fix: conn pool leak & placeholder feature so ci can compile

fix: placeholder feature so ci can compile
fix: time window filter expr use OR
2025-12-22 22:20:02 +00:00 · 2025-04-10 15:01:07 +08:00 · 2025-04-08 14:37:55 +08:00 · 2025-04-07 16:50:17 +08:00 · 2025-04-07 16:50:17 +08:00 · 2025-04-07 16:50:17 +08:00
79 changed files with 3949 additions and 230 deletions
--- a/Cargo.lock
+++ b/Cargo.lock
@@ -432,7 +432,7 @@ dependencies = [
 "arrow-schema",
 "chrono",
 "half",
- "indexmap 2.6.0",
+ "indexmap 2.7.1",
 "lexical-core",
 "num",
 "serde",
@@ -1475,7 +1475,7 @@ version = "0.13.7"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "6026d8cd82ada8bbcfe337805dd1eb6afdc9e80fa4d57e977b3a36315e0c5525"
 dependencies = [
- "indexmap 2.6.0",
+ "indexmap 2.7.1",
 "lazy_static",
 "num-traits",
 "regex",
@@ -2009,6 +2009,7 @@ dependencies = [
 name = "common-function"
 version = "0.12.0"
 dependencies = [
+ "ahash 0.8.11",
 "api",
 "approx 0.5.1",
 "arc-swap",
@@ -2031,6 +2032,7 @@ dependencies = [
 "geo-types",
 "geohash",
 "h3o",
+ "hyperloglogplus",
 "jsonb",
 "nalgebra 0.33.2",
 "num",
@@ -2974,7 +2976,7 @@ dependencies = [
 "chrono",
 "half",
 "hashbrown 0.14.5",
- "indexmap 2.6.0",
+ "indexmap 2.7.1",
 "libc",
 "object_store",
 "parquet",
@@ -3034,7 +3036,7 @@ dependencies = [
 "datafusion-functions-aggregate-common",
 "datafusion-functions-window-common",
 "datafusion-physical-expr-common",
- "indexmap 2.6.0",
+ "indexmap 2.7.1",
 "paste",
 "recursive",
 "serde_json",
@@ -3156,7 +3158,7 @@ dependencies = [
 "datafusion-physical-expr-common",
 "datafusion-physical-plan",
 "half",
- "indexmap 2.6.0",
+ "indexmap 2.7.1",
 "log",
 "parking_lot 0.12.3",
 "paste",
@@ -3207,7 +3209,7 @@ dependencies = [
 "datafusion-common",
 "datafusion-expr",
 "datafusion-physical-expr",
- "indexmap 2.6.0",
+ "indexmap 2.7.1",
 "itertools 0.13.0",
 "log",
 "recursive",
@@ -3232,7 +3234,7 @@ dependencies = [
 "datafusion-physical-expr-common",
 "half",
 "hashbrown 0.14.5",
- "indexmap 2.6.0",
+ "indexmap 2.7.1",
 "itertools 0.13.0",
 "log",
 "paste",
@@ -3291,7 +3293,7 @@ dependencies = [
 "futures",
 "half",
 "hashbrown 0.14.5",
- "indexmap 2.6.0",
+ "indexmap 2.7.1",
 "itertools 0.13.0",
 "log",
 "once_cell",
@@ -3311,7 +3313,7 @@ dependencies = [
 "arrow-schema",
 "datafusion-common",
 "datafusion-expr",
- "indexmap 2.6.0",
+ "indexmap 2.7.1",
 "log",
 "recursive",
 "regex",
@@ -4163,6 +4165,7 @@ dependencies = [
 "bytes",
 "cache",
 "catalog",
+ "chrono",
 "client",
 "common-base",
 "common-catalog",
@@ -4697,7 +4700,7 @@ dependencies = [
 [[package]]
 name = "greptime-proto"
 version = "0.1.0"
-source = "git+https://github.com/GreptimeTeam/greptime-proto.git?rev=a25adc8a01340231121646d8f0a29d0e92f45461#a25adc8a01340231121646d8f0a29d0e92f45461"
+source = "git+https://github.com/GreptimeTeam/greptime-proto.git?rev=072ce580502e015df1a6b03a185b60309a7c2a7a#072ce580502e015df1a6b03a185b60309a7c2a7a"
 dependencies = [
 "prost 0.13.3",
 "serde",
@@ -4720,7 +4723,7 @@ dependencies = [
 "futures-sink",
 "futures-util",
 "http 0.2.12",
- "indexmap 2.6.0",
+ "indexmap 2.7.1",
 "slab",
 "tokio",
 "tokio-util",
@@ -4739,7 +4742,7 @@ dependencies = [
 "futures-core",
 "futures-sink",
 "http 1.1.0",
- "indexmap 2.6.0",
+ "indexmap 2.7.1",
 "slab",
 "tokio",
 "tokio-util",
@@ -5289,6 +5292,15 @@ dependencies = [
 "tracing",
 ]

+[[package]]
+name = "hyperloglogplus"
+version = "0.4.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "621debdf94dcac33e50475fdd76d34d5ea9c0362a834b9db08c3024696c1fbe3"
+dependencies = [
+ "serde",
+]
+
 [[package]]
 name = "i_float"
 version = "1.3.1"
@@ -5577,9 +5589,9 @@ dependencies = [

 [[package]]
 name = "indexmap"
-version = "2.6.0"
+version = "2.7.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "707907fe3c25f5424cce2cb7e1cbcafee6bdbe735ca90ef77c29e84591e5b9da"
+checksum = "8c9c992b02b5b4c94ea26e32fe5bccb7aa7d9f390ab5c1221ff895bc7ea8b652"
 dependencies = [
 "equivalent",
 "hashbrown 0.15.2",
@@ -5593,7 +5605,7 @@ source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "232929e1d75fe899576a3d5c7416ad0d88dbfbb3c3d6aa00873a7408a50ddb88"
 dependencies = [
 "ahash 0.8.11",
- "indexmap 2.6.0",
+ "indexmap 2.7.1",
 "is-terminal",
 "itoa",
 "log",
@@ -5940,7 +5952,7 @@ version = "0.4.2"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "4ee7893dab2e44ae5f9d0173f26ff4aa327c10b01b06a72b52dd9405b628640d"
 dependencies = [
- "indexmap 2.6.0",
+ "indexmap 2.7.1",
 ]

 [[package]]
@@ -6423,7 +6435,7 @@ dependencies = [
 "cactus",
 "cfgrammar",
 "filetime",
- "indexmap 2.6.0",
+ "indexmap 2.7.1",
 "lazy_static",
 "lrtable",
 "num-traits",
@@ -7664,7 +7676,7 @@ checksum = "1e32339a5dc40459130b3bd269e9892439f55b33e772d2a9d402a789baaf4e8a"
 dependencies = [
 "futures-core",
 "futures-sink",
- "indexmap 2.6.0",
+ "indexmap 2.7.1",
 "js-sys",
 "once_cell",
 "pin-project-lite",
@@ -8236,7 +8248,7 @@ source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "b4c5cc86750666a3ed20bdaf5ca2a0344f9c67674cae0515bec2da16fbaa47db"
 dependencies = [
 "fixedbitset",
- "indexmap 2.6.0",
+ "indexmap 2.7.1",
 ]

 [[package]]
@@ -10327,7 +10339,7 @@ version = "1.0.137"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "930cfb6e6abf99298aaad7d29abbef7a9999a9a8806a40088f55f0dcec03146b"
 dependencies = [
- "indexmap 2.6.0",
+ "indexmap 2.7.1",
 "itoa",
 "memchr",
 "ryu",
@@ -10398,7 +10410,7 @@ dependencies = [
 "chrono",
 "hex",
 "indexmap 1.9.3",
- "indexmap 2.6.0",
+ "indexmap 2.7.1",
 "serde",
 "serde_derive",
 "serde_json",
@@ -10424,7 +10436,7 @@ version = "0.9.34+deprecated"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "6a8b1a1a2ebf674015cc02edccce75287f1a0130d394307b36743c2f5d504b47"
 dependencies = [
- "indexmap 2.6.0",
+ "indexmap 2.7.1",
 "itoa",
 "ryu",
 "serde",
@@ -10485,6 +10497,7 @@ dependencies = [
 "humantime",
 "humantime-serde",
 "hyper 1.4.1",
+ "indexmap 2.7.1",
 "influxdb_line_protocol",
 "itertools 0.10.5",
 "json5",
@@ -11026,7 +11039,7 @@ dependencies = [
 "futures-util",
 "hashbrown 0.15.2",
 "hashlink",
- "indexmap 2.6.0",
+ "indexmap 2.7.1",
 "log",
 "memchr",
 "once_cell",
@@ -12322,7 +12335,7 @@ version = "0.19.15"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "1b5bb770da30e5cbfde35a2d7b9b8a2c4b8ef89548a7a6aeab5c9a576e3e7421"
 dependencies = [
- "indexmap 2.6.0",
+ "indexmap 2.7.1",
 "toml_datetime",
 "winnow 0.5.40",
 ]
@@ -12333,7 +12346,7 @@ version = "0.22.22"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "4ae48d6208a266e853d946088ed816055e556cc6028c5e8e2b84d9fa5dd7c7f5"
 dependencies = [
- "indexmap 2.6.0",
+ "indexmap 2.7.1",
 "serde",
 "serde_spanned",
 "toml_datetime",
@@ -12471,7 +12484,7 @@ dependencies = [
 "futures-core",
 "futures-util",
 "hdrhistogram",
- "indexmap 2.6.0",
+ "indexmap 2.7.1",
 "pin-project-lite",
 "slab",
 "sync_wrapper 1.0.1",
--- a/Cargo.toml
+++ b/Cargo.toml
@@ -129,7 +129,7 @@ etcd-client = "0.14"
 fst = "0.4.7"
 futures = "0.3"
 futures-util = "0.3"
-greptime-proto = { git = "https://github.com/GreptimeTeam/greptime-proto.git", rev = "a25adc8a01340231121646d8f0a29d0e92f45461" }
+greptime-proto = { git = "https://github.com/GreptimeTeam/greptime-proto.git", rev = "072ce580502e015df1a6b03a185b60309a7c2a7a" }
 hex = "0.4"
 http = "1"
 humantime = "2.1"
--- a/config/config.md
+++ b/config/config.md
@@ -319,6 +319,7 @@
 | `selector` | String | `round_robin` | Datanode selector type.<br/>- `round_robin` (default value)<br/>- `lease_based`<br/>- `load_based`<br/>For details, please see "https://docs.greptime.com/developer-guide/metasrv/selector". |
 | `use_memory_store` | Bool | `false` | Store data in memory. |
 | `enable_region_failover` | Bool | `false` | Whether to enable region failover.<br/>This feature is only available on GreptimeDB running on cluster mode and<br/>- Using Remote WAL<br/>- Using shared storage (e.g., s3). |
+| `node_max_idle_time` | String | `24hours` | Max allowed idle time before removing node info from metasrv memory. |
 | `enable_telemetry` | Bool | `true` | Whether to enable greptimedb telemetry. Enabled by default. |
 | `runtime` | -- | -- | The runtime options. |
 | `runtime.global_rt_size` | Integer | `8` | The number of threads to execute the runtime for global read operations. |
--- a/config/metasrv.example.toml
+++ b/config/metasrv.example.toml
@@ -50,6 +50,9 @@ use_memory_store = false
 ## - Using shared storage (e.g., s3).
 enable_region_failover = false

+## Max allowed idle time before removing node info from metasrv memory.
+node_max_idle_time = "24hours"
+
 ## Whether to enable greptimedb telemetry. Enabled by default.
 #+ enable_telemetry = true

--- a/src/client/src/lib.rs
+++ b/src/client/src/lib.rs
@@ -16,7 +16,6 @@

 mod client;
 pub mod client_manager;
-#[cfg(feature = "testing")]
 mod database;
 pub mod error;
 pub mod flow;
@@ -34,7 +33,6 @@ pub use common_recordbatch::{RecordBatches, SendableRecordBatchStream};
 use snafu::OptionExt;

 pub use self::client::Client;
-#[cfg(feature = "testing")]
 pub use self::database::Database;
 pub use self::error::{Error, Result};
 use crate::error::{IllegalDatabaseResponseSnafu, ServerSnafu};
--- a/src/cmd/src/flownode.rs
+++ b/src/cmd/src/flownode.rs
@@ -32,7 +32,7 @@ use common_meta::key::TableMetadataManager;
 use common_telemetry::info;
 use common_telemetry::logging::TracingOptions;
 use common_version::{short_version, version};
-use flow::{FlownodeBuilder, FlownodeInstance, FrontendInvoker};
+use flow::{FlownodeBuilder, FlownodeInstance, FrontendClient, FrontendInvoker};
 use meta_client::{MetaClientOptions, MetaClientType};
 use servers::Mode;
 use snafu::{OptionExt, ResultExt};
@@ -317,6 +317,8 @@ impl StartCommand {
            Arc::new(executor),
        );

+        let frontend_client = FrontendClient::from_meta_client(meta_client.clone());
+
        let flow_metadata_manager = Arc::new(FlowMetadataManager::new(cached_meta_backend.clone()));
        let flownode_builder = FlownodeBuilder::new(
            opts,
@@ -324,6 +326,7 @@ impl StartCommand {
            table_metadata_manager,
            catalog_manager.clone(),
            flow_metadata_manager,
+            Arc::new(frontend_client),
        )
        .with_heartbeat_task(heartbeat_task);

--- a/src/cmd/src/standalone.rs
+++ b/src/cmd/src/standalone.rs
@@ -54,7 +54,10 @@ use datanode::config::{DatanodeOptions, ProcedureConfig, RegionEngineConfig, Sto
 use datanode::datanode::{Datanode, DatanodeBuilder};
 use datanode::region_server::RegionServer;
 use file_engine::config::EngineConfig as FileEngineConfig;
-use flow::{FlowConfig, FlowWorkerManager, FlownodeBuilder, FlownodeOptions, FrontendInvoker};
+use flow::{
+    FlowConfig, FlowWorkerManager, FlownodeBuilder, FlownodeOptions, FrontendClient,
+    FrontendInvoker,
+};
 use frontend::frontend::FrontendOptions;
 use frontend::instance::builder::FrontendBuilder;
 use frontend::instance::{FrontendInstance, Instance as FeInstance, StandaloneDatanodeManager};
@@ -533,12 +536,16 @@ impl StartCommand {
            flow: opts.flow.clone(),
            ..Default::default()
        };
+
+        let fe_server_addr = fe_opts.grpc.bind_addr.clone();
+        let frontend_client = FrontendClient::from_static_grpc_addr(fe_server_addr);
        let flow_builder = FlownodeBuilder::new(
            flownode_options,
            plugins.clone(),
            table_metadata_manager.clone(),
            catalog_manager.clone(),
            flow_metadata_manager.clone(),
+            Arc::new(frontend_client),
        );
        let flownode = Arc::new(
            flow_builder
--- a/src/common/function/Cargo.toml
+++ b/src/common/function/Cargo.toml
@@ -12,6 +12,7 @@ default = ["geo"]
 geo = ["geohash", "h3o", "s2", "wkt", "geo-types", "dep:geo"]

 [dependencies]
+ahash = "0.8"
 api.workspace = true
 arc-swap = "1.0"
 async-trait.workspace = true
@@ -33,6 +34,7 @@ geo = { version = "0.29", optional = true }
 geo-types = { version = "0.7", optional = true }
 geohash = { version = "0.13", optional = true }
 h3o = { version = "0.6", optional = true }
+hyperloglogplus = "0.4"
 jsonb.workspace = true
 nalgebra.workspace = true
 num = "0.4"
--- a/src/common/function/src/aggr.rs
+++ b/src/common/function/src/aggr.rs
@@ -12,6 +12,9 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.

+mod hll;
 mod uddsketch_state;

