cores-arm
           - ec2-c6g.xlarge-arm64 # 4C8G

.gitignore

@@ -47,6 +47,10 @@ benchmarks/data
 
 venv/
 
-# Fuzz tests 
+# Fuzz tests
 tests-fuzz/artifacts/
 tests-fuzz/corpus/
+
+# Nix
+.direnv
+.envrc

Cargo.toml

@@ -68,7 +68,7 @@ members = [
 resolver = "2"
 
 [workspace.package]
-version = "0.11.0"
+version = "0.11.1"
 edition = "2021"
 license = "Apache-2.0"
 
@@ -180,6 +180,7 @@ sysinfo = "0.30"
 # on branch v0.44.x
 sqlparser = { git = "https://github.com/GreptimeTeam/sqlparser-rs.git", rev = "54a267ac89c09b11c0c88934690530807185d3e7", features = [
     "visitor",
+    "serde",
 ] }
 strum = { version = "0.25", features = ["derive"] }
 tempfile = "3"

config/config.md

@@ -13,11 +13,11 @@
 | Key | Type | Default | Descriptions |
 | --- | -----| ------- | ----------- |
 | `mode` | String | `standalone` | The running mode of the datanode. It can be `standalone` or `distributed`. |
-| `enable_telemetry` | Bool | `true` | Enable telemetry to collect anonymous usage data. |
 | `default_timezone` | String | Unset | The default timezone of the server. |
 | `init_regions_in_background` | Bool | `false` | Initialize all regions in the background during the startup.<br/>By default, it provides services after all regions have been initialized. |
 | `init_regions_parallelism` | Integer | `16` | Parallelism of initializing regions. |
 | `max_concurrent_queries` | Integer | `0` | The maximum current queries allowed to be executed. Zero means unlimited. |
+| `enable_telemetry` | Bool | `true` | Enable telemetry to collect anonymous usage data. 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. |
@@ -61,9 +61,9 @@
 | `wal` | -- | -- | The WAL options. |
 | `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 | `256MB` | The size of the WAL segment file.<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 | `4GB` | The threshold of the WAL size 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 flush.<br/>**It's only used when the provider is `raft_engine`**. |
-| `wal.purge_interval` | String | `10m` | The interval 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.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`**. |
@@ -150,6 +150,7 @@
 | `region_engine.mito.inverted_index.intermediate_path` | String | `""` | Deprecated, use `region_engine.mito.index.aux_path` instead. |
 | `region_engine.mito.inverted_index.metadata_cache_size` | String | `64MiB` | Cache size for inverted index metadata. |
 | `region_engine.mito.inverted_index.content_cache_size` | String | `128MiB` | Cache size for inverted index content. |
+| `region_engine.mito.inverted_index.content_cache_page_size` | String | `8MiB` | Page size for inverted index content cache. |
 | `region_engine.mito.fulltext_index` | -- | -- | The options for full-text index in Mito engine. |
 | `region_engine.mito.fulltext_index.create_on_flush` | String | `auto` | Whether to create the index on flush.<br/>- `auto`: automatically (default)<br/>- `disable`: never |
 | `region_engine.mito.fulltext_index.create_on_compaction` | String | `auto` | Whether to create the index on compaction.<br/>- `auto`: automatically (default)<br/>- `disable`: never |
@@ -286,12 +287,12 @@
 | `bind_addr` | String | `127.0.0.1:3002` | The bind address of metasrv. |
 | `server_addr` | String | `127.0.0.1:3002` | The communication server address for frontend and datanode to connect to metasrv,  "127.0.0.1:3002" by default for localhost. |
 | `store_addrs` | Array | -- | Store server address default to etcd store. |
+| `store_key_prefix` | String | `""` | If it's not empty, the metasrv will store all data with this key prefix. |
+| `backend` | String | `EtcdStore` | The datastore for meta server. |
 | `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_telemetry` | Bool | `true` | Whether to enable greptimedb telemetry. |
-| `store_key_prefix` | String | `""` | If it's not empty, the metasrv will store all data with this key prefix. |
 | `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). |
-| `backend` | String | `EtcdStore` | The datastore for meta server. |
+| `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. |
@@ -356,7 +357,6 @@
 | `node_id` | Integer | Unset | The datanode identifier and should be unique in the cluster. |
 | `require_lease_before_startup` | Bool | `false` | Start services after regions have obtained leases.<br/>It will block the datanode start if it can't receive leases in the heartbeat from metasrv. |
 | `init_regions_in_background` | Bool | `false` | Initialize all regions in the background during the startup.<br/>By default, it provides services after all regions have been initialized. |
-| `enable_telemetry` | Bool | `true` | Enable telemetry to collect anonymous usage data. |
 | `init_regions_parallelism` | Integer | `16` | Parallelism of initializing regions. |
 | `max_concurrent_queries` | Integer | `0` | The maximum current queries allowed to be executed. Zero means unlimited. |
 | `rpc_addr` | String | Unset | Deprecated, use `grpc.addr` instead. |
@@ -364,6 +364,7 @@
 | `rpc_runtime_size` | Integer | Unset | Deprecated, use `grpc.runtime_size` instead. |
 | `rpc_max_recv_message_size` | String | Unset | Deprecated, use `grpc.rpc_max_recv_message_size` instead. |
 | `rpc_max_send_message_size` | String | Unset | Deprecated, use `grpc.rpc_max_send_message_size` instead. |
+| `enable_telemetry` | Bool | `true` | Enable telemetry to collect anonymous usage data. Enabled by default. |
 | `http` | -- | -- | The HTTP server options. |
 | `http.addr` | String | `127.0.0.1:4000` | The address to bind the HTTP server. |
 | `http.timeout` | String | `30s` | HTTP request timeout. Set to 0 to disable timeout. |
@@ -398,9 +399,9 @@
 | `wal` | -- | -- | The WAL options. |
 | `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 | `256MB` | The size of the WAL segment file.<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 | `4GB` | The threshold of the WAL size 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 flush.<br/>**It's only used when the provider is `raft_engine`**. |
-| `wal.purge_interval` | String | `10m` | The interval 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.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`**. |
@@ -475,6 +476,9 @@
 | `region_engine.mito.inverted_index.apply_on_query` | String | `auto` | Whether to apply the index on query<br/>- `auto`: automatically (default)<br/>- `disable`: never |
 | `region_engine.mito.inverted_index.mem_threshold_on_create` | String | `auto` | Memory threshold for performing an external sort during index creation.<br/>- `auto`: automatically determine the threshold based on the system memory size (default)<br/>- `unlimited`: no memory limit<br/>- `[size]` e.g. `64MB`: fixed memory threshold |
 | `region_engine.mito.inverted_index.intermediate_path` | String | `""` | Deprecated, use `region_engine.mito.index.aux_path` instead. |
+| `region_engine.mito.inverted_index.metadata_cache_size` | String | `64MiB` | Cache size for inverted index metadata. |
+| `region_engine.mito.inverted_index.content_cache_size` | String | `128MiB` | Cache size for inverted index content. |
+| `region_engine.mito.inverted_index.content_cache_page_size` | String | `8MiB` | Page size for inverted index content cache. |
 | `region_engine.mito.fulltext_index` | -- | -- | The options for full-text index in Mito engine. |
 | `region_engine.mito.fulltext_index.create_on_flush` | String | `auto` | Whether to create the index on flush.<br/>- `auto`: automatically (default)<br/>- `disable`: never |
 | `region_engine.mito.fulltext_index.create_on_compaction` | String | `auto` | Whether to create the index on compaction.<br/>- `auto`: automatically (default)<br/>- `disable`: never |

config/datanode.example.toml

@@ -13,9 +13,6 @@ require_lease_before_startup = false
 ## By default, it provides services after all regions have been initialized.
 init_regions_in_background = false
 
-## Enable telemetry to collect anonymous usage data.
-enable_telemetry = true
-
 ## Parallelism of initializing regions.
 init_regions_parallelism = 16
 
@@ -42,6 +39,8 @@ rpc_max_recv_message_size = "512MB"
 ## @toml2docs:none-default
 rpc_max_send_message_size = "512MB"
 
+## Enable telemetry to collect anonymous usage data. Enabled by default.
+#+ enable_telemetry = true
 
 ## The HTTP server options.
 [http]
@@ -143,15 +142,15 @@ dir = "/tmp/greptimedb/wal"
 
 ## The size of the WAL segment file.
 ## **It's only used when the provider is `raft_engine`**.
-file_size = "256MB"
+file_size = "128MB"
 
 ## The threshold of the WAL size to trigger a flush.
 ## **It's only used when the provider is `raft_engine`**.
-purge_threshold = "4GB"
+purge_threshold = "1GB"
 
 ## The interval to trigger a flush.
 ## **It's only used when the provider is `raft_engine`**.
-purge_interval = "10m"
+purge_interval = "1m"
 
 ## The read batch size.
 ## **It's only used when the provider is `raft_engine`**.
@@ -544,6 +543,15 @@ mem_threshold_on_create = "auto"
 ## Deprecated, use `region_engine.mito.index.aux_path` instead.
 intermediate_path = ""
 
+## Cache size for inverted index metadata.
+metadata_cache_size = "64MiB"
+
+## Cache size for inverted index content.
+content_cache_size = "128MiB"
+
+## Page size for inverted index content cache.
+content_cache_page_size = "8MiB"
+
 ## The options for full-text index in Mito engine.
 [region_engine.mito.fulltext_index]
 

config/metasrv.example.toml

@@ -10,6 +10,12 @@ server_addr = "127.0.0.1:3002"
 ## Store server address default to etcd store.
 store_addrs = ["127.0.0.1:2379"]
 
+## If it's not empty, the metasrv will store all data with this key prefix.
+store_key_prefix = ""
+
+## The datastore for meta server.
+backend = "EtcdStore"
+
 ## Datanode selector type.
 ## - `round_robin` (default value)
 ## - `lease_based`
@@ -20,20 +26,14 @@ selector = "round_robin"
 ## Store data in memory.
 use_memory_store = false
 
-## Whether to enable greptimedb telemetry.
-enable_telemetry = true
-
-## If it's not empty, the metasrv will store all data with this key prefix.
-store_key_prefix = ""
-
 ## Whether to enable region failover.
 ## This feature is only available on GreptimeDB running on cluster mode and
 ## - Using Remote WAL
 ## - Using shared storage (e.g., s3).
 enable_region_failover = false
 
-## The datastore for meta server.
+## Whether to enable greptimedb telemetry. Enabled by default.
-backend = "EtcdStore"
+#+ enable_telemetry = true
 
 ## The runtime options.
 #+ [runtime]

config/standalone.example.toml

@@ -1,9 +1,6 @@
 ## The running mode of the datanode. It can be `standalone` or `distributed`.
 mode = "standalone"
 
-## Enable telemetry to collect anonymous usage data.
-enable_telemetry = true
-
 ## The default timezone of the server.
 ## @toml2docs:none-default
 default_timezone = "UTC"
@@ -18,6 +15,9 @@ init_regions_parallelism = 16
 ## The maximum current queries allowed to be executed. Zero means unlimited.
 max_concurrent_queries = 0
 
+## Enable telemetry to collect anonymous usage data. Enabled by default.
+#+ enable_telemetry = true
+
 ## The runtime options.
 #+ [runtime]
 ## The number of threads to execute the runtime for global read operations.
@@ -147,15 +147,15 @@ dir = "/tmp/greptimedb/wal"
 
 ## The size of the WAL segment file.
 ## **It's only used when the provider is `raft_engine`**.
-file_size = "256MB"
+file_size = "128MB"
 
 ## The threshold of the WAL size to trigger a flush.
 ## **It's only used when the provider is `raft_engine`**.
-purge_threshold = "4GB"
+purge_threshold = "1GB"
 
 ## The interval to trigger a flush.
 ## **It's only used when the provider is `raft_engine`**.
-purge_interval = "10m"
+purge_interval = "1m"
 
 ## The read batch size.
 ## **It's only used when the provider is `raft_engine`**.
@@ -588,6 +588,9 @@ metadata_cache_size = "64MiB"
 ## Cache size for inverted index content.
 content_cache_size = "128MiB"
 
+## Page size for inverted index content cache.
+content_cache_page_size = "8MiB"
+
 ## The options for full-text index in Mito engine.
 [region_engine.mito.fulltext_index]
 

docker/dev-builder/ubuntu/Dockerfile

@@ -15,8 +15,8 @@ RUN apt-get update && \
 RUN apt-get update && DEBIAN_FRONTEND=noninteractive apt-get install -y \
     libssl-dev \
     tzdata \
-    protobuf-compiler \
     curl \
+    unzip \
     ca-certificates \
     git \
     build-essential \
@@ -24,6 +24,20 @@ RUN apt-get update && DEBIAN_FRONTEND=noninteractive apt-get install -y \
     python3.10 \
     python3.10-dev
 
+ARG TARGETPLATFORM
+RUN echo "target platform: $TARGETPLATFORM"
+
+# Install protobuf, because the one in the apt is too old (v3.12).
+RUN if [ "$TARGETPLATFORM" = "linux/arm64" ]; then \
+    curl -OL https://github.com/protocolbuffers/protobuf/releases/download/v29.1/protoc-29.1-linux-aarch_64.zip && \
+    unzip protoc-29.1-linux-aarch_64.zip -d protoc3; \
+elif [ "$TARGETPLATFORM" = "linux/amd64" ]; then \
+    curl -OL https://github.com/protocolbuffers/protobuf/releases/download/v29.1/protoc-29.1-linux-x86_64.zip && \
+    unzip protoc-29.1-linux-x86_64.zip -d protoc3; \
+fi
+RUN mv protoc3/bin/* /usr/local/bin/
+RUN mv protoc3/include/* /usr/local/include/
+
 # https://github.com/GreptimeTeam/greptimedb/actions/runs/10935485852/job/30357457188#step:3:7106
 # `aws-lc-sys` require gcc >= 10.3.0 to work, hence alias to use gcc-10
 RUN apt-get remove -y gcc-9 g++-9 cpp-9 && \
@@ -49,7 +63,7 @@ RUN apt-get -y purge python3.8 && \
 # wildcard here. However, that requires the git's config files and the submodules all owned by the very same user.
 # It's troublesome to do this since the dev build runs in Docker, which is under user "root"; while outside the Docker,
 # it can be a different user that have prepared the submodules.
-RUN git config --global --add safe.directory *
+RUN git config --global --add safe.directory '*'
 
 # Install Python dependencies.
 COPY $DOCKER_BUILD_ROOT/docker/python/requirements.txt /etc/greptime/requirements.txt

rust-toolchain.toml

@@ -1,2 +1,3 @@
 [toolchain]
 channel = "nightly-2024-10-19"
+components = ["rust-analyzer"]

scripts/check-snafu.py

@@ -58,8 +58,10 @@ def main():
         if not check_snafu_in_files(branch_name, other_rust_files)
     ]
 
-    for name in unused_snafu:
+    if unused_snafu:
-        print(name)
+        print("Unused error variants:")
+        for name in unused_snafu:
+            print(name)
 
     if unused_snafu:
         raise SystemExit(1)

shell.nix

@@ -0,0 +1,27 @@
+let
+  nixpkgs = fetchTarball "https://github.com/NixOS/nixpkgs/tarball/nixos-unstable";
+  fenix = import (fetchTarball "https://github.com/nix-community/fenix/archive/main.tar.gz") {};
+  pkgs = import nixpkgs { config = {}; overlays = []; };
+in
+
+pkgs.mkShell rec {
+  nativeBuildInputs = with pkgs; [
+    pkg-config
+    git
+    clang
+    gcc
+    protobuf
+    mold
+    (fenix.fromToolchainFile {
+      dir = ./.;
+    })
+    cargo-nextest
+    taplo
+  ];
+
+  buildInputs = with pkgs; [
+    libgit2
+  ];
+
+  LD_LIBRARY_PATH = pkgs.lib.makeLibraryPath buildInputs;
+}

src/api/src/v1/column_def.rs

@@ -16,7 +16,7 @@ use std::collections::HashMap;
 
 use datatypes::schema::{
     ColumnDefaultConstraint, ColumnSchema, FulltextAnalyzer, FulltextOptions, COMMENT_KEY,
-    FULLTEXT_KEY, INVERTED_INDEX_KEY,
+    FULLTEXT_KEY, INVERTED_INDEX_KEY, SKIPPING_INDEX_KEY,
 };
 use greptime_proto::v1::Analyzer;
 use snafu::ResultExt;
@@ -29,6 +29,8 @@ use crate::v1::{ColumnDef, ColumnOptions, SemanticType};
 const FULLTEXT_GRPC_KEY: &str = "fulltext";
 /// Key used to store inverted index options in gRPC column options.
 const INVERTED_INDEX_GRPC_KEY: &str = "inverted_index";
+/// Key used to store skip index options in gRPC column options.
+const SKIPPING_INDEX_GRPC_KEY: &str = "skipping_index";
 
 /// Tries to construct a `ColumnSchema` from the given  `ColumnDef`.
 pub fn try_as_column_schema(column_def: &ColumnDef) -> Result<ColumnSchema> {
@@ -60,6 +62,9 @@ pub fn try_as_column_schema(column_def: &ColumnDef) -> Result<ColumnSchema> {
         if let Some(inverted_index) = options.options.get(INVERTED_INDEX_GRPC_KEY) {
             metadata.insert(INVERTED_INDEX_KEY.to_string(), inverted_index.clone());
         }
+        if let Some(skipping_index) = options.options.get(SKIPPING_INDEX_GRPC_KEY) {
+            metadata.insert(SKIPPING_INDEX_KEY.to_string(), skipping_index.clone());
+        }
     }
 
     ColumnSchema::new(&column_def.name, data_type.into(), column_def.is_nullable)
@@ -84,6 +89,11 @@ pub fn options_from_column_schema(column_schema: &ColumnSchema) -> Option<Column
             .options
             .insert(INVERTED_INDEX_GRPC_KEY.to_string(), inverted_index.clone());
     }
+    if let Some(skipping_index) = column_schema.metadata().get(SKIPPING_INDEX_KEY) {
+        options
+            .options
+            .insert(SKIPPING_INDEX_GRPC_KEY.to_string(), skipping_index.clone());
+    }
 
     (!options.options.is_empty()).then_some(options)
 }

src/cache/Cargo.toml

@@ -11,4 +11,3 @@ common-macro.workspace = true
 common-meta.workspace = true
 moka.workspace = true
 snafu.workspace = true
-substrait.workspace = true

src/catalog/Cargo.toml

@@ -18,7 +18,6 @@ async-stream.workspace = true
 async-trait = "0.1"
 bytes.workspace = true
 common-catalog.workspace = true
-common-config.workspace = true
 common-error.workspace = true
 common-macro.workspace = true
 common-meta.workspace = true
@@ -58,7 +57,5 @@ catalog = { workspace = true, features = ["testing"] }
 chrono.workspace = true
 common-meta = { workspace = true, features = ["testing"] }
 common-query = { workspace = true, features = ["testing"] }
-common-test-util.workspace = true
-log-store.workspace = true
 object-store.workspace = true
 tokio.workspace = true

src/catalog/src/error.rs

@@ -64,6 +64,13 @@ pub enum Error {
         source: BoxedError,
     },
 
+    #[snafu(display("Failed to list flow stats"))]
+    ListFlowStats {
+        #[snafu(implicit)]
+        location: Location,
+        source: BoxedError,
+    },
+
     #[snafu(display("Failed to list flows in catalog {catalog}"))]
     ListFlows {
         #[snafu(implicit)]
@@ -326,6 +333,7 @@ impl ErrorExt for Error {
             | Error::ListSchemas { source, .. }
             | Error::ListTables { source, .. }
             | Error::ListFlows { source, .. }
+            | Error::ListFlowStats { source, .. }
             | Error::ListProcedures { source, .. }
             | Error::ListRegionStats { source, .. }
             | Error::ConvertProtoData { source, .. } => source.status_code(),

src/catalog/src/information_extension.rs

@@ -17,6 +17,7 @@ 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::rpc::procedure;
 use common_procedure::{ProcedureInfo, ProcedureState};
 use meta_client::MetaClientRef;
@@ -89,4 +90,12 @@ impl InformationExtension for DistributedInformationExtension {
             .map_err(BoxedError::new)
             .context(error::ListRegionStatsSnafu)
     }
+
+    async fn flow_stats(&self) -> std::result::Result<Option<FlowStat>, Self::Error> {
+        self.meta_client
+            .list_flow_stats()
+            .await
+            .map_err(BoxedError::new)
+            .context(crate::error::ListFlowStatsSnafu)
+    }
 }

src/catalog/src/system_schema/information_schema.rs

@@ -35,6 +35,7 @@ use common_catalog::consts::{self, DEFAULT_CATALOG_NAME, INFORMATION_SCHEMA_NAME
 use common_error::ext::ErrorExt;
 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_procedure::ProcedureInfo;
 use common_recordbatch::SendableRecordBatchStream;
@@ -192,6 +193,7 @@ impl SystemSchemaProviderInner for InformationSchemaProvider {
             )) as _),
             FLOWS => Some(Arc::new(InformationSchemaFlows::new(
                 self.catalog_name.clone(),
+                self.catalog_manager.clone(),
                 self.flow_metadata_manager.clone(),
             )) as _),
             PROCEDURE_INFO => Some(
@@ -338,6 +340,9 @@ pub trait InformationExtension {
 
     /// Gets the region statistics.
     async fn region_stats(&self) -> std::result::Result<Vec<RegionStat>, Self::Error>;
+
+    /// Get the flow statistics. If no flownode is available, return `None`.
+    async fn flow_stats(&self) -> std::result::Result<Option<FlowStat>, Self::Error>;
 }
 
 pub struct NoopInformationExtension;
@@ -357,4 +362,8 @@ impl InformationExtension for NoopInformationExtension {
     async fn region_stats(&self) -> std::result::Result<Vec<RegionStat>, Self::Error> {
         Ok(vec![])
     }
+
+    async fn flow_stats(&self) -> std::result::Result<Option<FlowStat>, Self::Error> {
+        Ok(None)
+    }
 }

src/catalog/src/system_schema/information_schema/flows.rs

@@ -12,11 +12,12 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
-use std::sync::Arc;
+use std::sync::{Arc, Weak};
 
 use common_catalog::consts::INFORMATION_SCHEMA_FLOW_TABLE_ID;
 use common_error::ext::BoxedError;
 use common_meta::key::flow::flow_info::FlowInfoValue;
+use common_meta::key::flow::flow_state::FlowStat;
 use common_meta::key::flow::FlowMetadataManager;
 use common_meta::key::FlowId;
 use common_recordbatch::adapter::RecordBatchStreamAdapter;
@@ -28,7 +29,9 @@ use datatypes::prelude::ConcreteDataType as CDT;
 use datatypes::scalars::ScalarVectorBuilder;
 use datatypes::schema::{ColumnSchema, Schema, SchemaRef};
 use datatypes::value::Value;
-use datatypes::vectors::{Int64VectorBuilder, StringVectorBuilder, UInt32VectorBuilder, VectorRef};
+use datatypes::vectors::{
+    Int64VectorBuilder, StringVectorBuilder, UInt32VectorBuilder, UInt64VectorBuilder, VectorRef,
+};
 use futures::TryStreamExt;
 use snafu::{OptionExt, ResultExt};
 use store_api::storage::{ScanRequest, TableId};
@@ -38,6 +41,8 @@ use crate::error::{
 };
 use crate::information_schema::{Predicates, FLOWS};
 use crate::system_schema::information_schema::InformationTable;
+use crate::system_schema::utils;
+use crate::CatalogManager;
 
 const INIT_CAPACITY: usize = 42;
 
@@ -45,6 +50,7 @@ const INIT_CAPACITY: usize = 42;
 // pk is (flow_name, flow_id, table_catalog)
 pub const FLOW_NAME: &str = "flow_name";
 pub const FLOW_ID: &str = "flow_id";
+pub const STATE_SIZE: &str = "state_size";
 pub const TABLE_CATALOG: &str = "table_catalog";
 pub const FLOW_DEFINITION: &str = "flow_definition";
 pub const COMMENT: &str = "comment";
@@ -55,20 +61,24 @@ pub const FLOWNODE_IDS: &str = "flownode_ids";
 pub const OPTIONS: &str = "options";
 
 /// The `information_schema.flows` to provides information about flows in databases.
+///
 pub(super) struct InformationSchemaFlows {
     schema: SchemaRef,
     catalog_name: String,
+    catalog_manager: Weak<dyn CatalogManager>,
     flow_metadata_manager: Arc<FlowMetadataManager>,
 }
 
 impl InformationSchemaFlows {
     pub(super) fn new(
         catalog_name: String,
+        catalog_manager: Weak<dyn CatalogManager>,
         flow_metadata_manager: Arc<FlowMetadataManager>,
     ) -> Self {
         Self {
             schema: Self::schema(),
             catalog_name,
+            catalog_manager,
             flow_metadata_manager,
         }
     }
@@ -80,6 +90,7 @@ impl InformationSchemaFlows {
             vec![
                 (FLOW_NAME, CDT::string_datatype(), false),
                 (FLOW_ID, CDT::uint32_datatype(), false),
+                (STATE_SIZE, CDT::uint64_datatype(), true),
                 (TABLE_CATALOG, CDT::string_datatype(), false),
                 (FLOW_DEFINITION, CDT::string_datatype(), false),
                 (COMMENT, CDT::string_datatype(), true),
@@ -99,6 +110,7 @@ impl InformationSchemaFlows {
         InformationSchemaFlowsBuilder::new(
             self.schema.clone(),
             self.catalog_name.clone(),
+            self.catalog_manager.clone(),
             &self.flow_metadata_manager,
         )
     }
@@ -144,10 +156,12 @@ impl InformationTable for InformationSchemaFlows {
 struct InformationSchemaFlowsBuilder {
     schema: SchemaRef,
     catalog_name: String,
+    catalog_manager: Weak<dyn CatalogManager>,
     flow_metadata_manager: Arc<FlowMetadataManager>,
 
     flow_names: StringVectorBuilder,
     flow_ids: UInt32VectorBuilder,
+    state_sizes: UInt64VectorBuilder,
     table_catalogs: StringVectorBuilder,
     raw_sqls: StringVectorBuilder,
     comments: StringVectorBuilder,
@@ -162,15 +176,18 @@ impl InformationSchemaFlowsBuilder {
     fn new(
         schema: SchemaRef,
         catalog_name: String,
+        catalog_manager: Weak<dyn CatalogManager>,
         flow_metadata_manager: &Arc<FlowMetadataManager>,
     ) -> Self {
         Self {
             schema,
             catalog_name,
+            catalog_manager,
             flow_metadata_manager: flow_metadata_manager.clone(),
 
             flow_names: StringVectorBuilder::with_capacity(INIT_CAPACITY),
             flow_ids: UInt32VectorBuilder::with_capacity(INIT_CAPACITY),
+            state_sizes: UInt64VectorBuilder::with_capacity(INIT_CAPACITY),
             table_catalogs: StringVectorBuilder::with_capacity(INIT_CAPACITY),
             raw_sqls: StringVectorBuilder::with_capacity(INIT_CAPACITY),
             comments: StringVectorBuilder::with_capacity(INIT_CAPACITY),
@@ -195,6 +212,11 @@ impl InformationSchemaFlowsBuilder {
             .flow_names(&catalog_name)
             .await;
 
+        let flow_stat = {
+            let information_extension = utils::information_extension(&self.catalog_manager)?;
+            information_extension.flow_stats().await?
+        };
+
         while let Some((flow_name, flow_id)) = stream
             .try_next()
             .await
@@ -213,7 +235,7 @@ impl InformationSchemaFlowsBuilder {
                     catalog_name: catalog_name.to_string(),
                     flow_name: flow_name.to_string(),
                 })?;
-            self.add_flow(&predicates, flow_id.flow_id(), flow_info)?;
+            self.add_flow(&predicates, flow_id.flow_id(), flow_info, &flow_stat)?;
         }
 
         self.finish()
@@ -224,6 +246,7 @@ impl InformationSchemaFlowsBuilder {
         predicates: &Predicates,
         flow_id: FlowId,
         flow_info: FlowInfoValue,
+        flow_stat: &Option<FlowStat>,
     ) -> Result<()> {
         let row = [
             (FLOW_NAME, &Value::from(flow_info.flow_name().to_string())),
@@ -238,6 +261,11 @@ impl InformationSchemaFlowsBuilder {
         }
         self.flow_names.push(Some(flow_info.flow_name()));
         self.flow_ids.push(Some(flow_id));
+        self.state_sizes.push(
+            flow_stat
+                .as_ref()
+                .and_then(|state| state.state_size.get(&flow_id).map(|v| *v as u64)),
+        );
         self.table_catalogs.push(Some(flow_info.catalog_name()));
         self.raw_sqls.push(Some(flow_info.raw_sql()));
         self.comments.push(Some(flow_info.comment()));
@@ -270,6 +298,7 @@ impl InformationSchemaFlowsBuilder {
         let columns: Vec<VectorRef> = vec![
             Arc::new(self.flow_names.finish()),
             Arc::new(self.flow_ids.finish()),
+            Arc::new(self.state_sizes.finish()),
             Arc::new(self.table_catalogs.finish()),
             Arc::new(self.raw_sqls.finish()),
             Arc::new(self.comments.finish()),

src/catalog/src/system_schema/information_schema/key_column_usage.rs

@@ -54,6 +54,10 @@ const INIT_CAPACITY: usize = 42;
 pub(crate) const PRI_CONSTRAINT_NAME: &str = "PRIMARY";
 /// Time index constraint name
 pub(crate) const TIME_INDEX_CONSTRAINT_NAME: &str = "TIME INDEX";
+/// Inverted index constraint name
+pub(crate) const INVERTED_INDEX_CONSTRAINT_NAME: &str = "INVERTED INDEX";
+/// Fulltext index constraint name
+pub(crate) const FULLTEXT_INDEX_CONSTRAINT_NAME: &str = "FULLTEXT INDEX";
 
 /// The virtual table implementation for `information_schema.KEY_COLUMN_USAGE`.
 pub(super) struct InformationSchemaKeyColumnUsage {
@@ -216,14 +220,13 @@ impl InformationSchemaKeyColumnUsageBuilder {
             let mut stream = catalog_manager.tables(&catalog_name, &schema_name, None);
 
             while let Some(table) = stream.try_next().await? {
-                let mut primary_constraints = vec![];
-
                 let table_info = table.table_info();
                 let table_name = &table_info.name;
                 let keys = &table_info.meta.primary_key_indices;
                 let schema = table.schema();
 
                 for (idx, column) in schema.column_schemas().iter().enumerate() {
+                    let mut constraints = vec![];
                     if column.is_time_index() {
                         self.add_key_column_usage(
                             &predicates,
@@ -236,30 +239,31 @@ impl InformationSchemaKeyColumnUsageBuilder {
                             1, //always 1 for time index
                         );
                     }
-                    if keys.contains(&idx) {
-                        primary_constraints.push((
-                            catalog_name.clone(),
-                            schema_name.clone(),
-                            table_name.to_string(),
-                            column.name.clone(),
-                        ));
-                    }
                     // TODO(dimbtp): foreign key constraint not supported yet
-                }
+                    if keys.contains(&idx) {
+                        constraints.push(PRI_CONSTRAINT_NAME);
+                    }
+                    if column.is_inverted_indexed() {
+                        constraints.push(INVERTED_INDEX_CONSTRAINT_NAME);
+                    }
 
-                for (i, (catalog_name, schema_name, table_name, column_name)) in
+                    if column.has_fulltext_index_key() {
-                    primary_constraints.into_iter().enumerate()
+                        constraints.push(FULLTEXT_INDEX_CONSTRAINT_NAME);
-                {
+                    }
-                    self.add_key_column_usage(
+
-                        &predicates,
+                    if !constraints.is_empty() {
-                        &schema_name,
+                        let aggregated_constraints = constraints.join(", ");
-                        PRI_CONSTRAINT_NAME,
+                        self.add_key_column_usage(
-                        &catalog_name,
+                            &predicates,
-                        &schema_name,
+                            &schema_name,
-                        &table_name,
+                            &aggregated_constraints,
-                        &column_name,
+                            &catalog_name,
-                        i as u32 + 1,
+                            &schema_name,
-                    );
+                            table_name,
+                            &column.name,
+                            idx as u32 + 1,
+                        );
+                    }
                 }
             }
         }

src/cli/Cargo.toml

@@ -23,7 +23,6 @@ common-error.workspace = true
 common-grpc.workspace = true
 common-macro.workspace = true
 common-meta.workspace = true
-common-options.workspace = true
 common-procedure.workspace = true
 common-query.workspace = true
 common-recordbatch.workspace = true
@@ -61,5 +60,4 @@ client = { workspace = true, features = ["testing"] }
 common-test-util.workspace = true
 common-version.workspace = true
 serde.workspace = true
-temp-env = "0.3"
 tempfile.workspace = true

src/cli/src/repl.rs

@@ -34,7 +34,7 @@ use common_query::Output;
 use common_recordbatch::RecordBatches;
 use common_telemetry::debug;
 use either::Either;
-use meta_client::client::MetaClientBuilder;
+use meta_client::client::{ClusterKvBackend, MetaClientBuilder};
 use query::datafusion::DatafusionQueryEngine;
 use query::parser::QueryLanguageParser;
 use query::query_engine::{DefaultSerializer, QueryEngineState};

src/client/Cargo.toml

@@ -42,8 +42,6 @@ tonic.workspace = true
 
 [dev-dependencies]
 common-grpc-expr.workspace = true
-datanode.workspace = true
-derive-new = "0.5"
 tracing = "0.1"
 
 [dev-dependencies.substrait_proto]

src/cmd/src/datanode.rs

@@ -59,10 +59,6 @@ impl Instance {
         }
     }
 
-    pub fn datanode_mut(&mut self) -> &mut Datanode {
-        &mut self.datanode
-    }
-
     pub fn datanode(&self) -> &Datanode {
         &self.datanode
     }

src/cmd/src/flownode.rs

@@ -63,10 +63,6 @@ impl Instance {
         }
     }
 
-    pub fn flownode_mut(&mut self) -> &mut FlownodeInstance {
-        &mut self.flownode
-    }
-
     pub fn flownode(&self) -> &FlownodeInstance {
         &self.flownode
     }

src/cmd/src/standalone.rs

@@ -34,6 +34,7 @@ use common_meta::ddl::flow_meta::{FlowMetadataAllocator, FlowMetadataAllocatorRe
 use common_meta::ddl::table_meta::{TableMetadataAllocator, TableMetadataAllocatorRef};
 use common_meta::ddl::{DdlContext, NoopRegionFailureDetectorControl, ProcedureExecutorRef};
 use common_meta::ddl_manager::DdlManager;
+use common_meta::key::flow::flow_state::FlowStat;
 use common_meta::key::flow::{FlowMetadataManager, FlowMetadataManagerRef};
 use common_meta::key::{TableMetadataManager, TableMetadataManagerRef};
 use common_meta::kv_backend::KvBackendRef;
@@ -70,7 +71,7 @@ use servers::http::HttpOptions;
 use servers::tls::{TlsMode, TlsOption};
 use servers::Mode;
 use snafu::ResultExt;
-use tokio::sync::broadcast;
+use tokio::sync::{broadcast, RwLock};
 use tracing_appender::non_blocking::WorkerGuard;
 
 use crate::error::{
@@ -507,7 +508,7 @@ impl StartCommand {
             procedure_manager.clone(),
         ));
         let catalog_manager = KvBackendCatalogManager::new(
-            information_extension,
+            information_extension.clone(),
             kv_backend.clone(),
             layered_cache_registry.clone(),
             Some(procedure_manager.clone()),
@@ -532,6 +533,14 @@ impl StartCommand {
                 .context(OtherSnafu)?,
         );
 
+        // set the ref to query for the local flow state
+        {
+            let flow_worker_manager = flownode.flow_worker_manager();
+            information_extension
+                .set_flow_worker_manager(flow_worker_manager.clone())
+                .await;
+        }
+
         let node_manager = Arc::new(StandaloneDatanodeManager {
             region_server: datanode.region_server(),
             flow_server: flownode.flow_worker_manager(),
@@ -669,6 +678,7 @@ pub struct StandaloneInformationExtension {
     region_server: RegionServer,
     procedure_manager: ProcedureManagerRef,
     start_time_ms: u64,
+    flow_worker_manager: RwLock<Option<Arc<FlowWorkerManager>>>,
 }
 
 impl StandaloneInformationExtension {
@@ -677,8 +687,15 @@ impl StandaloneInformationExtension {
             region_server,
             procedure_manager,
             start_time_ms: common_time::util::current_time_millis() as u64,
+            flow_worker_manager: RwLock::new(None),
         }
     }
+
+    /// Set the flow worker manager for the standalone instance.
+    pub async fn set_flow_worker_manager(&self, flow_worker_manager: Arc<FlowWorkerManager>) {
+        let mut guard = self.flow_worker_manager.write().await;
+        *guard = Some(flow_worker_manager);
+    }
 }
 
 #[async_trait::async_trait]
@@ -750,6 +767,18 @@ impl InformationExtension for StandaloneInformationExtension {
             .collect::<Vec<_>>();
         Ok(stats)
     }
+
+    async fn flow_stats(&self) -> std::result::Result<Option<FlowStat>, Self::Error> {
+        Ok(Some(
+            self.flow_worker_manager
+                .read()
+                .await
+                .as_ref()
+                .unwrap()
+                .gen_state_report()
+                .await,
+        ))
+    }
 }
 
 #[cfg(test)]

src/common/base/Cargo.toml

@@ -17,6 +17,7 @@ common-macro.workspace = true
 futures.workspace = true
 paste = "1.0"
 pin-project.workspace = true
+rand.workspace = true
 serde = { version = "1.0", features = ["derive"] }
 snafu.workspace = true
 tokio.workspace = true

src/common/base/src/range_read.rs

@@ -36,6 +36,11 @@ pub struct Metadata {
 /// `RangeReader` reads a range of bytes from a source.
 #[async_trait]
 pub trait RangeReader: Send + Unpin {
+    /// Sets the file size hint for the reader.
+    ///
+    /// It's used to optimize the reading process by reducing the number of remote requests.
+    fn with_file_size_hint(&mut self, file_size_hint: u64);
+
     /// Returns the metadata of the source.
     async fn metadata(&mut self) -> io::Result<Metadata>;
 
@@ -70,6 +75,10 @@ pub trait RangeReader: Send + Unpin {
 
 #[async_trait]
 impl<R: ?Sized + RangeReader> RangeReader for &mut R {
+    fn with_file_size_hint(&mut self, file_size_hint: u64) {
+        (*self).with_file_size_hint(file_size_hint)
+    }
+
     async fn metadata(&mut self) -> io::Result<Metadata> {
         (*self).metadata().await
     }
@@ -186,15 +195,17 @@ impl<R: RangeReader + 'static> AsyncRead for AsyncReadAdapter<R> {
 
 #[async_trait]
 impl RangeReader for Vec<u8> {
+    fn with_file_size_hint(&mut self, _file_size_hint: u64) {
+        // do nothing
+    }
+
     async fn metadata(&mut self) -> io::Result<Metadata> {
         Ok(Metadata {
             content_length: self.len() as u64,
         })
     }
 
-    async fn read(&mut self, mut range: Range<u64>) -> io::Result<Bytes> {
+    async fn read(&mut self, range: Range<u64>) -> io::Result<Bytes> {
-        range.end = range.end.min(self.len() as u64);
-
         let bytes = Bytes::copy_from_slice(&self[range.start as usize..range.end as usize]);
         Ok(bytes)
     }
@@ -222,6 +233,10 @@ impl FileReader {
 
 #[async_trait]
 impl RangeReader for FileReader {
+    fn with_file_size_hint(&mut self, _file_size_hint: u64) {
+        // do nothing
+    }
+
     async fn metadata(&mut self) -> io::Result<Metadata> {
         Ok(Metadata {
             content_length: self.content_length,

src/common/base/src/readable_size.rs

@@ -19,7 +19,7 @@ pub const GIB: u64 = MIB * BINARY_DATA_MAGNITUDE;
 pub const TIB: u64 = GIB * BINARY_DATA_MAGNITUDE;
 pub const PIB: u64 = TIB * BINARY_DATA_MAGNITUDE;
 
-#[derive(Clone, Copy, PartialEq, Eq, Ord, PartialOrd)]
+#[derive(Clone, Copy, PartialEq, Eq, Ord, PartialOrd, Default)]
 pub struct ReadableSize(pub u64);
 
 impl ReadableSize {

src/common/catalog/Cargo.toml

@@ -8,10 +8,5 @@ license.workspace = true
 workspace = true
 
 [dependencies]
-common-error.workspace = true
-common-macro.workspace = true
-snafu.workspace = true
 
 [dev-dependencies]
-chrono.workspace = true
-tokio.workspace = true

src/common/datasource/Cargo.toml

@@ -48,5 +48,4 @@ url = "2.3"
 [dev-dependencies]
 common-telemetry.workspace = true
 common-test-util.workspace = true
-dotenv.workspace = true
 uuid.workspace = true

src/common/frontend/Cargo.toml

@@ -5,12 +5,7 @@ edition.workspace = true
 license.workspace = true
 
 [dependencies]
-api.workspace = true
 async-trait.workspace = true
-common-base.workspace = true
 common-error.workspace = true
 common-macro.workspace = true
-common-query.workspace = true
-session.workspace = true
 snafu.workspace = true
-sql.workspace = true

src/common/function/Cargo.toml

@@ -51,6 +51,5 @@ wkt = { version = "0.11", optional = true }
 
 [dev-dependencies]
 approx = "0.5"
-ron = "0.7"
 serde = { version = "1.0", features = ["derive"] }
 tokio.workspace = true

src/common/function/src/scalars/vector.rs

@@ -15,6 +15,8 @@
 mod convert;
 mod distance;
 pub(crate) mod impl_conv;
+mod scalar_add;
+mod scalar_mul;
 
 use std::sync::Arc;
 
@@ -32,5 +34,9 @@ impl VectorFunction {
         registry.register(Arc::new(distance::CosDistanceFunction));
         registry.register(Arc::new(distance::DotProductFunction));
         registry.register(Arc::new(distance::L2SqDistanceFunction));
+
+        // scalar calculation
+        registry.register(Arc::new(scalar_add::ScalarAddFunction));
+        registry.register(Arc::new(scalar_mul::ScalarMulFunction));
     }
 }

src/common/function/src/scalars/vector/impl_conv.rs

@@ -109,7 +109,6 @@ pub fn parse_veclit_from_strlit(s: &str) -> Result<Vec<f32>> {
         })
 }
 