+pub(crate) use hll::HllStateType;
+pub use hll::{HllState, HLL_MERGE_NAME, HLL_NAME};
 pub use uddsketch_state::{UddSketchState, UDDSKETCH_STATE_NAME};
--- a/src/common/function/src/aggr/hll.rs
+++ b/src/common/function/src/aggr/hll.rs
@@ -0,0 +1,319 @@
+// Copyright 2023 Greptime Team
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+use std::sync::Arc;
+
+use common_query::prelude::*;
+use common_telemetry::trace;
+use datafusion::arrow::array::ArrayRef;
+use datafusion::common::cast::{as_binary_array, as_string_array};
+use datafusion::common::not_impl_err;
+use datafusion::error::{DataFusionError, Result as DfResult};
+use datafusion::logical_expr::function::AccumulatorArgs;
+use datafusion::logical_expr::{Accumulator as DfAccumulator, AggregateUDF};
+use datafusion::prelude::create_udaf;
+use datatypes::arrow::datatypes::DataType;
+use hyperloglogplus::{HyperLogLog, HyperLogLogPlus};
+
+use crate::utils::FixedRandomState;
+
+pub const HLL_NAME: &str = "hll";
+pub const HLL_MERGE_NAME: &str = "hll_merge";
+
+const DEFAULT_PRECISION: u8 = 14;
+
+pub(crate) type HllStateType = HyperLogLogPlus<String, FixedRandomState>;
+
+pub struct HllState {
+    hll: HllStateType,
+}
+
+impl std::fmt::Debug for HllState {
+    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
+        write!(f, "HllState<Opaque>")
+    }
+}
+
+impl Default for HllState {
+    fn default() -> Self {
+        Self::new()
+    }
+}
+
+impl HllState {
+    pub fn new() -> Self {
+        Self {
+            // Safety: the DEFAULT_PRECISION is fixed and valid
+            hll: HllStateType::new(DEFAULT_PRECISION, FixedRandomState::new()).unwrap(),
+        }
+    }
+
+    /// Create a UDF for the `hll` function.
+    ///
+    /// `hll` accepts a string column and aggregates the
+    /// values into a HyperLogLog state.
+    pub fn state_udf_impl() -> AggregateUDF {
+        create_udaf(
+            HLL_NAME,
+            vec![DataType::Utf8],
+            Arc::new(DataType::Binary),
+            Volatility::Immutable,
+            Arc::new(Self::create_accumulator),
+            Arc::new(vec![DataType::Binary]),
+        )
+    }
+
+    /// Create a UDF for the `hll_merge` function.
+    ///
+    /// `hll_merge` accepts a binary column of states generated by `hll`
+    /// and merges them into a single state.
+    pub fn merge_udf_impl() -> AggregateUDF {
+        create_udaf(
+            HLL_MERGE_NAME,
+            vec![DataType::Binary],
+            Arc::new(DataType::Binary),
+            Volatility::Immutable,
+            Arc::new(Self::create_merge_accumulator),
+            Arc::new(vec![DataType::Binary]),
+        )
+    }
+
+    fn update(&mut self, value: &str) {
+        self.hll.insert(value);
+    }
+
+    fn merge(&mut self, raw: &[u8]) {
+        if let Ok(serialized) = bincode::deserialize::<HllStateType>(raw) {
+            if let Ok(()) = self.hll.merge(&serialized) {
+                return;
+            }
+        }
+        trace!("Warning: Failed to merge HyperLogLog from {:?}", raw);
+    }
+
+    fn create_accumulator(acc_args: AccumulatorArgs) -> DfResult<Box<dyn DfAccumulator>> {
+        let data_type = acc_args.exprs[0].data_type(acc_args.schema)?;
+
+        match data_type {
+            DataType::Utf8 => Ok(Box::new(HllState::new())),
+            other => not_impl_err!("{HLL_NAME} does not support data type: {other}"),
+        }
+    }
+
+    fn create_merge_accumulator(acc_args: AccumulatorArgs) -> DfResult<Box<dyn DfAccumulator>> {
+        let data_type = acc_args.exprs[0].data_type(acc_args.schema)?;
+
+        match data_type {
+            DataType::Binary => Ok(Box::new(HllState::new())),
+            other => not_impl_err!("{HLL_MERGE_NAME} does not support data type: {other}"),
+        }
+    }
+}
+
+impl DfAccumulator for HllState {
+    fn update_batch(&mut self, values: &[ArrayRef]) -> DfResult<()> {
+        let array = &values[0];
+
+        match array.data_type() {
+            DataType::Utf8 => {
+                let string_array = as_string_array(array)?;
+                for value in string_array.iter().flatten() {
+                    self.update(value);
+                }
+            }
+            DataType::Binary => {
+                let binary_array = as_binary_array(array)?;
+                for v in binary_array.iter().flatten() {
+                    self.merge(v);
+                }
+            }
+            _ => {
+                return not_impl_err!(
+                    "HLL functions do not support data type: {}",
+                    array.data_type()
+                )
+            }
+        }
+
+        Ok(())
+    }
+
+    fn evaluate(&mut self) -> DfResult<ScalarValue> {
+        Ok(ScalarValue::Binary(Some(
+            bincode::serialize(&self.hll).map_err(|e| {
+                DataFusionError::Internal(format!("Failed to serialize HyperLogLog: {}", e))
+            })?,
+        )))
+    }
+
+    fn size(&self) -> usize {
+        std::mem::size_of_val(&self.hll)
+    }
+
+    fn state(&mut self) -> DfResult<Vec<ScalarValue>> {
+        Ok(vec![ScalarValue::Binary(Some(
+            bincode::serialize(&self.hll).map_err(|e| {
+                DataFusionError::Internal(format!("Failed to serialize HyperLogLog: {}", e))
+            })?,
+        ))])
+    }
+
+    fn merge_batch(&mut self, states: &[ArrayRef]) -> DfResult<()> {
+        let array = &states[0];
+        let binary_array = as_binary_array(array)?;
+        for v in binary_array.iter().flatten() {
+            self.merge(v);
+        }
+
+        Ok(())
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use datafusion::arrow::array::{BinaryArray, StringArray};
+
+    use super::*;
+
+    #[test]
+    fn test_hll_basic() {
+        let mut state = HllState::new();
+        state.update("1");
+        state.update("2");
+        state.update("3");
+
+        let result = state.evaluate().unwrap();
+        if let ScalarValue::Binary(Some(bytes)) = result {
+            let mut hll: HllStateType = bincode::deserialize(&bytes).unwrap();
+            assert_eq!(hll.count().trunc() as u32, 3);
+        } else {
+            panic!("Expected binary scalar value");
+        }
+    }
+
+    #[test]
+    fn test_hll_roundtrip() {
+        let mut state = HllState::new();
+        state.update("1");
+        state.update("2");
+
+        // Serialize
+        let serialized = state.evaluate().unwrap();
+
+        // Create new state and merge the serialized data
+        let mut new_state = HllState::new();
+        if let ScalarValue::Binary(Some(bytes)) = &serialized {
+            new_state.merge(bytes);
+
+            // Verify the merged state matches original
+            let result = new_state.evaluate().unwrap();
+            if let ScalarValue::Binary(Some(new_bytes)) = result {
+                let mut original: HllStateType = bincode::deserialize(bytes).unwrap();
+                let mut merged: HllStateType = bincode::deserialize(&new_bytes).unwrap();
+                assert_eq!(original.count(), merged.count());
+            } else {
+                panic!("Expected binary scalar value");
+            }
+        } else {
+            panic!("Expected binary scalar value");
+        }
+    }
+
+    #[test]
+    fn test_hll_batch_update() {
+        let mut state = HllState::new();
+
+        // Test string values
+        let str_values = vec!["a", "b", "c", "d", "e", "f", "g", "h", "i"];
+        let str_array = Arc::new(StringArray::from(str_values)) as ArrayRef;
+        state.update_batch(&[str_array]).unwrap();
+
+        let result = state.evaluate().unwrap();
+        if let ScalarValue::Binary(Some(bytes)) = result {
+            let mut hll: HllStateType = bincode::deserialize(&bytes).unwrap();
+            assert_eq!(hll.count().trunc() as u32, 9);
+        } else {
+            panic!("Expected binary scalar value");
+        }
+    }
+
+    #[test]
+    fn test_hll_merge_batch() {
+        let mut state1 = HllState::new();
+        state1.update("1");
+        let state1_binary = state1.evaluate().unwrap();
+
+        let mut state2 = HllState::new();
+        state2.update("2");
+        let state2_binary = state2.evaluate().unwrap();
+
+        let mut merged_state = HllState::new();
+        if let (ScalarValue::Binary(Some(bytes1)), ScalarValue::Binary(Some(bytes2))) =
+            (&state1_binary, &state2_binary)
+        {
+            let binary_array = Arc::new(BinaryArray::from(vec![
+                bytes1.as_slice(),
+                bytes2.as_slice(),
+            ])) as ArrayRef;
+            merged_state.merge_batch(&[binary_array]).unwrap();
+
+            let result = merged_state.evaluate().unwrap();
+            if let ScalarValue::Binary(Some(bytes)) = result {
+                let mut hll: HllStateType = bincode::deserialize(&bytes).unwrap();
+                assert_eq!(hll.count().trunc() as u32, 2);
+            } else {
+                panic!("Expected binary scalar value");
+            }
+        } else {
+            panic!("Expected binary scalar values");
+        }
+    }
+
+    #[test]
+    fn test_hll_merge_function() {
+        // Create two HLL states with different values
+        let mut state1 = HllState::new();
+        state1.update("1");
+        state1.update("2");
+        let state1_binary = state1.evaluate().unwrap();
+
+        let mut state2 = HllState::new();
+        state2.update("2");
+        state2.update("3");
+        let state2_binary = state2.evaluate().unwrap();
+
+        // Create a merge state and merge both states
+        let mut merge_state = HllState::new();
+        if let (ScalarValue::Binary(Some(bytes1)), ScalarValue::Binary(Some(bytes2))) =
+            (&state1_binary, &state2_binary)
+        {
+            let binary_array = Arc::new(BinaryArray::from(vec![
+                bytes1.as_slice(),
+                bytes2.as_slice(),
+            ])) as ArrayRef;
+            merge_state.update_batch(&[binary_array]).unwrap();
+
+            let result = merge_state.evaluate().unwrap();
+            if let ScalarValue::Binary(Some(bytes)) = result {
+                let mut hll: HllStateType = bincode::deserialize(&bytes).unwrap();
+                // Should have 3 unique values: "1", "2", "3"
+                assert_eq!(hll.count().trunc() as u32, 3);
+            } else {
+                panic!("Expected binary scalar value");
+            }
+        } else {
+            panic!("Expected binary scalar values");
+        }
+    }
+}
--- a/src/common/function/src/function_registry.rs
+++ b/src/common/function/src/function_registry.rs
@@ -22,6 +22,7 @@ use crate::function::{AsyncFunctionRef, FunctionRef};
 use crate::scalars::aggregate::{AggregateFunctionMetaRef, AggregateFunctions};
 use crate::scalars::date::DateFunction;
 use crate::scalars::expression::ExpressionFunction;
+use crate::scalars::hll_count::HllCalcFunction;
 use crate::scalars::json::JsonFunction;
 use crate::scalars::matches::MatchesFunction;
 use crate::scalars::math::MathFunction;
@@ -107,6 +108,7 @@ pub static FUNCTION_REGISTRY: Lazy<Arc<FunctionRegistry>> = Lazy::new(|| {
    DateFunction::register(&function_registry);
    ExpressionFunction::register(&function_registry);
    UddSketchCalcFunction::register(&function_registry);
+    HllCalcFunction::register(&function_registry);

    // Aggregate functions
    AggregateFunctions::register(&function_registry);
--- a/src/common/function/src/scalars.rs
+++ b/src/common/function/src/scalars.rs
@@ -22,6 +22,7 @@ pub mod matches;
 pub mod math;
 pub mod vector;

+pub(crate) mod hll_count;
 #[cfg(test)]
 pub(crate) mod test;
 pub(crate) mod timestamp;
--- a/src/common/function/src/scalars/hll_count.rs
+++ b/src/common/function/src/scalars/hll_count.rs
@@ -0,0 +1,175 @@
+// 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.
+
+//! Implementation of the scalar function `hll_count`.
+
+use std::fmt;
+use std::fmt::Display;
+use std::sync::Arc;
+
+use common_query::error::{DowncastVectorSnafu, InvalidFuncArgsSnafu, Result};
+use common_query::prelude::{Signature, Volatility};
+use datatypes::data_type::ConcreteDataType;
+use datatypes::prelude::Vector;
+use datatypes::scalars::{ScalarVector, ScalarVectorBuilder};
+use datatypes::vectors::{BinaryVector, MutableVector, UInt64VectorBuilder, VectorRef};
+use hyperloglogplus::HyperLogLog;
+use snafu::OptionExt;
+
+use crate::aggr::HllStateType;
+use crate::function::{Function, FunctionContext};
+use crate::function_registry::FunctionRegistry;
+
+const NAME: &str = "hll_count";
+
+/// HllCalcFunction implements the scalar function `hll_count`.
+///
+/// It accepts one argument:
+/// 1. The serialized HyperLogLogPlus state, as produced by the aggregator (binary).
+///
+/// For each row, it deserializes the sketch and returns the estimated cardinality.
+#[derive(Debug, Default)]
+pub struct HllCalcFunction;
+
+impl HllCalcFunction {
+    pub fn register(registry: &FunctionRegistry) {
+        registry.register(Arc::new(HllCalcFunction));
+    }
+}
+
+impl Display for HllCalcFunction {
+    fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
+        write!(f, "{}", NAME.to_ascii_uppercase())
+    }
+}
+
+impl Function for HllCalcFunction {
+    fn name(&self) -> &str {
+        NAME
+    }
+
+    fn return_type(&self, _input_types: &[ConcreteDataType]) -> Result<ConcreteDataType> {
+        Ok(ConcreteDataType::uint64_datatype())
+    }
+
+    fn signature(&self) -> Signature {
+        // Only argument: HyperLogLogPlus state (binary)
+        Signature::exact(
+            vec![ConcreteDataType::binary_datatype()],
+            Volatility::Immutable,
+        )
+    }
+
+    fn eval(&self, _func_ctx: FunctionContext, columns: &[VectorRef]) -> Result<VectorRef> {
+        if columns.len() != 1 {
+            return InvalidFuncArgsSnafu {
+                err_msg: format!("hll_count expects 1 argument, got {}", columns.len()),
+            }
+            .fail();
+        }
+
+        let hll_vec = columns[0]
+            .as_any()
+            .downcast_ref::<BinaryVector>()
+            .with_context(|| DowncastVectorSnafu {
+                err_msg: format!("expect BinaryVector, got {}", columns[0].vector_type_name()),
+            })?;
+        let len = hll_vec.len();
+        let mut builder = UInt64VectorBuilder::with_capacity(len);
+
+        for i in 0..len {
+            let hll_opt = hll_vec.get_data(i);
+
+            if hll_opt.is_none() {
+                builder.push_null();
+                continue;
+            }
+
+            let hll_bytes = hll_opt.unwrap();
+
+            // Deserialize the HyperLogLogPlus from its bincode representation
+            let mut hll: HllStateType = match bincode::deserialize(hll_bytes) {
+                Ok(h) => h,
+                Err(e) => {
+                    common_telemetry::trace!("Failed to deserialize HyperLogLogPlus: {}", e);
+                    builder.push_null();
+                    continue;
+                }
+            };
+
+            builder.push(Some(hll.count().round() as u64));
+        }
+
+        Ok(builder.to_vector())
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use datatypes::vectors::BinaryVector;
+
+    use super::*;
+    use crate::utils::FixedRandomState;
+
+    #[test]
+    fn test_hll_count_function() {
+        let function = HllCalcFunction;
+        assert_eq!("hll_count", function.name());
+        assert_eq!(
+            ConcreteDataType::uint64_datatype(),
+            function
+                .return_type(&[ConcreteDataType::uint64_datatype()])
+                .unwrap()
+        );
+
+        // Create a test HLL
+        let mut hll = HllStateType::new(14, FixedRandomState::new()).unwrap();
+        for i in 1..=10 {
+            hll.insert(&i.to_string());
+        }
+
+        let serialized_bytes = bincode::serialize(&hll).unwrap();
+        let args: Vec<VectorRef> = vec![Arc::new(BinaryVector::from(vec![Some(serialized_bytes)]))];
+
+        let result = function.eval(FunctionContext::default(), &args).unwrap();
+        assert_eq!(result.len(), 1);
+
+        // Test cardinality estimate
+        if let datatypes::value::Value::UInt64(v) = result.get(0) {
+            assert_eq!(v, 10);
+        } else {
+            panic!("Expected uint64 value");
+        }
+    }
+
+    #[test]
+    fn test_hll_count_function_errors() {
+        let function = HllCalcFunction;
+
+        // Test with invalid number of arguments
+        let args: Vec<VectorRef> = vec![];
+        let result = function.eval(FunctionContext::default(), &args);
+        assert!(result.is_err());
+        assert!(result
+            .unwrap_err()
+            .to_string()
+            .contains("hll_count expects 1 argument"));
+
+        // Test with invalid binary data
+        let args: Vec<VectorRef> = vec![Arc::new(BinaryVector::from(vec![Some(vec![1, 2, 3])]))]; // Invalid binary data
+        let result = function.eval(FunctionContext::default(), &args).unwrap();
+        assert_eq!(result.len(), 1);
+        assert!(matches!(result.get(0), datatypes::value::Value::Null));
+    }
+}
--- a/src/common/function/src/utils.rs
+++ b/src/common/function/src/utils.rs
@@ -12,6 +12,11 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.

+use std::hash::BuildHasher;
+
+use ahash::RandomState;
+use serde::{Deserialize, Serialize};
+
 /// Escapes special characters in the provided pattern string for `LIKE`.
 ///
 /// Specifically, it prefixes the backslash (`\`), percent (`%`), and underscore (`_`)
@@ -32,6 +37,71 @@ pub fn escape_like_pattern(pattern: &str) -> String {
        })
        .collect::<String>()
 }
+
+/// A random state with fixed seeds.
+///
+/// This is used to ensure that the hash values are consistent across
+/// different processes, and easy to serialize and deserialize.
+#[derive(Debug)]
+pub struct FixedRandomState {
+    state: RandomState,
+}
+
+impl FixedRandomState {
+    // some random seeds
+    const RANDOM_SEED_0: u64 = 0x517cc1b727220a95;
+    const RANDOM_SEED_1: u64 = 0x428a2f98d728ae22;
+    const RANDOM_SEED_2: u64 = 0x7137449123ef65cd;
+    const RANDOM_SEED_3: u64 = 0xb5c0fbcfec4d3b2f;
+
+    pub fn new() -> Self {
+        Self {
+            state: ahash::RandomState::with_seeds(
+                Self::RANDOM_SEED_0,
+                Self::RANDOM_SEED_1,
+                Self::RANDOM_SEED_2,
+                Self::RANDOM_SEED_3,
+            ),
+        }
+    }
+}
+
+impl Default for FixedRandomState {
+    fn default() -> Self {
+        Self::new()
+    }
+}
+
+impl BuildHasher for FixedRandomState {
+    type Hasher = ahash::AHasher;
+
+    fn build_hasher(&self) -> Self::Hasher {
+        self.state.build_hasher()
+    }
+
+    fn hash_one<T: std::hash::Hash>(&self, x: T) -> u64 {
+        self.state.hash_one(x)
+    }
+}
+
+impl Serialize for FixedRandomState {
+    fn serialize<S>(&self, serializer: S) -> Result<S::Ok, S::Error>
+    where
+        S: serde::Serializer,
+    {
+        serializer.serialize_unit()
+    }
+}
+
+impl<'de> Deserialize<'de> for FixedRandomState {
+    fn deserialize<D>(_deserializer: D) -> Result<Self, D::Error>
+    where
+        D: serde::Deserializer<'de>,
+    {
+        Ok(Self::new())
+    }
+}
+
 #[cfg(test)]
 mod tests {
    use super::*;
--- a/src/common/grpc/src/channel_manager.rs
+++ b/src/common/grpc/src/channel_manager.rs
@@ -445,10 +445,16 @@ impl Pool {

 async fn recycle_channel_in_loop(pool: Arc<Pool>, interval_secs: u64) {
    let mut interval = tokio::time::interval(Duration::from_secs(interval_secs));
-
+    // use weak ref here to prevent pool being leaked
+    let pool_weak = Arc::downgrade(&pool);
    loop {
        let _ = interval.tick().await;
-        pool.retain_channel(|_, c| c.access.swap(0, Ordering::Relaxed) != 0)
+        if let Some(pool) = pool_weak.upgrade() {
+            pool.retain_channel(|_, c| c.access.swap(0, Ordering::Relaxed) != 0)
+        } else {
+            // no one is using this pool, so we can also let go
+            break;
+        }
    }
 }

--- a/src/common/meta/src/cluster.rs
+++ b/src/common/meta/src/cluster.rs
@@ -57,12 +57,10 @@ pub trait ClusterInfo {
 }

 /// The key of [NodeInfo] in the storage. The format is `__meta_cluster_node_info-{cluster_id}-{role}-{node_id}`.
-///
-/// This key cannot be used to describe the `Metasrv` because the `Metasrv` does not have
-/// a `cluster_id`, it serves multiple clusters.
 #[derive(Debug, Clone, Copy, Eq, Hash, PartialEq, Serialize, Deserialize)]
 pub struct NodeInfoKey {
    /// The cluster id.
+    // todo(hl): remove cluster_id as it is not assigned anywhere.
    pub cluster_id: ClusterId,
    /// The role of the node. It can be `[Role::Datanode]` or `[Role::Frontend]`.
    pub role: Role,
@@ -232,8 +230,8 @@ impl TryFrom<Vec<u8>> for NodeInfoKey {
    }
 }

-impl From<NodeInfoKey> for Vec<u8> {
-    fn from(key: NodeInfoKey) -> Self {
+impl From<&NodeInfoKey> for Vec<u8> {
+    fn from(key: &NodeInfoKey) -> Self {
        format!(
            "{}-{}-{}-{}",
            CLUSTER_NODE_INFO_PREFIX,
@@ -315,7 +313,7 @@ mod tests {
            node_id: 2,
        };

-        let key_bytes: Vec<u8> = key.into();
+        let key_bytes: Vec<u8> = (&key).into();
        let new_key: NodeInfoKey = key_bytes.try_into().unwrap();

        assert_eq!(1, new_key.cluster_id);
--- a/src/common/meta/src/ddl/create_flow.rs
+++ b/src/common/meta/src/ddl/create_flow.rs
@@ -343,6 +343,7 @@ pub enum FlowType {
 impl FlowType {
    pub const RECORDING_RULE: &str = "recording_rule";
    pub const STREAMING: &str = "streaming";
+    pub const FLOW_TYPE_KEY: &str = "flow_type";
 }

 impl Default for FlowType {
@@ -398,7 +399,8 @@ impl From<&CreateFlowData> for CreateRequest {
        };

        let flow_type = value.flow_type.unwrap_or_default().to_string();
-        req.flow_options.insert("flow_type".to_string(), flow_type);
+        req.flow_options
+            .insert(FlowType::FLOW_TYPE_KEY.to_string(), flow_type);
        req
    }
 }
@@ -430,7 +432,7 @@ impl From<&CreateFlowData> for (FlowInfoValue, Vec<(FlowPartitionId, FlowRouteVa
            .collect::<Vec<_>>();

        let flow_type = value.flow_type.unwrap_or_default().to_string();
-        options.insert("flow_type".to_string(), flow_type);
+        options.insert(FlowType::FLOW_TYPE_KEY.to_string(), flow_type);

        let flow_info = FlowInfoValue {
            source_table_ids: value.source_table_ids.clone(),
--- a/src/common/meta/src/ddl/drop_database/executor.rs
+++ b/src/common/meta/src/ddl/drop_database/executor.rs
@@ -128,7 +128,7 @@ impl State for DropDatabaseExecutor {
            .await?;
        executor.invalidate_table_cache(ddl_ctx).await?;
        executor
-            .on_drop_regions(ddl_ctx, &self.physical_region_routes)
+            .on_drop_regions(ddl_ctx, &self.physical_region_routes, true)
            .await?;
        info!("Table: {}({}) is dropped", self.table_name, self.table_id);