-#[allow(unused)]
 /// Convert a vector literal to a binary literal.
 pub fn veclit_to_binlit(vec: &[f32]) -> Vec<u8> {
     if cfg!(target_endian = "little") {

src/common/function/src/scalars/vector/scalar_add.rs

@@ -0,0 +1,173 @@
+// Copyright 2023 Greptime Team
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+use std::borrow::Cow;
+use std::fmt::Display;
+
+use common_query::error::{InvalidFuncArgsSnafu, Result};
+use common_query::prelude::Signature;
+use datatypes::prelude::ConcreteDataType;
+use datatypes::scalars::ScalarVectorBuilder;
+use datatypes::vectors::{BinaryVectorBuilder, MutableVector, VectorRef};
+use nalgebra::DVectorView;
+use snafu::ensure;
+
+use crate::function::{Function, FunctionContext};
+use crate::helper;
+use crate::scalars::vector::impl_conv::{as_veclit, as_veclit_if_const, veclit_to_binlit};
+
+const NAME: &str = "vec_scalar_add";
+
+/// Adds a scalar to each element of a vector.
+///
+/// # Example
+///
+/// ```sql
+/// SELECT vec_to_string(vec_scalar_add(1, "[1, 2, 3]")) as result;
+///
+/// +---------+
+/// | result  |
+/// +---------+
+/// | [2,3,4] |
+/// +---------+
+///
+/// -- Negative scalar to simulate subtraction
+/// SELECT vec_to_string(vec_scalar_add(-1, "[1, 2, 3]")) as result;
+///
+/// +---------+
+/// | result  |
+/// +---------+
+/// | [0,1,2] |
+/// +---------+
+/// ```
+#[derive(Debug, Clone, Default)]
+pub struct ScalarAddFunction;
+
+impl Function for ScalarAddFunction {
+    fn name(&self) -> &str {
+        NAME
+    }
+
+    fn return_type(&self, _input_types: &[ConcreteDataType]) -> Result<ConcreteDataType> {
+        Ok(ConcreteDataType::binary_datatype())
+    }
+
+    fn signature(&self) -> Signature {
+        helper::one_of_sigs2(
+            vec![ConcreteDataType::float64_datatype()],
+            vec![
+                ConcreteDataType::string_datatype(),
+                ConcreteDataType::binary_datatype(),
+            ],
+        )
+    }
+
+    fn eval(&self, _func_ctx: FunctionContext, columns: &[VectorRef]) -> Result<VectorRef> {
+        ensure!(
+            columns.len() == 2,
+            InvalidFuncArgsSnafu {
+                err_msg: format!(
+                    "The length of the args is not correct, expect exactly two, have: {}",
+                    columns.len()
+                ),
+            }
+        );
+        let arg0 = &columns[0];
+        let arg1 = &columns[1];
+
+        let len = arg0.len();
+        let mut result = BinaryVectorBuilder::with_capacity(len);
+        if len == 0 {
+            return Ok(result.to_vector());
+        }
+
+        let arg1_const = as_veclit_if_const(arg1)?;
+
+        for i in 0..len {
+            let arg0 = arg0.get(i).as_f64_lossy();
+            let Some(arg0) = arg0 else {
+                result.push_null();
+                continue;
+            };
+
+            let arg1 = match arg1_const.as_ref() {
+                Some(arg1) => Some(Cow::Borrowed(arg1.as_ref())),
+                None => as_veclit(arg1.get_ref(i))?,
+            };
+            let Some(arg1) = arg1 else {
+                result.push_null();
+                continue;
+            };
+
+            let vec = DVectorView::from_slice(&arg1, arg1.len());
+            let vec_res = vec.add_scalar(arg0 as _);
+
+            let veclit = vec_res.as_slice();
+            let binlit = veclit_to_binlit(veclit);
+            result.push(Some(&binlit));
+        }
+
+        Ok(result.to_vector())
+    }
+}
+
+impl Display for ScalarAddFunction {
+    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
+        write!(f, "{}", NAME.to_ascii_uppercase())
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use std::sync::Arc;
+
+    use datatypes::vectors::{Float32Vector, StringVector};
+
+    use super::*;
+
+    #[test]
+    fn test_scalar_add() {
+        let func = ScalarAddFunction;
+
+        let input0 = Arc::new(Float32Vector::from(vec![
+            Some(1.0),
+            Some(-1.0),
+            None,
+            Some(3.0),
+        ]));
+        let input1 = Arc::new(StringVector::from(vec![
+            Some("[1.0,2.0,3.0]".to_string()),
+            Some("[4.0,5.0,6.0]".to_string()),
+            Some("[7.0,8.0,9.0]".to_string()),
+            None,
+        ]));
+
+        let result = func
+            .eval(FunctionContext::default(), &[input0, input1])
+            .unwrap();
+
+        let result = result.as_ref();
+        assert_eq!(result.len(), 4);
+        assert_eq!(
+            result.get_ref(0).as_binary().unwrap(),
+            Some(veclit_to_binlit(&[2.0, 3.0, 4.0]).as_slice())
+        );
+        assert_eq!(
+            result.get_ref(1).as_binary().unwrap(),
+            Some(veclit_to_binlit(&[3.0, 4.0, 5.0]).as_slice())
+        );
+        assert!(result.get_ref(2).is_null());
+        assert!(result.get_ref(3).is_null());
+    }
+}

src/common/function/src/scalars/vector/scalar_mul.rs

@@ -0,0 +1,173 @@
+// Copyright 2023 Greptime Team
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+use std::borrow::Cow;
+use std::fmt::Display;
+
+use common_query::error::{InvalidFuncArgsSnafu, Result};
+use common_query::prelude::Signature;
+use datatypes::prelude::ConcreteDataType;
+use datatypes::scalars::ScalarVectorBuilder;
+use datatypes::vectors::{BinaryVectorBuilder, MutableVector, VectorRef};
+use nalgebra::DVectorView;
+use snafu::ensure;
+
+use crate::function::{Function, FunctionContext};
+use crate::helper;
+use crate::scalars::vector::impl_conv::{as_veclit, as_veclit_if_const, veclit_to_binlit};
+
+const NAME: &str = "vec_scalar_mul";
+
+/// Multiples a scalar to each element of a vector.
+///
+/// # Example
+///
+/// ```sql
+/// SELECT vec_to_string(vec_scalar_mul(2, "[1, 2, 3]")) as result;
+///
+/// +---------+
+/// | result  |
+/// +---------+
+/// | [2,4,6] |
+/// +---------+
+///
+/// -- 1/scalar to simulate division
+/// SELECT vec_to_string(vec_scalar_mul(0.5, "[2, 4, 6]")) as result;
+///
+/// +---------+
+/// | result  |
+/// +---------+
+/// | [1,2,3] |
+/// +---------+
+/// ```
+#[derive(Debug, Clone, Default)]
+pub struct ScalarMulFunction;
+
+impl Function for ScalarMulFunction {
+    fn name(&self) -> &str {
+        NAME
+    }
+
+    fn return_type(&self, _input_types: &[ConcreteDataType]) -> Result<ConcreteDataType> {
+        Ok(ConcreteDataType::binary_datatype())
+    }
+
+    fn signature(&self) -> Signature {
+        helper::one_of_sigs2(
+            vec![ConcreteDataType::float64_datatype()],
+            vec![
+                ConcreteDataType::string_datatype(),
+                ConcreteDataType::binary_datatype(),
+            ],
+        )
+    }
+
+    fn eval(&self, _func_ctx: FunctionContext, columns: &[VectorRef]) -> Result<VectorRef> {
+        ensure!(
+            columns.len() == 2,
+            InvalidFuncArgsSnafu {
+                err_msg: format!(
+                    "The length of the args is not correct, expect exactly two, have: {}",
+                    columns.len()
+                ),
+            }
+        );
+        let arg0 = &columns[0];
+        let arg1 = &columns[1];
+
+        let len = arg0.len();
+        let mut result = BinaryVectorBuilder::with_capacity(len);
+        if len == 0 {
+            return Ok(result.to_vector());
+        }
+
+        let arg1_const = as_veclit_if_const(arg1)?;
+
+        for i in 0..len {
+            let arg0 = arg0.get(i).as_f64_lossy();
+            let Some(arg0) = arg0 else {
+                result.push_null();
+                continue;
+            };
+
+            let arg1 = match arg1_const.as_ref() {
+                Some(arg1) => Some(Cow::Borrowed(arg1.as_ref())),
+                None => as_veclit(arg1.get_ref(i))?,
+            };
+            let Some(arg1) = arg1 else {
+                result.push_null();
+                continue;
+            };
+
+            let vec = DVectorView::from_slice(&arg1, arg1.len());
+            let vec_res = vec.scale(arg0 as _);
+
+            let veclit = vec_res.as_slice();
+            let binlit = veclit_to_binlit(veclit);
+            result.push(Some(&binlit));
+        }
+
+        Ok(result.to_vector())
+    }
+}
+
+impl Display for ScalarMulFunction {
+    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
+        write!(f, "{}", NAME.to_ascii_uppercase())
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use std::sync::Arc;
+
+    use datatypes::vectors::{Float32Vector, StringVector};
+
+    use super::*;
+
+    #[test]
+    fn test_scalar_mul() {
+        let func = ScalarMulFunction;
+
+        let input0 = Arc::new(Float32Vector::from(vec![
+            Some(2.0),
+            Some(-0.5),
+            None,
+            Some(3.0),
+        ]));
+        let input1 = Arc::new(StringVector::from(vec![
+            Some("[1.0,2.0,3.0]".to_string()),
+            Some("[8.0,10.0,12.0]".to_string()),
+            Some("[7.0,8.0,9.0]".to_string()),
+            None,
+        ]));
+
+        let result = func
+            .eval(FunctionContext::default(), &[input0, input1])
+            .unwrap();
+
+        let result = result.as_ref();
+        assert_eq!(result.len(), 4);
+        assert_eq!(
+            result.get_ref(0).as_binary().unwrap(),
+            Some(veclit_to_binlit(&[2.0, 4.0, 6.0]).as_slice())
+        );
+        assert_eq!(
+            result.get_ref(1).as_binary().unwrap(),
+            Some(veclit_to_binlit(&[-4.0, -5.0, -6.0]).as_slice())
+        );
+        assert!(result.get_ref(2).is_null());
+        assert!(result.get_ref(3).is_null());
+    }
+}

src/common/meta/src/cache/table/table_route.rs

@@ -49,14 +49,6 @@ impl TableRoute {
             TableRoute::Logical(_) => None,
         }
     }
-
-    /// Returns [LogicalTableRouteValue] reference if it's [TableRoute::Logical]; Otherwise it returns [None].
-    pub fn as_logical_table_route_ref(&self) -> Option<&Arc<LogicalTableRouteValue>> {
-        match self {
-            TableRoute::Physical(_) => None,
-            TableRoute::Logical(table_route) => Some(table_route),
-        }
-    }
 }
 
 /// [TableRouteCache] caches the [TableId] to [TableRoute] mapping.

src/common/meta/src/key.rs

@@ -137,6 +137,7 @@ use self::schema_name::{SchemaManager, SchemaNameKey, SchemaNameValue};
 use self::table_route::{TableRouteManager, TableRouteValue};
 use self::tombstone::TombstoneManager;
 use crate::error::{self, Result, SerdeJsonSnafu};
+use crate::key::flow::flow_state::FlowStateValue;
 use crate::key::node_address::NodeAddressValue;
 use crate::key::table_route::TableRouteKey;
 use crate::key::txn_helper::TxnOpGetResponseSet;
@@ -1262,7 +1263,8 @@ impl_metadata_value! {
     FlowRouteValue,
     TableFlowValue,
     NodeAddressValue,
-    SchemaNameValue
+    SchemaNameValue,
+    FlowStateValue
 }
 
 impl_optional_metadata_value! {

src/common/meta/src/key/flow.rs

@@ -15,6 +15,7 @@
 pub mod flow_info;
 pub(crate) mod flow_name;
 pub(crate) mod flow_route;
+pub mod flow_state;
 pub(crate) mod flownode_flow;
 pub(crate) mod table_flow;
 
@@ -35,6 +36,7 @@ use crate::ensure_values;
 use crate::error::{self, Result};
 use crate::key::flow::flow_info::FlowInfoManager;
 use crate::key::flow::flow_name::FlowNameManager;
+use crate::key::flow::flow_state::FlowStateManager;
 use crate::key::flow::flownode_flow::FlownodeFlowManager;
 pub use crate::key::flow::table_flow::{TableFlowManager, TableFlowManagerRef};
 use crate::key::txn_helper::TxnOpGetResponseSet;
@@ -102,6 +104,8 @@ pub struct FlowMetadataManager {
     flownode_flow_manager: FlownodeFlowManager,
     table_flow_manager: TableFlowManager,
     flow_name_manager: FlowNameManager,
+    /// only metasrv have access to itself's memory backend, so for other case it should be None
+    flow_state_manager: Option<FlowStateManager>,
     kv_backend: KvBackendRef,
 }
 
@@ -114,6 +118,7 @@ impl FlowMetadataManager {
             flow_name_manager: FlowNameManager::new(kv_backend.clone()),
             flownode_flow_manager: FlownodeFlowManager::new(kv_backend.clone()),
             table_flow_manager: TableFlowManager::new(kv_backend.clone()),
+            flow_state_manager: None,
             kv_backend,
         }
     }
@@ -123,6 +128,10 @@ impl FlowMetadataManager {
         &self.flow_name_manager
     }
 
+    pub fn flow_state_manager(&self) -> Option<&FlowStateManager> {
+        self.flow_state_manager.as_ref()
+    }
+
     /// Returns the [`FlowInfoManager`].
     pub fn flow_info_manager(&self) -> &FlowInfoManager {
         &self.flow_info_manager

src/common/meta/src/key/flow/flow_state.rs

@@ -0,0 +1,162 @@
+// Copyright 2023 Greptime Team
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+use std::collections::BTreeMap;
+use std::sync::Arc;
+
+use serde::{Deserialize, Serialize};
+
+use crate::error::{self, Result};
+use crate::key::flow::FlowScoped;
+use crate::key::{FlowId, MetadataKey, MetadataValue};
+use crate::kv_backend::KvBackendRef;
+use crate::rpc::store::PutRequest;
+
+/// The entire FlowId to Flow Size's Map is stored directly in the value part of the key.
+const FLOW_STATE_KEY: &str = "state";
+
+/// The key of flow state.
+#[derive(Debug, Clone, Copy, PartialEq)]
+struct FlowStateKeyInner;
+
+impl FlowStateKeyInner {
+    pub fn new() -> Self {
+        Self
+    }
+}
+
+impl<'a> MetadataKey<'a, FlowStateKeyInner> for FlowStateKeyInner {
+    fn to_bytes(&self) -> Vec<u8> {
+        FLOW_STATE_KEY.as_bytes().to_vec()
+    }
+
+    fn from_bytes(bytes: &'a [u8]) -> Result<FlowStateKeyInner> {
+        let key = std::str::from_utf8(bytes).map_err(|e| {
+            error::InvalidMetadataSnafu {
+                err_msg: format!(
+                    "FlowInfoKeyInner '{}' is not a valid UTF8 string: {e}",
+                    String::from_utf8_lossy(bytes)
+                ),
+            }
+            .build()
+        })?;
+        if key != FLOW_STATE_KEY {
+            return Err(error::InvalidMetadataSnafu {
+                err_msg: format!("Invalid FlowStateKeyInner '{key}'"),
+            }
+            .build());
+        }
+        Ok(FlowStateKeyInner::new())
+    }
+}
+
+/// The key stores the state size of the flow.
+///
+/// The layout: `__flow/state`.
+pub struct FlowStateKey(FlowScoped<FlowStateKeyInner>);
+
+impl FlowStateKey {
+    /// Returns the [FlowStateKey].
+    pub fn new() -> FlowStateKey {
+        let inner = FlowStateKeyInner::new();
+        FlowStateKey(FlowScoped::new(inner))
+    }
+}
+
+impl Default for FlowStateKey {
+    fn default() -> Self {
+        Self::new()
+    }
+}
+
+impl<'a> MetadataKey<'a, FlowStateKey> for FlowStateKey {
+    fn to_bytes(&self) -> Vec<u8> {
+        self.0.to_bytes()
+    }
+
+    fn from_bytes(bytes: &'a [u8]) -> Result<FlowStateKey> {
+        Ok(FlowStateKey(FlowScoped::<FlowStateKeyInner>::from_bytes(
+            bytes,
+        )?))
+    }
+}
+
+/// The value of flow state size
+#[derive(Debug, Clone, Serialize, Deserialize, PartialEq)]
+pub struct FlowStateValue {
+    /// For each key, the bytes of the state in memory
+    pub state_size: BTreeMap<FlowId, usize>,
+}
+
+impl FlowStateValue {
+    pub fn new(state_size: BTreeMap<FlowId, usize>) -> Self {
+        Self { state_size }
+    }
+}
+
+pub type FlowStateManagerRef = Arc<FlowStateManager>;
+
+/// The manager of [FlowStateKey]. Since state size changes frequently, we store it in memory.
+///
+/// This is only used in distributed mode. When meta-srv use heartbeat to update the flow stat report
+/// and frontned use get to get the latest flow stat report.
+pub struct FlowStateManager {
+    in_memory: KvBackendRef,
+}
+
+impl FlowStateManager {
+    pub fn new(in_memory: KvBackendRef) -> Self {
+        Self { in_memory }
+    }
+
+    pub async fn get(&self) -> Result<Option<FlowStateValue>> {
+        let key = FlowStateKey::new().to_bytes();
+        self.in_memory
+            .get(&key)
+            .await?
+            .map(|x| FlowStateValue::try_from_raw_value(&x.value))
+            .transpose()
+    }
+
+    pub async fn put(&self, value: FlowStateValue) -> Result<()> {
+        let key = FlowStateKey::new().to_bytes();
+        let value = value.try_as_raw_value()?;
+        let req = PutRequest::new().with_key(key).with_value(value);
+        self.in_memory.put(req).await?;
+        Ok(())
+    }
+}
+
+/// Flow's state report, send regularly through heartbeat message
+#[derive(Debug, Clone)]
+pub struct FlowStat {
+    /// For each key, the bytes of the state in memory
+    pub state_size: BTreeMap<u32, usize>,
+}
+
+impl From<FlowStateValue> for FlowStat {
+    fn from(value: FlowStateValue) -> Self {
+        Self {
+            state_size: value.state_size,
+        }
+    }
+}
+
+impl From<FlowStat> for FlowStateValue {
+    fn from(value: FlowStat) -> Self {
+        Self {
+            state_size: value.state_size,
+        }
+    }
+}

src/common/meta/src/key/table_route.rs

@@ -290,28 +290,6 @@ impl TableRouteManager {
         }
     }
 
-    /// Returns the [`PhysicalTableRouteValue`] in the first level,
-    /// It won't follow the [`LogicalTableRouteValue`] to find the next level [`PhysicalTableRouteValue`].
-    ///
-    /// Returns an error if the first level value is not a [`PhysicalTableRouteValue`].
-    pub async fn try_get_physical_table_route(
-        &self,
-        table_id: TableId,
-    ) -> Result<Option<PhysicalTableRouteValue>> {
-        match self.storage.get(table_id).await? {
-            Some(route) => {
-                ensure!(
-                    route.is_physical(),
-                    UnexpectedLogicalRouteTableSnafu {
-                        err_msg: format!("{route:?} is a non-physical TableRouteValue.")
-                    }
-                );
-                Ok(Some(route.into_physical_table_route()))
-            }
-            None => Ok(None),
-        }
-    }
-
     /// Returns the [TableId] recursively.
     ///
     /// Returns a [TableRouteNotFound](crate::error::Error::TableRouteNotFound) Error if:
@@ -569,37 +547,6 @@ impl TableRouteStorage {
             .transpose()
     }
 
-    /// Returns the physical `DeserializedValueWithBytes<TableRouteValue>` recursively.
-    ///
-    /// Returns a [TableRouteNotFound](crate::error::Error::TableRouteNotFound) Error if:
-    /// - the physical table(`logical_or_physical_table_id`) does not exist
-    /// - the corresponding physical table of the logical table(`logical_or_physical_table_id`) does not exist.
-    pub async fn get_physical_table_route_with_raw_bytes(
-        &self,
-        logical_or_physical_table_id: TableId,
-    ) -> Result<(TableId, DeserializedValueWithBytes<TableRouteValue>)> {
-        let table_route = self
-            .get_with_raw_bytes(logical_or_physical_table_id)
-            .await?
-            .context(TableRouteNotFoundSnafu {
-                table_id: logical_or_physical_table_id,
-            })?;
-
-        match table_route.get_inner_ref() {
-            TableRouteValue::Physical(_) => Ok((logical_or_physical_table_id, table_route)),
-            TableRouteValue::Logical(x) => {
-                let physical_table_id = x.physical_table_id();
-                let physical_table_route = self
-                    .get_with_raw_bytes(physical_table_id)
-                    .await?
-                    .context(TableRouteNotFoundSnafu {
-                        table_id: physical_table_id,
-                    })?;
-                Ok((physical_table_id, physical_table_route))
-            }
-        }
-    }
-
     /// Returns batch of [`TableRouteValue`] that respects the order of `table_ids`.
     pub async fn batch_get(&self, table_ids: &[TableId]) -> Result<Vec<Option<TableRouteValue>>> {
         let mut table_routes = self.batch_get_inner(table_ids).await?;

src/common/meta/src/kv_backend.rs

@@ -36,7 +36,7 @@ pub mod postgres;
 pub mod test;
 pub mod txn;
 
-pub type KvBackendRef = Arc<dyn KvBackend<Error = Error> + Send + Sync>;
+pub type KvBackendRef<E = Error> = Arc<dyn KvBackend<Error = E> + Send + Sync>;
 
 #[async_trait]
 pub trait KvBackend: TxnService
@@ -161,6 +161,9 @@ where
     Self::Error: ErrorExt,
 {
     fn reset(&self);
+
+    /// Upcast as `KvBackendRef`. Since https://github.com/rust-lang/rust/issues/65991 is not yet stable.
+    fn as_kv_backend_ref(self: Arc<Self>) -> KvBackendRef<Self::Error>;
 }
 
-pub type ResettableKvBackendRef = Arc<dyn ResettableKvBackend<Error = Error> + Send + Sync>;
+pub type ResettableKvBackendRef<E = Error> = Arc<dyn ResettableKvBackend<Error = E> + Send + Sync>;

src/common/meta/src/kv_backend/etcd.rs

@@ -15,6 +15,7 @@
 use std::any::Any;
 use std::sync::Arc;
 
+use common_telemetry::info;
 use etcd_client::{
     Client, DeleteOptions, GetOptions, PutOptions, Txn, TxnOp, TxnOpResponse, TxnResponse,
 };
@@ -55,6 +56,7 @@ impl EtcdStore {
     }
 
     pub fn with_etcd_client(client: Client, max_txn_ops: usize) -> KvBackendRef {
+        info!("Connected to etcd");
         Arc::new(Self {
             client,
             max_txn_ops,

src/common/meta/src/kv_backend/memory.rs

@@ -16,13 +16,13 @@ use std::any::Any;
 use std::collections::BTreeMap;
 use std::fmt::{Display, Formatter};
 use std::marker::PhantomData;
-use std::sync::RwLock;
+use std::sync::{Arc, RwLock};
 
 use async_trait::async_trait;
 use common_error::ext::ErrorExt;
 use serde::Serializer;
 
-use super::ResettableKvBackend;
+use super::{KvBackendRef, ResettableKvBackend};
 use crate::kv_backend::txn::{Txn, TxnOp, TxnOpResponse, TxnRequest, TxnResponse};
 use crate::kv_backend::{KvBackend, TxnService};
 use crate::metrics::METRIC_META_TXN_REQUEST;
@@ -311,6 +311,10 @@ impl<T: ErrorExt + Send + Sync + 'static> ResettableKvBackend for MemoryKvBacken
     fn reset(&self) {
         self.clear();
     }
+
+    fn as_kv_backend_ref(self: Arc<Self>) -> KvBackendRef<T> {
+        self
+    }
 }
 
 #[cfg(test)]

src/common/meta/src/rpc/router.rs

@@ -89,39 +89,6 @@ pub fn convert_to_region_leader_map(region_routes: &[RegionRoute]) -> HashMap<Re
         .collect::<HashMap<_, _>>()
 }
 
-/// Returns the HashMap<[RegionNumber], HashSet<DatanodeId>>
-pub fn convert_to_region_peer_map(
-    region_routes: &[RegionRoute],
-) -> HashMap<RegionNumber, HashSet<u64>> {
-    region_routes
-        .iter()
-        .map(|x| {
-            let set = x
-                .follower_peers
-                .iter()
-                .map(|p| p.id)
-                .chain(x.leader_peer.as_ref().map(|p| p.id))
-                .collect::<HashSet<_>>();
-
-            (x.region.id.region_number(), set)
-        })
-        .collect::<HashMap<_, _>>()
-}
-
-/// Returns the HashMap<[RegionNumber], [LeaderState]>;
-pub fn convert_to_region_leader_state_map(
-    region_routes: &[RegionRoute],
-) -> HashMap<RegionNumber, LeaderState> {
-    region_routes
-        .iter()
-        .filter_map(|x| {
-            x.leader_state
-                .as_ref()
-                .map(|state| (x.region.id.region_number(), *state))
-        })
-        .collect::<HashMap<_, _>>()
-}
-
 pub fn find_region_leader(
     region_routes: &[RegionRoute],
     region_number: RegionNumber,
@@ -147,19 +114,6 @@ pub fn find_leader_regions(region_routes: &[RegionRoute], datanode: &Peer) -> Ve
         .collect()
 }
 
-pub fn extract_all_peers(region_routes: &[RegionRoute]) -> Vec<Peer> {
-    let mut peers = region_routes
-        .iter()
-        .flat_map(|x| x.leader_peer.iter().chain(x.follower_peers.iter()))
-        .collect::<HashSet<_>>()
-        .into_iter()
-        .cloned()
-        .collect::<Vec<_>>();
-    peers.sort_by_key(|x| x.id);
-
-    peers
-}
-
 impl TableRoute {
     pub fn new(table: Table, region_routes: Vec<RegionRoute>) -> Self {
         let region_leaders = region_routes

src/common/recordbatch/src/lib.rs

@@ -26,7 +26,6 @@ use std::sync::Arc;
 
 use adapter::RecordBatchMetrics;
 use arc_swap::ArcSwapOption;
-use datafusion::physical_plan::memory::MemoryStream;
 pub use datafusion::physical_plan::SendableRecordBatchStream as DfSendableRecordBatchStream;
 use datatypes::arrow::compute::SortOptions;
 pub use datatypes::arrow::record_batch::RecordBatch as DfRecordBatch;
@@ -170,19 +169,6 @@ impl RecordBatches {
             index: 0,
         })
     }
-
-    pub fn into_df_stream(self) -> DfSendableRecordBatchStream {
-        let df_record_batches = self
-            .batches
-            .into_iter()
-            .map(|batch| batch.into_df_record_batch())
-            .collect();
-        // unwrap safety: `MemoryStream::try_new` won't fail
-        Box::pin(
-            MemoryStream::try_new(df_record_batches, self.schema.arrow_schema().clone(), None)
-                .unwrap(),
-        )
-    }
 }
 
 impl IntoIterator for RecordBatches {

src/common/runtime/Cargo.toml

@@ -35,8 +35,6 @@ serde_json.workspace = true
 snafu.workspace = true
 tempfile.workspace = true
 tokio.workspace = true
-tokio-metrics = "0.3"
-tokio-metrics-collector = { git = "https://github.com/MichaelScofield/tokio-metrics-collector.git", rev = "89d692d5753d28564a7aac73c6ac5aba22243ba0" }
 tokio-util.workspace = true
 
 [dev-dependencies]

src/common/time/src/util.rs

@@ -29,10 +29,6 @@ pub fn format_utc_datetime(utc: &NaiveDateTime, pattern: &str) -> String {
     }
 }
 
-pub fn system_datetime_to_utc(local: &NaiveDateTime) -> LocalResult<NaiveDateTime> {
-    datetime_to_utc(local, get_timezone(None))
-}
-
 /// Cast a [`NaiveDateTime`] with the given timezone.
 pub fn datetime_to_utc(
     datetime: &NaiveDateTime,

src/common/wal/src/config/raft_engine.rs

@@ -49,9 +49,9 @@ impl Default for RaftEngineConfig {
     fn default() -> Self {
         Self {
             dir: None,
-            file_size: ReadableSize::mb(256),
+            file_size: ReadableSize::mb(128),
-            purge_threshold: ReadableSize::gb(4),
+            purge_threshold: ReadableSize::gb(1),
-            purge_interval: Duration::from_secs(600),
+            purge_interval: Duration::from_secs(60),
             read_batch_size: 128,
             sync_write: false,
             enable_log_recycle: true,

src/datatypes/src/data_type.rs

@@ -370,6 +370,51 @@ impl ConcreteDataType {
             _ => None,
         }
     }
+
+    /// Return the datatype name in postgres type system
+    pub fn postgres_datatype_name(&self) -> &'static str {
+        match self {
+            &ConcreteDataType::Null(_) => "UNKNOWN",
+            &ConcreteDataType::Boolean(_) => "BOOL",
+            &ConcreteDataType::Int8(_) | &ConcreteDataType::UInt8(_) => "CHAR",
+            &ConcreteDataType::Int16(_) | &ConcreteDataType::UInt16(_) => "INT2",
+            &ConcreteDataType::Int32(_) | &ConcreteDataType::UInt32(_) => "INT4",
+            &ConcreteDataType::Int64(_) | &ConcreteDataType::UInt64(_) => "INT8",
+            &ConcreteDataType::Float32(_) => "FLOAT4",
+            &ConcreteDataType::Float64(_) => "FLOAT8",
+            &ConcreteDataType::Binary(_) | &ConcreteDataType::Vector(_) => "BYTEA",
+            &ConcreteDataType::String(_) => "VARCHAR",
+            &ConcreteDataType::Date(_) => "DATE",
+            &ConcreteDataType::DateTime(_) | &ConcreteDataType::Timestamp(_) => "TIMESTAMP",
+            &ConcreteDataType::Time(_) => "TIME",
+            &ConcreteDataType::Interval(_) => "INTERVAL",
+            &ConcreteDataType::Decimal128(_) => "NUMERIC",
+            &ConcreteDataType::Json(_) => "JSON",
+            ConcreteDataType::List(list) => match list.item_type() {
+                &ConcreteDataType::Null(_) => "UNKNOWN",
+                &ConcreteDataType::Boolean(_) => "_BOOL",
+                &ConcreteDataType::Int8(_) | &ConcreteDataType::UInt8(_) => "_CHAR",
+                &ConcreteDataType::Int16(_) | &ConcreteDataType::UInt16(_) => "_INT2",
+                &ConcreteDataType::Int32(_) | &ConcreteDataType::UInt32(_) => "_INT4",
+                &ConcreteDataType::Int64(_) | &ConcreteDataType::UInt64(_) => "_INT8",
+                &ConcreteDataType::Float32(_) => "_FLOAT4",
+                &ConcreteDataType::Float64(_) => "_FLOAT8",
+                &ConcreteDataType::Binary(_) => "_BYTEA",
+                &ConcreteDataType::String(_) => "_VARCHAR",
+                &ConcreteDataType::Date(_) => "_DATE",
+                &ConcreteDataType::DateTime(_) | &ConcreteDataType::Timestamp(_) => "_TIMESTAMP",
+                &ConcreteDataType::Time(_) => "_TIME",
+                &ConcreteDataType::Interval(_) => "_INTERVAL",
+                &ConcreteDataType::Decimal128(_) => "_NUMERIC",
+                &ConcreteDataType::Json(_) => "_JSON",
+                &ConcreteDataType::Duration(_)
+                | &ConcreteDataType::Dictionary(_)
+                | &ConcreteDataType::Vector(_)
+                | &ConcreteDataType::List(_) => "UNKNOWN",
+            },
+            &ConcreteDataType::Duration(_) | &ConcreteDataType::Dictionary(_) => "UNKNOWN",
+        }
+    }
 }
 
 impl From<&ConcreteDataType> for ConcreteDataType {

src/datatypes/src/error.rs

@@ -232,6 +232,12 @@ pub enum Error {
         #[snafu(implicit)]
         location: Location,
     },
+    #[snafu(display("Invalid skipping index option: {}", msg))]
+    InvalidSkippingIndexOption {
+        msg: String,
+        #[snafu(implicit)]
+        location: Location,
+    },
 }
 
 impl ErrorExt for Error {
@@ -252,7 +258,8 @@ impl ErrorExt for Error {
             | InvalidPrecisionOrScale { .. }
             | InvalidJson { .. }
             | InvalidVector { .. }
-            | InvalidFulltextOption { .. } => StatusCode::InvalidArguments,
+            | InvalidFulltextOption { .. }
+            | InvalidSkippingIndexOption { .. } => StatusCode::InvalidArguments,
 
             ValueExceedsPrecision { .. }
             | CastType { .. }

src/datatypes/src/schema.rs

@@ -28,10 +28,11 @@ use snafu::{ensure, ResultExt};
 use crate::error::{self, DuplicateColumnSnafu, Error, ProjectArrowSchemaSnafu, Result};
 use crate::prelude::ConcreteDataType;
 pub use crate::schema::column_schema::{
-    ColumnSchema, FulltextAnalyzer, FulltextOptions, Metadata,
+    ColumnSchema, FulltextAnalyzer, FulltextOptions, Metadata, SkippingIndexOptions,
     COLUMN_FULLTEXT_CHANGE_OPT_KEY_ENABLE, COLUMN_FULLTEXT_OPT_KEY_ANALYZER,
-    COLUMN_FULLTEXT_OPT_KEY_CASE_SENSITIVE, COMMENT_KEY, FULLTEXT_KEY, INVERTED_INDEX_KEY,
+    COLUMN_FULLTEXT_OPT_KEY_CASE_SENSITIVE, COLUMN_SKIPPING_INDEX_OPT_KEY_GRANULARITY,
-    TIME_INDEX_KEY,
+    COLUMN_SKIPPING_INDEX_OPT_KEY_TYPE, COMMENT_KEY, FULLTEXT_KEY, INVERTED_INDEX_KEY,
+    SKIPPING_INDEX_KEY, TIME_INDEX_KEY,
 };
 pub use crate::schema::constraint::ColumnDefaultConstraint;
 pub use crate::schema::raw::RawSchema;

src/datatypes/src/schema/column_schema.rs

@@ -39,12 +39,20 @@ const DEFAULT_CONSTRAINT_KEY: &str = "greptime:default_constraint";
 pub const FULLTEXT_KEY: &str = "greptime:fulltext";
 /// Key used to store whether the column has inverted index in arrow field's metadata.
 pub const INVERTED_INDEX_KEY: &str = "greptime:inverted_index";
+/// Key used to store skip options in arrow field's metadata.
+pub const SKIPPING_INDEX_KEY: &str = "greptime:skipping_index";
 
 /// Keys used in fulltext options
 pub const COLUMN_FULLTEXT_CHANGE_OPT_KEY_ENABLE: &str = "enable";
 pub const COLUMN_FULLTEXT_OPT_KEY_ANALYZER: &str = "analyzer";
 pub const COLUMN_FULLTEXT_OPT_KEY_CASE_SENSITIVE: &str = "case_sensitive";
 
+/// Keys used in SKIPPING index options
+pub const COLUMN_SKIPPING_INDEX_OPT_KEY_GRANULARITY: &str = "granularity";
+pub const COLUMN_SKIPPING_INDEX_OPT_KEY_TYPE: &str = "type";
+
+pub const DEFAULT_GRANULARITY: u32 = 10240;
+
 /// Schema of a column, used as an immutable struct.
 #[derive(Clone, PartialEq, Eq, Serialize, Deserialize)]
 pub struct ColumnSchema {
@@ -156,6 +164,10 @@ impl ColumnSchema {
             .unwrap_or(false)
     }
 
+    pub fn has_fulltext_index_key(&self) -> bool {
+        self.metadata.contains_key(FULLTEXT_KEY)
+    }
+
     pub fn has_inverted_index_key(&self) -> bool {
         self.metadata.contains_key(INVERTED_INDEX_KEY)
     }
@@ -298,6 +310,34 @@ impl ColumnSchema {
         );
         Ok(())
     }
+
+    /// Retrieves the skipping index options for the column.
+    pub fn skipping_index_options(&self) -> Result<Option<SkippingIndexOptions>> {
+        match self.metadata.get(SKIPPING_INDEX_KEY) {
+            None => Ok(None),
+            Some(json) => {
+                let options =
+                    serde_json::from_str(json).context(error::DeserializeSnafu { json })?;
+                Ok(Some(options))
+            }
+        }
+    }
+
+    pub fn with_skipping_options(mut self, options: SkippingIndexOptions) -> Result<Self> {
+        self.metadata.insert(
+            SKIPPING_INDEX_KEY.to_string(),
+            serde_json::to_string(&options).context(error::SerializeSnafu)?,
+        );
+        Ok(self)
+    }
+
+    pub fn set_skipping_options(&mut self, options: &SkippingIndexOptions) -> Result<()> {
+        self.metadata.insert(
+            SKIPPING_INDEX_KEY.to_string(),
+            serde_json::to_string(options).context(error::SerializeSnafu)?,
+        );
+        Ok(())
+    }
 }
 
 /// Column extended type set in column schema's metadata.
@@ -495,6 +535,76 @@ impl fmt::Display for FulltextAnalyzer {
     }
 }
 
+/// Skipping options for a column.
+#[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize, Default, Visit, VisitMut)]
+#[serde(rename_all = "kebab-case")]
+pub struct SkippingIndexOptions {
+    /// The granularity of the skip index.
+    pub granularity: u32,
+    /// The type of the skip index.
+    #[serde(default)]
+    pub index_type: SkipIndexType,
+}
+
+impl fmt::Display for SkippingIndexOptions {
+    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
+        write!(f, "granularity={}", self.granularity)?;
+        write!(f, ", index_type={}", self.index_type)?;
+        Ok(())
+    }
+}
+
+/// Skip index types.
+#[derive(Debug, Default, Clone, PartialEq, Eq, Serialize, Deserialize, Visit, VisitMut)]
+pub enum SkipIndexType {
+    #[default]
+    BloomFilter,
+}
+
+impl fmt::Display for SkipIndexType {
+    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
+        match self {
+            SkipIndexType::BloomFilter => write!(f, "BLOOM"),
+        }
+    }
+}
+
+impl TryFrom<HashMap<String, String>> for SkippingIndexOptions {
+    type Error = Error;
+
+    fn try_from(options: HashMap<String, String>) -> Result<Self> {
+        // Parse granularity with default value 1
+        let granularity = match options.get(COLUMN_SKIPPING_INDEX_OPT_KEY_GRANULARITY) {
+            Some(value) => value.parse::<u32>().map_err(|_| {
+                error::InvalidSkippingIndexOptionSnafu {
+                    msg: format!("Invalid granularity: {value}, expected: positive integer"),
+                }
+                .build()
+            })?,
+            None => DEFAULT_GRANULARITY,
+        };
+
+        // Parse index type with default value BloomFilter
+        let index_type = match options.get(COLUMN_SKIPPING_INDEX_OPT_KEY_TYPE) {
+            Some(typ) => match typ.to_ascii_uppercase().as_str() {
+                "BLOOM" => SkipIndexType::BloomFilter,
+                _ => {
+                    return error::InvalidSkippingIndexOptionSnafu {
+                        msg: format!("Invalid index type: {typ}, expected: 'BLOOM'"),
+                    }
+                    .fail();
+                }
+            },
+            None => SkipIndexType::default(),
+        };
+
+        Ok(SkippingIndexOptions {
+            granularity,
+            index_type,
+        })
+    }
+}
+
 #[cfg(test)]
 mod tests {
     use std::sync::Arc;

src/file-engine/Cargo.toml

@@ -38,5 +38,4 @@ tokio.workspace = true
 
 [dev-dependencies]
 api.workspace = true
-common-procedure-test.workspace = true
 common-test-util.workspace = true

src/flow/Cargo.toml

@@ -40,6 +40,8 @@ datatypes.workspace = true
 enum-as-inner = "0.6.0"
 enum_dispatch = "0.3"
 futures = "0.3"
+get-size-derive2 = "0.1.2"
+get-size2 = "0.1.2"
 greptime-proto.workspace = true
 # This fork of hydroflow is simply for keeping our dependency in our org, and pin the version
 # otherwise it is the same with upstream repo
@@ -47,7 +49,6 @@ hydroflow = { git = "https://github.com/GreptimeTeam/hydroflow.git", branch = "m
 itertools.workspace = true
 lazy_static.workspace = true
 meta-client.workspace = true
-minstant = "0.1.7"
 nom = "7.1.3"
 num-traits = "0.2"
 operator.workspace = true

src/flow/src/adapter.rs

@@ -60,6 +60,7 @@ use crate::repr::{self, DiffRow, Row, BATCH_SIZE};
 
 mod flownode_impl;
 mod parse_expr;
+mod stat;
 #[cfg(test)]
 mod tests;
 mod util;
@@ -69,6 +70,7 @@ pub(crate) mod node_context;
 mod table_source;
 
 use crate::error::Error;
+use crate::utils::StateReportHandler;
 use crate::FrontendInvoker;
 
 // `GREPTIME_TIMESTAMP` is not used to distinguish when table is created automatically by flow
@@ -137,6 +139,8 @@ pub struct FlowWorkerManager {
     ///
     /// So that a series of event like `inserts -> flush` can be handled correctly
     flush_lock: RwLock<()>,
+    /// receive a oneshot sender to send state size report
+    state_report_handler: RwLock<Option<StateReportHandler>>,
 }
 
 /// Building FlownodeManager
@@ -170,9 +174,15 @@ impl FlowWorkerManager {
             tick_manager,
             node_id,
             flush_lock: RwLock::new(()),
+            state_report_handler: RwLock::new(None),
         }
     }
 
+    pub async fn with_state_report_handler(self, handler: StateReportHandler) -> Self {
+        *self.state_report_handler.write().await = Some(handler);
+        self
+    }
+
     /// Create a flownode manager with one worker
     pub fn new_with_worker<'s>(
         node_id: Option<u32>,
@@ -206,28 +216,6 @@ impl DiffRequest {
     }
 }
 
-/// iterate through the diff row and form continuous diff row with same diff type
-pub fn diff_row_to_request(rows: Vec<DiffRow>) -> Vec<DiffRequest> {
-    let mut reqs = Vec::new();
-    for (row, ts, diff) in rows {
-        let last = reqs.last_mut();
-        match (last, diff) {
-            (Some(DiffRequest::Insert(rows)), 1) => {
-                rows.push((row, ts));
-            }
-            (Some(DiffRequest::Insert(_)), -1) => reqs.push(DiffRequest::Delete(vec![(row, ts)])),
-            (Some(DiffRequest::Delete(rows)), -1) => {
-                rows.push((row, ts));
-            }
-            (Some(DiffRequest::Delete(_)), 1) => reqs.push(DiffRequest::Insert(vec![(row, ts)])),
-            (None, 1) => reqs.push(DiffRequest::Insert(vec![(row, ts)])),
-            (None, -1) => reqs.push(DiffRequest::Delete(vec![(row, ts)])),
-            _ => {}
-        }
-    }
-    reqs
-}
-
 pub fn batches_to_rows_req(batches: Vec<Batch>) -> Result<Vec<DiffRequest>, Error> {
     let mut reqs = Vec::new();
     for batch in batches {
@@ -522,6 +510,27 @@ impl FlowWorkerManager {
 
 /// Flow Runtime related methods
 impl FlowWorkerManager {
+    /// Start state report handler, which will receive a sender from HeartbeatTask to send state size report back
+    ///
+    /// if heartbeat task is shutdown, this future will exit too
+    async fn start_state_report_handler(self: Arc<Self>) -> Option<JoinHandle<()>> {
+        let state_report_handler = self.state_report_handler.write().await.take();
+        if let Some(mut handler) = state_report_handler {
+            let zelf = self.clone();
+            let handler = common_runtime::spawn_global(async move {
+                while let Some(ret_handler) = handler.recv().await {
+                    let state_report = zelf.gen_state_report().await;
+                    ret_handler.send(state_report).unwrap_or_else(|err| {
+                        common_telemetry::error!(err; "Send state size report error");
+                    });
+                }
+            });
+            Some(handler)
+        } else {
+            None
+        }
+    }
+
     /// run in common_runtime background runtime
     pub fn run_background(
         self: Arc<Self>,
@@ -529,6 +538,7 @@ impl FlowWorkerManager {
     ) -> JoinHandle<()> {
         info!("Starting flownode manager's background task");
         common_runtime::spawn_global(async move {
+            let _state_report_handler = self.clone().start_state_report_handler().await;
             self.run(shutdown).await;
         })
     }
@@ -555,6 +565,8 @@ impl FlowWorkerManager {
         let default_interval = Duration::from_secs(1);
         let mut avg_spd = 0; // rows/sec
         let mut since_last_run = tokio::time::Instant::now();
+        let run_per_trace = 10;
+        let mut run_cnt = 0;
         loop {
             // TODO(discord9): only run when new inputs arrive or scheduled to
             let row_cnt = self.run_available(true).await.unwrap_or_else(|err| {
@@ -597,10 +609,19 @@ impl FlowWorkerManager {
             } else {
                 (9 * avg_spd + cur_spd) / 10
             };
-            trace!("avg_spd={} r/s, cur_spd={} r/s", avg_spd, cur_spd);
             let new_wait = BATCH_SIZE * 1000 / avg_spd.max(1); //in ms
             let new_wait = Duration::from_millis(new_wait as u64).min(default_interval);
-            trace!("Wait for {} ms, row_cnt={}", new_wait.as_millis(), row_cnt);
+
+            // print trace every `run_per_trace` times so that we can see if there is something wrong
+            // but also not get flooded with trace
+            if run_cnt >= run_per_trace {
+                trace!("avg_spd={} r/s, cur_spd={} r/s", avg_spd, cur_spd);
+                trace!("Wait for {} ms, row_cnt={}", new_wait.as_millis(), row_cnt);
+                run_cnt = 0;
+            } else {
+                run_cnt += 1;
+            }
+
             METRIC_FLOW_RUN_INTERVAL_MS.set(new_wait.as_millis() as i64);
             since_last_run = tokio::time::Instant::now();
             tokio::time::sleep(new_wait).await;
@@ -660,13 +681,18 @@ impl FlowWorkerManager {
         &self,
         region_id: RegionId,
         rows: Vec<DiffRow>,
+        batch_datatypes: &[ConcreteDataType],
     ) -> Result<(), Error> {
         let rows_len = rows.len();
         let table_id = region_id.table_id();
         let _timer = METRIC_FLOW_INSERT_ELAPSED
             .with_label_values(&[table_id.to_string().as_str()])
             .start_timer();
-        self.node_context.read().await.send(table_id, rows).await?;
+        self.node_context
+            .read()
+            .await
+            .send(table_id, rows, batch_datatypes)
+            .await?;
         trace!(
             "Handling write request for table_id={} with {} rows",
             table_id,

src/flow/src/adapter/flownode_impl.rs

@@ -28,6 +28,7 @@ use itertools::Itertools;
 use snafu::{OptionExt, ResultExt};
 use store_api::storage::RegionId;
 