--- a/src/common/meta/src/ddl/drop_table.rs
+++ b/src/common/meta/src/ddl/drop_table.rs
@@ -156,7 +156,7 @@ impl DropTableProcedure {

    pub async fn on_datanode_drop_regions(&mut self) -> Result<Status> {
        self.executor
-            .on_drop_regions(&self.context, &self.data.physical_region_routes)
+            .on_drop_regions(&self.context, &self.data.physical_region_routes, false)
            .await?;
        self.data.state = DropTableState::DeleteTombstone;
        Ok(Status::executing(true))
--- a/src/common/meta/src/ddl/drop_table/executor.rs
+++ b/src/common/meta/src/ddl/drop_table/executor.rs
@@ -214,6 +214,7 @@ impl DropTableExecutor {
        &self,
        ctx: &DdlContext,
        region_routes: &[RegionRoute],
+        fast_path: bool,
    ) -> Result<()> {
        let leaders = find_leaders(region_routes);
        let mut drop_region_tasks = Vec::with_capacity(leaders.len());
@@ -236,6 +237,7 @@ impl DropTableExecutor {
                    }),
                    body: Some(region_request::Body::Drop(PbDropRegionRequest {
                        region_id: region_id.as_u64(),
+                        fast_path,
                    })),
                };
                let datanode = datanode.clone();
--- a/src/common/meta/src/lib.rs
+++ b/src/common/meta/src/lib.rs
@@ -34,6 +34,7 @@ pub mod kv_backend;
 pub mod leadership_notifier;
 pub mod lock_key;
 pub mod metrics;
+pub mod node_expiry_listener;
 pub mod node_manager;
 pub mod peer;
 pub mod range_stream;
--- a/src/common/meta/src/node_expiry_listener.rs
+++ b/src/common/meta/src/node_expiry_listener.rs
@@ -0,0 +1,152 @@
+// Copyright 2023 Greptime Team
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+use std::sync::Mutex;
+use std::time::Duration;
+
+use common_telemetry::{debug, error, info, warn};
+use tokio::task::JoinHandle;
+use tokio::time::{interval, MissedTickBehavior};
+
+use crate::cluster::{NodeInfo, NodeInfoKey};
+use crate::error;
+use crate::kv_backend::ResettableKvBackendRef;
+use crate::leadership_notifier::LeadershipChangeListener;
+use crate::rpc::store::RangeRequest;
+use crate::rpc::KeyValue;
+
+/// [NodeExpiryListener] periodically checks all node info in memory and removes
+/// expired node info to prevent memory leak.
+pub struct NodeExpiryListener {
+    handle: Mutex<Option<JoinHandle<()>>>,
+    max_idle_time: Duration,
+    in_memory: ResettableKvBackendRef,
+}
+
+impl Drop for NodeExpiryListener {
+    fn drop(&mut self) {
+        self.stop();
+    }
+}
+
+impl NodeExpiryListener {
+    pub fn new(max_idle_time: Duration, in_memory: ResettableKvBackendRef) -> Self {
+        Self {
+            handle: Mutex::new(None),
+            max_idle_time,
+            in_memory,
+        }
+    }
+
+    async fn start(&self) {
+        let mut handle = self.handle.lock().unwrap();
+        if handle.is_none() {
+            let in_memory = self.in_memory.clone();
+
+            let max_idle_time = self.max_idle_time;
+            let ticker_loop = tokio::spawn(async move {
+                // Run clean task every minute.
+                let mut interval = interval(Duration::from_secs(60));
+                interval.set_missed_tick_behavior(MissedTickBehavior::Skip);
+                loop {
+                    interval.tick().await;
+                    if let Err(e) = Self::clean_expired_nodes(&in_memory, max_idle_time).await {
+                        error!(e; "Failed to clean expired node");
+                    }
+                }
+            });
+            *handle = Some(ticker_loop);
+        }
+    }
+
+    fn stop(&self) {
+        if let Some(handle) = self.handle.lock().unwrap().take() {
+            handle.abort();
+            info!("Node expiry listener stopped")
+        }
+    }
+
+    /// Cleans expired nodes from memory.
+    async fn clean_expired_nodes(
+        in_memory: &ResettableKvBackendRef,
+        max_idle_time: Duration,
+    ) -> error::Result<()> {
+        let node_keys = Self::list_expired_nodes(in_memory, max_idle_time).await?;
+        for key in node_keys {
+            let key_bytes: Vec<u8> = (&key).into();
+            if let Err(e) = in_memory.delete(&key_bytes, false).await {
+                warn!(e; "Failed to delete expired node: {:?}", key_bytes);
+            } else {
+                debug!("Deleted expired node key: {:?}", key);
+            }
+        }
+        Ok(())
+    }
+
+    /// Lists expired nodes that have been inactive more than `max_idle_time`.
+    async fn list_expired_nodes(
+        in_memory: &ResettableKvBackendRef,
+        max_idle_time: Duration,
+    ) -> error::Result<impl Iterator<Item = NodeInfoKey>> {
+        let prefix = NodeInfoKey::key_prefix_with_cluster_id(0);
+        let req = RangeRequest::new().with_prefix(prefix);
+        let current_time_millis = common_time::util::current_time_millis();
+        let resp = in_memory.range(req).await?;
+        Ok(resp
+            .kvs
+            .into_iter()
+            .filter_map(move |KeyValue { key, value }| {
+                let Ok(info) = NodeInfo::try_from(value).inspect_err(|e| {
+                    warn!(e; "Unrecognized node info value");
+                }) else {
+                    return None;
+                };
+                if (current_time_millis - info.last_activity_ts) > max_idle_time.as_millis() as i64
+                {
+                    NodeInfoKey::try_from(key)
+                        .inspect_err(|e| {
+                            warn!(e; "Unrecognized node info key: {:?}", info.peer);
+                        })
+                        .ok()
+                        .inspect(|node_key| {
+                            debug!("Found expired node: {:?}", node_key);
+                        })
+                } else {
+                    None
+                }
+            }))
+    }
+}
+
+#[async_trait::async_trait]
+impl LeadershipChangeListener for NodeExpiryListener {
+    fn name(&self) -> &str {
+        "NodeExpiryListener"
+    }
+
+    async fn on_leader_start(&self) -> error::Result<()> {
+        self.start().await;
+        info!(
+            "On leader start, node expiry listener started with max idle time: {:?}",
+            self.max_idle_time
+        );
+        Ok(())
+    }
+
+    async fn on_leader_stop(&self) -> error::Result<()> {
+        self.stop();
+        info!("On leader stop, node expiry listener stopped");
+        Ok(())
+    }
+}
--- a/src/datanode/src/region_server.rs
+++ b/src/datanode/src/region_server.rs
@@ -1218,7 +1218,10 @@ mod tests {
        );

        let response = mock_region_server
-            .handle_request(region_id, RegionRequest::Drop(RegionDropRequest {}))
+            .handle_request(
+                region_id,
+                RegionRequest::Drop(RegionDropRequest { fast_path: false }),
+            )
            .await
            .unwrap();
        assert_eq!(response.affected_rows, 0);
@@ -1310,7 +1313,10 @@ mod tests {
            .insert(region_id, RegionEngineWithStatus::Ready(engine.clone()));

        mock_region_server
-            .handle_request(region_id, RegionRequest::Drop(RegionDropRequest {}))
+            .handle_request(
+                region_id,
+                RegionRequest::Drop(RegionDropRequest { fast_path: false }),
+            )
            .await
            .unwrap_err();

--- a/src/datatypes/src/lib.rs
+++ b/src/datatypes/src/lib.rs
@@ -32,5 +32,5 @@ pub mod types;
 pub mod value;
 pub mod vectors;

-pub use arrow;
+pub use arrow::{self, compute};
 pub use error::{Error, Result};
--- a/src/flow/Cargo.toml
+++ b/src/flow/Cargo.toml
@@ -16,6 +16,7 @@ async-trait.workspace = true
 bytes.workspace = true
 cache.workspace = true
 catalog.workspace = true
+chrono.workspace = true
 client.workspace = true
 common-base.workspace = true
 common-config.workspace = true
--- a/src/flow/src/adapter.rs
+++ b/src/flow/src/adapter.rs
@@ -49,12 +49,13 @@ pub(crate) use crate::adapter::node_context::FlownodeContext;
 use crate::adapter::refill::RefillTask;
 use crate::adapter::table_source::ManagedTableSource;
 use crate::adapter::util::relation_desc_to_column_schemas_with_fallback;
-pub(crate) use crate::adapter::worker::{create_worker, Worker, WorkerHandle};
+pub(crate) use crate::adapter::worker::{create_worker, WorkerHandle};
 use crate::compute::ErrCollector;
 use crate::df_optimizer::sql_to_flow_plan;
 use crate::error::{EvalSnafu, ExternalSnafu, InternalSnafu, InvalidQuerySnafu, UnexpectedSnafu};
 use crate::expr::Batch;
 use crate::metrics::{METRIC_FLOW_INSERT_ELAPSED, METRIC_FLOW_ROWS, METRIC_FLOW_RUN_INTERVAL_MS};
+use crate::recording_rules::RecordingRuleEngine;
 use crate::repr::{self, DiffRow, RelationDesc, Row, BATCH_SIZE};

 mod flownode_impl;
@@ -63,7 +64,7 @@ pub(crate) mod refill;
 mod stat;
 #[cfg(test)]
 mod tests;
-mod util;
+pub(crate) mod util;
 mod worker;

 pub(crate) mod node_context;
@@ -171,6 +172,8 @@ pub struct FlowWorkerManager {
    flush_lock: RwLock<()>,
    /// receive a oneshot sender to send state size report
    state_report_handler: RwLock<Option<StateReportHandler>>,
+    /// engine for recording rule
+    rule_engine: RecordingRuleEngine,
 }

 /// Building FlownodeManager
@@ -185,6 +188,7 @@ impl FlowWorkerManager {
        node_id: Option<u32>,
        query_engine: Arc<dyn QueryEngine>,
        table_meta: TableMetadataManagerRef,
+        rule_engine: RecordingRuleEngine,
    ) -> Self {
        let srv_map = ManagedTableSource::new(
            table_meta.table_info_manager().clone(),
@@ -207,6 +211,7 @@ impl FlowWorkerManager {
            node_id,
            flush_lock: RwLock::new(()),
            state_report_handler: RwLock::new(None),
+            rule_engine,
        }
    }

@@ -215,25 +220,6 @@ impl FlowWorkerManager {
        self
    }

-    /// Create a flownode manager with one worker
-    pub fn new_with_workers<'s>(
-        node_id: Option<u32>,
-        query_engine: Arc<dyn QueryEngine>,
-        table_meta: TableMetadataManagerRef,
-        num_workers: usize,
-    ) -> (Self, Vec<Worker<'s>>) {
-        let mut zelf = Self::new(node_id, query_engine, table_meta);
-
-        let workers: Vec<_> = (0..num_workers)
-            .map(|_| {
-                let (handle, worker) = create_worker();
-                zelf.add_worker_handle(handle);
-                worker
-            })
-            .collect();
-        (zelf, workers)
-    }
-
    /// add a worker handler to manager, meaning this corresponding worker is under it's manage
    pub fn add_worker_handle(&mut self, handle: WorkerHandle) {
        self.worker_handles.push(handle);
@@ -751,7 +737,11 @@ pub struct CreateFlowArgs {
 /// Create&Remove flow
 impl FlowWorkerManager {
    /// remove a flow by it's id
+    #[allow(unreachable_code)]
    pub async fn remove_flow(&self, flow_id: FlowId) -> Result<(), Error> {
+        // TODO(discord9): reroute some back to streaming engine later
+        return self.rule_engine.remove_flow(flow_id).await;
+
        for handle in self.worker_handles.iter() {
            if handle.contains_flow(flow_id).await? {
                handle.remove_flow(flow_id).await?;
@@ -767,8 +757,10 @@ impl FlowWorkerManager {
    /// steps to create task:
    /// 1. parse query into typed plan(and optional parse expire_after expr)
    /// 2. render source/sink with output table id and used input table id
-    #[allow(clippy::too_many_arguments)]
+    #[allow(clippy::too_many_arguments, unreachable_code)]
    pub async fn create_flow(&self, args: CreateFlowArgs) -> Result<Option<FlowId>, Error> {
+        // TODO(discord9): reroute some back to streaming engine later
+        return self.rule_engine.create_flow(args).await;
        let CreateFlowArgs {
            flow_id,
            sink_table_name,
--- a/src/flow/src/adapter/flownode_impl.rs
+++ b/src/flow/src/adapter/flownode_impl.rs
@@ -153,7 +153,13 @@ impl Flownode for FlowWorkerManager {
        }
    }

+    #[allow(unreachable_code, unused)]
    async fn handle_inserts(&self, request: InsertRequests) -> Result<FlowResponse> {
+        return self
+            .rule_engine
+            .handle_inserts(request)
+            .await
+            .map_err(to_meta_err(snafu::location!()));
        // using try_read to ensure two things:
        // 1. flush wouldn't happen until inserts before it is inserted
        // 2. inserts happening concurrently with flush wouldn't be block by flush
@@ -206,15 +212,15 @@ impl Flownode for FlowWorkerManager {
                    .collect_vec();
                let table_col_names = table_schema.relation_desc.names;
                let table_col_names = table_col_names
-                    .iter().enumerate()
-                    .map(|(idx,name)| match name {
-                        Some(name) => Ok(name.clone()),
-                        None => InternalSnafu {
-                            reason: format!("Expect column {idx} of table id={table_id} to have name in table schema, found None"),
-                        }
-                        .fail().map_err(BoxedError::new).context(ExternalSnafu),
-                    })
-                    .collect::<Result<Vec<_>>>()?;
+                        .iter().enumerate()
+                        .map(|(idx,name)| match name {
+                            Some(name) => Ok(name.clone()),
+                            None => InternalSnafu {
+                                reason: format!("Expect column {idx} of table id={table_id} to have name in table schema, found None"),
+                            }
+                            .fail().map_err(BoxedError::new).context(ExternalSnafu),
+                        })
+                        .collect::<Result<Vec<_>>>()?;
                let name_to_col = HashMap::<_, _>::from_iter(
                    insert_schema
                        .iter()
--- a/src/flow/src/adapter/util.rs
+++ b/src/flow/src/adapter/util.rs
@@ -12,6 +12,8 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.

+//! Some utility functions
+
 use std::sync::Arc;

 use api::helper::ColumnDataTypeWrapper;
--- a/src/flow/src/error.rs
+++ b/src/flow/src/error.rs
@@ -16,6 +16,7 @@

 use std::any::Any;

+use arrow_schema::ArrowError;
 use common_error::ext::BoxedError;
 use common_error::{define_into_tonic_status, from_err_code_msg_to_header};
 use common_macro::stack_trace_debug;
@@ -53,6 +54,13 @@ pub enum Error {
        location: Location,
    },

+    #[snafu(display("Time error"))]
+    Time {
+        source: common_time::error::Error,
+        #[snafu(implicit)]
+        location: Location,
+    },
+
    #[snafu(display("External error"))]
    External {
        source: BoxedError,
@@ -156,6 +164,15 @@ pub enum Error {
        location: Location,
    },

+    #[snafu(display("Arrow error: {raw:?} in context: {context}"))]
+    Arrow {
+        #[snafu(source)]
+        raw: ArrowError,
+        context: String,
+        #[snafu(implicit)]
+        location: Location,
+    },
+
    #[snafu(display("Datafusion error: {raw:?} in context: {context}"))]
    Datafusion {
        #[snafu(source)]
@@ -230,6 +247,7 @@ impl ErrorExt for Error {
        match self {
            Self::Eval { .. }
            | Self::JoinTask { .. }
+            | Self::Arrow { .. }
            | Self::Datafusion { .. }
            | Self::InsertIntoFlow { .. } => StatusCode::Internal,
            Self::FlowAlreadyExist { .. } => StatusCode::TableAlreadyExists,
@@ -238,7 +256,9 @@ impl ErrorExt for Error {
            | Self::FlowNotFound { .. }
            | Self::ListFlows { .. } => StatusCode::TableNotFound,
            Self::Plan { .. } | Self::Datatypes { .. } => StatusCode::PlanQuery,
-            Self::InvalidQuery { .. } | Self::CreateFlow { .. } => StatusCode::EngineExecuteQuery,
+            Self::InvalidQuery { .. } | Self::CreateFlow { .. } | Self::Time { .. } => {
+                StatusCode::EngineExecuteQuery
+            }
            Self::Unexpected { .. } => StatusCode::Unexpected,
            Self::NotImplemented { .. } | Self::UnsupportedTemporalFilter { .. } => {
                StatusCode::Unsupported
--- a/src/flow/src/expr/utils.rs
+++ b/src/flow/src/expr/utils.rs
@@ -238,6 +238,7 @@ mod test {

        for (sql, current, expected) in &testcases {
            let plan = sql_to_substrait(engine.clone(), sql).await;
+
            let mut ctx = create_test_ctx();
            let flow_plan = TypedPlan::from_substrait_plan(&mut ctx, &plan)
                .await
--- a/src/flow/src/heartbeat.rs
+++ b/src/flow/src/heartbeat.rs
@@ -103,6 +103,11 @@ impl HeartbeatTask {
            warn!("Heartbeat task started multiple times");
            return Ok(());
        }
+
+        self.create_streams().await
+    }
+
+    async fn create_streams(&self) -> Result<(), Error> {
        info!("Start to establish the heartbeat connection to metasrv.");
        let (req_sender, resp_stream) = self
            .meta_client
@@ -125,13 +130,6 @@ impl HeartbeatTask {

    pub fn shutdown(&self) {
        info!("Close heartbeat task for flownode");
-        if self
-            .running
-            .compare_exchange(true, false, Ordering::AcqRel, Ordering::Acquire)
-            .is_err()
-        {
-            warn!("Call close heartbeat task multiple times");
-        }
    }

    fn new_heartbeat_request(
@@ -231,6 +229,8 @@ impl HeartbeatTask {
                // 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;
            }
+
+            info!("flownode heartbeat task stopped.");
        });
    }

@@ -274,7 +274,7 @@ impl HeartbeatTask {

            info!("Try to re-establish the heartbeat connection to metasrv.");

-            if self.start().await.is_ok() {
+            if self.create_streams().await.is_ok() {
                break;
            }
        }
--- a/src/flow/src/lib.rs
+++ b/src/flow/src/lib.rs
@@ -33,6 +33,7 @@ mod expr;
 pub mod heartbeat;
 mod metrics;
 mod plan;
+mod recording_rules;
 mod repr;
 mod server;
 mod transform;
@@ -43,4 +44,5 @@ mod test_utils;