+use super::util::from_proto_to_data_type;
 use crate::adapter::{CreateFlowArgs, FlowWorkerManager};
 use crate::error::InternalSnafu;
 use crate::metrics::METRIC_FLOW_TASK_COUNT;
@@ -206,9 +207,17 @@ impl Flownode for FlowWorkerManager {
                 })
                 .map(|r| (r, now, 1))
                 .collect_vec();
-            self.handle_write_request(region_id.into(), rows)
+            let batch_datatypes = insert_schema
-                .await
+                .iter()
+                .map(from_proto_to_data_type)
+                .collect::<std::result::Result<Vec<_>, _>>()
                 .map_err(to_meta_err)?;
+            self.handle_write_request(region_id.into(), rows, &batch_datatypes)
+                .await
+                .map_err(|err| {
+                    common_telemetry::error!(err;"Failed to handle write request");
+                    to_meta_err(err)
+                })?;
         }
         Ok(Default::default())
     }

src/flow/src/adapter/node_context.rs

@@ -19,6 +19,7 @@ use std::sync::atomic::{AtomicUsize, Ordering};
 use std::sync::Arc;
 
 use common_telemetry::trace;
+use datatypes::prelude::ConcreteDataType;
 use session::context::QueryContext;
 use snafu::{OptionExt, ResultExt};
 use table::metadata::TableId;
@@ -131,7 +132,11 @@ impl SourceSender {
     }
 
     /// return number of rows it actual send(including what's in the buffer)
-    pub async fn send_rows(&self, rows: Vec<DiffRow>) -> Result<usize, Error> {
+    pub async fn send_rows(
+        &self,
+        rows: Vec<DiffRow>,
+        batch_datatypes: &[ConcreteDataType],
+    ) -> Result<usize, Error> {
         METRIC_FLOW_INPUT_BUF_SIZE.add(rows.len() as _);
         while self.send_buf_row_cnt.load(Ordering::SeqCst) >= BATCH_SIZE * 4 {
             tokio::task::yield_now().await;
@@ -139,8 +144,11 @@ impl SourceSender {
         // row count metrics is approx so relaxed order is ok
         self.send_buf_row_cnt
             .fetch_add(rows.len(), Ordering::SeqCst);
-        let batch = Batch::try_from_rows(rows.into_iter().map(|(row, _, _)| row).collect())
+        let batch = Batch::try_from_rows_with_types(
-            .context(EvalSnafu)?;
+            rows.into_iter().map(|(row, _, _)| row).collect(),
+            batch_datatypes,
+        )
+        .context(EvalSnafu)?;
         common_telemetry::trace!("Send one batch to worker with {} rows", batch.row_count());
         self.send_buf_tx.send(batch).await.map_err(|e| {
             crate::error::InternalSnafu {
@@ -157,14 +165,19 @@ impl FlownodeContext {
     /// return number of rows it actual send(including what's in the buffer)
     ///
     /// TODO(discord9): make this concurrent
-    pub async fn send(&self, table_id: TableId, rows: Vec<DiffRow>) -> Result<usize, Error> {
+    pub async fn send(
+        &self,
+        table_id: TableId,
+        rows: Vec<DiffRow>,
+        batch_datatypes: &[ConcreteDataType],
+    ) -> Result<usize, Error> {
         let sender = self
             .source_sender
             .get(&table_id)
             .with_context(|| TableNotFoundSnafu {
                 name: table_id.to_string(),
             })?;
-        sender.send_rows(rows).await
+        sender.send_rows(rows, batch_datatypes).await
     }
 
     /// flush all sender's buf

src/flow/src/adapter/stat.rs

@@ -0,0 +1,40 @@
+// Copyright 2023 Greptime Team
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+use std::collections::BTreeMap;
+
+use common_meta::key::flow::flow_state::FlowStat;
+
+use crate::FlowWorkerManager;
+
+impl FlowWorkerManager {
+    pub async fn gen_state_report(&self) -> FlowStat {
+        let mut full_report = BTreeMap::new();
+        for worker in self.worker_handles.iter() {
+            let worker = worker.lock().await;
+            match worker.get_state_size().await {
+                Ok(state_size) => {
+                    full_report.extend(state_size.into_iter().map(|(k, v)| (k as u32, v)))
+                }
+                Err(err) => {
+                    common_telemetry::error!(err; "Get flow stat size error");
+                }
+            }
+        }
+
+        FlowStat {
+            state_size: full_report,
+        }
+    }
+}

src/flow/src/adapter/util.rs

@@ -16,12 +16,27 @@ use api::helper::ColumnDataTypeWrapper;
 use api::v1::column_def::options_from_column_schema;
 use api::v1::{ColumnDataType, ColumnDataTypeExtension, SemanticType};
 use common_error::ext::BoxedError;
+use datatypes::prelude::ConcreteDataType;
 use datatypes::schema::ColumnSchema;
 use itertools::Itertools;
 use snafu::ResultExt;
 
 use crate::error::{Error, ExternalSnafu};
 
+pub fn from_proto_to_data_type(
+    column_schema: &api::v1::ColumnSchema,
+) -> Result<ConcreteDataType, Error> {
+    let wrapper = ColumnDataTypeWrapper::try_new(
+        column_schema.datatype,
+        column_schema.datatype_extension.clone(),
+    )
+    .map_err(BoxedError::new)
+    .context(ExternalSnafu)?;
+    let cdt = ConcreteDataType::from(wrapper);
+
+    Ok(cdt)
+}
+
 /// convert `ColumnSchema` lists to it's corresponding proto type
 pub fn column_schemas_to_proto(
     column_schemas: Vec<ColumnSchema>,

src/flow/src/adapter/worker.rs

@@ -197,6 +197,21 @@ impl WorkerHandle {
             .fail()
         }
     }
+
+    pub async fn get_state_size(&self) -> Result<BTreeMap<FlowId, usize>, Error> {
+        let ret = self
+            .itc_client
+            .call_with_resp(Request::QueryStateSize)
+            .await?;
+        ret.into_query_state_size().map_err(|ret| {
+            InternalSnafu {
+                reason: format!(
+                    "Flow Node/Worker itc failed, expect Response::QueryStateSize, found {ret:?}"
+                ),
+            }
+            .build()
+        })
+    }
 }
 
 impl Drop for WorkerHandle {
@@ -361,6 +376,13 @@ impl<'s> Worker<'s> {
                 Some(Response::ContainTask { result: ret })
             }
             Request::Shutdown => return Err(()),
+            Request::QueryStateSize => {
+                let mut ret = BTreeMap::new();
+                for (flow_id, task_state) in self.task_states.iter() {
+                    ret.insert(*flow_id, task_state.state.get_state_size());
+                }
+                Some(Response::QueryStateSize { result: ret })
+            }
         };
         Ok(ret)
     }
@@ -391,6 +413,7 @@ pub enum Request {
         flow_id: FlowId,
     },
     Shutdown,
+    QueryStateSize,
 }
 
 #[derive(Debug, EnumAsInner)]
@@ -406,6 +429,10 @@ enum Response {
         result: bool,
     },
     RunAvail,
+    QueryStateSize {
+        /// each flow tasks' state size
+        result: BTreeMap<FlowId, usize>,
+    },
 }
 
 fn create_inter_thread_call() -> (InterThreadCallClient, InterThreadCallServer) {
@@ -423,10 +450,12 @@ struct InterThreadCallClient {
 }
 
 impl InterThreadCallClient {
+    /// call without response
     fn call_no_resp(&self, req: Request) -> Result<(), Error> {
         self.arg_sender.send((req, None)).map_err(from_send_error)
     }
 
+    /// call with response
     async fn call_with_resp(&self, req: Request) -> Result<Response, Error> {
         let (tx, rx) = oneshot::channel();
         self.arg_sender
@@ -527,6 +556,7 @@ mod test {
         );
         tx.send(Batch::empty()).unwrap();
         handle.run_available(0, true).await.unwrap();
+        assert_eq!(handle.get_state_size().await.unwrap().len(), 1);
         assert_eq!(sink_rx.recv().await.unwrap(), Batch::empty());
         drop(handle);
         worker_thread_handle.join().unwrap();

src/flow/src/compute/render.rs

@@ -30,7 +30,7 @@ use crate::compute::types::{Collection, CollectionBundle, ErrCollector, Toff};
 use crate::error::{Error, InvalidQuerySnafu, NotImplementedSnafu};
 use crate::expr::{self, Batch, GlobalId, LocalId};
 use crate::plan::{Plan, TypedPlan};
-use crate::repr::{self, DiffRow};
+use crate::repr::{self, DiffRow, RelationType};
 
 mod map;
 mod reduce;
@@ -124,10 +124,10 @@ impl Context<'_, '_> {
     /// Like `render_plan` but in Batch Mode
     pub fn render_plan_batch(&mut self, plan: TypedPlan) -> Result<CollectionBundle<Batch>, Error> {
         match plan.plan {
-            Plan::Constant { rows } => Ok(self.render_constant_batch(rows)),
+            Plan::Constant { rows } => Ok(self.render_constant_batch(rows, &plan.schema.typ)),
             Plan::Get { id } => self.get_batch_by_id(id),
             Plan::Let { id, value, body } => self.eval_batch_let(id, value, body),
-            Plan::Mfp { input, mfp } => self.render_mfp_batch(input, mfp),
+            Plan::Mfp { input, mfp } => self.render_mfp_batch(input, mfp, &plan.schema.typ),
             Plan::Reduce {
                 input,
                 key_val_plan,
@@ -172,7 +172,11 @@ impl Context<'_, '_> {
     /// render Constant, take all rows that have a timestamp not greater than the current time
     /// This function is primarily used for testing
     /// Always assume input is sorted by timestamp
-    pub fn render_constant_batch(&mut self, rows: Vec<DiffRow>) -> CollectionBundle<Batch> {
+    pub fn render_constant_batch(
+        &mut self,
+        rows: Vec<DiffRow>,
+        output_type: &RelationType,
+    ) -> CollectionBundle<Batch> {
         let (send_port, recv_port) = self.df.make_edge::<_, Toff<Batch>>("constant_batch");
         let mut per_time: BTreeMap<repr::Timestamp, Vec<DiffRow>> = Default::default();
         for (key, group) in &rows.into_iter().group_by(|(_row, ts, _diff)| *ts) {
@@ -185,6 +189,8 @@ impl Context<'_, '_> {
         let scheduler_inner = scheduler.clone();
         let err_collector = self.err_collector.clone();
 
+        let output_type = output_type.clone();
+
         let subgraph_id =
             self.df
                 .add_subgraph_source("ConstantBatch", send_port, move |_ctx, send_port| {
@@ -199,7 +205,14 @@ impl Context<'_, '_> {
                     not_great_than_now.into_iter().for_each(|(_ts, rows)| {
                         err_collector.run(|| {
                             let rows = rows.into_iter().map(|(row, _ts, _diff)| row).collect();
-                            let batch = Batch::try_from_rows(rows)?;
+                            let batch = Batch::try_from_rows_with_types(
+                                rows,
+                                &output_type
+                                    .column_types
+                                    .iter()
+                                    .map(|ty| ty.scalar_type().clone())
+                                    .collect_vec(),
+                            )?;
                             send_port.give(vec![batch]);
                             Ok(())
                         });

src/flow/src/compute/render/map.rs

@@ -25,7 +25,7 @@ use crate::compute::types::{Arranged, Collection, CollectionBundle, ErrCollector
 use crate::error::{Error, PlanSnafu};
 use crate::expr::{Batch, EvalError, MapFilterProject, MfpPlan, ScalarExpr};
 use crate::plan::TypedPlan;
-use crate::repr::{self, DiffRow, KeyValDiffRow, Row};
+use crate::repr::{self, DiffRow, KeyValDiffRow, RelationType, Row};
 use crate::utils::ArrangeHandler;
 
 impl Context<'_, '_> {
@@ -34,6 +34,7 @@ impl Context<'_, '_> {
         &mut self,
         input: Box<TypedPlan>,
         mfp: MapFilterProject,
+        _output_type: &RelationType,
     ) -> Result<CollectionBundle<Batch>, Error> {
         let input = self.render_plan_batch(*input)?;
 

src/flow/src/compute/render/reduce.rs

@@ -87,6 +87,8 @@ impl Context<'_, '_> {
         })?;
         let key_val_plan = key_val_plan.clone();
 
+        let output_type = output_type.clone();
+
         let now = self.compute_state.current_time_ref();
 
         let err_collector = self.err_collector.clone();
@@ -118,6 +120,7 @@ impl Context<'_, '_> {
                     src_data,
                     &key_val_plan,
                     &accum_plan,
+                    &output_type,
                     SubgraphArg {
                         now,
                         err_collector: &err_collector,
@@ -354,6 +357,7 @@ fn reduce_batch_subgraph(
     src_data: impl IntoIterator<Item = Batch>,
     key_val_plan: &KeyValPlan,
     accum_plan: &AccumulablePlan,
+    output_type: &RelationType,
     SubgraphArg {
         now,
         err_collector,
@@ -535,17 +539,13 @@ fn reduce_batch_subgraph(
     // this output part is not supposed to be resource intensive
     // (because for every batch there wouldn't usually be as many output row?),
     // so we can do some costly operation here
-    let output_types = all_output_dict.first_entry().map(|entry| {
+    let output_types = output_type
-        entry
+        .column_types
-            .key()
+        .iter()
-            .iter()
+        .map(|t| t.scalar_type.clone())
-            .chain(entry.get().iter())
+        .collect_vec();
-            .map(|v| v.data_type())
-            .collect::<Vec<ConcreteDataType>>()
-    });
 
-    if let Some(output_types) = output_types {
+    err_collector.run(|| {
-        err_collector.run(|| {
             let column_cnt = output_types.len();
             let row_cnt = all_output_dict.len();
 
@@ -585,7 +585,6 @@ fn reduce_batch_subgraph(
 
             Ok(())
         });
-    }
 }
 
 /// reduce subgraph, reduce the input data into a single row
@@ -1516,7 +1515,9 @@ mod test {
         let mut ctx = harness_test_ctx(&mut df, &mut state);
 
         let rows = vec![
-            (Row::new(vec![1i64.into()]), 1, 1),
+            (Row::new(vec![Value::Null]), -1, 1),
+            (Row::new(vec![1i64.into()]), 0, 1),
+            (Row::new(vec![Value::Null]), 1, 1),
             (Row::new(vec![2i64.into()]), 2, 1),
             (Row::new(vec![3i64.into()]), 3, 1),
             (Row::new(vec![1i64.into()]), 4, 1),
@@ -1558,13 +1559,15 @@ mod test {
                 Box::new(input_plan.with_types(typ.into_unnamed())),
                 &key_val_plan,
                 &reduce_plan,
-                &RelationType::empty(),
+                &RelationType::new(vec![ColumnType::new(CDT::int64_datatype(), true)]),
             )
             .unwrap();
 
         {
             let now_inner = now.clone();
             let expected = BTreeMap::<i64, Vec<i64>>::from([
+                (-1, vec![]),
+                (0, vec![1i64]),
                 (1, vec![1i64]),
                 (2, vec![3i64]),
                 (3, vec![6i64]),
@@ -1581,7 +1584,11 @@ mod test {
 
                     if let Some(expected) = expected.get(&now) {
                         let batch = expected.iter().map(|v| Value::from(*v)).collect_vec();
-                        let batch = Batch::try_from_rows(vec![batch.into()]).unwrap();
+                        let batch = Batch::try_from_rows_with_types(
+                            vec![batch.into()],
+                            &[CDT::int64_datatype()],
+                        )
+                        .unwrap();
                         assert_eq!(res.first(), Some(&batch));
                     }
                 });

src/flow/src/compute/render/src_sink.rs

@@ -14,7 +14,7 @@
 
 //! Source and Sink for the dataflow
 
-use std::collections::{BTreeMap, VecDeque};
+use std::collections::BTreeMap;
 
 use common_telemetry::{debug, trace};
 use hydroflow::scheduled::graph_ext::GraphExt;
@@ -28,7 +28,7 @@ use crate::compute::types::{Arranged, Collection, CollectionBundle, Toff};
 use crate::error::{Error, PlanSnafu};
 use crate::expr::error::InternalSnafu;
 use crate::expr::{Batch, EvalError};
-use crate::repr::{DiffRow, Row, BROADCAST_CAP};
+use crate::repr::{DiffRow, Row};
 
 #[allow(clippy::mutable_key_type)]
 impl Context<'_, '_> {
@@ -242,44 +242,4 @@ impl Context<'_, '_> {
             },
         );
     }
-
-    /// Render a sink which send updates to broadcast channel, have internal buffer in case broadcast channel is full
-    pub fn render_sink(&mut self, bundle: CollectionBundle, sender: broadcast::Sender<DiffRow>) {
-        let CollectionBundle {
-            collection,
-            arranged: _,
-        } = bundle;
-        let mut buf = VecDeque::with_capacity(1000);
-
-        let schd = self.compute_state.get_scheduler();
-        let inner_schd = schd.clone();
-        let now = self.compute_state.current_time_ref();
-
-        let sink = self
-            .df
-            .add_subgraph_sink("Sink", collection.into_inner(), move |_ctx, recv| {
-                let data = recv.take_inner();
-                buf.extend(data.into_iter().flat_map(|i| i.into_iter()));
-                if sender.len() >= BROADCAST_CAP {
-                    return;
-                } else {
-                    while let Some(row) = buf.pop_front() {
-                        // if the sender is full, stop sending
-                        if sender.len() >= BROADCAST_CAP {
-                            break;
-                        }
-                        // TODO(discord9): handling tokio broadcast error
-                        let _ = sender.send(row);
-                    }
-                }
-
-                // if buffer is not empty, schedule the next run at next tick
-                // so the buffer can be drained as soon as possible
-                if !buf.is_empty() {
-                    inner_schd.schedule_at(*now.borrow() + 1);
-                }
-            });
-
-        schd.set_cur_subgraph(sink);
-    }
 }

src/flow/src/compute/state.rs

@@ -16,6 +16,7 @@ use std::cell::RefCell;
 use std::collections::{BTreeMap, VecDeque};
 use std::rc::Rc;
 
+use get_size2::GetSize;
 use hydroflow::scheduled::graph::Hydroflow;
 use hydroflow::scheduled::SubgraphId;
 
@@ -109,6 +110,10 @@ impl DataflowState {
     pub fn expire_after(&self) -> Option<Timestamp> {
         self.expire_after
     }
+
+    pub fn get_state_size(&self) -> usize {
+        self.arrange_used.iter().map(|x| x.read().get_size()).sum()
+    }
 }
 
 #[derive(Debug, Clone)]

src/flow/src/compute/types.rs

@@ -82,22 +82,6 @@ impl Arranged {
                 writer: self.writer.clone(),
             })
     }
-
-    /// Copy the full arrangement, including the future and the current updates.
-    ///
-    /// Internally `Rc-ed` so it's cheap to copy
-    pub fn try_copy_full(&self) -> Option<Self> {
-        self.arrangement
-            .clone_full_arrange()
-            .map(|arrangement| Arranged {
-                arrangement,
-                readers: self.readers.clone(),
-                writer: self.writer.clone(),
-            })
-    }
-    pub fn add_reader(&self, id: SubgraphId) {
-        self.readers.borrow_mut().push(id)
-    }
 }
 
 /// A bundle of the various ways a collection can be represented.

src/flow/src/expr.rs

@@ -24,7 +24,7 @@ mod scalar;
 mod signature;
 
 use arrow::compute::FilterBuilder;
-use datatypes::prelude::DataType;
+use datatypes::prelude::{ConcreteDataType, DataType};
 use datatypes::value::Value;
 use datatypes::vectors::{BooleanVector, Helper, VectorRef};
 pub(crate) use df_func::{DfScalarFunction, RawDfScalarFn};
@@ -85,16 +85,18 @@ impl Default for Batch {
 }
 
 impl Batch {
-    pub fn try_from_rows(rows: Vec<crate::repr::Row>) -> Result<Self, EvalError> {
+    /// Get batch from rows, will try best to determine data type
+    pub fn try_from_rows_with_types(
+        rows: Vec<crate::repr::Row>,
+        batch_datatypes: &[ConcreteDataType],
+    ) -> Result<Self, EvalError> {
         if rows.is_empty() {
             return Ok(Self::empty());
         }
         let len = rows.len();
-        let mut builder = rows
+        let mut builder = batch_datatypes
-            .first()
-            .unwrap()
             .iter()
-            .map(|v| v.data_type().create_mutable_vector(len))
+            .map(|ty| ty.create_mutable_vector(len))
             .collect_vec();
         for row in rows {
             ensure!(
@@ -221,10 +223,25 @@ impl Batch {
             return Ok(());
         }
 