 pub use adapter::{FlowConfig, FlowWorkerManager, FlowWorkerManagerRef, FlownodeOptions};
 pub use error::{Error, Result};
+pub use recording_rules::FrontendClient;
 pub use server::{FlownodeBuilder, FlownodeInstance, FlownodeServer, FrontendInvoker};
--- a/src/flow/src/metrics.rs
+++ b/src/flow/src/metrics.rs
@@ -28,6 +28,32 @@ lazy_static! {
        &["table_id"]
    )
    .unwrap();
+    pub static ref METRIC_FLOW_RULE_ENGINE_QUERY_TIME: HistogramVec = register_histogram_vec!(
+        "greptime_flow_rule_engine_query_time",
+        "flow rule engine query time",
+        &["flow_id"],
+        vec![
+            0.0,
+            1.,
+            3.,
+            5.,
+            10.,
+            20.,
+            30.,
+            60.,
+            2. * 60.,
+            5. * 60.,
+            10. * 60.
+        ]
+    )
+    .unwrap();
+    pub static ref METRIC_FLOW_RULE_ENGINE_SLOW_QUERY: HistogramVec = register_histogram_vec!(
+        "greptime_flow_rule_engine_slow_query",
+        "flow rule engine slow query",
+        &["flow_id", "sql", "peer"],
+        vec![60., 2. * 60., 3. * 60., 5. * 60., 10. * 60.]
+    )
+    .unwrap();
    pub static ref METRIC_FLOW_RUN_INTERVAL_MS: IntGauge =
        register_int_gauge!("greptime_flow_run_interval_ms", "flow run interval in ms").unwrap();
    pub static ref METRIC_FLOW_ROWS: IntCounterVec = register_int_counter_vec!(
--- a/src/flow/src/recording_rules.rs
+++ b/src/flow/src/recording_rules.rs
@@ -0,0 +1,940 @@
+// 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.
+
+//! Run flow as recording rule which is time-window-aware normal query triggered every tick set by user
+
+mod engine;
+mod frontend_client;
+
+use std::collections::BTreeSet;
+use std::sync::Arc;
+
+use api::helper::pb_value_to_value_ref;
+use catalog::CatalogManagerRef;
+use common_error::ext::BoxedError;
+use common_recordbatch::DfRecordBatch;
+use common_telemetry::warn;
+use common_time::timestamp::TimeUnit;
+use common_time::Timestamp;
+use datafusion::error::Result as DfResult;
+use datafusion::logical_expr::Expr;
+use datafusion::physical_planner::{DefaultPhysicalPlanner, PhysicalPlanner};
+use datafusion::prelude::SessionContext;
+use datafusion::sql::unparser::Unparser;
+use datafusion_common::tree_node::{Transformed, TreeNode, TreeNodeRecursion, TreeNodeRewriter};
+use datafusion_common::{DFSchema, TableReference};
+use datafusion_expr::{ColumnarValue, LogicalPlan};
+use datafusion_physical_expr::PhysicalExprRef;
+use datatypes::prelude::{ConcreteDataType, DataType};
+use datatypes::scalars::ScalarVector;
+use datatypes::schema::TIME_INDEX_KEY;
+use datatypes::value::Value;
+use datatypes::vectors::{
+    TimestampMicrosecondVector, TimestampMillisecondVector, TimestampNanosecondVector,
+    TimestampSecondVector, Vector,
+};
+pub use engine::RecordingRuleEngine;
+pub use frontend_client::FrontendClient;
+use itertools::Itertools;
+use query::parser::QueryLanguageParser;
+use query::QueryEngineRef;
+use session::context::QueryContextRef;
+use snafu::{ensure, OptionExt, ResultExt};
+
+use crate::adapter::util::from_proto_to_data_type;
+use crate::df_optimizer::apply_df_optimizer;
+use crate::error::{ArrowSnafu, DatafusionSnafu, DatatypesSnafu, ExternalSnafu, UnexpectedSnafu};
+use crate::expr::error::DataTypeSnafu;
+use crate::Error;
+
+#[derive(Debug, Clone)]
+pub struct TimeWindowExpr {
+    phy_expr: PhysicalExprRef,
+    column_name: String,
+    logical_expr: Expr,
+    df_schema: DFSchema,
+}
+
+impl TimeWindowExpr {
+    pub fn from_expr(expr: &Expr, column_name: &str, df_schema: &DFSchema) -> Result<Self, Error> {
+        let phy_planner = DefaultPhysicalPlanner::default();
+
+        let phy_expr: PhysicalExprRef = phy_planner
+            .create_physical_expr(expr, df_schema, &SessionContext::new().state())
+            .with_context(|_e| DatafusionSnafu {
+                context: format!(
+                    "Failed to create physical expression from {expr:?} using {df_schema:?}"
+                ),
+            })?;
+        Ok(Self {
+            phy_expr,
+            column_name: column_name.to_string(),
+            logical_expr: expr.clone(),
+            df_schema: df_schema.clone(),
+        })
+    }
+
+    pub fn eval(
+        &self,
+        current: Timestamp,
+    ) -> Result<(Option<Timestamp>, Option<Timestamp>), Error> {
+        let lower_bound =
+            find_expr_time_window_lower_bound(&self.logical_expr, &self.df_schema, current)?;
+        let upper_bound =
+            find_expr_time_window_upper_bound(&self.logical_expr, &self.df_schema, current)?;
+        Ok((lower_bound, upper_bound))
+    }
+
+    /// Find timestamps from rows using time window expr
+    pub async fn handle_rows(
+        &self,
+        rows_list: Vec<api::v1::Rows>,
+    ) -> Result<BTreeSet<Timestamp>, Error> {
+        let mut time_windows = BTreeSet::new();
+
+        for rows in rows_list {
+            // pick the time index column and use it to eval on `self.expr`
+            let ts_col_index = rows
+                .schema
+                .iter()
+                .map(|col| col.column_name.clone())
+                .position(|name| name == self.column_name);
+            let Some(ts_col_index) = ts_col_index else {
+                warn!("can't found time index column in schema: {:?}", rows.schema);
+                continue;
+            };
+            let col_schema = &rows.schema[ts_col_index];
+            let cdt = from_proto_to_data_type(col_schema)?;
+
+            let column_values = rows
+                .rows
+                .iter()
+                .map(|row| &row.values[ts_col_index])
+                .collect_vec();
+
+            let mut vector = cdt.create_mutable_vector(column_values.len());
+            for value in column_values {
+                let value = pb_value_to_value_ref(value, &None);
+                vector.try_push_value_ref(value).context(DataTypeSnafu {
+                    msg: "Failed to convert rows to columns",
+                })?;
+            }
+            let vector = vector.to_vector();
+
+            let df_schema = create_df_schema_for_ts_column(&self.column_name, cdt)?;
+
+            let rb =
+                DfRecordBatch::try_new(df_schema.inner().clone(), vec![vector.to_arrow_array()])
+                    .with_context(|_e| ArrowSnafu {
+                        context: format!(
+                            "Failed to create record batch from {df_schema:?} and {vector:?}"
+                        ),
+                    })?;
+
+            let eval_res = self
+                .phy_expr
+                .evaluate(&rb)
+                .with_context(|_| DatafusionSnafu {
+                    context: format!(
+                        "Failed to evaluate physical expression {:?} on {rb:?}",
+                        self.phy_expr
+                    ),
+                })?;
+
+            let res = columnar_to_ts_vector(&eval_res)?;
+
+            for ts in res.into_iter().flatten() {
+                time_windows.insert(ts);
+            }
+        }
+
+        Ok(time_windows)
+    }
+}
+
+fn create_df_schema_for_ts_column(name: &str, cdt: ConcreteDataType) -> Result<DFSchema, Error> {
+    let arrow_schema = Arc::new(arrow_schema::Schema::new(vec![arrow_schema::Field::new(
+        name,
+        cdt.as_arrow_type(),
+        false,
+    )]));
+
+    let df_schema = DFSchema::from_field_specific_qualified_schema(
+        vec![Some(TableReference::bare("TimeIndexOnlyTable"))],
+        &arrow_schema,
+    )
+    .with_context(|_e| DatafusionSnafu {
+        context: format!("Failed to create DFSchema from arrow schema {arrow_schema:?}"),
+    })?;
+
+    Ok(df_schema)
+}
+
+/// Convert `ColumnarValue` to `Vec<Option<Timestamp>>`
+fn columnar_to_ts_vector(columnar: &ColumnarValue) -> Result<Vec<Option<Timestamp>>, Error> {
+    let val = match columnar {
+        datafusion_expr::ColumnarValue::Array(array) => {
+            let ty = array.data_type();
+            let ty = ConcreteDataType::from_arrow_type(ty);
+            let time_unit = if let ConcreteDataType::Timestamp(ty) = ty {
+                ty.unit()
+            } else {
+                return UnexpectedSnafu {
+                    reason: format!("Non-timestamp type: {ty:?}"),
+                }
+                .fail();
+            };
+
+            match time_unit {
+                TimeUnit::Second => TimestampSecondVector::try_from_arrow_array(array.clone())
+                    .with_context(|_| DatatypesSnafu {
+                        extra: format!("Failed to create vector from arrow array {array:?}"),
+                    })?
+                    .iter_data()
+                    .map(|d| d.map(|d| d.0))
+                    .collect_vec(),
+                TimeUnit::Millisecond => {
+                    TimestampMillisecondVector::try_from_arrow_array(array.clone())
+                        .with_context(|_| DatatypesSnafu {
+                            extra: format!("Failed to create vector from arrow array {array:?}"),
+                        })?
+                        .iter_data()
+                        .map(|d| d.map(|d| d.0))
+                        .collect_vec()
+                }
+                TimeUnit::Microsecond => {
+                    TimestampMicrosecondVector::try_from_arrow_array(array.clone())
+                        .with_context(|_| DatatypesSnafu {
+                            extra: format!("Failed to create vector from arrow array {array:?}"),
+                        })?
+                        .iter_data()
+                        .map(|d| d.map(|d| d.0))
+                        .collect_vec()
+                }
+                TimeUnit::Nanosecond => {
+                    TimestampNanosecondVector::try_from_arrow_array(array.clone())
+                        .with_context(|_| DatatypesSnafu {
+                            extra: format!("Failed to create vector from arrow array {array:?}"),
+                        })?
+                        .iter_data()
+                        .map(|d| d.map(|d| d.0))
+                        .collect_vec()
+                }
+            }
+        }
+        datafusion_expr::ColumnarValue::Scalar(scalar) => {
+            let value = Value::try_from(scalar.clone()).with_context(|_| DatatypesSnafu {
+                extra: format!("Failed to convert scalar {scalar:?} to value"),
+            })?;
+            let ts = value.as_timestamp().context(UnexpectedSnafu {
+                reason: format!("Expect Timestamp, found {:?}", value),
+            })?;
+            vec![Some(ts)]
+        }
+    };
+    Ok(val)
+}
+
+/// Convert sql to datafusion logical plan
+pub async fn sql_to_df_plan(
+    query_ctx: QueryContextRef,
+    engine: QueryEngineRef,
+    sql: &str,
+    optimize: bool,
+) -> Result<LogicalPlan, Error> {
+    let stmt = QueryLanguageParser::parse_sql(sql, &query_ctx)
+        .map_err(BoxedError::new)
+        .context(ExternalSnafu)?;
+    let plan = engine
+        .planner()
+        .plan(&stmt, query_ctx)
+        .await
+        .map_err(BoxedError::new)
+        .context(ExternalSnafu)?;
+    let plan = if optimize {
+        apply_df_optimizer(plan).await?
+    } else {
+        plan
+    };
+    Ok(plan)
+}
+
+/// Return (the column name of time index column, the time window expr, the expected time unit of time index column, the expr's schema for evaluating the time window)
+async fn find_time_window_expr(
+    plan: &LogicalPlan,
+    catalog_man: CatalogManagerRef,
+    query_ctx: QueryContextRef,
+) -> Result<(String, Option<datafusion_expr::Expr>, TimeUnit, DFSchema), Error> {
+    // TODO(discord9): find the expr that do time window
+
+    let mut table_name = None;
+
+    // first find the table source in the logical plan
+    plan.apply(|plan| {
+        let LogicalPlan::TableScan(table_scan) = plan else {
+            return Ok(TreeNodeRecursion::Continue);
+        };
+        table_name = Some(table_scan.table_name.clone());
+        Ok(TreeNodeRecursion::Stop)
+    })
+    .with_context(|_| DatafusionSnafu {
+        context: format!("Can't find table source in plan {plan:?}"),
+    })?;
+    let Some(table_name) = table_name else {
+        UnexpectedSnafu {
+            reason: format!("Can't find table source in plan {plan:?}"),
+        }
+        .fail()?
+    };
+
+    let current_schema = query_ctx.current_schema();
+
+    let catalog_name = table_name.catalog().unwrap_or(query_ctx.current_catalog());
+    let schema_name = table_name.schema().unwrap_or(&current_schema);
+    let table_name = table_name.table();
+
+    let Some(table_ref) = catalog_man
+        .table(catalog_name, schema_name, table_name, Some(&query_ctx))
+        .await
+        .map_err(BoxedError::new)
+        .context(ExternalSnafu)?
+    else {
+        UnexpectedSnafu {
+            reason: format!(
+                "Can't find table {table_name:?} in catalog {catalog_name:?}/{schema_name:?}"
+            ),
+        }
+        .fail()?
+    };
+
+    let schema = &table_ref.table_info().meta.schema;
+
+    let ts_index = schema.timestamp_column().context(UnexpectedSnafu {
+        reason: format!("Can't find timestamp column in table {table_name:?}"),
+    })?;
+
+    let ts_col_name = ts_index.name.clone();
+
+    let expected_time_unit = ts_index.data_type.as_timestamp().with_context(|| UnexpectedSnafu {
+        reason: format!(
+            "Expected timestamp column {ts_col_name:?} in table {table_name:?} to be timestamp, but got {ts_index:?}"
+        ),
+    })?.unit();
+
+    let arrow_schema = Arc::new(arrow_schema::Schema::new(vec![arrow_schema::Field::new(
+        ts_col_name.clone(),
+        ts_index.data_type.as_arrow_type(),
+        false,
+    )]));
+
+    let df_schema = DFSchema::from_field_specific_qualified_schema(
+        vec![Some(TableReference::bare(table_name))],
+        &arrow_schema,
+    )
+    .with_context(|_e| DatafusionSnafu {
+        context: format!("Failed to create DFSchema from arrow schema {arrow_schema:?}"),
+    })?;
+
+    // find the time window expr which refers to the time index column
+    let mut aggr_expr = None;
+    let mut time_window_expr: Option<Expr> = None;
+
+    let find_inner_aggr_expr = |plan: &LogicalPlan| {
+        if let LogicalPlan::Aggregate(aggregate) = plan {
+            aggr_expr = Some(aggregate.clone());
+        };
+
+        Ok(TreeNodeRecursion::Continue)
+    };
+    plan.apply(find_inner_aggr_expr)
+        .with_context(|_| DatafusionSnafu {
+            context: format!("Can't find aggr expr in plan {plan:?}"),
+        })?;
+
+    if let Some(aggregate) = aggr_expr {
+        for group_expr in &aggregate.group_expr {
+            let refs = group_expr.column_refs();
+            if refs.len() != 1 {
+                continue;
+            }
+            let ref_col = refs.iter().next().unwrap();
+
+            let index = aggregate.input.schema().maybe_index_of_column(ref_col);
+            let Some(index) = index else {
+                continue;
+            };
+            let field = aggregate.input.schema().field(index);
+
+            let is_time_index = field.metadata().get(TIME_INDEX_KEY) == Some(&"true".to_string());
+
+            if is_time_index {
+                let rewrite_column = group_expr.clone();
+                let rewritten = rewrite_column
+                    .rewrite(&mut RewriteColumn {
+                        table_name: table_name.to_string(),
+                    })
+                    .with_context(|_| DatafusionSnafu {
+                        context: format!("Rewrite expr failed, expr={:?}", group_expr),
+                    })?
+                    .data;
+                struct RewriteColumn {
+                    table_name: String,
+                }
+
+                impl TreeNodeRewriter for RewriteColumn {
+                    type Node = Expr;
+                    fn f_down(&mut self, node: Self::Node) -> DfResult<Transformed<Self::Node>> {
+                        let Expr::Column(mut column) = node else {
+                            return Ok(Transformed::no(node));
+                        };
+
+                        column.relation = Some(TableReference::bare(self.table_name.clone()));
+
+                        Ok(Transformed::yes(Expr::Column(column)))
+                    }
+                }
+
+                time_window_expr = Some(rewritten);
+                break;
+            }
+        }
+        Ok((ts_col_name, time_window_expr, expected_time_unit, df_schema))
+    } else {
+        // can't found time window expr, return None
+        Ok((ts_col_name, None, expected_time_unit, df_schema))
+    }
+}
+
+/// Find nearest lower bound for time `current` in given `plan` for the time window expr.
+/// i.e. for time window expr being `date_bin(INTERVAL '5 minutes', ts) as time_window` and `current="2021-07-01 00:01:01.000"`,
+/// return `Some("2021-07-01 00:00:00.000")`
+/// if `plan` doesn't contain a `TIME INDEX` column, return `None`
+///
+/// Time window expr is a expr that:
+/// 1. ref only to a time index column
+/// 2. is monotonic increasing
+/// 3. show up in GROUP BY clause
+///
+/// note this plan should only contain one TableScan
+pub async fn find_plan_time_window_bound(
+    plan: &LogicalPlan,
+    current: Timestamp,
+    query_ctx: QueryContextRef,
+    engine: QueryEngineRef,
+) -> Result<(String, Option<Timestamp>, Option<Timestamp>), Error> {
+    // TODO(discord9): find the expr that do time window
+    let catalog_man = engine.engine_state().catalog_manager();
+
+    let (ts_col_name, time_window_expr, expected_time_unit, df_schema) =
+        find_time_window_expr(plan, catalog_man.clone(), query_ctx).await?;
+    // cast current to ts_index's type
+    let new_current = current
+        .convert_to(expected_time_unit)
+        .with_context(|| UnexpectedSnafu {
+            reason: format!("Failed to cast current timestamp {current:?} to {expected_time_unit}"),
+        })?;
+
+    // if no time_window_expr is found, return None
+    if let Some(time_window_expr) = time_window_expr {
+        let lower_bound =
+            find_expr_time_window_lower_bound(&time_window_expr, &df_schema, new_current)?;
+        let upper_bound =
+            find_expr_time_window_upper_bound(&time_window_expr, &df_schema, new_current)?;
+        Ok((ts_col_name, lower_bound, upper_bound))
+    } else {
+        Ok((ts_col_name, None, None))
+    }
+}
+
+/// Find the lower bound of time window in given `expr` and `current` timestamp.
+///
+/// i.e. for `current="2021-07-01 00:01:01.000"` and `expr=date_bin(INTERVAL '5 minutes', ts) as time_window` and `ts_col=ts`,
+/// return `Some("2021-07-01 00:00:00.000")` since it's the lower bound
+/// return `Some("2021-07-01 00:00:00.000")` since it's the lower bound
+/// of current time window given the current timestamp
+///
+/// if return None, meaning this time window have no lower bound
+fn find_expr_time_window_lower_bound(
+    expr: &Expr,
+    df_schema: &DFSchema,
+    current: Timestamp,
+) -> Result<Option<Timestamp>, Error> {
+    let phy_planner = DefaultPhysicalPlanner::default();
+
+    let phy_expr: PhysicalExprRef = phy_planner
+        .create_physical_expr(expr, df_schema, &SessionContext::new().state())
+        .with_context(|_e| DatafusionSnafu {
+            context: format!(
+                "Failed to create physical expression from {expr:?} using {df_schema:?}"
+            ),
+        })?;
+
+    let cur_time_window = eval_ts_to_ts(&phy_expr, df_schema, current)?;
+    let input_time_unit = cur_time_window.unit();
+    Ok(cur_time_window.convert_to(input_time_unit))
+}
+
+/// Find the upper bound for time window expression
+fn find_expr_time_window_upper_bound(
+    expr: &Expr,
+    df_schema: &DFSchema,
+    current: Timestamp,
+) -> Result<Option<Timestamp>, Error> {
+    use std::cmp::Ordering;
+
+    let phy_planner = DefaultPhysicalPlanner::default();
+
+    let phy_expr: PhysicalExprRef = phy_planner
+        .create_physical_expr(expr, df_schema, &SessionContext::new().state())
+        .with_context(|_e| DatafusionSnafu {
+            context: format!(
+                "Failed to create physical expression from {expr:?} using {df_schema:?}"
+            ),
+        })?;
+
+    let cur_time_window = eval_ts_to_ts(&phy_expr, df_schema, current)?;
+
+    // search to find the lower bound
+    let mut offset: i64 = 1;
+    let mut lower_bound = Some(current);
+    let upper_bound;
+    // first expontial probe to found a range for binary search
+    loop {
+        let Some(next_val) = current.value().checked_add(offset) else {
+            // no upper bound if overflow
+            return Ok(None);
+        };
+
+        let next_time_probe = common_time::Timestamp::new(next_val, current.unit());
+
+        let next_time_window = eval_ts_to_ts(&phy_expr, df_schema, next_time_probe)?;
+
+        match next_time_window.cmp(&cur_time_window) {
+            Ordering::Less => {UnexpectedSnafu {
+                reason: format!(
+                    "Unsupported time window expression, expect monotonic increasing for time window expression {expr:?}"
+                ),
+            }
+            .fail()?
+            }
+            Ordering::Equal => {
+                lower_bound = Some(next_time_probe);
+            }
+            Ordering::Greater => {
+                upper_bound = Some(next_time_probe);
+                break
+            }
+        }
+
+        let Some(new_offset) = offset.checked_mul(2) else {
+            // no upper bound if overflow
+            return Ok(None);
+        };
+        offset = new_offset;
+    }
+
+    // binary search for the exact upper bound
+
+    ensure!(lower_bound.map(|v|v.unit())==upper_bound.map(|v|v.unit()), UnexpectedSnafu{
+        reason: format!(" unit mismatch for time window expression {expr:?}, found {lower_bound:?} and {upper_bound:?}"),
+    });
+
+    let output_unit = upper_bound
+        .context(UnexpectedSnafu {
+            reason: "should have lower bound",
+        })?
+        .unit();
+
+    let mut low = lower_bound
+        .context(UnexpectedSnafu {
+            reason: "should have lower bound",
+        })?
+        .value();
+    let mut high = upper_bound
+        .context(UnexpectedSnafu {
+            reason: "should have upper bound",
+        })?
+        .value();
+    while low < high {
+        let mid = (low + high) / 2;
+        let mid_probe = common_time::Timestamp::new(mid, output_unit);
+        let mid_time_window = eval_ts_to_ts(&phy_expr, df_schema, mid_probe)?;
+
+        match mid_time_window.cmp(&cur_time_window) {
+            Ordering::Less => UnexpectedSnafu {
+                reason: format!("Binary search failed for time window expression {expr:?}"),
+            }
+            .fail()?,
+            Ordering::Equal => low = mid + 1,
+            Ordering::Greater => high = mid,
+        }
+    }
+
+    let final_upper_bound_for_time_window = common_time::Timestamp::new(high, output_unit);
+
+    Ok(Some(final_upper_bound_for_time_window))
+}
+
+fn eval_ts_to_ts(
+    phy: &PhysicalExprRef,
+    df_schema: &DFSchema,
+    input_value: Timestamp,
+) -> Result<Timestamp, Error> {
+    let schema_ty = df_schema.field(0).data_type();
+    let schema_cdt = ConcreteDataType::from_arrow_type(schema_ty);
+    let schema_unit = if let ConcreteDataType::Timestamp(ts) = schema_cdt {
+        ts.unit()
+    } else {
+        return UnexpectedSnafu {
+            reason: format!("Expect Timestamp, found {:?}", schema_cdt),
+        }
+        .fail();
+    };
+    let input_value = input_value
+        .convert_to(schema_unit)
+        .with_context(|| UnexpectedSnafu {
+            reason: format!("Failed to convert timestamp {input_value:?} to {schema_unit}"),
+        })?;
+    let ts_vector = match schema_unit {
+        TimeUnit::Second => {
+            TimestampSecondVector::from_vec(vec![input_value.value()]).to_arrow_array()
+        }
+        TimeUnit::Millisecond => {
+            TimestampMillisecondVector::from_vec(vec![input_value.value()]).to_arrow_array()
+        }
+        TimeUnit::Microsecond => {
+            TimestampMicrosecondVector::from_vec(vec![input_value.value()]).to_arrow_array()
+        }
+        TimeUnit::Nanosecond => {
+            TimestampNanosecondVector::from_vec(vec![input_value.value()]).to_arrow_array()
+        }
+    };
+
+    let rb = DfRecordBatch::try_new(df_schema.inner().clone(), vec![ts_vector.clone()])
+        .with_context(|_| ArrowSnafu {
+            context: format!("Failed to create record batch from {df_schema:?} and {ts_vector:?}"),
+        })?;
+
+    let eval_res = phy.evaluate(&rb).with_context(|_| DatafusionSnafu {
+        context: format!("Failed to evaluate physical expression {phy:?} on {rb:?}"),
+    })?;
+
+    if let Some(Some(ts)) = columnar_to_ts_vector(&eval_res)?.first() {
+        Ok(*ts)
+    } else {
+        UnexpectedSnafu {
+            reason: format!(
+                "Expected timestamp in expression {phy:?} but got {:?}",
+                eval_res
+            ),
+        }
+        .fail()?
+    }
+}
+
+// TODO(discord9): a method to found out the precise time window
+
+/// Find out the `Filter` Node corresponding to outermost `WHERE` and add a new filter expr to it
+#[derive(Debug)]
+pub struct AddFilterRewriter {
+    extra_filter: Expr,
+    is_rewritten: bool,
+}
+
+impl AddFilterRewriter {
+    fn new(filter: Expr) -> Self {
+        Self {
+            extra_filter: filter,
+            is_rewritten: false,
+        }
+    }
+}
+
+impl TreeNodeRewriter for AddFilterRewriter {
+    type Node = LogicalPlan;
+    fn f_up(&mut self, node: Self::Node) -> DfResult<Transformed<Self::Node>> {
+        if self.is_rewritten {
+            return Ok(Transformed::no(node));
+        }
+        match node {
+            LogicalPlan::Filter(mut filter) if !filter.having => {
+                filter.predicate = filter.predicate.and(self.extra_filter.clone());
+                self.is_rewritten = true;
+                Ok(Transformed::yes(LogicalPlan::Filter(filter)))
+            }
+            LogicalPlan::TableScan(_) => {
+                // add a new filter
+                let filter =
+                    datafusion_expr::Filter::try_new(self.extra_filter.clone(), Arc::new(node))?;
+                self.is_rewritten = true;
+                Ok(Transformed::yes(LogicalPlan::Filter(filter)))
+            }
+            _ => Ok(Transformed::no(node)),
+        }
+    }
+}
+
+fn df_plan_to_sql(plan: &LogicalPlan) -> Result<String, Error> {
+    /// A dialect that forces all identifiers to be quoted
+    struct ForceQuoteIdentifiers;
+    impl datafusion::sql::unparser::dialect::Dialect for ForceQuoteIdentifiers {
+        fn identifier_quote_style(&self, identifier: &str) -> Option<char> {
+            if identifier.to_lowercase() != identifier {
+                Some('"')
+            } else {
+                None
+            }
+        }
+    }
+    let unparser = Unparser::new(&ForceQuoteIdentifiers);
+    // first make all column qualified
+    let sql = unparser
+        .plan_to_sql(plan)
+        .with_context(|_e| DatafusionSnafu {
+            context: format!("Failed to unparse logical plan {plan:?}"),
+        })?;
+    Ok(sql.to_string())
+}
+
+#[cfg(test)]
+mod test {
+    use datafusion_common::tree_node::TreeNode;
+    use pretty_assertions::assert_eq;
+    use session::context::QueryContext;
+
+    use super::{sql_to_df_plan, *};
+    use crate::recording_rules::{df_plan_to_sql, AddFilterRewriter};
+    use crate::test_utils::create_test_query_engine;
+
+    #[tokio::test]
+    async fn test_sql_plan_convert() {
+        let query_engine = create_test_query_engine();
+        let ctx = QueryContext::arc();
+        let old = r#"SELECT "NUMBER" FROM "UPPERCASE_NUMBERS_WITH_TS""#;
+        let new = sql_to_df_plan(ctx.clone(), query_engine.clone(), old, false)
+            .await
+            .unwrap();
+        let new_sql = df_plan_to_sql(&new).unwrap();
+
+        assert_eq!(
+            r#"SELECT "UPPERCASE_NUMBERS_WITH_TS"."NUMBER" FROM "UPPERCASE_NUMBERS_WITH_TS""#,
+            new_sql
+        );
+    }
+
+    #[tokio::test]
+    async fn test_add_filter() {
+        let testcases = vec![
+            (
+                "SELECT number FROM numbers_with_ts GROUP BY number","SELECT numbers_with_ts.number FROM numbers_with_ts WHERE (number > 4) GROUP BY numbers_with_ts.number"
+            ),
+            (
+                "SELECT number FROM numbers_with_ts WHERE number < 2 OR number >10",
+                "SELECT numbers_with_ts.number FROM numbers_with_ts WHERE ((numbers_with_ts.number < 2) OR (numbers_with_ts.number > 10)) AND (number > 4)"
+            ),
+            (
+                "SELECT date_bin('5 minutes', ts) as time_window FROM numbers_with_ts GROUP BY time_window",
+                "SELECT date_bin('5 minutes', numbers_with_ts.ts) AS time_window FROM numbers_with_ts WHERE (number > 4) GROUP BY date_bin('5 minutes', numbers_with_ts.ts)"
+            )
+        ];
+        use datafusion_expr::{col, lit};
+        let query_engine = create_test_query_engine();
+        let ctx = QueryContext::arc();
+
+        for (before, after) in testcases {
+            let sql = before;
+            let plan = sql_to_df_plan(ctx.clone(), query_engine.clone(), sql, false)
+                .await
+                .unwrap();
+
+            let mut add_filter = AddFilterRewriter::new(col("number").gt(lit(4u32)));
+            let plan = plan.rewrite(&mut add_filter).unwrap().data;
+            let new_sql = df_plan_to_sql(&plan).unwrap();
+            assert_eq!(after, new_sql);
+        }
+    }
+
+    #[tokio::test]
+    async fn test_plan_time_window_lower_bound() {
+        use datafusion_expr::{col, lit};
+        let query_engine = create_test_query_engine();
+        let ctx = QueryContext::arc();
+
+        let testcases = [
+            // same alias is not same column
+            (
+                "SELECT arrow_cast(date_bin(INTERVAL '1 MINS', numbers_with_ts.ts), 'Timestamp(Second, None)') AS ts FROM numbers_with_ts GROUP BY ts;",
+                Timestamp::new(1740394109, TimeUnit::Second),
+                (
+                    "ts".to_string(),
+                    Some(Timestamp::new(1740394109000, TimeUnit::Millisecond)),
+                    Some(Timestamp::new(1740394109001, TimeUnit::Millisecond)),
+                ),
+                r#"SELECT arrow_cast(date_bin(INTERVAL '1 MINS', numbers_with_ts.ts), 'Timestamp(Second, None)') AS ts FROM numbers_with_ts WHERE ((ts >= CAST('2025-02-24 10:48:29' AS TIMESTAMP)) AND (ts <= CAST('2025-02-24 10:48:29.001' AS TIMESTAMP))) GROUP BY numbers_with_ts.ts"#
+            ),
+            // complex time window index
+            (
+                "SELECT arrow_cast(date_bin(INTERVAL '1 MINS', numbers_with_ts.ts), 'Timestamp(Second, None)') AS time_window FROM numbers_with_ts GROUP BY time_window;",
+                Timestamp::new(1740394109, TimeUnit::Second),
+                (
+                    "ts".to_string(),
+                    Some(Timestamp::new(1740394080, TimeUnit::Second)),
+                    Some(Timestamp::new(1740394140, TimeUnit::Second)),
+                ),
+                "SELECT arrow_cast(date_bin(INTERVAL '1 MINS', numbers_with_ts.ts), 'Timestamp(Second, None)') AS time_window FROM numbers_with_ts WHERE ((ts >= CAST('2025-02-24 10:48:00' AS TIMESTAMP)) AND (ts <= CAST('2025-02-24 10:49:00' AS TIMESTAMP))) GROUP BY arrow_cast(date_bin(INTERVAL '1 MINS', numbers_with_ts.ts), 'Timestamp(Second, None)')"
+            ),
+            // no time index
+            (
+                "SELECT date_bin('5 minutes', ts) FROM numbers_with_ts;",
+                Timestamp::new(23, TimeUnit::Millisecond),
+                ("ts".to_string(), None, None),
+                "SELECT date_bin('5 minutes', ts) FROM numbers_with_ts;"
+            ),
+            // time index
+            (
+                "SELECT date_bin('5 minutes', ts) as time_window FROM numbers_with_ts GROUP BY time_window;",
+                Timestamp::new(23, TimeUnit::Nanosecond),
+                (
+                    "ts".to_string(),
+                    Some(Timestamp::new(0, TimeUnit::Millisecond)),
+                    Some(Timestamp::new(300000, TimeUnit::Millisecond)),
+                ),
+                "SELECT date_bin('5 minutes', numbers_with_ts.ts) AS time_window FROM numbers_with_ts WHERE ((ts >= CAST('1970-01-01 00:00:00' AS TIMESTAMP)) AND (ts <= CAST('1970-01-01 00:05:00' AS TIMESTAMP))) GROUP BY date_bin('5 minutes', numbers_with_ts.ts)"
+            ),
+            // on spot
+            (
+                "SELECT date_bin('5 minutes', ts) as time_window FROM numbers_with_ts GROUP BY time_window;",
+                Timestamp::new(0, TimeUnit::Nanosecond),
+                (
+                    "ts".to_string(),
+                    Some(Timestamp::new(0, TimeUnit::Millisecond)),
+                    Some(Timestamp::new(300000, TimeUnit::Millisecond)),
+                ),
+                "SELECT date_bin('5 minutes', numbers_with_ts.ts) AS time_window FROM numbers_with_ts WHERE ((ts >= CAST('1970-01-01 00:00:00' AS TIMESTAMP)) AND (ts <= CAST('1970-01-01 00:05:00' AS TIMESTAMP))) GROUP BY date_bin('5 minutes', numbers_with_ts.ts)"
+            ),
+            // different time unit
+            (
+                "SELECT date_bin('5 minutes', ts) as time_window FROM numbers_with_ts GROUP BY time_window;",
+                Timestamp::new(23_000_000, TimeUnit::Nanosecond),
+                (
+                    "ts".to_string(),
+                    Some(Timestamp::new(0, TimeUnit::Millisecond)),
+                    Some(Timestamp::new(300000, TimeUnit::Millisecond)),
+                ),
+                "SELECT date_bin('5 minutes', numbers_with_ts.ts) AS time_window FROM numbers_with_ts WHERE ((ts >= CAST('1970-01-01 00:00:00' AS TIMESTAMP)) AND (ts <= CAST('1970-01-01 00:05:00' AS TIMESTAMP))) GROUP BY date_bin('5 minutes', numbers_with_ts.ts)"
+            ),
+            // time index with other fields
+            (
+                "SELECT sum(number) as sum_up, date_bin('5 minutes', ts) as time_window FROM numbers_with_ts GROUP BY time_window;",
+                Timestamp::new(23, TimeUnit::Millisecond),
+                (
+                    "ts".to_string(),
+                    Some(Timestamp::new(0, TimeUnit::Millisecond)),
+                    Some(Timestamp::new(300000, TimeUnit::Millisecond)),
+                ),
+                "SELECT sum(numbers_with_ts.number) AS sum_up, date_bin('5 minutes', numbers_with_ts.ts) AS time_window FROM numbers_with_ts WHERE ((ts >= CAST('1970-01-01 00:00:00' AS TIMESTAMP)) AND (ts <= CAST('1970-01-01 00:05:00' AS TIMESTAMP))) GROUP BY date_bin('5 minutes', numbers_with_ts.ts)"
+            ),
+            // time index with other pks
+            (
+                "SELECT number, date_bin('5 minutes', ts) as time_window FROM numbers_with_ts GROUP BY time_window, number;",
+                Timestamp::new(23, TimeUnit::Millisecond),
+                (
+                    "ts".to_string(),
+                    Some(Timestamp::new(0, TimeUnit::Millisecond)),
+                    Some(Timestamp::new(300000, TimeUnit::Millisecond)),
+                ),
+                "SELECT numbers_with_ts.number, date_bin('5 minutes', numbers_with_ts.ts) AS time_window FROM numbers_with_ts WHERE ((ts >= CAST('1970-01-01 00:00:00' AS TIMESTAMP)) AND (ts <= CAST('1970-01-01 00:05:00' AS TIMESTAMP))) GROUP BY date_bin('5 minutes', numbers_with_ts.ts), numbers_with_ts.number"
+            ),
+            // subquery
+            (
+                "SELECT number, time_window FROM (SELECT number, date_bin('5 minutes', ts) as time_window FROM numbers_with_ts GROUP BY time_window, number);",
+                Timestamp::new(23, TimeUnit::Millisecond),
+                (
+                    "ts".to_string(),
+                    Some(Timestamp::new(0, TimeUnit::Millisecond)),
+                    Some(Timestamp::new(300000, TimeUnit::Millisecond)),
+                ),
+                "SELECT numbers_with_ts.number, time_window FROM (SELECT numbers_with_ts.number, date_bin('5 minutes', numbers_with_ts.ts) AS time_window FROM numbers_with_ts WHERE ((ts >= CAST('1970-01-01 00:00:00' AS TIMESTAMP)) AND (ts <= CAST('1970-01-01 00:05:00' AS TIMESTAMP))) GROUP BY date_bin('5 minutes', numbers_with_ts.ts), numbers_with_ts.number)"
+            ),
+            // cte
+            (
+                "with cte as (select number, date_bin('5 minutes', ts) as time_window from numbers_with_ts GROUP BY time_window, number) select number, time_window from cte;",
+                Timestamp::new(23, TimeUnit::Millisecond),
+                (
+                    "ts".to_string(),
+                    Some(Timestamp::new(0, TimeUnit::Millisecond)),
+                    Some(Timestamp::new(300000, TimeUnit::Millisecond)),
+                ),
+                "SELECT cte.number, cte.time_window FROM (SELECT numbers_with_ts.number, date_bin('5 minutes', numbers_with_ts.ts) AS time_window FROM numbers_with_ts WHERE ((ts >= CAST('1970-01-01 00:00:00' AS TIMESTAMP)) AND (ts <= CAST('1970-01-01 00:05:00' AS TIMESTAMP))) GROUP BY date_bin('5 minutes', numbers_with_ts.ts), numbers_with_ts.number) AS cte"
+            ),
+            // complex subquery without alias
+            (
+                "SELECT sum(number), number, date_bin('5 minutes', ts) as time_window, bucket_name FROM (SELECT number, ts, case when number < 5 THEN 'bucket_0_5' when number >= 5 THEN 'bucket_5_inf' END as bucket_name FROM numbers_with_ts) GROUP BY number, time_window, bucket_name;",
+                Timestamp::new(23, TimeUnit::Millisecond),
+                (
+                    "ts".to_string(),
+                    Some(Timestamp::new(0, TimeUnit::Millisecond)),
+                    Some(Timestamp::new(300000, TimeUnit::Millisecond)),
+                ),
+                "SELECT sum(numbers_with_ts.number), numbers_with_ts.number, date_bin('5 minutes', numbers_with_ts.ts) AS time_window, bucket_name FROM (SELECT numbers_with_ts.number, numbers_with_ts.ts, CASE WHEN (numbers_with_ts.number < 5) THEN 'bucket_0_5' WHEN (numbers_with_ts.number >= 5) THEN 'bucket_5_inf' END AS bucket_name FROM numbers_with_ts WHERE ((ts >= CAST('1970-01-01 00:00:00' AS TIMESTAMP)) AND (ts <= CAST('1970-01-01 00:05:00' AS TIMESTAMP)))) GROUP BY numbers_with_ts.number, date_bin('5 minutes', numbers_with_ts.ts), bucket_name"
+            ),
+            // complex subquery alias
+            (
+                "SELECT sum(number), number, date_bin('5 minutes', ts) as time_window, bucket_name FROM (SELECT number, ts, case when number < 5 THEN 'bucket_0_5' when number >= 5 THEN 'bucket_5_inf' END as bucket_name FROM numbers_with_ts) as cte GROUP BY number, time_window, bucket_name;",
+                Timestamp::new(23, TimeUnit::Millisecond),
+                (
+                    "ts".to_string(),
+                    Some(Timestamp::new(0, TimeUnit::Millisecond)),
+                    Some(Timestamp::new(300000, TimeUnit::Millisecond)),
+                ),
+                "SELECT sum(cte.number), cte.number, date_bin('5 minutes', cte.ts) AS time_window, cte.bucket_name FROM (SELECT numbers_with_ts.number, numbers_with_ts.ts, CASE WHEN (numbers_with_ts.number < 5) THEN 'bucket_0_5' WHEN (numbers_with_ts.number >= 5) THEN 'bucket_5_inf' END AS bucket_name FROM numbers_with_ts WHERE ((ts >= CAST('1970-01-01 00:00:00' AS TIMESTAMP)) AND (ts <= CAST('1970-01-01 00:05:00' AS TIMESTAMP)))) AS cte GROUP BY cte.number, date_bin('5 minutes', cte.ts), cte.bucket_name"
+            ),
+        ];
+
+        for (sql, current, expected, expected_unparsed) in testcases {
+            let plan = sql_to_df_plan(ctx.clone(), query_engine.clone(), sql, true)
+                .await
+                .unwrap();
+
+            let real =
+                find_plan_time_window_bound(&plan, current, ctx.clone(), query_engine.clone())
+                    .await
+                    .unwrap();
+            assert_eq!(expected, real);
+
+            let plan = sql_to_df_plan(ctx.clone(), query_engine.clone(), sql, false)
+                .await
+                .unwrap();
+            let (col_name, lower, upper) = real;
+            let new_sql = if lower.is_some() {
+                let to_df_literal = |value| {
+                    let value = Value::from(value);
+
+                    value.try_to_scalar_value(&value.data_type()).unwrap()
+                };
+                let lower = to_df_literal(lower.unwrap());
+                let upper = to_df_literal(upper.unwrap());
+                let expr = col(&col_name)
+                    .gt_eq(lit(lower))
+                    .and(col(&col_name).lt_eq(lit(upper)));
+                let mut add_filter = AddFilterRewriter::new(expr);
+                let plan = plan.rewrite(&mut add_filter).unwrap().data;
+                df_plan_to_sql(&plan).unwrap()
+            } else {
+                sql.to_string()
+            };
+            assert_eq!(expected_unparsed, new_sql);
+        }
+    }
+}
--- a/src/flow/src/recording_rules/engine.rs
+++ b/src/flow/src/recording_rules/engine.rs
@@ -0,0 +1,815 @@
+// 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, HashMap, HashSet};
+use std::sync::Arc;
+use std::time::{Duration, SystemTime, UNIX_EPOCH};
+
+use api::v1::flow::FlowResponse;
+use common_error::ext::BoxedError;
+use common_meta::ddl::create_flow::FlowType;
+use common_meta::key::flow::FlowMetadataManagerRef;
+use common_meta::key::table_info::TableInfoManager;
+use common_meta::key::TableMetadataManagerRef;
+use common_telemetry::tracing::warn;
+use common_telemetry::{debug, info};
+use common_time::Timestamp;
+use datafusion::sql::unparser::expr_to_sql;
+use datafusion_common::tree_node::TreeNode;
+use datatypes::value::Value;
+use query::QueryEngineRef;
+use session::context::QueryContextRef;
+use snafu::{ensure, OptionExt, ResultExt};
+use store_api::storage::RegionId;
+use table::metadata::TableId;
+use tokio::sync::oneshot::error::TryRecvError;
+use tokio::sync::{oneshot, RwLock};
+use tokio::time::Instant;
+
+use super::frontend_client::FrontendClient;
+use super::{df_plan_to_sql, AddFilterRewriter, TimeWindowExpr};
+use crate::adapter::{CreateFlowArgs, FlowId, TableName};
+use crate::error::{
+    DatafusionSnafu, DatatypesSnafu, ExternalSnafu, FlowAlreadyExistSnafu, InternalSnafu,
+    TimeSnafu, UnexpectedSnafu,
+};
+use crate::metrics::{METRIC_FLOW_RULE_ENGINE_QUERY_TIME, METRIC_FLOW_RULE_ENGINE_SLOW_QUERY};
+use crate::recording_rules::{find_time_window_expr, sql_to_df_plan};
+use crate::Error;
+
+/// TODO(discord9): make those constants configurable
+/// The default rule engine query timeout is 10 minutes
+pub const DEFAULT_RULE_ENGINE_QUERY_TIMEOUT: Duration = Duration::from_secs(10 * 60);
+
+/// will output a warn log for any query that runs for more that 1 minutes, and also every 1 minutes when that query is still running
+pub const SLOW_QUERY_THRESHOLD: Duration = Duration::from_secs(60);
+
+/// TODO(discord9): determine how to configure refresh rate
+pub struct RecordingRuleEngine {
+    tasks: RwLock<BTreeMap<FlowId, RecordingRuleTask>>,
+    shutdown_txs: RwLock<BTreeMap<FlowId, oneshot::Sender<()>>>,
+    frontend_client: Arc<FrontendClient>,
+    flow_metadata_manager: FlowMetadataManagerRef,
+    table_meta: TableMetadataManagerRef,
+    engine: QueryEngineRef,
+}
+
+impl RecordingRuleEngine {
+    pub fn new(
+        frontend_client: Arc<FrontendClient>,
+        engine: QueryEngineRef,
+        flow_metadata_manager: FlowMetadataManagerRef,
+        table_meta: TableMetadataManagerRef,
+    ) -> Self {
+        Self {
+            tasks: Default::default(),
+            shutdown_txs: Default::default(),
+            frontend_client,
+            flow_metadata_manager,
+            table_meta,
+            engine,
+        }
+    }
+
+    pub async fn handle_inserts(
+        &self,
+        request: api::v1::region::InsertRequests,
+    ) -> Result<FlowResponse, Error> {
+        let table_info_mgr = self.table_meta.table_info_manager();
+        let mut group_by_table_name: HashMap<TableName, Vec<api::v1::Rows>> = HashMap::new();
+        for r in request.requests {
+            let tid = RegionId::from(r.region_id).table_id();
+            let name = get_table_name(table_info_mgr, &tid).await?;
+            let entry = group_by_table_name.entry(name).or_default();
+            if let Some(rows) = r.rows {
+                entry.push(rows);
+            }
+        }
+
+        for (_flow_id, task) in self.tasks.read().await.iter() {
+            let src_table_names = &task.source_table_names;
+
+            for src_table_name in src_table_names {
+                if let Some(entry) = group_by_table_name.get(src_table_name) {
+                    let Some(expr) = &task.time_window_expr else {
+                        continue;
+                    };
+                    let involved_time_windows = expr.handle_rows(entry.clone()).await?;
+                    let mut state = task.state.write().await;
+                    state
+                        .dirty_time_windows
+                        .add_lower_bounds(involved_time_windows.into_iter());
+                }
+            }
+        }
+
+        Ok(Default::default())
+    }
+}
+
+async fn get_table_name(zelf: &TableInfoManager, table_id: &TableId) -> Result<TableName, Error> {
+    zelf.get(*table_id)
+        .await
+        .map_err(BoxedError::new)
+        .context(ExternalSnafu)?
+        .with_context(|| UnexpectedSnafu {
+            reason: format!("Table id = {:?}, couldn't found table name", table_id),
+        })
+        .map(|name| name.table_name())
+        .map(|name| [name.catalog_name, name.schema_name, name.table_name])
+}
+
+const MIN_REFRESH_DURATION: Duration = Duration::new(5, 0);
+
+impl RecordingRuleEngine {
+    pub async fn create_flow(&self, args: CreateFlowArgs) -> Result<Option<FlowId>, Error> {
+        let CreateFlowArgs {
+            flow_id,
+            sink_table_name,
+            source_table_ids,
+            create_if_not_exists,
+            or_replace,
+            expire_after,
+            comment: _,
+            sql,
+            flow_options,
+            query_ctx,
+        } = args;
+
+        // or replace logic
+        {
+            let is_exist = self.tasks.read().await.contains_key(&flow_id);
+            match (create_if_not_exists, or_replace, is_exist) {
+                // if replace, ignore that old flow exists
+                (_, true, true) => {
+                    info!("Replacing flow with id={}", flow_id);
+                }
+                (false, false, true) => FlowAlreadyExistSnafu { id: flow_id }.fail()?,
+                // already exists, and not replace, return None
+                (true, false, true) => {
+                    info!("Flow with id={} already exists, do nothing", flow_id);
+                    return Ok(None);
+                }
+
+                // continue as normal
+                (_, _, false) => (),
+            }
+        }
+
+        let flow_type = flow_options.get(FlowType::FLOW_TYPE_KEY);
+
+        ensure!(
+            flow_type == Some(&FlowType::RecordingRule.to_string()) || flow_type.is_none(),
+            UnexpectedSnafu {
+                reason: format!("Flow type is not RecordingRule nor None, got {flow_type:?}")
+            }
+        );
+
+        let Some(query_ctx) = query_ctx else {
+            UnexpectedSnafu {
+                reason: "Query context is None".to_string(),
+            }
+            .fail()?
+        };
+        let query_ctx = Arc::new(query_ctx);
+        let mut source_table_names = Vec::new();
+        for src_id in source_table_ids {
+            let table_name = self
+                .table_meta
+                .table_info_manager()
+                .get(src_id)
+                .await
+                .map_err(BoxedError::new)
+                .context(ExternalSnafu)?
+                .with_context(|| UnexpectedSnafu {
+                    reason: format!("Table id = {:?}, couldn't found table name", src_id),
+                })
+                .map(|name| name.table_name())
+                .map(|name| [name.catalog_name, name.schema_name, name.table_name])?;
+            source_table_names.push(table_name);
+        }
+
+        let (tx, rx) = oneshot::channel();
+
+        let plan = sql_to_df_plan(query_ctx.clone(), self.engine.clone(), &sql, true).await?;
+        let (column_name, time_window_expr, _, df_schema) = find_time_window_expr(
+            &plan,
+            self.engine.engine_state().catalog_manager().clone(),
+            query_ctx.clone(),
+        )
+        .await?;
+
+        let phy_expr = time_window_expr
+            .map(|expr| TimeWindowExpr::from_expr(&expr, &column_name, &df_schema))
+            .transpose()?;
+
+        info!("Flow id={}, found time window expr={:?}", flow_id, phy_expr);
+
+        let task = RecordingRuleTask::new(
+            flow_id,
+            &sql,
+            phy_expr,
+            expire_after,
+            sink_table_name,
+            source_table_names,
+            query_ctx,
+            rx,
+        );
+
+        let task_inner = task.clone();
+        let engine = self.engine.clone();
+        let frontend = self.frontend_client.clone();
+
+        // TODO(discord9): also save handle & use time wheel or what for better
+        let _handle = common_runtime::spawn_global(async move {
+            match task_inner.start_executing(engine, frontend).await {
+                Ok(()) => info!("Flow {} shutdown", task_inner.flow_id),
+                Err(err) => common_telemetry::error!(
+                    "Flow {} encounter unrecoverable error: {err:?}",
+                    task_inner.flow_id
+                ),
+            }
+        });
+
+        // TODO(discord9): deal with replace logic
+        let replaced_old_task_opt = self.tasks.write().await.insert(flow_id, task);
+        drop(replaced_old_task_opt);
+
+        self.shutdown_txs.write().await.insert(flow_id, tx);
+
+        Ok(Some(flow_id))
+    }
+
+    pub async fn remove_flow(&self, flow_id: FlowId) -> Result<(), Error> {
+        if self.tasks.write().await.remove(&flow_id).is_none() {
+            warn!("Flow {flow_id} not found in tasks")
+        }
+        let Some(tx) = self.shutdown_txs.write().await.remove(&flow_id) else {
+            UnexpectedSnafu {
+                reason: format!("Can't found shutdown tx for flow {flow_id}"),
+            }
+            .fail()?
+        };
+        if tx.send(()).is_err() {
+            warn!("Fail to shutdown flow {flow_id} due to receiver already dropped, maybe flow {flow_id} is already dropped?")
+        }
+        Ok(())
+    }
+}
+
+#[derive(Debug, Clone)]