-        let dts = if self.batch.is_empty() {
+        let dts = {
-            other.batch.iter().map(|v| v.data_type()).collect_vec()
+            let max_len = self.batch.len().max(other.batch.len());
-        } else {
+            let mut dts = Vec::with_capacity(max_len);
-            self.batch.iter().map(|v| v.data_type()).collect_vec()
+            for i in 0..max_len {
+                if let Some(v) = self.batch().get(i)
+                    && !v.data_type().is_null()
+                {
+                    dts.push(v.data_type())
+                } else if let Some(v) = other.batch().get(i)
+                    && !v.data_type().is_null()
+                {
+                    dts.push(v.data_type())
+                } else {
+                    // both are null, so we will push null type
+                    dts.push(datatypes::prelude::ConcreteDataType::null_datatype())
+                }
+            }
+
+            dts
         };
 
         let batch_builders = dts

src/flow/src/expr/error.rs

@@ -21,11 +21,6 @@ use datafusion_common::DataFusionError;
 use datatypes::data_type::ConcreteDataType;
 use snafu::{Location, Snafu};
 
-fn is_send_sync() {
-    fn check<T: Send + Sync>() {}
-    check::<EvalError>();
-}
-
 /// EvalError is about errors happen on columnar evaluation
 ///
 /// TODO(discord9): add detailed location of column/operator(instead of code) to errors tp help identify related column

src/flow/src/expr/linear.rs

@@ -359,14 +359,6 @@ impl MapFilterProject {
         )
     }
 
-    /// Convert the `MapFilterProject` into a staged evaluation plan.
-    ///
-    /// The main behavior is extract temporal predicates, which cannot be evaluated
-    /// using the standard machinery.
-    pub fn into_plan(self) -> Result<MfpPlan, Error> {
-        MfpPlan::create_from(self)
-    }
-
     /// Lists input columns whose values are used in outputs.
     ///
     /// It is entirely appropriate to determine the demand of an instance
@@ -602,26 +594,6 @@ impl SafeMfpPlan {
         }
     }
 
-    /// A version of `evaluate` which produces an iterator over `Datum`
-    /// as output.
-    ///
-    /// This version can be useful when one wants to capture the resulting
-    /// datums without packing and then unpacking a row.
-    #[inline(always)]
-    pub fn evaluate_iter<'a>(
-        &'a self,
-        datums: &'a mut Vec<Value>,
-    ) -> Result<Option<impl Iterator<Item = Value> + 'a>, EvalError> {
-        let passed_predicates = self.evaluate_inner(datums)?;
-        if !passed_predicates {
-            Ok(None)
-        } else {
-            Ok(Some(
-                self.mfp.projection.iter().map(move |i| datums[*i].clone()),
-            ))
-        }
-    }
-
     /// Populates `values` with `self.expressions` and tests `self.predicates`.
     ///
     /// This does not apply `self.projection`, which is up to the calling method.
@@ -936,20 +908,33 @@ mod test {
             .unwrap()
             .unwrap();
         assert_eq!(ret, Row::pack(vec![Value::from(false), Value::from(true)]));
-
+        let ty = [
+            ConcreteDataType::int32_datatype(),
+            ConcreteDataType::int32_datatype(),
+            ConcreteDataType::int32_datatype(),
+        ];
         // batch mode
-        let mut batch = Batch::try_from_rows(vec![Row::from(vec![
+        let mut batch = Batch::try_from_rows_with_types(
-            Value::from(4),
+            vec![Row::from(vec![
-            Value::from(2),
+                Value::from(4),
-            Value::from(3),
+                Value::from(2),
-        ])])
+                Value::from(3),
+            ])],
+            &ty,
+        )
         .unwrap();
         let ret = safe_mfp.eval_batch_into(&mut batch).unwrap();
 
         assert_eq!(
             ret,
-            Batch::try_from_rows(vec![Row::from(vec![Value::from(false), Value::from(true)])])
+            Batch::try_from_rows_with_types(
-                .unwrap()
+                vec![Row::from(vec![Value::from(false), Value::from(true)])],
+                &[
+                    ConcreteDataType::boolean_datatype(),
+                    ConcreteDataType::boolean_datatype(),
+                ],
+            )
+            .unwrap()
         );
     }
 
@@ -984,7 +969,15 @@ mod test {
             .unwrap();
         assert_eq!(ret, None);
 
-        let mut input1_batch = Batch::try_from_rows(vec![Row::new(input1)]).unwrap();
+        let input_type = [
+            ConcreteDataType::int32_datatype(),
+            ConcreteDataType::int32_datatype(),
+            ConcreteDataType::int32_datatype(),
+            ConcreteDataType::string_datatype(),
+        ];
+
+        let mut input1_batch =
+            Batch::try_from_rows_with_types(vec![Row::new(input1)], &input_type).unwrap();
         let ret_batch = safe_mfp.eval_batch_into(&mut input1_batch).unwrap();
         assert_eq!(
             ret_batch,
@@ -1002,7 +995,8 @@ mod test {
             .unwrap();
         assert_eq!(ret, Some(Row::pack(vec![Value::from(11)])));
 
-        let mut input2_batch = Batch::try_from_rows(vec![Row::new(input2)]).unwrap();
+        let mut input2_batch =
+            Batch::try_from_rows_with_types(vec![Row::new(input2)], &input_type).unwrap();
         let ret_batch = safe_mfp.eval_batch_into(&mut input2_batch).unwrap();
         assert_eq!(
             ret_batch,
@@ -1055,7 +1049,14 @@ mod test {
         let ret = safe_mfp.evaluate_into(&mut input1.clone(), &mut Row::empty());
         assert!(matches!(ret, Err(EvalError::InvalidArgument { .. })));
 
-        let mut input1_batch = Batch::try_from_rows(vec![Row::new(input1)]).unwrap();
+        let input_type = [
+            ConcreteDataType::int64_datatype(),
+            ConcreteDataType::int32_datatype(),
+            ConcreteDataType::int32_datatype(),
+            ConcreteDataType::int32_datatype(),
+        ];
+        let mut input1_batch =
+            Batch::try_from_rows_with_types(vec![Row::new(input1)], &input_type).unwrap();
         let ret_batch = safe_mfp.eval_batch_into(&mut input1_batch);
         assert!(matches!(ret_batch, Err(EvalError::InvalidArgument { .. })));
 
@@ -1065,7 +1066,13 @@ mod test {
             .unwrap();
         assert_eq!(ret, Some(Row::new(input2.clone())));
 
-        let input2_batch = Batch::try_from_rows(vec![Row::new(input2)]).unwrap();
+        let input_type = [
+            ConcreteDataType::int64_datatype(),
+            ConcreteDataType::int32_datatype(),
+            ConcreteDataType::int32_datatype(),
+        ];
+        let input2_batch =
+            Batch::try_from_rows_with_types(vec![Row::new(input2)], &input_type).unwrap();
         let ret_batch = safe_mfp.eval_batch_into(&mut input2_batch.clone()).unwrap();
         assert_eq!(ret_batch, input2_batch);
 
@@ -1075,7 +1082,8 @@ mod test {
             .unwrap();
         assert_eq!(ret, None);
 
-        let input3_batch = Batch::try_from_rows(vec![Row::new(input3)]).unwrap();
+        let input3_batch =
+            Batch::try_from_rows_with_types(vec![Row::new(input3)], &input_type).unwrap();
         let ret_batch = safe_mfp.eval_batch_into(&mut input3_batch.clone()).unwrap();
         assert_eq!(
             ret_batch,
@@ -1111,7 +1119,13 @@ mod test {
         let ret = safe_mfp.evaluate_into(&mut input1.clone(), &mut Row::empty());
         assert_eq!(ret.unwrap(), Some(Row::new(vec![Value::from(false)])));
 
-        let mut input1_batch = Batch::try_from_rows(vec![Row::new(input1)]).unwrap();
+        let input_type = [
+            ConcreteDataType::int32_datatype(),
+            ConcreteDataType::int32_datatype(),
+            ConcreteDataType::int32_datatype(),
+        ];
+        let mut input1_batch =
+            Batch::try_from_rows_with_types(vec![Row::new(input1)], &input_type).unwrap();
         let ret_batch = safe_mfp.eval_batch_into(&mut input1_batch).unwrap();
 
         assert_eq!(

src/flow/src/heartbeat.rs

@@ -24,6 +24,7 @@ use common_meta::heartbeat::handler::{
 };
 use common_meta::heartbeat::mailbox::{HeartbeatMailbox, MailboxRef, OutgoingMessage};
 use common_meta::heartbeat::utils::outgoing_message_to_mailbox_message;
+use common_meta::key::flow::flow_state::FlowStat;
 use common_telemetry::{debug, error, info, warn};
 use greptime_proto::v1::meta::NodeInfo;
 use meta_client::client::{HeartbeatSender, HeartbeatStream, MetaClient};
@@ -34,8 +35,27 @@ use tokio::sync::mpsc;
 use tokio::time::Duration;
 
 use crate::error::ExternalSnafu;
+use crate::utils::SizeReportSender;
 use crate::{Error, FlownodeOptions};
 
+async fn query_flow_state(
+    query_stat_size: &Option<SizeReportSender>,
+    timeout: Duration,
+) -> Option<FlowStat> {
+    if let Some(report_requester) = query_stat_size.as_ref() {
+        let ret = report_requester.query(timeout).await;
+        match ret {
+            Ok(latest) => Some(latest),
+            Err(err) => {
+                error!(err; "Failed to get query stat size");
+                None
+            }
+        }
+    } else {
+        None
+    }
+}
+
 /// The flownode heartbeat task which sending `[HeartbeatRequest]` to Metasrv periodically in background.
 #[derive(Clone)]
 pub struct HeartbeatTask {
@@ -47,9 +67,14 @@ pub struct HeartbeatTask {
     resp_handler_executor: HeartbeatResponseHandlerExecutorRef,
     start_time_ms: u64,
     running: Arc<AtomicBool>,
+    query_stat_size: Option<SizeReportSender>,
 }
 
 impl HeartbeatTask {
+    pub fn with_query_stat_size(mut self, query_stat_size: SizeReportSender) -> Self {
+        self.query_stat_size = Some(query_stat_size);
+        self
+    }
     pub fn new(
         opts: &FlownodeOptions,
         meta_client: Arc<MetaClient>,
@@ -65,6 +90,7 @@ impl HeartbeatTask {
             resp_handler_executor,
             start_time_ms: common_time::util::current_time_millis() as u64,
             running: Arc::new(AtomicBool::new(false)),
+            query_stat_size: None,
         }
     }
 
@@ -112,6 +138,7 @@ impl HeartbeatTask {
         message: Option<OutgoingMessage>,
         peer: Option<Peer>,
         start_time_ms: u64,
+        latest_report: &Option<FlowStat>,
     ) -> Option<HeartbeatRequest> {
         let mailbox_message = match message.map(outgoing_message_to_mailbox_message) {
             Some(Ok(message)) => Some(message),
@@ -121,11 +148,22 @@ impl HeartbeatTask {
             }
             None => None,
         };
+        let flow_stat = latest_report
+            .as_ref()
+            .map(|report| {
+                report
+                    .state_size
+                    .iter()
+                    .map(|(k, v)| (*k, *v as u64))
+                    .collect()
+            })
+            .map(|f| api::v1::meta::FlowStat { flow_stat_size: f });
 
         Some(HeartbeatRequest {
             mailbox_message,
             peer,
             info: Self::build_node_info(start_time_ms),
+            flow_stat,
             ..Default::default()
         })
     }
@@ -151,24 +189,27 @@ impl HeartbeatTask {
             addr: self.peer_addr.clone(),
         });
 
+        let query_stat_size = self.query_stat_size.clone();
+
         common_runtime::spawn_hb(async move {
             // note that using interval will cause it to first immediately send
             // a heartbeat
             let mut interval = tokio::time::interval(report_interval);
             interval.set_missed_tick_behavior(tokio::time::MissedTickBehavior::Delay);
+            let mut latest_report = None;
 
             loop {
                 let req = tokio::select! {
                     message = outgoing_rx.recv() => {
                         if let Some(message) = message {
-                            Self::create_heartbeat_request(Some(message), self_peer.clone(), start_time_ms)
+                            Self::create_heartbeat_request(Some(message), self_peer.clone(), start_time_ms, &latest_report)
                         } else {
                             // Receives None that means Sender was dropped, we need to break the current loop
                             break
                         }
                     }
                     _ = interval.tick() => {
-                        Self::create_heartbeat_request(None, self_peer.clone(), start_time_ms)
+                        Self::create_heartbeat_request(None, self_peer.clone(), start_time_ms, &latest_report)
                     }
                 };
 
@@ -180,6 +221,10 @@ impl HeartbeatTask {
                         debug!("Send a heartbeat request to metasrv, content: {:?}", req);
                     }
                 }
+                // after sending heartbeat, try to get the latest report
+                // TODO(discord9): consider a better place to update the size report
+                // set the timeout to half of the report interval so that it wouldn't delay heartbeat if something went horribly wrong
+                latest_report = query_flow_state(&query_stat_size, report_interval / 2).await;
             }
         });
     }

src/flow/src/plan.rs

@@ -18,10 +18,8 @@
 mod join;
 mod reduce;
 
-use std::collections::BTreeSet;
-
 use crate::error::Error;
-use crate::expr::{GlobalId, Id, LocalId, MapFilterProject, SafeMfpPlan, TypedExpr};
+use crate::expr::{Id, LocalId, MapFilterProject, SafeMfpPlan, TypedExpr};
 use crate::plan::join::JoinPlan;
 pub(crate) use crate::plan::reduce::{AccumulablePlan, AggrWithIndex, KeyValPlan, ReducePlan};
 use crate::repr::{DiffRow, RelationDesc};
@@ -186,48 +184,6 @@ pub enum Plan {
     },
 }
 
-impl Plan {
-    /// Find all the used collection in the plan
-    pub fn find_used_collection(&self) -> BTreeSet<GlobalId> {
-        fn recur_find_use(plan: &Plan, used: &mut BTreeSet<GlobalId>) {
-            match plan {
-                Plan::Get { id } => {
-                    match id {
-                        Id::Local(_) => (),
-                        Id::Global(g) => {
-                            used.insert(*g);
-                        }
-                    };
-                }
-                Plan::Let { value, body, .. } => {
-                    recur_find_use(&value.plan, used);
-                    recur_find_use(&body.plan, used);
-                }
-                Plan::Mfp { input, .. } => {
-                    recur_find_use(&input.plan, used);
-                }
-                Plan::Reduce { input, .. } => {
-                    recur_find_use(&input.plan, used);
-                }
-                Plan::Join { inputs, .. } => {
-                    for input in inputs {
-                        recur_find_use(&input.plan, used);
-                    }
-                }
-                Plan::Union { inputs, .. } => {
-                    for input in inputs {
-                        recur_find_use(&input.plan, used);
-                    }
-                }
-                _ => {}
-            }
-        }
-        let mut ret = Default::default();
-        recur_find_use(self, &mut ret);
-        ret
-    }
-}
-
 impl Plan {
     pub fn with_types(self, schema: RelationDesc) -> TypedPlan {
         TypedPlan { schema, plan: self }

src/flow/src/repr.rs

@@ -22,12 +22,14 @@ use api::v1::Row as ProtoRow;
 use datatypes::data_type::ConcreteDataType;
 use datatypes::types::cast;
 use datatypes::value::Value;
+use get_size2::GetSize;
 use itertools::Itertools;
 pub(crate) use relation::{ColumnType, Key, RelationDesc, RelationType};
 use serde::{Deserialize, Serialize};
 use snafu::ResultExt;
 
 use crate::expr::error::{CastValueSnafu, EvalError, InvalidArgumentSnafu};
+use crate::utils::get_value_heap_size;
 
 /// System-wide Record count difference type. Useful for capture data change
 ///
@@ -105,6 +107,12 @@ pub struct Row {
     pub inner: Vec<Value>,
 }
 
+impl GetSize for Row {
+    fn get_heap_size(&self) -> usize {
+        self.inner.iter().map(get_value_heap_size).sum()
+    }
+}
+
 impl Row {
     /// Create an empty row
     pub fn empty() -> Self {

src/flow/src/repr/relation.rs

@@ -46,14 +46,6 @@ impl Key {
         self.column_indices.push(col);
     }
 
-    /// Add columns to Key
-    pub fn add_cols<I>(&mut self, cols: I)
-    where
-        I: IntoIterator<Item = usize>,
-    {
-        self.column_indices.extend(cols);
-    }
-
     /// Remove a column from Key
     pub fn remove_col(&mut self, col: usize) {
         self.column_indices.retain(|&r| r != col);

src/flow/src/server.rs

@@ -55,6 +55,7 @@ use crate::error::{
 };
 use crate::heartbeat::HeartbeatTask;
 use crate::transform::register_function_to_query_engine;
+use crate::utils::{SizeReportSender, StateReportHandler};
 use crate::{Error, FlowWorkerManager, FlownodeOptions};
 
 pub const FLOW_NODE_SERVER_NAME: &str = "FLOW_NODE_SERVER";
@@ -236,6 +237,8 @@ pub struct FlownodeBuilder {
     catalog_manager: CatalogManagerRef,
     flow_metadata_manager: FlowMetadataManagerRef,
     heartbeat_task: Option<HeartbeatTask>,
+    /// receive a oneshot sender to send state size report
+    state_report_handler: Option<StateReportHandler>,
 }
 
 impl FlownodeBuilder {
@@ -254,17 +257,20 @@ impl FlownodeBuilder {
             catalog_manager,
             flow_metadata_manager,
             heartbeat_task: None,
+            state_report_handler: None,
         }
     }
 
     pub fn with_heartbeat_task(self, heartbeat_task: HeartbeatTask) -> Self {
+        let (sender, receiver) = SizeReportSender::new();
         Self {
-            heartbeat_task: Some(heartbeat_task),
+            heartbeat_task: Some(heartbeat_task.with_query_stat_size(sender)),
+            state_report_handler: Some(receiver),
             ..self
         }
     }
 
-    pub async fn build(self) -> Result<FlownodeInstance, Error> {
+    pub async fn build(mut self) -> Result<FlownodeInstance, Error> {
         // TODO(discord9): does this query engine need those?
         let query_engine_factory = QueryEngineFactory::new_with_plugins(
             // query engine in flownode is only used for translate plan with resolved table source.
@@ -383,7 +389,7 @@ impl FlownodeBuilder {
     /// build [`FlowWorkerManager`], note this doesn't take ownership of `self`,
     /// nor does it actually start running the worker.
     async fn build_manager(
-        &self,
+        &mut self,
         query_engine: Arc<dyn QueryEngine>,
     ) -> Result<FlowWorkerManager, Error> {
         let table_meta = self.table_meta.clone();
@@ -402,12 +408,15 @@ impl FlownodeBuilder {
                 info!("Flow Worker started in new thread");
                 worker.run();
             });
-        let man = rx.await.map_err(|_e| {
+        let mut man = rx.await.map_err(|_e| {
             UnexpectedSnafu {
                 reason: "sender is dropped, failed to create flow node manager",
             }
             .build()
         })?;
+        if let Some(handler) = self.state_report_handler.take() {
+            man = man.with_state_report_handler(handler).await;
+        }
         info!("Flow Node Manager started");
         Ok(man)
     }

src/flow/src/utils.rs

@@ -18,16 +18,73 @@ use std::collections::{BTreeMap, BTreeSet};
 use std::ops::Bound;
 use std::sync::Arc;
 
+use common_meta::key::flow::flow_state::FlowStat;
 use common_telemetry::trace;
+use datatypes::value::Value;
+use get_size2::GetSize;
 use smallvec::{smallvec, SmallVec};
-use tokio::sync::RwLock;
+use tokio::sync::{mpsc, oneshot, RwLock};
+use tokio::time::Instant;
 
+use crate::error::InternalSnafu;
 use crate::expr::{EvalError, ScalarExpr};
 use crate::repr::{value_to_internal_ts, DiffRow, Duration, KeyValDiffRow, Row, Timestamp};
 
 /// A batch of updates, arranged by key
 pub type Batch = BTreeMap<Row, SmallVec<[DiffRow; 2]>>;
 
+/// Get a estimate of heap size of a value
+pub fn get_value_heap_size(v: &Value) -> usize {
+    match v {
+        Value::Binary(bin) => bin.len(),
+        Value::String(s) => s.len(),
+        Value::List(list) => list.items().iter().map(get_value_heap_size).sum(),
+        _ => 0,
+    }
+}
+
+#[derive(Clone)]
+pub struct SizeReportSender {
+    inner: mpsc::Sender<oneshot::Sender<FlowStat>>,
+}
+
+impl SizeReportSender {
+    pub fn new() -> (Self, StateReportHandler) {
+        let (tx, rx) = mpsc::channel(1);
+        let zelf = Self { inner: tx };
+        (zelf, rx)
+    }
+
+    /// Query the size report, will timeout after one second if no response
+    pub async fn query(&self, timeout: std::time::Duration) -> crate::Result<FlowStat> {
+        let (tx, rx) = oneshot::channel();
+        self.inner.send(tx).await.map_err(|_| {
+            InternalSnafu {
+                reason: "failed to send size report request due to receiver dropped",
+            }
+            .build()
+        })?;
+        let timeout = tokio::time::timeout(timeout, rx);
+        timeout
+            .await
+            .map_err(|_elapsed| {
+                InternalSnafu {
+                    reason: "failed to receive size report after one second timeout",
+                }
+                .build()
+            })?
+            .map_err(|_| {
+                InternalSnafu {
+                    reason: "failed to receive size report due to sender dropped",
+                }
+                .build()
+            })
+    }
+}
+
+/// Handle the size report request, and send the report back
+pub type StateReportHandler = mpsc::Receiver<oneshot::Sender<FlowStat>>;
+
 /// A spine of batches, arranged by timestamp
 /// TODO(discord9): consider internally index by key, value, and timestamp for faster lookup
 pub type Spine = BTreeMap<Timestamp, Batch>;
@@ -49,6 +106,24 @@ pub struct KeyExpiryManager {
     event_timestamp_from_row: Option<ScalarExpr>,
 }
 
+impl GetSize for KeyExpiryManager {
+    fn get_heap_size(&self) -> usize {
+        let row_size = if let Some(row_size) = &self
+            .event_ts_to_key
+            .first_key_value()
+            .map(|(_, v)| v.first().get_heap_size())
+        {
+            *row_size
+        } else {
+            0
+        };
+        self.event_ts_to_key
+            .values()
+            .map(|v| v.len() * row_size + std::mem::size_of::<i64>())
+            .sum::<usize>()
+    }
+}
+
 impl KeyExpiryManager {
     pub fn new(
         key_expiration_duration: Option<Duration>,
@@ -154,7 +229,7 @@ impl KeyExpiryManager {
 ///
 /// Note the two way arrow between reduce operator and arrange, it's because reduce operator need to query existing state
 /// and also need to update existing state.
-#[derive(Debug, Clone, Default, Eq, PartialEq, Ord, PartialOrd)]
+#[derive(Debug, Clone, Eq, PartialEq, Ord, PartialOrd)]
 pub struct Arrangement {
     /// A name or identifier for the arrangement which can be used for debugging or logging purposes.
     /// This field is not critical to the functionality but aids in monitoring and management of arrangements.
@@ -196,6 +271,61 @@ pub struct Arrangement {
 
     /// The time that the last compaction happened, also known as the current time.
     last_compaction_time: Option<Timestamp>,
+
+    /// Estimated size of the arrangement in heap size.
+    estimated_size: usize,
+    last_size_update: Instant,
+    size_update_interval: tokio::time::Duration,
+}
+
+impl Arrangement {
+    fn compute_size(&self) -> usize {
+        self.spine
+            .values()
+            .map(|v| {
+                let per_entry_size = v
+                    .first_key_value()
+                    .map(|(k, v)| {
+                        k.get_heap_size()
+                            + v.len() * v.first().map(|r| r.get_heap_size()).unwrap_or(0)
+                    })
+                    .unwrap_or(0);
+                std::mem::size_of::<i64>() + v.len() * per_entry_size
+            })
+            .sum::<usize>()
+            + self.expire_state.get_heap_size()
+            + self.name.get_heap_size()
+    }
+
+    fn update_and_fetch_size(&mut self) -> usize {
+        if self.last_size_update.elapsed() > self.size_update_interval {
+            self.estimated_size = self.compute_size();
+            self.last_size_update = Instant::now();
+        }
+        self.estimated_size
+    }
+}
+
+impl GetSize for Arrangement {
+    fn get_heap_size(&self) -> usize {
+        self.estimated_size
+    }
+}
+
+impl Default for Arrangement {
+    fn default() -> Self {
+        Self {
+            spine: Default::default(),
+            full_arrangement: false,
+            is_written: false,
+            expire_state: None,
+            last_compaction_time: None,
+            name: Vec::new(),
+            estimated_size: 0,
+            last_size_update: Instant::now(),
+            size_update_interval: tokio::time::Duration::from_secs(3),
+        }
+    }
 }
 
 impl Arrangement {
@@ -207,6 +337,9 @@ impl Arrangement {
             expire_state: None,
             last_compaction_time: None,
             name,
+            estimated_size: 0,
+            last_size_update: Instant::now(),
+            size_update_interval: tokio::time::Duration::from_secs(3),
         }
     }
 
@@ -269,6 +402,7 @@ impl Arrangement {
             // without changing the order of updates within same tick
             key_updates.sort_by_key(|(_val, ts, _diff)| *ts);
         }
+        self.update_and_fetch_size();
         Ok(max_expired_by)
     }
 
@@ -390,6 +524,7 @@ impl Arrangement {
 
         // insert the compacted batch into spine with key being `now`
         self.spine.insert(now, compacting_batch);
+        self.update_and_fetch_size();
         Ok(max_expired_by)
     }
 

src/frontend/Cargo.toml

@@ -25,7 +25,6 @@ common-catalog.workspace = true
 common-config.workspace = true
 common-datasource.workspace = true
 common-error.workspace = true
-common-frontend.workspace = true
 common-function.workspace = true
 common-grpc.workspace = true
 common-macro.workspace = true
@@ -71,7 +70,6 @@ common-test-util.workspace = true
 datanode.workspace = true
 datatypes.workspace = true
 futures = "0.3"
-meta-srv = { workspace = true, features = ["mock"] }
 serde_json.workspace = true
 strfmt = "0.2"
 tower.workspace = true

src/frontend/src/instance/log_handler.rs

@@ -19,14 +19,16 @@ use async_trait::async_trait;
 use auth::{PermissionChecker, PermissionCheckerRef, PermissionReq};
 use client::Output;
 use common_error::ext::BoxedError;
+use pipeline::pipeline_operator::PipelineOperator;
 use pipeline::{GreptimeTransformer, Pipeline, PipelineInfo, PipelineVersion};
 use servers::error::{
     AuthSnafu, Error as ServerError, ExecuteGrpcRequestSnafu, PipelineSnafu, Result as ServerResult,
 };
 use servers::interceptor::{LogIngestInterceptor, LogIngestInterceptorRef};
 use servers::query_handler::PipelineHandler;
-use session::context::QueryContextRef;
+use session::context::{QueryContext, QueryContextRef};
 use snafu::ResultExt;
+use table::Table;
 
 use crate::instance::Instance;
 
@@ -84,6 +86,22 @@ impl PipelineHandler for Instance {
             .await
             .context(PipelineSnafu)
     }
+
+    async fn get_table(
+        &self,
+        table: &str,
+        query_ctx: &QueryContext,
+    ) -> std::result::Result<Option<Arc<Table>>, catalog::error::Error> {
+        let catalog = query_ctx.current_catalog();
+        let schema = query_ctx.current_schema();
+        self.catalog_manager
+            .table(catalog, &schema, table, None)
+            .await
+    }
+
+    fn build_pipeline(&self, pipeline: &str) -> ServerResult<Pipeline<GreptimeTransformer>> {
+        PipelineOperator::build_pipeline(pipeline).context(PipelineSnafu)
+    }
 }
 
 impl Instance {

src/index/Cargo.toml

@@ -17,6 +17,7 @@ common-error.workspace = true
 common-macro.workspace = true
 common-runtime.workspace = true
 common-telemetry.workspace = true
+fastbloom = "0.8"
 fst.workspace = true
 futures.workspace = true
 greptime-proto.workspace = true
@@ -26,6 +27,7 @@ prost.workspace = true
 regex.workspace = true
 regex-automata.workspace = true
 serde.workspace = true
+serde_json.workspace = true
 snafu.workspace = true
 tantivy = { version = "0.22", features = ["zstd-compression"] }
 tantivy-jieba = "0.11.0"

src/index/src/bloom_filter.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 serde::{Deserialize, Serialize};
+
+pub mod creator;
+mod error;
+
+pub type Bytes = Vec<u8>;
+pub type BytesRef<'a> = &'a [u8];
+
+/// The Meta information of the bloom filter stored in the file.
+#[derive(Debug, Default, Serialize, Deserialize)]
+pub struct BloomFilterMeta {
+    /// The number of rows per segment.
+    pub rows_per_segment: usize,
+
+    /// The number of segments.
+    pub seg_count: usize,
+
+    /// The number of total rows.
+    pub row_count: usize,
+
+    /// The size of the bloom filter excluding the meta information.
+    pub bloom_filter_segments_size: usize,
+
+    /// Offset and size of bloom filters in the file.
+    pub bloom_filter_segments: Vec<BloomFilterSegmentLocation>,
+}
+
+/// The location of the bloom filter segment in the file.
+#[derive(Debug, Serialize, Deserialize)]
+pub struct BloomFilterSegmentLocation {
+    /// The offset of the bloom filter segment in the file.
+    pub offset: u64,
+
+    /// The size of the bloom filter segment in the file.
+    pub size: u64,
+
+    /// The number of elements in the bloom filter segment.
+    pub elem_count: usize,
+}

src/index/src/bloom_filter/creator.rs

@@ -0,0 +1,294 @@
+// Copyright 2023 Greptime Team
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+use std::collections::HashSet;
+
+use fastbloom::BloomFilter;
+use futures::{AsyncWrite, AsyncWriteExt};
+use snafu::ResultExt;
+
+use super::error::{IoSnafu, SerdeJsonSnafu};
+use crate::bloom_filter::error::Result;
+use crate::bloom_filter::{BloomFilterMeta, BloomFilterSegmentLocation, Bytes};
+
+/// The seed used for the Bloom filter.
+const SEED: u128 = 42;
+
+/// The false positive rate of the Bloom filter.
+const FALSE_POSITIVE_RATE: f64 = 0.01;
+
+/// `BloomFilterCreator` is responsible for creating and managing bloom filters
+/// for a set of elements. It divides the rows into segments and creates
+/// bloom filters for each segment.
+///
+/// # Format
+///
+/// The bloom filter creator writes the following format to the writer:
+///
+/// ```text
+/// +--------------------+--------------------+-----+----------------------+----------------------+
+/// | Bloom filter 0     | Bloom filter 1     | ... | BloomFilterMeta      | Meta size            |
+/// +--------------------+--------------------+-----+----------------------+----------------------+
+/// |<- bytes (size 0) ->|<- bytes (size 1) ->| ... |<- json (meta size) ->|<- u32 LE (4 bytes) ->|
+/// ```
+///
+pub struct BloomFilterCreator {
+    /// The number of rows per segment set by the user.
+    rows_per_segment: usize,
+
+    /// Row count that added to the bloom filter so far.
+    accumulated_row_count: usize,
+
+    /// A set of distinct elements in the current segment.
+    cur_seg_distinct_elems: HashSet<Bytes>,
+
+    /// The memory usage of the current segment's distinct elements.
+    cur_seg_distinct_elems_mem_usage: usize,
+
+    /// Storage for finalized Bloom filters.
+    finalized_bloom_filters: FinalizedBloomFilterStorage,
+}
+
+impl BloomFilterCreator {
+    /// Creates a new `BloomFilterCreator` with the specified number of rows per segment.
+    ///
+    /// # PANICS
+    ///
+    /// `rows_per_segment` <= 0
+    pub fn new(rows_per_segment: usize) -> Self {
+        assert!(
+            rows_per_segment > 0,
+            "rows_per_segment must be greater than 0"
+        );
+
+        Self {
+            rows_per_segment,
+            accumulated_row_count: 0,
+            cur_seg_distinct_elems: HashSet::default(),
+            cur_seg_distinct_elems_mem_usage: 0,
+            finalized_bloom_filters: FinalizedBloomFilterStorage::default(),
+        }
+    }
+
+    /// Adds a row of elements to the bloom filter. If the number of accumulated rows
+    /// reaches `rows_per_segment`, it finalizes the current segment.
+    pub fn push_row_elems(&mut self, elems: impl IntoIterator<Item = Bytes>) {
+        self.accumulated_row_count += 1;
+        for elem in elems.into_iter() {
+            let len = elem.len();
+            let is_new = self.cur_seg_distinct_elems.insert(elem);
+            if is_new {
+                self.cur_seg_distinct_elems_mem_usage += len;
+            }
+        }
+
+        if self.accumulated_row_count % self.rows_per_segment == 0 {
+            self.finalize_segment();
+        }
+    }
+
+    /// Finalizes any remaining segments and writes the bloom filters and metadata to the provided writer.
+    pub async fn finish(&mut self, mut writer: impl AsyncWrite + Unpin) -> Result<()> {
+        if !self.cur_seg_distinct_elems.is_empty() {
+            self.finalize_segment();
+        }
+
+        let mut meta = BloomFilterMeta {
+            rows_per_segment: self.rows_per_segment,
+            seg_count: self.finalized_bloom_filters.len(),
+            row_count: self.accumulated_row_count,
+            ..Default::default()
+        };
+
+        let mut buf = Vec::new();
+        for segment in self.finalized_bloom_filters.drain() {
+            let slice = segment.bloom_filter.as_slice();
+            buf.clear();
+            write_u64_slice(&mut buf, slice);
+            writer.write_all(&buf).await.context(IoSnafu)?;
+
+            let size = buf.len();
+            meta.bloom_filter_segments.push(BloomFilterSegmentLocation {
+                offset: meta.bloom_filter_segments_size as _,
+                size: size as _,
+                elem_count: segment.element_count,
+            });
+            meta.bloom_filter_segments_size += size;
+        }
+
+        let meta_bytes = serde_json::to_vec(&meta).context(SerdeJsonSnafu)?;
+        writer.write_all(&meta_bytes).await.context(IoSnafu)?;
+
+        let meta_size = meta_bytes.len() as u32;
+        writer
+            .write_all(&meta_size.to_le_bytes())
+            .await
+            .context(IoSnafu)?;
+        writer.flush().await.unwrap();
+
+        Ok(())
+    }
+
+    /// Returns the memory usage of the creating bloom filter.
+    pub fn memory_usage(&self) -> usize {
+        self.cur_seg_distinct_elems_mem_usage + self.finalized_bloom_filters.memory_usage()
+    }
+
+    fn finalize_segment(&mut self) {
+        let elem_count = self.cur_seg_distinct_elems.len();
+        self.finalized_bloom_filters
+            .add(self.cur_seg_distinct_elems.drain(), elem_count);
+        self.cur_seg_distinct_elems_mem_usage = 0;
+    }
+}
+
+/// Storage for finalized Bloom filters.
+///
+/// TODO(zhongzc): Add support for storing intermediate bloom filters on disk to control memory usage.
+#[derive(Debug, Default)]
+struct FinalizedBloomFilterStorage {
+    /// Bloom filters that are stored in memory.
+    in_memory: Vec<FinalizedBloomFilterSegment>,
+}
+
+impl FinalizedBloomFilterStorage {
+    fn memory_usage(&self) -> usize {
+        self.in_memory.iter().map(|s| s.size).sum()
+    }
+
+    /// Adds a new finalized Bloom filter to the storage.
+    ///
+    /// TODO(zhongzc): Add support for flushing to disk.
+    fn add(&mut self, elems: impl IntoIterator<Item = Bytes>, elem_count: usize) {
+        let mut bf = BloomFilter::with_false_pos(FALSE_POSITIVE_RATE)
+            .seed(&SEED)
+            .expected_items(elem_count);
+        for elem in elems.into_iter() {
+            bf.insert(&elem);
+        }
+
+        let cbf = FinalizedBloomFilterSegment::new(bf, elem_count);
+        self.in_memory.push(cbf);
+    }
+
+    fn len(&self) -> usize {
+        self.in_memory.len()
+    }
+
+    fn drain(&mut self) -> impl Iterator<Item = FinalizedBloomFilterSegment> + '_ {
+        self.in_memory.drain(..)
+    }
+}
+
+/// A finalized Bloom filter segment.
+#[derive(Debug)]
+struct FinalizedBloomFilterSegment {
+    /// The underlying Bloom filter.
+    bloom_filter: BloomFilter,
+
+    /// The number of elements in the Bloom filter.
+    element_count: usize,
+
+    /// The occupied memory size of the Bloom filter.
+    size: usize,
+}
+
+impl FinalizedBloomFilterSegment {
+    fn new(bloom_filter: BloomFilter, elem_count: usize) -> Self {
+        let memory_usage = std::mem::size_of_val(bloom_filter.as_slice());
+        Self {
+            bloom_filter,
+            element_count: elem_count,
+            size: memory_usage,
+        }
+    }
+}
+
+/// Writes a slice of `u64` to the buffer in little-endian order.
+fn write_u64_slice(buf: &mut Vec<u8>, slice: &[u64]) {
+    buf.reserve(std::mem::size_of_val(slice));
+    for &x in slice {
+        buf.extend_from_slice(&x.to_le_bytes());
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use futures::io::Cursor;
+
+    use super::*;
+
+    fn u64_vec_from_bytes(bytes: &[u8]) -> Vec<u64> {
+        bytes
+            .chunks_exact(std::mem::size_of::<u64>())
+            .map(|chunk| u64::from_le_bytes(chunk.try_into().unwrap()))
+            .collect()
+    }
+
+    #[tokio::test]
+    async fn test_bloom_filter_creator() {
+        let mut writer = Cursor::new(Vec::new());
+        let mut creator = BloomFilterCreator::new(2);
+
+        creator.push_row_elems(vec![b"a".to_vec(), b"b".to_vec()]);
+        assert!(creator.cur_seg_distinct_elems_mem_usage > 0);
+        assert!(creator.memory_usage() > 0);
+
+        creator.push_row_elems(vec![b"c".to_vec(), b"d".to_vec()]);
+        // Finalize the first segment
+        assert!(creator.cur_seg_distinct_elems_mem_usage == 0);
+        assert!(creator.memory_usage() > 0);
+
+        creator.push_row_elems(vec![b"e".to_vec(), b"f".to_vec()]);
+        assert!(creator.cur_seg_distinct_elems_mem_usage > 0);
+        assert!(creator.memory_usage() > 0);
+
+        creator.finish(&mut writer).await.unwrap();
+
+        let bytes = writer.into_inner();
+        let total_size = bytes.len();
+        let meta_size_offset = total_size - 4;
+        let meta_size = u32::from_le_bytes((&bytes[meta_size_offset..]).try_into().unwrap());
+
+        let meta_bytes = &bytes[total_size - meta_size as usize - 4..total_size - 4];
+        let meta: BloomFilterMeta = serde_json::from_slice(meta_bytes).unwrap();
+
+        assert_eq!(meta.rows_per_segment, 2);
+        assert_eq!(meta.seg_count, 2);
+        assert_eq!(meta.row_count, 3);
+        assert_eq!(
+            meta.bloom_filter_segments_size + meta_bytes.len() + 4,
+            total_size
+        );
+
+        let mut bfs = Vec::new();
+        for segment in meta.bloom_filter_segments {
+            let bloom_filter_bytes =
+                &bytes[segment.offset as usize..(segment.offset + segment.size) as usize];
+            let v = u64_vec_from_bytes(bloom_filter_bytes);
+            let bloom_filter = BloomFilter::from_vec(v)
+                .seed(&SEED)
+                .expected_items(segment.elem_count);
+            bfs.push(bloom_filter);
+        }
+
+        assert_eq!(bfs.len(), 2);
+        assert!(bfs[0].contains(&b"a"));
+        assert!(bfs[0].contains(&b"b"));
+        assert!(bfs[0].contains(&b"c"));
+        assert!(bfs[0].contains(&b"d"));
+        assert!(bfs[1].contains(&b"e"));
+        assert!(bfs[1].contains(&b"f"));
+    }
+}

src/index/src/bloom_filter/error.rs

@@ -0,0 +1,66 @@
+// Copyright 2023 Greptime Team
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+use std::any::Any;
+
+use common_error::ext::{BoxedError, 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("IO error"))]
+    Io {
+        #[snafu(source)]
+        error: std::io::Error,
+        #[snafu(implicit)]
+        location: Location,
+    },
+
+    #[snafu(display("Failed to serde json"))]
+    SerdeJson {
+        #[snafu(source)]
+        error: serde_json::error::Error,
+        #[snafu(implicit)]
+        location: Location,
+    },
+
+    #[snafu(display("External error"))]
+    External {
+        source: BoxedError,
+        #[snafu(implicit)]
+        location: Location,
+    },
+}
+
+impl ErrorExt for Error {
+    fn status_code(&self) -> StatusCode {
+        use Error::*;
+
+        match self {
+            Io { .. } | Self::SerdeJson { .. } => StatusCode::Unexpected,
+
+            External { source, .. } => source.status_code(),
+        }
+    }
+
+    fn as_any(&self) -> &dyn Any {
+        self
+    }
+}
+
+pub type Result<T> = std::result::Result<T, Error>;

src/index/src/inverted_index/error.rs

@@ -26,14 +26,6 @@ use crate::inverted_index::search::predicate::Predicate;
 #[snafu(visibility(pub))]
 #[stack_trace_debug]
 pub enum Error {
-    #[snafu(display("Failed to seek"))]
-    Seek {
-        #[snafu(source)]
-        error: IoError,
-        #[snafu(implicit)]
-        location: Location,
-    },
-
     #[snafu(display("Failed to read"))]
     Read {
         #[snafu(source)]
@@ -76,6 +68,18 @@ pub enum Error {
         location: Location,
     },
 
+    #[snafu(display("Blob size too small"))]
+    BlobSizeTooSmall {
+        #[snafu(implicit)]
+        location: Location,
+    },
+
+    #[snafu(display("Invalid footer payload size"))]
+    InvalidFooterPayloadSize {
+        #[snafu(implicit)]
+        location: Location,
+    },
+
     #[snafu(display("Unexpected inverted index footer payload size, max: {max_payload_size}, actual: {actual_payload_size}"))]
     UnexpectedFooterPayloadSize {
         max_payload_size: u64,
@@ -215,8 +219,7 @@ impl ErrorExt for Error {
     fn status_code(&self) -> StatusCode {
         use Error::*;
         match self {
-            Seek { .. }
+            Read { .. }
-            | Read { .. }
             | Write { .. }
             | Flush { .. }
             | Close { .. }
@@ -229,7 +232,9 @@ impl ErrorExt for Error {
             | KeysApplierUnexpectedPredicates { .. }
             | CommonIo { .. }
             | UnknownIntermediateCodecMagic { .. }
-            | FstCompile { .. } => StatusCode::Unexpected,
+            | FstCompile { .. }
+            | InvalidFooterPayloadSize { .. }
+            | BlobSizeTooSmall { .. } => StatusCode::Unexpected,
 