+pub struct RecordingRuleTask {
+    pub flow_id: FlowId,
+    query: String,
+    pub time_window_expr: Option<TimeWindowExpr>,
+    /// in seconds
+    pub expire_after: Option<i64>,
+    sink_table_name: [String; 3],
+    source_table_names: HashSet<[String; 3]>,
+    state: Arc<RwLock<RecordingRuleState>>,
+}
+
+impl RecordingRuleTask {
+    #[allow(clippy::too_many_arguments)]
+    pub fn new(
+        flow_id: FlowId,
+        query: &str,
+        time_window_expr: Option<TimeWindowExpr>,
+        expire_after: Option<i64>,
+        sink_table_name: [String; 3],
+        source_table_names: Vec<[String; 3]>,
+        query_ctx: QueryContextRef,
+        shutdown_rx: oneshot::Receiver<()>,
+    ) -> Self {
+        Self {
+            flow_id,
+            query: query.to_string(),
+            time_window_expr,
+            expire_after,
+            sink_table_name,
+            source_table_names: source_table_names.into_iter().collect(),
+            state: Arc::new(RwLock::new(RecordingRuleState::new(query_ctx, shutdown_rx))),
+        }
+    }
+}
+impl RecordingRuleTask {
+    /// This should be called in a new tokio task
+    pub async fn start_executing(
+        &self,
+        engine: QueryEngineRef,
+        frontend_client: Arc<FrontendClient>,
+    ) -> Result<(), Error> {
+        // only first query don't need upper bound
+        let mut is_first = true;
+
+        loop {
+            // FIXME(discord9): test if need upper bound also works
+            let new_query = self.gen_query_with_time_window(engine.clone()).await?;
+
+            let insert_into = if let Some(new_query) = new_query {
+                format!(
+                    "INSERT INTO {}.{}.{} {}",
+                    self.sink_table_name[0],
+                    self.sink_table_name[1],
+                    self.sink_table_name[2],
+                    new_query
+                )
+            } else {
+                tokio::time::sleep(MIN_REFRESH_DURATION).await;
+                continue;
+            };
+
+            if is_first {
+                is_first = false;
+            }
+
+            let instant = Instant::now();
+            let flow_id = self.flow_id;
+            let db_client = frontend_client.get_database_client().await?;
+            let peer_addr = db_client.peer.addr;
+            debug!(
+                "Executing flow {flow_id}(expire_after={:?} secs) on {:?} with query {}",
+                self.expire_after, peer_addr, &insert_into
+            );
+
+            let timer = METRIC_FLOW_RULE_ENGINE_QUERY_TIME
+                .with_label_values(&[flow_id.to_string().as_str()])
+                .start_timer();
+
+            let res = db_client.database.sql(&insert_into).await;
+            drop(timer);
+
+            let elapsed = instant.elapsed();
+            if let Ok(res1) = &res {
+                debug!(
+                    "Flow {flow_id} executed, result: {res1:?}, elapsed: {:?}",
+                    elapsed
+                );
+            } else if let Err(res) = &res {
+                warn!(
+                    "Failed to execute Flow {flow_id} on frontend {}, result: {res:?}, elapsed: {:?} with query: {}",
+                    peer_addr, elapsed, &insert_into
+                );
+            }
+
+            // record slow query
+            if elapsed >= SLOW_QUERY_THRESHOLD {
+                warn!(
+                    "Flow {flow_id} on frontend {} executed for {:?} before complete, query: {}",
+                    peer_addr, elapsed, &insert_into
+                );
+                METRIC_FLOW_RULE_ENGINE_SLOW_QUERY
+                    .with_label_values(&[flow_id.to_string().as_str(), &insert_into, &peer_addr])
+                    .observe(elapsed.as_secs_f64());
+            }
+
+            self.state
+                .write()
+                .await
+                .after_query_exec(elapsed, res.is_ok());
+            // drop the result to free client-related resources
+            drop(res);
+
+            let sleep_until = {
+                let mut state = self.state.write().await;
+                match state.shutdown_rx.try_recv() {
+                    Ok(()) => break Ok(()),
+                    Err(TryRecvError::Closed) => {
+                        warn!("Unexpected shutdown flow {flow_id}, shutdown anyway");
+                        break Ok(());
+                    }
+                    Err(TryRecvError::Empty) => (),
+                }
+                state.get_next_start_query_time(None)
+            };
+            tokio::time::sleep_until(sleep_until).await;
+        }
+    }
+
+    /// will merge and use the first ten time window in query
+    async fn gen_query_with_time_window(
+        &self,
+        engine: QueryEngineRef,
+    ) -> Result<Option<String>, Error> {
+        let query_ctx = self.state.read().await.query_ctx.clone();
+        let start = SystemTime::now();
+        let since_the_epoch = start
+            .duration_since(UNIX_EPOCH)
+            .expect("Time went backwards");
+        let low_bound = self
+            .expire_after
+            .map(|e| since_the_epoch.as_secs() - e as u64)
+            .unwrap_or(u64::MIN);
+
+        let low_bound = Timestamp::new_second(low_bound as i64);
+
+        // TODO(discord9): use time window expr to get the precise expire lower bound
+        let expire_time_window_bound = self
+            .time_window_expr
+            .as_ref()
+            .map(|expr| expr.eval(low_bound))
+            .transpose()?;
+
+        let new_sql = {
+            let expr = {
+                match expire_time_window_bound {
+                    Some((Some(l), Some(u))) => {
+                        let window_size = u.sub(&l).with_context(|| UnexpectedSnafu {
+                            reason: format!("Can't get window size from {u:?} - {l:?}"),
+                        })?;
+                        let col_name = self
+                            .time_window_expr
+                            .as_ref()
+                            .map(|expr| expr.column_name.clone())
+                            .with_context(|| UnexpectedSnafu {
+                                reason: format!(
+                                    "Flow id={:?}, Failed to get column name from time window expr",
+                                    self.flow_id
+                                ),
+                            })?;
+
+                        self.state
+                            .write()
+                            .await
+                            .dirty_time_windows
+                            .gen_filter_exprs(&col_name, Some(l), window_size, self)?
+                    }
+                    _ => {
+                        debug!(
+                            "Flow id = {:?}, can't get window size: precise_lower_bound={expire_time_window_bound:?}, using the same query", self.flow_id
+                        );
+                        // since no time window lower/upper bound is found, just return the original query
+                        return Ok(Some(self.query.clone()));
+                    }
+                }
+            };
+
+            debug!(
+                "Flow id={:?}, Generated filter expr: {:?}",
+                self.flow_id,
+                expr.as_ref()
+                    .map(|expr| expr_to_sql(expr).with_context(|_| DatafusionSnafu {
+                        context: format!("Failed to generate filter expr from {expr:?}"),
+                    }))
+                    .transpose()?
+                    .map(|s| s.to_string())
+            );
+
+            let Some(expr) = expr else {
+                // no new data, hence no need to update
+                debug!("Flow id={:?}, no new data, not update", self.flow_id);
+                return Ok(None);
+            };
+
+            let mut add_filter = AddFilterRewriter::new(expr);
+            // make a not optimized plan for clearer unparse
+            let plan =
+                sql_to_df_plan(query_ctx.clone(), engine.clone(), &self.query, false).await?;
+            let plan = plan
+                .clone()
+                .rewrite(&mut add_filter)
+                .with_context(|_| DatafusionSnafu {
+                    context: format!("Failed to rewrite plan {plan:?}"),
+                })?
+                .data;
+            df_plan_to_sql(&plan)?
+        };
+
+        Ok(Some(new_sql))
+    }
+}
+
+#[derive(Debug)]
+pub struct RecordingRuleState {
+    query_ctx: QueryContextRef,
+    /// last query complete time
+    last_update_time: Instant,
+    /// last time query duration
+    last_query_duration: Duration,
+    /// Dirty Time windows need to be updated
+    /// mapping of `start -> end` and non-overlapping
+    dirty_time_windows: DirtyTimeWindows,
+    exec_state: ExecState,
+    shutdown_rx: oneshot::Receiver<()>,
+}
+
+#[derive(Debug, Clone, Default)]
+pub struct DirtyTimeWindows {
+    windows: BTreeMap<Timestamp, Option<Timestamp>>,
+}
+
+fn to_df_literal(value: Timestamp) -> Result<datafusion_common::ScalarValue, Error> {
+    let value = Value::from(value);
+    let value = value
+        .try_to_scalar_value(&value.data_type())
+        .with_context(|_| DatatypesSnafu {
+            extra: format!("Failed to convert to scalar value: {}", value),
+        })?;
+    Ok(value)
+}
+
+impl DirtyTimeWindows {
+    /// Time window merge distance
+    const MERGE_DIST: i32 = 3;
+
+    /// Maximum number of filters allowed in a single query
+    const MAX_FILTER_NUM: usize = 20;
+
+    /// Add lower bounds to the dirty time windows. Upper bounds are ignored.
+    ///
+    /// # Arguments
+    ///
+    /// * `lower_bounds` - An iterator of lower bounds to be added.
+    pub fn add_lower_bounds(&mut self, lower_bounds: impl Iterator<Item = Timestamp>) {
+        for lower_bound in lower_bounds {
+            let entry = self.windows.entry(lower_bound);
+            entry.or_insert(None);
+        }
+    }
+
+    /// Generate all filter expressions consuming all time windows
+    pub fn gen_filter_exprs(
+        &mut self,
+        col_name: &str,
+        expire_lower_bound: Option<Timestamp>,
+        window_size: chrono::Duration,
+        task_ctx: &RecordingRuleTask,
+    ) -> Result<Option<datafusion_expr::Expr>, Error> {
+        debug!(
+            "expire_lower_bound: {:?}, window_size: {:?}",
+            expire_lower_bound.map(|t| t.to_iso8601_string()),
+            window_size
+        );
+        self.merge_dirty_time_windows(window_size, expire_lower_bound)?;
+
+        if self.windows.len() > Self::MAX_FILTER_NUM {
+            let first_time_window = self.windows.first_key_value();
+            let last_time_window = self.windows.last_key_value();
+            warn!(
+                "Flow id = {:?}, too many time windows: {}, only the first {} are taken for this query, the group by expression might be wrong. Time window expr={:?}, expire_after={:?}, first_time_window={:?}, last_time_window={:?}, the original query: {:?}",
+                task_ctx.flow_id,
+                self.windows.len(),
+                Self::MAX_FILTER_NUM,
+                task_ctx.time_window_expr,
+                task_ctx.expire_after,
+                first_time_window,
+                last_time_window,
+                task_ctx.query
+            );
+        }
+
+        // get the first `MAX_FILTER_NUM` time windows
+        let nth = self
+            .windows
+            .iter()
+            .nth(Self::MAX_FILTER_NUM)
+            .map(|(key, _)| *key);
+        let first_nth = {
+            if let Some(nth) = nth {
+                let mut after = self.windows.split_off(&nth);
+                std::mem::swap(&mut self.windows, &mut after);
+
+                after
+            } else {
+                std::mem::take(&mut self.windows)
+            }
+        };
+
+        let mut expr_lst = vec![];
+        for (start, end) in first_nth.into_iter() {
+            debug!(
+                "Time window start: {:?}, end: {:?}",
+                start.to_iso8601_string(),
+                end.map(|t| t.to_iso8601_string())
+            );
+
+            use datafusion_expr::{col, lit};
+            let lower = to_df_literal(start)?;
+            let upper = end.map(to_df_literal).transpose()?;
+            let expr = if let Some(upper) = upper {
+                col(col_name)
+                    .gt_eq(lit(lower))
+                    .and(col(col_name).lt(lit(upper)))
+            } else {
+                col(col_name).gt_eq(lit(lower))
+            };
+            expr_lst.push(expr);
+        }
+        let expr = expr_lst.into_iter().reduce(|a, b| a.or(b));
+        Ok(expr)
+    }
+
+    /// Merge time windows that overlaps or get too close
+    pub fn merge_dirty_time_windows(
+        &mut self,
+        window_size: chrono::Duration,
+        expire_lower_bound: Option<Timestamp>,
+    ) -> Result<(), Error> {
+        let mut new_windows = BTreeMap::new();
+
+        let mut prev_tw = None;
+        for (lower_bound, upper_bound) in std::mem::take(&mut self.windows) {
+            // filter out expired time window
+            if let Some(expire_lower_bound) = expire_lower_bound {
+                if lower_bound <= expire_lower_bound {
+                    continue;
+                }
+            }
+
+            let Some(prev_tw) = &mut prev_tw else {
+                prev_tw = Some((lower_bound, upper_bound));
+                continue;
+            };
+
+            let std_window_size = window_size.to_std().map_err(|e| {
+                InternalSnafu {
+                    reason: e.to_string(),
+                }
+                .build()
+            })?;
+
+            // if cur.lower - prev.upper <= window_size * 2, merge
+            let prev_upper = prev_tw
+                .1
+                .unwrap_or(prev_tw.0.add_duration(std_window_size).context(TimeSnafu)?);
+            prev_tw.1 = Some(prev_upper);
+
+            let cur_upper = upper_bound.unwrap_or(
+                lower_bound
+                    .add_duration(std_window_size)
+                    .context(TimeSnafu)?,
+            );
+
+            if lower_bound
+                .sub(&prev_upper)
+                .map(|dist| dist <= window_size * Self::MERGE_DIST)
+                .unwrap_or(false)
+            {
+                prev_tw.1 = Some(cur_upper);
+            } else {
+                new_windows.insert(prev_tw.0, prev_tw.1);
+                *prev_tw = (lower_bound, Some(cur_upper));
+            }
+        }
+
+        if let Some(prev_tw) = prev_tw {
+            new_windows.insert(prev_tw.0, prev_tw.1);
+        }
+
+        self.windows = new_windows;
+
+        Ok(())
+    }
+}
+
+impl RecordingRuleState {
+    pub fn new(query_ctx: QueryContextRef, shutdown_rx: oneshot::Receiver<()>) -> Self {
+        Self {
+            query_ctx,
+            last_update_time: Instant::now(),
+            last_query_duration: Duration::from_secs(0),
+            dirty_time_windows: Default::default(),
+            exec_state: ExecState::Idle,
+            shutdown_rx,
+        }
+    }
+
+    /// called after last query is done
+    /// `is_succ` indicate whether the last query is successful
+    pub fn after_query_exec(&mut self, elapsed: Duration, _is_succ: bool) {
+        self.exec_state = ExecState::Idle;
+        self.last_query_duration = elapsed;
+        self.last_update_time = Instant::now();
+    }
+
+    /// wait for at least `last_query_duration`, at most `max_timeout` to start next query
+    pub fn get_next_start_query_time(&self, max_timeout: Option<Duration>) -> Instant {
+        let next_duration = max_timeout
+            .unwrap_or(self.last_query_duration)
+            .min(self.last_query_duration);
+        let next_duration = next_duration.max(MIN_REFRESH_DURATION);
+
+        self.last_update_time + next_duration
+    }
+}
+
+#[derive(Debug, Clone)]
+enum ExecState {
+    Idle,
+    Executing,
+}
+
+#[cfg(test)]
+mod test {
+    use pretty_assertions::assert_eq;
+
+    use super::*;
+
+    #[test]
+    fn test_merge_dirty_time_windows() {
+        let mut dirty = DirtyTimeWindows::default();
+        dirty.add_lower_bounds(
+            vec![
+                Timestamp::new_second(0),
+                Timestamp::new_second((1 + DirtyTimeWindows::MERGE_DIST as i64) * 5 * 60),
+            ]
+            .into_iter(),
+        );
+        dirty
+            .merge_dirty_time_windows(chrono::Duration::seconds(5 * 60), None)
+            .unwrap();
+        // just enough to merge
+        assert_eq!(
+            dirty.windows,
+            BTreeMap::from([(
+                Timestamp::new_second(0),
+                Some(Timestamp::new_second(
+                    (2 + DirtyTimeWindows::MERGE_DIST as i64) * 5 * 60
+                ))
+            )])
+        );
+
+        // separate time window
+        let mut dirty = DirtyTimeWindows::default();
+        dirty.add_lower_bounds(
+            vec![
+                Timestamp::new_second(0),
+                Timestamp::new_second((2 + DirtyTimeWindows::MERGE_DIST as i64) * 5 * 60),
+            ]
+            .into_iter(),
+        );
+        dirty
+            .merge_dirty_time_windows(chrono::Duration::seconds(5 * 60), None)
+            .unwrap();
+        // just enough to merge
+        assert_eq!(
+            BTreeMap::from([
+                (
+                    Timestamp::new_second(0),
+                    Some(Timestamp::new_second(5 * 60))
+                ),
+                (
+                    Timestamp::new_second((2 + DirtyTimeWindows::MERGE_DIST as i64) * 5 * 60),
+                    Some(Timestamp::new_second(
+                        (3 + DirtyTimeWindows::MERGE_DIST as i64) * 5 * 60
+                    ))
+                )
+            ]),
+            dirty.windows
+        );
+
+        // overlapping
+        let mut dirty = DirtyTimeWindows::default();
+        dirty.add_lower_bounds(
+            vec![
+                Timestamp::new_second(0),
+                Timestamp::new_second((DirtyTimeWindows::MERGE_DIST as i64) * 5 * 60),
+            ]
+            .into_iter(),
+        );
+        dirty
+            .merge_dirty_time_windows(chrono::Duration::seconds(5 * 60), None)
+            .unwrap();
+        // just enough to merge
+        assert_eq!(
+            BTreeMap::from([(
+                Timestamp::new_second(0),
+                Some(Timestamp::new_second(
+                    (1 + DirtyTimeWindows::MERGE_DIST as i64) * 5 * 60
+                ))
+            ),]),
+            dirty.windows
+        );
+
+        // expired
+        let mut dirty = DirtyTimeWindows::default();
+        dirty.add_lower_bounds(
+            vec![
+                Timestamp::new_second(0),
+                Timestamp::new_second((DirtyTimeWindows::MERGE_DIST as i64) * 5 * 60),
+            ]
+            .into_iter(),
+        );
+        dirty
+            .merge_dirty_time_windows(
+                chrono::Duration::seconds(5 * 60),
+                Some(Timestamp::new_second(
+                    (DirtyTimeWindows::MERGE_DIST as i64) * 6 * 60,
+                )),
+            )
+            .unwrap();
+        // just enough to merge
+        assert_eq!(BTreeMap::from([]), dirty.windows);
+    }
+}
--- a/src/flow/src/recording_rules/frontend_client.rs
+++ b/src/flow/src/recording_rules/frontend_client.rs
@@ -0,0 +1,163 @@
+// 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.
+
+//! Frontend client to run flow as recording rule which is time-window-aware normal query triggered every tick set by user
+
+use std::sync::Arc;
+
+use client::{Client, Database, DEFAULT_CATALOG_NAME, DEFAULT_SCHEMA_NAME};
+use common_error::ext::BoxedError;
+use common_grpc::channel_manager::{ChannelConfig, ChannelManager};
+use common_meta::cluster::{NodeInfo, NodeInfoKey, Role};
+use common_meta::peer::Peer;
+use common_meta::rpc::store::RangeRequest;
+use meta_client::client::MetaClient;
+use snafu::ResultExt;
+
+use crate::error::{ExternalSnafu, UnexpectedSnafu};
+use crate::recording_rules::engine::DEFAULT_RULE_ENGINE_QUERY_TIMEOUT;
+use crate::Error;
+
+fn default_channel_mgr() -> ChannelManager {
+    let cfg = ChannelConfig::new().timeout(DEFAULT_RULE_ENGINE_QUERY_TIMEOUT);
+    ChannelManager::with_config(cfg)
+}
+
+fn client_from_urls(addrs: Vec<String>) -> Client {
+    Client::with_manager_and_urls(default_channel_mgr(), addrs)
+}
+
+/// A simple frontend client able to execute sql using grpc protocol
+#[derive(Debug)]
+pub enum FrontendClient {
+    Distributed {
+        meta_client: Arc<MetaClient>,
+        channel_mgr: ChannelManager,
+    },
+    Standalone {
+        /// for the sake of simplicity still use grpc even in standalone mode
+        /// notice the client here should all be lazy, so that can wait after frontend is booted then make conn
+        /// TODO(discord9): not use grpc under standalone mode
+        database_client: DatabaseWithPeer,
+    },
+}
+
+#[derive(Debug, Clone)]
+pub struct DatabaseWithPeer {
+    pub database: Database,
+    pub peer: Peer,
+}
+
+impl DatabaseWithPeer {
+    fn new(database: Database, peer: Peer) -> Self {
+        Self { database, peer }
+    }
+}
+
+impl FrontendClient {
+    pub fn from_meta_client(meta_client: Arc<MetaClient>) -> Self {
+        Self::Distributed {
+            meta_client,
+            channel_mgr: default_channel_mgr(),
+        }
+    }
+
+    pub fn from_static_grpc_addr(addr: String) -> Self {
+        let peer = Peer {
+            id: 0,
+            addr: addr.clone(),
+        };
+
+        let mgr = default_channel_mgr();
+        let client = Client::with_manager_and_urls(mgr.clone(), vec![addr]);
+        let database = Database::new(DEFAULT_CATALOG_NAME, DEFAULT_SCHEMA_NAME, client);
+        Self::Standalone {
+            database_client: DatabaseWithPeer::new(database, peer),
+        }
+    }
+}
+
+impl FrontendClient {
+    async fn scan_for_frontend(&self) -> Result<Vec<(NodeInfoKey, NodeInfo)>, Error> {
+        let Self::Distributed { meta_client, .. } = self else {
+            return Ok(vec![]);
+        };
+        let cluster_client = meta_client
+            .cluster_client()
+            .map_err(BoxedError::new)
+            .context(ExternalSnafu)?;
+        let cluster_id = meta_client.id().0;
+        let prefix = NodeInfoKey::key_prefix_with_role(cluster_id, Role::Frontend);
+        let req = RangeRequest::new().with_prefix(prefix);
+        let resp = cluster_client
+            .range(req)
+            .await
+            .map_err(BoxedError::new)
+            .context(ExternalSnafu)?;
+        let mut res = Vec::with_capacity(resp.kvs.len());
+        for kv in resp.kvs {
+            let key = NodeInfoKey::try_from(kv.key)
+                .map_err(BoxedError::new)
+                .context(ExternalSnafu)?;
+
+            let val = NodeInfo::try_from(kv.value)
+                .map_err(BoxedError::new)
+                .context(ExternalSnafu)?;
+            res.push((key, val));
+        }
+        Ok(res)
+    }
+
+    /// Get the database with max `last_activity_ts`
+    async fn get_last_active_frontend(&self) -> Result<DatabaseWithPeer, Error> {
+        if let Self::Standalone { database_client } = self {
+            return Ok(database_client.clone());
+        }
+        match &self {
+            Self::Standalone { database_client } => Ok(database_client.clone()),
+            Self::Distributed {
+                meta_client: _,
+                channel_mgr,
+            } => {
+                let frontends = self.scan_for_frontend().await?;
+                let mut last_activity_ts = i64::MIN;
+                let mut peer = None;
+                for (_key, val) in frontends.iter() {
+                    if val.last_activity_ts > last_activity_ts {
+                        last_activity_ts = val.last_activity_ts;
+                        peer = Some(val.peer.clone());
+                    }
+                }
+                let Some(peer) = peer else {
+                    UnexpectedSnafu {
+                        reason: format!("No frontend available: {:?}", frontends),
+                    }
+                    .fail()?
+                };
+                let client =
+                    Client::with_manager_and_urls(channel_mgr.clone(), vec![peer.addr.clone()]);
+                let database = Database::new(DEFAULT_CATALOG_NAME, DEFAULT_SCHEMA_NAME, client);
+                Ok(DatabaseWithPeer::new(database, peer))
+            }
+        }
+    }
+
+    /// Get a database client, and possibly update it before returning.
+    pub async fn get_database_client(&self) -> Result<DatabaseWithPeer, Error> {
+        match self {
+            Self::Standalone { database_client } => Ok(database_client.clone()),
+            Self::Distributed { meta_client: _, .. } => self.get_last_active_frontend().await,
+        }
+    }
+}
--- a/src/flow/src/server.rs
+++ b/src/flow/src/server.rs
@@ -57,6 +57,7 @@ use crate::error::{
 };
 use crate::heartbeat::HeartbeatTask;
 use crate::metrics::{METRIC_FLOW_PROCESSING_TIME, METRIC_FLOW_ROWS};
+use crate::recording_rules::{FrontendClient, RecordingRuleEngine};
 use crate::transform::register_function_to_query_engine;
 use crate::utils::{SizeReportSender, StateReportHandler};
 use crate::{Error, FlowWorkerManager, FlownodeOptions};
@@ -245,6 +246,7 @@ impl FlownodeInstance {
        self.server.shutdown().await.context(ShutdownServerSnafu)?;

        if let Some(task) = &self.heartbeat_task {
+            info!("Close heartbeat task for flownode");
            task.shutdown();
        }

@@ -271,6 +273,8 @@ pub struct FlownodeBuilder {
    heartbeat_task: Option<HeartbeatTask>,
    /// receive a oneshot sender to send state size report
    state_report_handler: Option<StateReportHandler>,
+    /// Client to send sql to frontend
+    frontend_client: Arc<FrontendClient>,
 }

 impl FlownodeBuilder {
@@ -281,6 +285,7 @@ impl FlownodeBuilder {
        table_meta: TableMetadataManagerRef,
        catalog_manager: CatalogManagerRef,
        flow_metadata_manager: FlowMetadataManagerRef,
+        frontend_client: Arc<FrontendClient>,
    ) -> Self {
        Self {
            opts,
@@ -290,6 +295,7 @@ impl FlownodeBuilder {
            flow_metadata_manager,
            heartbeat_task: None,
            state_report_handler: None,
+            frontend_client,
        }
    }

@@ -447,7 +453,14 @@ impl FlownodeBuilder {

        let node_id = self.opts.node_id.map(|id| id as u32);

-        let mut man = FlowWorkerManager::new(node_id, query_engine, table_meta);
+        let rule_engine = RecordingRuleEngine::new(
+            self.frontend_client.clone(),
+            query_engine.clone(),
+            self.flow_metadata_manager.clone(),
+            table_meta.clone(),
+        );
+
+        let mut man = FlowWorkerManager::new(node_id, query_engine, table_meta, rule_engine);
        for worker_id in 0..num_workers {
            let (tx, rx) = oneshot::channel();

--- a/src/flow/src/test_utils.rs
+++ b/src/flow/src/test_utils.rs
@@ -86,7 +86,8 @@ pub fn create_test_query_engine() -> Arc<dyn QueryEngine> {

    let schema = vec![
        datatypes::schema::ColumnSchema::new("number", CDT::uint32_datatype(), false),
-        datatypes::schema::ColumnSchema::new("ts", CDT::timestamp_millisecond_datatype(), false),
+        datatypes::schema::ColumnSchema::new("ts", CDT::timestamp_millisecond_datatype(), false)
+            .with_time_index(true),
    ];
    let mut columns = vec![];
    let numbers = (1..=10).collect_vec();
@@ -114,6 +115,37 @@ pub fn create_test_query_engine() -> Arc<dyn QueryEngine> {
    };
    catalog_list.register_table_sync(req_with_ts).unwrap();

+    let schema = vec![
+        datatypes::schema::ColumnSchema::new("NUMBER", CDT::uint32_datatype(), false),
+        datatypes::schema::ColumnSchema::new("ts", CDT::timestamp_millisecond_datatype(), false)
+            .with_time_index(true),
+    ];
+    let mut columns = vec![];
+    let numbers = (1..=10).collect_vec();
+    let column: VectorRef = Arc::new(<u32 as Scalar>::VectorType::from_vec(numbers));
+    columns.push(column);
+
+    let ts = (1..=10).collect_vec();
+    let mut builder = TimestampMillisecondVectorBuilder::with_capacity(10);
+    ts.into_iter()
+        .map(|v| builder.push(Some(TimestampMillisecond::new(v))))
+        .count();
+    let column: VectorRef = builder.to_vector_cloned();
+    columns.push(column);
+
+    let schema = Arc::new(Schema::new(schema));
+    let recordbatch = common_recordbatch::RecordBatch::new(schema, columns).unwrap();
+    let table = MemTable::table("UPPERCASE_NUMBERS_WITH_TS", recordbatch);
+
+    let req_with_ts = RegisterTableRequest {
+        catalog: DEFAULT_CATALOG_NAME.to_string(),
+        schema: DEFAULT_SCHEMA_NAME.to_string(),
+        table_name: "UPPERCASE_NUMBERS_WITH_TS".to_string(),
+        table_id: 1025,
+        table,
+    };
+    catalog_list.register_table_sync(req_with_ts).unwrap();
+
    let factory = query::QueryEngineFactory::new(catalog_list, None, None, None, None, false);

    let engine = factory.query_engine();
--- a/src/meta-client/src/client.rs
+++ b/src/meta-client/src/client.rs
@@ -112,6 +112,7 @@ impl MetaClientBuilder {
            .enable_store()
            .enable_heartbeat()
            .enable_procedure()
+            .enable_access_cluster_info()
    }

    pub fn enable_heartbeat(self) -> Self {
--- a/src/meta-client/src/client/heartbeat.rs
+++ b/src/meta-client/src/client/heartbeat.rs
@@ -198,13 +198,13 @@ impl Inner {
            }
        );

-        let leader = self
+        let leader_addr = self
            .ask_leader
            .as_ref()
            .unwrap()
            .get_leader()
            .context(error::NoLeaderSnafu)?;
-        let mut leader = self.make_client(leader)?;
+        let mut leader = self.make_client(&leader_addr)?;

        let (sender, receiver) = mpsc::channel::<HeartbeatRequest>(128);

@@ -236,7 +236,11 @@ impl Inner {
            .await
            .map_err(error::Error::from)?
            .context(error::CreateHeartbeatStreamSnafu)?;
-        info!("Success to create heartbeat stream to server: {:#?}", res);
+
+        info!(
+            "Success to create heartbeat stream to server: {}, response: {:#?}",
+            leader_addr, res
+        );

        Ok((
            HeartbeatSender::new(self.id, self.role, sender),
--- a/src/meta-srv/Cargo.toml
+++ b/src/meta-srv/Cargo.toml
@@ -7,6 +7,7 @@ license.workspace = true
 [features]
 mock = []
 pg_kvbackend = ["dep:tokio-postgres", "common-meta/pg_kvbackend"]
+mysql_kvbackend = []                                              # placeholder features so CI can compile

 [lints]
 workspace = true
--- a/src/meta-srv/src/handler/check_leader_handler.rs
+++ b/src/meta-srv/src/handler/check_leader_handler.rs
@@ -23,8 +23,8 @@ pub struct CheckLeaderHandler;

 #[async_trait::async_trait]
 impl HeartbeatHandler for CheckLeaderHandler {
-    fn is_acceptable(&self, role: Role) -> bool {
-        role == Role::Datanode
+    fn is_acceptable(&self, _role: Role) -> bool {
+        true
    }

    async fn handle(
--- a/src/meta-srv/src/handler/collect_cluster_info_handler.rs
+++ b/src/meta-srv/src/handler/collect_cluster_info_handler.rs
@@ -157,7 +157,7 @@ fn extract_base_info(request: &HeartbeatRequest) -> Option<(NodeInfoKey, Peer, P
 }

 async fn put_into_memory_store(ctx: &mut Context, key: NodeInfoKey, value: NodeInfo) -> Result<()> {
-    let key = key.into();
+    let key = (&key).into();
    let value = value.try_into().context(InvalidClusterInfoFormatSnafu)?;
    let put_req = PutRequest {
        key,
--- a/src/meta-srv/src/metasrv.rs
+++ b/src/meta-srv/src/metasrv.rs
@@ -32,6 +32,7 @@ use common_meta::kv_backend::{KvBackendRef, ResettableKvBackend, ResettableKvBac
 use common_meta::leadership_notifier::{
    LeadershipChangeNotifier, LeadershipChangeNotifierCustomizerRef,
 };
+use common_meta::node_expiry_listener::NodeExpiryListener;
 use common_meta::peer::Peer;
 use common_meta::region_keeper::MemoryRegionKeeperRef;
 use common_meta::wal_options_allocator::WalOptionsAllocatorRef;
@@ -151,6 +152,8 @@ pub struct MetasrvOptions {
    #[cfg(feature = "pg_kvbackend")]
    /// Lock id for meta kv election. Only effect when using pg_kvbackend.
    pub meta_election_lock_id: u64,
+    #[serde(with = "humantime_serde")]
+    pub node_max_idle_time: Duration,
 }

 const DEFAULT_METASRV_ADDR_PORT: &str = "3002";
@@ -192,6 +195,7 @@ impl Default for MetasrvOptions {
            meta_table_name: DEFAULT_META_TABLE_NAME.to_string(),
            #[cfg(feature = "pg_kvbackend")]
            meta_election_lock_id: DEFAULT_META_ELECTION_LOCK_ID,
+            node_max_idle_time: Duration::from_secs(24 * 60 * 60),
        }
    }
 }
@@ -442,6 +446,10 @@ impl Metasrv {
            leadership_change_notifier.add_listener(self.wal_options_allocator.clone());
            leadership_change_notifier
                .add_listener(Arc::new(ProcedureManagerListenerAdapter(procedure_manager)));
+            leadership_change_notifier.add_listener(Arc::new(NodeExpiryListener::new(
+                self.options.node_max_idle_time,
+                self.in_memory.clone(),
+            )));
            if let Some(region_supervisor_ticker) = &self.region_supervisor_ticker {
                leadership_change_notifier.add_listener(region_supervisor_ticker.clone() as _);
            }
--- a/src/meta-srv/src/service/heartbeat.rs
+++ b/src/meta-srv/src/service/heartbeat.rs
@@ -68,13 +68,15 @@ impl heartbeat_server::Heartbeat for Metasrv {
                        };

                        if pusher_id.is_none() {
-                            pusher_id = register_pusher(&handler_group, header, tx.clone()).await;
+                            pusher_id =
+                                Some(register_pusher(&handler_group, header, tx.clone()).await);
                        }
                        if let Some(k) = &pusher_id {
                            METRIC_META_HEARTBEAT_RECV.with_label_values(&[&k.to_string()]);
                        } else {
                            METRIC_META_HEARTBEAT_RECV.with_label_values(&["none"]);
                        }
+
                        let res = handler_group
                            .handle(req, ctx.clone())
                            .await
@@ -173,13 +175,13 @@ async fn register_pusher(
    handler_group: &HeartbeatHandlerGroup,
    header: &RequestHeader,
    sender: Sender<std::result::Result<HeartbeatResponse, tonic::Status>>,
-) -> Option<PusherId> {
+) -> PusherId {
    let role = header.role();
    let id = get_node_id(header);
    let pusher_id = PusherId::new(role, id);
    let pusher = Pusher::new(sender, header);
    handler_group.register_pusher(pusher_id, pusher).await;
-    Some(pusher_id)
+    pusher_id
 }