             ParseRegex { .. }
             | ParseDFA { .. }

src/index/src/inverted_index/format/reader.rs

@@ -12,9 +12,11 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
+use std::ops::Range;
 use std::sync::Arc;
 
 use async_trait::async_trait;
+use bytes::Bytes;
 use common_base::BitVec;
 use greptime_proto::v1::index::InvertedIndexMetas;
 use snafu::ResultExt;
@@ -30,23 +32,23 @@ mod footer;
 #[mockall::automock]
 #[async_trait]
 pub trait InvertedIndexReader: Send {
-    /// Reads all data to dest.
-    async fn read_all(&mut self, dest: &mut Vec<u8>) -> Result<usize>;
-
     /// Seeks to given offset and reads data with exact size as provided.
-    async fn seek_read(&mut self, offset: u64, size: u32) -> Result<Vec<u8>>;
+    async fn range_read(&mut self, offset: u64, size: u32) -> Result<Vec<u8>>;
+
+    /// Reads the bytes in the given ranges.
+    async fn read_vec(&mut self, ranges: &[Range<u64>]) -> Result<Vec<Bytes>>;
 
     /// Retrieves metadata of all inverted indices stored within the blob.
     async fn metadata(&mut self) -> Result<Arc<InvertedIndexMetas>>;
 
     /// Retrieves the finite state transducer (FST) map from the given offset and size.
     async fn fst(&mut self, offset: u64, size: u32) -> Result<FstMap> {
-        let fst_data = self.seek_read(offset, size).await?;
+        let fst_data = self.range_read(offset, size).await?;
         FstMap::new(fst_data).context(DecodeFstSnafu)
     }
 
     /// Retrieves the bitmap from the given offset and size.
     async fn bitmap(&mut self, offset: u64, size: u32) -> Result<BitVec> {
-        self.seek_read(offset, size).await.map(BitVec::from_vec)
+        self.range_read(offset, size).await.map(BitVec::from_vec)
     }
 }

src/index/src/inverted_index/format/reader/blob.rs

@@ -12,15 +12,18 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
+use std::ops::Range;
 use std::sync::Arc;
 
 use async_trait::async_trait;
+use bytes::Bytes;
 use common_base::range_read::RangeReader;
 use greptime_proto::v1::index::InvertedIndexMetas;
 use snafu::{ensure, ResultExt};
 
+use super::footer::DEFAULT_PREFETCH_SIZE;
 use crate::inverted_index::error::{CommonIoSnafu, Result, UnexpectedBlobSizeSnafu};
-use crate::inverted_index::format::reader::footer::InvertedIndeFooterReader;
+use crate::inverted_index::format::reader::footer::InvertedIndexFooterReader;
 use crate::inverted_index::format::reader::InvertedIndexReader;
 use crate::inverted_index::format::MIN_BLOB_SIZE;
 
@@ -49,16 +52,7 @@ impl<R> InvertedIndexBlobReader<R> {
 
 #[async_trait]
 impl<R: RangeReader> InvertedIndexReader for InvertedIndexBlobReader<R> {
-    async fn read_all(&mut self, dest: &mut Vec<u8>) -> Result<usize> {
+    async fn range_read(&mut self, offset: u64, size: u32) -> Result<Vec<u8>> {
-        let metadata = self.source.metadata().await.context(CommonIoSnafu)?;
-        self.source
-            .read_into(0..metadata.content_length, dest)
-            .await
-            .context(CommonIoSnafu)?;
-        Ok(metadata.content_length as usize)
-    }
-
-    async fn seek_read(&mut self, offset: u64, size: u32) -> Result<Vec<u8>> {
         let buf = self
             .source
             .read(offset..offset + size as u64)
@@ -67,12 +61,17 @@ impl<R: RangeReader> InvertedIndexReader for InvertedIndexBlobReader<R> {
         Ok(buf.into())
     }
 
+    async fn read_vec(&mut self, ranges: &[Range<u64>]) -> Result<Vec<Bytes>> {
+        self.source.read_vec(ranges).await.context(CommonIoSnafu)
+    }
+
     async fn metadata(&mut self) -> Result<Arc<InvertedIndexMetas>> {
         let metadata = self.source.metadata().await.context(CommonIoSnafu)?;
         let blob_size = metadata.content_length;
         Self::validate_blob_size(blob_size)?;
 
-        let mut footer_reader = InvertedIndeFooterReader::new(&mut self.source, blob_size);
+        let mut footer_reader = InvertedIndexFooterReader::new(&mut self.source, blob_size)
+            .with_prefetch_size(DEFAULT_PREFETCH_SIZE);
         footer_reader.metadata().await.map(Arc::new)
     }
 }

src/index/src/inverted_index/format/reader/footer.rs

@@ -18,53 +18,88 @@ use prost::Message;
 use snafu::{ensure, ResultExt};
 
 use crate::inverted_index::error::{
-    CommonIoSnafu, DecodeProtoSnafu, Result, UnexpectedFooterPayloadSizeSnafu,
+    BlobSizeTooSmallSnafu, CommonIoSnafu, DecodeProtoSnafu, InvalidFooterPayloadSizeSnafu, Result,
-    UnexpectedOffsetSizeSnafu, UnexpectedZeroSegmentRowCountSnafu,
+    UnexpectedFooterPayloadSizeSnafu, UnexpectedOffsetSizeSnafu,
+    UnexpectedZeroSegmentRowCountSnafu,
 };
 use crate::inverted_index::format::FOOTER_PAYLOAD_SIZE_SIZE;
 
-/// InvertedIndeFooterReader is for reading the footer section of the blob.
+pub const DEFAULT_PREFETCH_SIZE: u64 = 1024; // 1KiB
-pub struct InvertedIndeFooterReader<R> {
+
+/// InvertedIndexFooterReader is for reading the footer section of the blob.
+pub struct InvertedIndexFooterReader<R> {
     source: R,
     blob_size: u64,
+    prefetch_size: Option<u64>,
 }
 
-impl<R> InvertedIndeFooterReader<R> {
+impl<R> InvertedIndexFooterReader<R> {
     pub fn new(source: R, blob_size: u64) -> Self {
-        Self { source, blob_size }
+        Self {
+            source,
+            blob_size,
+            prefetch_size: None,
+        }
+    }
+
+    /// Set the prefetch size for the footer reader.
+    pub fn with_prefetch_size(mut self, prefetch_size: u64) -> Self {
+        self.prefetch_size = Some(prefetch_size.max(FOOTER_PAYLOAD_SIZE_SIZE));
+        self
+    }
+
+    pub fn prefetch_size(&self) -> u64 {
+        self.prefetch_size.unwrap_or(FOOTER_PAYLOAD_SIZE_SIZE)
     }
 }
 
-impl<R: RangeReader> InvertedIndeFooterReader<R> {
+impl<R: RangeReader> InvertedIndexFooterReader<R> {
     pub async fn metadata(&mut self) -> Result<InvertedIndexMetas> {
-        let payload_size = self.read_payload_size().await?;
+        ensure!(
-        let metas = self.read_payload(payload_size).await?;
+            self.blob_size >= FOOTER_PAYLOAD_SIZE_SIZE,
-        Ok(metas)
+            BlobSizeTooSmallSnafu
-    }
+        );
 
-    async fn read_payload_size(&mut self) -> Result<u64> {
+        let footer_start = self.blob_size.saturating_sub(self.prefetch_size());
-        let mut size_buf = [0u8; FOOTER_PAYLOAD_SIZE_SIZE as usize];
+        let suffix = self
-        let end = self.blob_size;
+            .source
-        let start = end - FOOTER_PAYLOAD_SIZE_SIZE;
+            .read(footer_start..self.blob_size)
-        self.source
-            .read_into(start..end, &mut &mut size_buf[..])
             .await
             .context(CommonIoSnafu)?;
+        let suffix_len = suffix.len();
+        let length = u32::from_le_bytes(Self::read_tailing_four_bytes(&suffix)?) as u64;
+        self.validate_payload_size(length)?;
 
-        let payload_size = u32::from_le_bytes(size_buf) as u64;
+        let footer_size = FOOTER_PAYLOAD_SIZE_SIZE;
-        self.validate_payload_size(payload_size)?;
 
-        Ok(payload_size)
+        // Did not fetch the entire file metadata in the initial read, need to make a second request.
+        if length > suffix_len as u64 - footer_size {
+            let metadata_start = self.blob_size - length - footer_size;
+            let meta = self
+                .source
+                .read(metadata_start..self.blob_size - footer_size)
+                .await
+                .context(CommonIoSnafu)?;
+            self.parse_payload(&meta, length)
+        } else {
+            let metadata_start = self.blob_size - length - footer_size - footer_start;
+            let meta = &suffix[metadata_start as usize..suffix_len - footer_size as usize];
+            self.parse_payload(meta, length)
+        }
     }
 
-    async fn read_payload(&mut self, payload_size: u64) -> Result<InvertedIndexMetas> {
+    fn read_tailing_four_bytes(suffix: &[u8]) -> Result<[u8; 4]> {
-        let end = self.blob_size - FOOTER_PAYLOAD_SIZE_SIZE;
+        let suffix_len = suffix.len();
-        let start = end - payload_size;
+        ensure!(suffix_len >= 4, InvalidFooterPayloadSizeSnafu);
-        let bytes = self.source.read(start..end).await.context(CommonIoSnafu)?;
+        let mut bytes = [0; 4];
+        bytes.copy_from_slice(&suffix[suffix_len - 4..suffix_len]);
 
-        let metas = InvertedIndexMetas::decode(&*bytes).context(DecodeProtoSnafu)?;
+        Ok(bytes)
+    }
+
+    fn parse_payload(&mut self, bytes: &[u8], payload_size: u64) -> Result<InvertedIndexMetas> {
+        let metas = InvertedIndexMetas::decode(bytes).context(DecodeProtoSnafu)?;
         self.validate_metas(&metas, payload_size)?;
-
         Ok(metas)
     }
 
@@ -113,9 +148,12 @@ impl<R: RangeReader> InvertedIndeFooterReader<R> {
 
 #[cfg(test)]
 mod tests {
+    use std::assert_matches::assert_matches;
+
     use prost::Message;
 
     use super::*;
+    use crate::inverted_index::error::Error;
 
     fn create_test_payload(meta: InvertedIndexMeta) -> Vec<u8> {
         let mut metas = InvertedIndexMetas {
@@ -141,14 +179,18 @@ mod tests {
 
         let mut payload_buf = create_test_payload(meta);
         let blob_size = payload_buf.len() as u64;
-        let mut reader = InvertedIndeFooterReader::new(&mut payload_buf, blob_size);
 
-        let payload_size = reader.read_payload_size().await.unwrap();
+        for prefetch in [0, blob_size / 2, blob_size, blob_size + 10] {
-        let metas = reader.read_payload(payload_size).await.unwrap();
+            let mut reader = InvertedIndexFooterReader::new(&mut payload_buf, blob_size);
+            if prefetch > 0 {
+                reader = reader.with_prefetch_size(prefetch);
+            }
 
-        assert_eq!(metas.metas.len(), 1);
+            let metas = reader.metadata().await.unwrap();
-        let index_meta = &metas.metas.get("test").unwrap();
+            assert_eq!(metas.metas.len(), 1);
-        assert_eq!(index_meta.name, "test");
+            let index_meta = &metas.metas.get("test").unwrap();
+            assert_eq!(index_meta.name, "test");
+        }
     }
 
     #[tokio::test]
@@ -157,14 +199,20 @@ mod tests {
             name: "test".to_string(),
             ..Default::default()
         };
-
         let mut payload_buf = create_test_payload(meta);
         payload_buf.push(0xff); // Add an extra byte to corrupt the footer
         let blob_size = payload_buf.len() as u64;
-        let mut reader = InvertedIndeFooterReader::new(&mut payload_buf, blob_size);
 
-        let payload_size_result = reader.read_payload_size().await;
+        for prefetch in [0, blob_size / 2, blob_size, blob_size + 10] {
-        assert!(payload_size_result.is_err());
+            let blob_size = payload_buf.len() as u64;
+            let mut reader = InvertedIndexFooterReader::new(&mut payload_buf, blob_size);
+            if prefetch > 0 {
+                reader = reader.with_prefetch_size(prefetch);
+            }
+
+            let result = reader.metadata().await;
+            assert_matches!(result, Err(Error::UnexpectedFooterPayloadSize { .. }));
+        }
     }
 
     #[tokio::test]
@@ -178,10 +226,15 @@ mod tests {
 
         let mut payload_buf = create_test_payload(meta);
         let blob_size = payload_buf.len() as u64;
-        let mut reader = InvertedIndeFooterReader::new(&mut payload_buf, blob_size);
 
-        let payload_size = reader.read_payload_size().await.unwrap();
+        for prefetch in [0, blob_size / 2, blob_size, blob_size + 10] {
-        let payload_result = reader.read_payload(payload_size).await;
+            let mut reader = InvertedIndexFooterReader::new(&mut payload_buf, blob_size);
-        assert!(payload_result.is_err());
+            if prefetch > 0 {
+                reader = reader.with_prefetch_size(prefetch);
+            }
+
+            let result = reader.metadata().await;
+            assert_matches!(result, Err(Error::UnexpectedOffsetSize { .. }));
+        }
     }
 }

src/index/src/lib.rs

@@ -13,6 +13,8 @@
 // limitations under the License.
 
 #![feature(iter_partition_in_place)]
+#![feature(assert_matches)]
 
+pub mod bloom_filter;
 pub mod fulltext_index;
 pub mod inverted_index;

src/meta-client/src/client.rs

@@ -25,6 +25,7 @@ use std::sync::Arc;
 
 use api::v1::meta::{ProcedureDetailResponse, Role};
 use cluster::Client as ClusterClient;
+pub use cluster::ClusterKvBackend;
 use common_error::ext::BoxedError;
 use common_grpc::channel_manager::{ChannelConfig, ChannelManager};
 use common_meta::cluster::{
@@ -33,6 +34,8 @@ use common_meta::cluster::{
 use common_meta::datanode::{DatanodeStatKey, DatanodeStatValue, RegionStat};
 use common_meta::ddl::{ExecutorContext, ProcedureExecutor};
 use common_meta::error::{self as meta_error, ExternalSnafu, Result as MetaResult};
+use common_meta::key::flow::flow_state::{FlowStat, FlowStateManager};
+use common_meta::kv_backend::KvBackendRef;
 use common_meta::range_stream::PaginationStream;
 use common_meta::rpc::ddl::{SubmitDdlTaskRequest, SubmitDdlTaskResponse};
 use common_meta::rpc::procedure::{
@@ -54,7 +57,8 @@ use store::Client as StoreClient;
 
 pub use self::heartbeat::{HeartbeatSender, HeartbeatStream};
 use crate::error::{
-    ConvertMetaRequestSnafu, ConvertMetaResponseSnafu, Error, NotStartedSnafu, Result,
+    ConvertMetaRequestSnafu, ConvertMetaResponseSnafu, Error, GetFlowStatSnafu, NotStartedSnafu,
+    Result,
 };
 
 pub type Id = (u64, u64);
@@ -347,6 +351,15 @@ fn decode_stats(kv: KeyValue) -> MetaResult<DatanodeStatValue> {
 }
 
 impl MetaClient {
+    pub async fn list_flow_stats(&self) -> Result<Option<FlowStat>> {
+        let cluster_backend = ClusterKvBackend::new(Arc::new(self.cluster_client()?));
+        let cluster_backend = Arc::new(cluster_backend) as KvBackendRef;
+        let flow_state_manager = FlowStateManager::new(cluster_backend);
+        let res = flow_state_manager.get().await.context(GetFlowStatSnafu)?;
+
+        Ok(res.map(|r| r.into()))
+    }
+
     pub fn new(id: Id) -> Self {
         Self {
             id,

src/meta-client/src/client/cluster.rs

@@ -40,8 +40,8 @@ use tonic::Status;
 use crate::client::ask_leader::AskLeader;
 use crate::client::{util, Id};
 use crate::error::{
-    ConvertMetaResponseSnafu, CreateChannelSnafu, Error, IllegalGrpcClientStateSnafu, Result,
+    ConvertMetaResponseSnafu, CreateChannelSnafu, Error, IllegalGrpcClientStateSnafu,
-    RetryTimesExceededSnafu,
+    ReadOnlyKvBackendSnafu, Result, RetryTimesExceededSnafu,
 };
 
 #[derive(Clone, Debug)]
@@ -308,3 +308,75 @@ impl Inner {
         .map(|res| (res.leader, res.followers))
     }
 }
+
+/// A client for the cluster info. Read only and corresponding to
+/// `in_memory` kvbackend in the meta-srv.
+#[derive(Clone, Debug)]
+pub struct ClusterKvBackend {
+    inner: Arc<Client>,
+}
+
+impl ClusterKvBackend {
+    pub fn new(client: Arc<Client>) -> Self {
+        Self { inner: client }
+    }
+
+    fn unimpl(&self) -> common_meta::error::Error {
+        let ret: common_meta::error::Result<()> = ReadOnlyKvBackendSnafu {
+            name: self.name().to_string(),
+        }
+        .fail()
+        .map_err(BoxedError::new)
+        .context(common_meta::error::ExternalSnafu);
+        ret.unwrap_err()
+    }
+}
+
+impl TxnService for ClusterKvBackend {
+    type Error = common_meta::error::Error;
+}
+
+#[async_trait::async_trait]
+impl KvBackend for ClusterKvBackend {
+    fn name(&self) -> &str {
+        "ClusterKvBackend"
+    }
+
+    fn as_any(&self) -> &dyn Any {
+        self
+    }
+
+    async fn range(&self, req: RangeRequest) -> common_meta::error::Result<RangeResponse> {
+        self.inner
+            .range(req)
+            .await
+            .map_err(BoxedError::new)
+            .context(common_meta::error::ExternalSnafu)
+    }
+
+    async fn batch_get(&self, _: BatchGetRequest) -> common_meta::error::Result<BatchGetResponse> {
+        Err(self.unimpl())
+    }
+
+    async fn put(&self, _: PutRequest) -> common_meta::error::Result<PutResponse> {
+        Err(self.unimpl())
+    }
+
+    async fn batch_put(&self, _: BatchPutRequest) -> common_meta::error::Result<BatchPutResponse> {
+        Err(self.unimpl())
+    }
+
+    async fn delete_range(
+        &self,
+        _: DeleteRangeRequest,
+    ) -> common_meta::error::Result<DeleteRangeResponse> {
+        Err(self.unimpl())
+    }
+
+    async fn batch_delete(
+        &self,
+        _: BatchDeleteRequest,
+    ) -> common_meta::error::Result<BatchDeleteResponse> {
+        Err(self.unimpl())
+    }
+}

src/meta-client/src/error.rs

@@ -99,8 +99,22 @@ pub enum Error {
         source: common_meta::error::Error,
     },
 
+    #[snafu(display("Failed to get flow stat"))]
+    GetFlowStat {
+        #[snafu(implicit)]
+        location: Location,
+        source: common_meta::error::Error,
+    },
+
     #[snafu(display("Retry exceeded max times({}), message: {}", times, msg))]
     RetryTimesExceeded { times: usize, msg: String },
+
+    #[snafu(display("Trying to write to a read-only kv backend: {}", name))]
+    ReadOnlyKvBackend {
+        name: String,
+        #[snafu(implicit)]
+        location: Location,
+    },
 }
 
 #[allow(dead_code)]
@@ -120,13 +134,15 @@ impl ErrorExt for Error {
             | Error::SendHeartbeat { .. }
             | Error::CreateHeartbeatStream { .. }
             | Error::CreateChannel { .. }
-            | Error::RetryTimesExceeded { .. } => StatusCode::Internal,
+            | Error::RetryTimesExceeded { .. }
+            | Error::ReadOnlyKvBackend { .. } => StatusCode::Internal,
 
             Error::MetaServer { code, .. } => *code,
 
             Error::InvalidResponseHeader { source, .. }
             | Error::ConvertMetaRequest { source, .. }
-            | Error::ConvertMetaResponse { source, .. } => source.status_code(),
+            | Error::ConvertMetaResponse { source, .. }
+            | Error::GetFlowStat { source, .. } => source.status_code(),
         }
     }
 }

src/meta-srv/src/bootstrap.rs

@@ -206,43 +206,41 @@ pub async fn metasrv_builder(
     plugins: Plugins,
     kv_backend: Option<KvBackendRef>,
 ) -> Result<MetasrvBuilder> {
-    let (kv_backend, election) = match (kv_backend, &opts.backend) {
+    let (mut kv_backend, election) = match (kv_backend, &opts.backend) {
         (Some(kv_backend), _) => (kv_backend, None),
         (None, BackendImpl::MemoryStore) => (Arc::new(MemoryKvBackend::new()) as _, None),
         (None, BackendImpl::EtcdStore) => {
             let etcd_client = create_etcd_client(opts).await?;
-            let kv_backend = {
+            let kv_backend = EtcdStore::with_etcd_client(etcd_client.clone(), opts.max_txn_ops);
-                let etcd_backend =
+            let election = EtcdElection::with_etcd_client(
-                    EtcdStore::with_etcd_client(etcd_client.clone(), opts.max_txn_ops);
+                &opts.server_addr,
-                if !opts.store_key_prefix.is_empty() {
+                etcd_client,
-                    Arc::new(ChrootKvBackend::new(
+                opts.store_key_prefix.clone(),
-                        opts.store_key_prefix.clone().into_bytes(),
-                        etcd_backend,
-                    ))
-                } else {
-                    etcd_backend
-                }
-            };
-            (
-                kv_backend,
-                Some(
-                    EtcdElection::with_etcd_client(
-                        &opts.server_addr,
-                        etcd_client.clone(),
-                        opts.store_key_prefix.clone(),
-                    )
-                    .await?,
-                ),
             )
+            .await?;
+
+            (kv_backend, Some(election))
         }
         #[cfg(feature = "pg_kvbackend")]
         (None, BackendImpl::PostgresStore) => {
             let pg_client = create_postgres_client(opts).await?;
             let kv_backend = PgStore::with_pg_client(pg_client).await.unwrap();
+            // TODO(jeremy, weny): implement election for postgres
             (kv_backend, None)
         }
     };
 
+    if !opts.store_key_prefix.is_empty() {
+        info!(
+            "using chroot kv backend with prefix: {prefix}",
+            prefix = opts.store_key_prefix
+        );
+        kv_backend = Arc::new(ChrootKvBackend::new(
+            opts.store_key_prefix.clone().into_bytes(),
+            kv_backend,
+        ))
+    }
+
     let in_memory = Arc::new(MemoryKvBackend::new()) as ResettableKvBackendRef;
 
     let selector = match opts.selector {