 #[cfg(test)]
--- a/src/meta-srv/src/service/procedure.rs
+++ b/src/meta-srv/src/service/procedure.rs
@@ -17,13 +17,15 @@ use std::time::Duration;

 use api::v1::meta::{
    procedure_service_server, DdlTaskRequest as PbDdlTaskRequest,
-    DdlTaskResponse as PbDdlTaskResponse, MigrateRegionRequest, MigrateRegionResponse,
+    DdlTaskResponse as PbDdlTaskResponse, Error, MigrateRegionRequest, MigrateRegionResponse,
    ProcedureDetailRequest, ProcedureDetailResponse, ProcedureStateResponse, QueryProcedureRequest,
+    ResponseHeader,
 };
 use common_meta::ddl::ExecutorContext;
 use common_meta::rpc::ddl::{DdlTask, SubmitDdlTaskRequest};
 use common_meta::rpc::procedure;
-use snafu::{ensure, OptionExt, ResultExt};
+use common_telemetry::warn;
+use snafu::{OptionExt, ResultExt};
 use tonic::{Request, Response};

 use super::GrpcResult;
@@ -37,6 +39,16 @@ impl procedure_service_server::ProcedureService for Metasrv {
        &self,
        request: Request<QueryProcedureRequest>,
    ) -> GrpcResult<ProcedureStateResponse> {
+        if !self.is_leader() {
+            let resp = ProcedureStateResponse {
+                header: Some(ResponseHeader::failed(0, Error::is_not_leader())),
+                ..Default::default()
+            };
+
+            warn!("The current meta is not leader, but a `query procedure state` request have reached the meta. Detail: {:?}.", request);
+            return Ok(Response::new(resp));
+        }
+
        let QueryProcedureRequest { header, pid, .. } = request.into_inner();
        let _header = header.context(error::MissingRequestHeaderSnafu)?;
        let pid = pid.context(error::MissingRequiredParameterSnafu { param: "pid" })?;
@@ -57,6 +69,16 @@ impl procedure_service_server::ProcedureService for Metasrv {
    }

    async fn ddl(&self, request: Request<PbDdlTaskRequest>) -> GrpcResult<PbDdlTaskResponse> {
+        if !self.is_leader() {
+            let resp = PbDdlTaskResponse {
+                header: Some(ResponseHeader::failed(0, Error::is_not_leader())),
+                ..Default::default()
+            };
+
+            warn!("The current meta is not leader, but a `ddl` request have reached the meta. Detail: {:?}.", request);
+            return Ok(Response::new(resp));
+        }
+
        let PbDdlTaskRequest {
            header,
            query_context,
@@ -99,12 +121,15 @@ impl procedure_service_server::ProcedureService for Metasrv {
        &self,
        request: Request<MigrateRegionRequest>,
    ) -> GrpcResult<MigrateRegionResponse> {
-        ensure!(
-            self.meta_peer_client().is_leader(),
-            error::UnexpectedSnafu {
-                violated: "Trying to submit a region migration procedure to non-leader meta server"
-            }
-        );
+        if !self.is_leader() {
+            let resp = MigrateRegionResponse {
+                header: Some(ResponseHeader::failed(0, Error::is_not_leader())),
+                ..Default::default()
+            };
+
+            warn!("The current meta is not leader, but a `migrate` request have reached the meta. Detail: {:?}.", request);
+            return Ok(Response::new(resp));
+        }

        let MigrateRegionRequest {
            header,
@@ -150,6 +175,16 @@ impl procedure_service_server::ProcedureService for Metasrv {
        &self,
        request: Request<ProcedureDetailRequest>,
    ) -> GrpcResult<ProcedureDetailResponse> {
+        if !self.is_leader() {
+            let resp = ProcedureDetailResponse {
+                header: Some(ResponseHeader::failed(0, Error::is_not_leader())),
+                ..Default::default()
+            };
+
+            warn!("The current meta is not leader, but a `procedure details` request have reached the meta. Detail: {:?}.", request);
+            return Ok(Response::new(resp));
+        }
+
        let ProcedureDetailRequest { header } = request.into_inner();
        let _header = header.context(error::MissingRequestHeaderSnafu)?;
        let metas = self
--- a/src/metric-engine/src/engine/drop.rs
+++ b/src/metric-engine/src/engine/drop.rs
@@ -30,9 +30,10 @@ impl MetricEngineInner {
    pub async fn drop_region(
        &self,
        region_id: RegionId,
-        _req: RegionDropRequest,
+        req: RegionDropRequest,
    ) -> Result<AffectedRows> {
        let data_region_id = utils::to_data_region_id(region_id);
+        let fast_path = req.fast_path;

        // enclose the guard in a block to prevent the guard from polluting the async context
        let (is_physical_region, is_physical_region_busy) = {
@@ -52,7 +53,7 @@ impl MetricEngineInner {

        if is_physical_region {
            // check if there is no logical region relates to this physical region
-            if is_physical_region_busy {
+            if is_physical_region_busy && !fast_path {
                // reject if there is any present logical region
                return Err(PhysicalRegionBusySnafu {
                    region_id: data_region_id,
@@ -60,9 +61,21 @@ impl MetricEngineInner {
                .build());
            }

-            self.drop_physical_region(data_region_id).await
+            return self.drop_physical_region(data_region_id).await;
+        }
+
+        if fast_path {
+            // for fast path, we don't delete the metadata in the metadata region.
+            // it only remove the logical region from the engine state.
+            //
+            // The drop database procedure will ensure the metadata region and data region are dropped eventually.
+            self.state
+                .write()
+                .unwrap()
+                .remove_logical_region(region_id)?;
+
+            Ok(0)
        } else {
-            // cannot merge these two `if` otherwise the stupid type checker will complain
            let metadata_region_id = self
                .state
                .read()
@@ -87,13 +100,16 @@ impl MetricEngineInner {
        // Since the physical regions are going to be dropped, we don't need to
        // update the contents in metadata region.
        self.mito
-            .handle_request(data_region_id, RegionRequest::Drop(RegionDropRequest {}))
+            .handle_request(
+                data_region_id,
+                RegionRequest::Drop(RegionDropRequest { fast_path: false }),
+            )
            .await
            .with_context(|_| CloseMitoRegionSnafu { region_id })?;
        self.mito
            .handle_request(
                metadata_region_id,
-                RegionRequest::Drop(RegionDropRequest {}),
+                RegionRequest::Drop(RegionDropRequest { fast_path: false }),
            )
            .await
            .with_context(|_| CloseMitoRegionSnafu { region_id })?;
--- a/src/mito2/src/cache/file_cache.rs
+++ b/src/mito2/src/cache/file_cache.rs
@@ -187,9 +187,12 @@ impl FileCache {
    }

    /// Removes a file from the cache explicitly.
+    /// It always tries to remove the file from the local store because we may not have the file
+    /// in the memory index if upload is failed.
    pub(crate) async fn remove(&self, key: IndexKey) {
        let file_path = self.cache_file_path(key);
        self.memory_index.remove(&key).await;
+        // Always delete the file from the local store.
        if let Err(e) = self.local_store.delete(&file_path).await {
            warn!(e; "Failed to delete a cached file {}", file_path);
        }
--- a/src/mito2/src/cache/write_cache.rs
+++ b/src/mito2/src/cache/write_cache.rs
@@ -22,6 +22,7 @@ use common_telemetry::{debug, info};
 use futures::AsyncWriteExt;
 use object_store::ObjectStore;
 use snafu::ResultExt;
+use store_api::storage::RegionId;

 use crate::access_layer::{
    new_fs_cache_store, FilePathProvider, RegionFilePathFactory, SstInfoArray, SstWriteRequest,
@@ -149,24 +150,41 @@ impl WriteCache {
            return Ok(sst_info);
        }

+        let mut upload_tracker = UploadTracker::new(region_id);
+        let mut err = None;
        let remote_store = &upload_request.remote_store;
        for sst in &sst_info {
            let parquet_key = IndexKey::new(region_id, sst.file_id, FileType::Parquet);
            let parquet_path = upload_request
                .dest_path_provider
                .build_sst_file_path(sst.file_id);
-            self.upload(parquet_key, &parquet_path, remote_store)
-                .await?;
+            if let Err(e) = self.upload(parquet_key, &parquet_path, remote_store).await {
+                err = Some(e);
+                break;
+            }
+            upload_tracker.push_uploaded_file(parquet_path);

            if sst.index_metadata.file_size > 0 {
                let puffin_key = IndexKey::new(region_id, sst.file_id, FileType::Puffin);
-                let puffin_path = &upload_request
+                let puffin_path = upload_request
                    .dest_path_provider
                    .build_index_file_path(sst.file_id);
-                self.upload(puffin_key, puffin_path, remote_store).await?;
+                if let Err(e) = self.upload(puffin_key, &puffin_path, remote_store).await {
+                    err = Some(e);
+                    break;
+                }
+                upload_tracker.push_uploaded_file(puffin_path);
            }
        }

+        if let Some(err) = err {
+            // Cleans files on failure.
+            upload_tracker
+                .clean(&sst_info, &self.file_cache, remote_store)
+                .await;
+            return Err(err);
+        }
+
        Ok(sst_info)
    }

@@ -332,6 +350,61 @@ pub struct SstUploadRequest {
    pub remote_store: ObjectStore,
 }

+/// A structs to track files to upload and clean them if upload failed.
+struct UploadTracker {
+    /// Id of the region to track.
+    region_id: RegionId,
+    /// Paths of files uploaded successfully.
+    files_uploaded: Vec<String>,
+}
+
+impl UploadTracker {
+    /// Creates a new instance of `UploadTracker` for a given region.
+    fn new(region_id: RegionId) -> Self {
+        Self {
+            region_id,
+            files_uploaded: Vec::new(),
+        }
+    }
+
+    /// Add a file path to the list of uploaded files.
+    fn push_uploaded_file(&mut self, path: String) {
+        self.files_uploaded.push(path);
+    }
+
+    /// Cleans uploaded files and files in the file cache at best effort.
+    async fn clean(
+        &self,
+        sst_info: &SstInfoArray,
+        file_cache: &FileCacheRef,
+        remote_store: &ObjectStore,
+    ) {
+        common_telemetry::info!(
+            "Start cleaning files on upload failure, region: {}, num_ssts: {}",
+            self.region_id,
+            sst_info.len()
+        );
+
+        // Cleans files in the file cache first.
+        for sst in sst_info {
+            let parquet_key = IndexKey::new(self.region_id, sst.file_id, FileType::Parquet);
+            file_cache.remove(parquet_key).await;
+
+            if sst.index_metadata.file_size > 0 {
+                let puffin_key = IndexKey::new(self.region_id, sst.file_id, FileType::Puffin);
+                file_cache.remove(puffin_key).await;
+            }
+        }
+
+        // Cleans uploaded files.
+        for file_path in &self.files_uploaded {
+            if let Err(e) = remote_store.delete(file_path).await {
+                common_telemetry::error!(e; "Failed to delete file {}", file_path);
+            }
+        }
+    }
+}
+
 #[cfg(test)]
 mod tests {
    use common_test_util::temp_dir::create_temp_dir;
--- a/src/mito2/src/engine/drop_test.rs
+++ b/src/mito2/src/engine/drop_test.rs
@@ -56,7 +56,10 @@ async fn test_engine_drop_region() {

    // It's okay to drop a region doesn't exist.
    engine
-        .handle_request(region_id, RegionRequest::Drop(RegionDropRequest {}))
+        .handle_request(
+            region_id,
+            RegionRequest::Drop(RegionDropRequest { fast_path: false }),
+        )
        .await
        .unwrap_err();

@@ -86,7 +89,10 @@ async fn test_engine_drop_region() {

    // drop the created region.
    engine
-        .handle_request(region_id, RegionRequest::Drop(RegionDropRequest {}))
+        .handle_request(
+            region_id,
+            RegionRequest::Drop(RegionDropRequest { fast_path: false }),
+        )
        .await
        .unwrap();
    assert!(!engine.is_region_exists(region_id));
@@ -192,7 +198,10 @@ async fn test_engine_drop_region_for_custom_store() {

    // Drop the custom region.
    engine
-        .handle_request(custom_region_id, RegionRequest::Drop(RegionDropRequest {}))
+        .handle_request(
+            custom_region_id,
+            RegionRequest::Drop(RegionDropRequest { fast_path: false }),
+        )
        .await
        .unwrap();
    assert!(!engine.is_region_exists(custom_region_id));
--- a/src/mito2/src/request.rs
+++ b/src/mito2/src/request.rs
@@ -35,8 +35,8 @@ use store_api::metadata::{ColumnMetadata, RegionMetadata, RegionMetadataRef};
 use store_api::region_engine::{SetRegionRoleStateResponse, SettableRegionRoleState};
 use store_api::region_request::{
    AffectedRows, RegionAlterRequest, RegionCatchupRequest, RegionCloseRequest,
-    RegionCompactRequest, RegionCreateRequest, RegionDropRequest, RegionFlushRequest,
-    RegionOpenRequest, RegionRequest, RegionTruncateRequest,
+    RegionCompactRequest, RegionCreateRequest, RegionFlushRequest, RegionOpenRequest,
+    RegionRequest, RegionTruncateRequest,
 };
 use store_api::storage::{RegionId, SequenceNumber};
 use tokio::sync::oneshot::{self, Receiver, Sender};
@@ -624,10 +624,10 @@ impl WorkerRequest {
                sender: sender.into(),
                request: DdlRequest::Create(v),
            }),
-            RegionRequest::Drop(v) => WorkerRequest::Ddl(SenderDdlRequest {
+            RegionRequest::Drop(_) => WorkerRequest::Ddl(SenderDdlRequest {
                region_id,
                sender: sender.into(),
-                request: DdlRequest::Drop(v),
+                request: DdlRequest::Drop,
            }),
            RegionRequest::Open(v) => WorkerRequest::Ddl(SenderDdlRequest {
                region_id,
@@ -690,7 +690,7 @@ impl WorkerRequest {
 #[derive(Debug)]
 pub(crate) enum DdlRequest {
    Create(RegionCreateRequest),
-    Drop(RegionDropRequest),
+    Drop,
    Open((RegionOpenRequest, Option<WalEntryReceiver>)),
    Close(RegionCloseRequest),
    Alter(RegionAlterRequest),
--- a/src/mito2/src/worker.rs
+++ b/src/mito2/src/worker.rs
@@ -836,7 +836,7 @@ impl<S: LogStore> RegionWorkerLoop<S> {
        for ddl in ddl_requests.drain(..) {
            let res = match ddl.request {
                DdlRequest::Create(req) => self.handle_create_request(ddl.region_id, req).await,
-                DdlRequest::Drop(_) => self.handle_drop_request(ddl.region_id).await,
+                DdlRequest::Drop => self.handle_drop_request(ddl.region_id).await,
                DdlRequest::Open((req, wal_entry_receiver)) => {
                    self.handle_open_request(ddl.region_id, req, wal_entry_receiver, ddl.sender)
                        .await;
--- a/src/operator/src/insert.rs
+++ b/src/operator/src/insert.rs
@@ -68,6 +68,7 @@ pub struct Inserter {
    catalog_manager: CatalogManagerRef,
    partition_manager: PartitionRuleManagerRef,
    node_manager: NodeManagerRef,
+    #[allow(unused)]
    table_flownode_set_cache: TableFlownodeSetCacheRef,
 }

@@ -338,6 +339,8 @@ impl Inserter {
            instant_requests,
        } = requests;

+        // TODO(discord9): mirror some
+
        // Mirror requests for source table to flownode asynchronously
        let flow_mirror_task = FlowMirrorTask::new(
            &self.table_flownode_set_cache,
@@ -817,12 +820,14 @@ struct CreateAlterTableResult {
    table_infos: HashMap<TableId, Arc<TableInfo>>,
 }

+#[allow(unused)]
 struct FlowMirrorTask {
    requests: HashMap<Peer, RegionInsertRequests>,
    num_rows: usize,
 }

 impl FlowMirrorTask {
+    #[allow(unused)]
    async fn new(
        cache: &TableFlownodeSetCacheRef,
        requests: impl Iterator<Item = &RegionInsertRequest>,
@@ -896,6 +901,7 @@ impl FlowMirrorTask {
        })
    }

+    #[allow(unused)]
    fn detach(self, node_manager: NodeManagerRef) -> Result<()> {
        crate::metrics::DIST_MIRROR_PENDING_ROW_COUNT.add(self.num_rows as i64);
        for (peer, inserts) in self.requests {
--- a/src/promql/src/extension_plan/histogram_fold.rs
+++ b/src/promql/src/extension_plan/histogram_fold.rs
@@ -583,7 +583,8 @@ impl HistogramFoldStream {
                    .expect("field column should not be nullable");
                counters.push(counter);
            }
-            let result = Self::evaluate_row(self.quantile, &bucket, &counters)?;
+            // ignore invalid data
+            let result = Self::evaluate_row(self.quantile, &bucket, &counters).unwrap_or(f64::NAN);
            self.output_buffer[self.field_column_index].push_value_ref(ValueRef::from(result));
            cursor += bucket_num;
            remaining_rows -= bucket_num;
@@ -672,7 +673,7 @@ impl HistogramFoldStream {
        if bucket.len() <= 1 {
            return Ok(f64::NAN);
        }
-        if *bucket.last().unwrap() != f64::INFINITY {
+        if bucket.last().unwrap().is_finite() {
            return Err(DataFusionError::Execution(
                "last bucket should be +Inf".to_string(),
            ));
@@ -692,8 +693,8 @@ impl HistogramFoldStream {
        }

        // check input value
-        debug_assert!(bucket.windows(2).all(|w| w[0] <= w[1]));
-        debug_assert!(counter.windows(2).all(|w| w[0] <= w[1]));
+        debug_assert!(bucket.windows(2).all(|w| w[0] <= w[1]), "{bucket:?}");
+        debug_assert!(counter.windows(2).all(|w| w[0] <= w[1]), "{counter:?}");

        let total = *counter.last().unwrap();
        let expected_pos = total * quantile;
--- a/src/promql/src/functions.rs
+++ b/src/promql/src/functions.rs
@@ -21,6 +21,7 @@ mod idelta;
 mod predict_linear;
 mod quantile;
 mod resets;
+mod round;
 #[cfg(test)]
 mod test_util;

@@ -39,6 +40,7 @@ pub use idelta::IDelta;
 pub use predict_linear::PredictLinear;
 pub use quantile::QuantileOverTime;
 pub use resets::Resets;
+pub use round::Round;

 pub(crate) fn extract_array(columnar_value: &ColumnarValue) -> Result<ArrayRef, DataFusionError> {
    if let ColumnarValue::Array(array) = columnar_value {
--- a/src/promql/src/functions/round.rs
+++ b/src/promql/src/functions/round.rs
@@ -0,0 +1,105 @@
+// Copyright 2023 Greptime Team
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+use std::sync::Arc;
+
+use datafusion::error::DataFusionError;
+use datafusion_expr::{create_udf, ColumnarValue, ScalarUDF, Volatility};
+use datatypes::arrow::array::AsArray;
+use datatypes::arrow::datatypes::{DataType, Float64Type};
+use datatypes::compute;
+
+use crate::functions::extract_array;
+
+pub struct Round {
+    nearest: f64,
+}
+
+impl Round {
+    fn new(nearest: f64) -> Self {
+        Self { nearest }
+    }
+
+    pub const fn name() -> &'static str {
+        "prom_round"
+    }
+
+    fn input_type() -> Vec<DataType> {
+        vec![DataType::Float64]
+    }
+
+    pub fn return_type() -> DataType {
+        DataType::Float64
+    }
+
+    pub fn scalar_udf(nearest: f64) -> ScalarUDF {
+        create_udf(
+            Self::name(),
+            Self::input_type(),
+            Self::return_type(),
+            Volatility::Immutable,
+            Arc::new(move |input: &_| Self::new(nearest).calc(input)) as _,
+        )
+    }
+
+    fn calc(&self, input: &[ColumnarValue]) -> Result<ColumnarValue, DataFusionError> {
+        assert_eq!(input.len(), 1);
+
+        let value_array = extract_array(&input[0])?;
+
+        if self.nearest == 0.0 {
+            let values = value_array.as_primitive::<Float64Type>();
+            let result = compute::unary::<_, _, Float64Type>(values, |a| a.round());
+            Ok(ColumnarValue::Array(Arc::new(result) as _))
+        } else {
+            let values = value_array.as_primitive::<Float64Type>();
+            let nearest = self.nearest;
+            let result =
+                compute::unary::<_, _, Float64Type>(values, |a| ((a / nearest).round() * nearest));
+            Ok(ColumnarValue::Array(Arc::new(result) as _))
+        }
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use datatypes::arrow::array::Float64Array;
+
+    use super::*;
+
+    fn test_round_f64(value: Vec<f64>, nearest: f64, expected: Vec<f64>) {
+        let round_udf = Round::scalar_udf(nearest);
+        let input = vec![ColumnarValue::Array(Arc::new(Float64Array::from(value)))];
+        let result = round_udf.invoke_batch(&input, 1).unwrap();
+        let result_array = extract_array(&result).unwrap();
+        assert_eq!(result_array.len(), 1);
+        assert_eq!(
+            result_array.as_primitive::<Float64Type>().values(),
+            &expected
+        );
+    }
+
+    #[test]
+    fn test_round() {
+        test_round_f64(vec![123.456], 0.001, vec![123.456]);
+        test_round_f64(vec![123.456], 0.01, vec![123.46000000000001]);
+        test_round_f64(vec![123.456], 0.1, vec![123.5]);
+        test_round_f64(vec![123.456], 0.0, vec![123.0]);
+        test_round_f64(vec![123.456], 1.0, vec![123.0]);
+        test_round_f64(vec![123.456], 10.0, vec![120.0]);
+        test_round_f64(vec![123.456], 100.0, vec![100.0]);
+        test_round_f64(vec![123.456], 105.0, vec![105.0]);
+        test_round_f64(vec![123.456], 1000.0, vec![0.0]);
+    }
+}
--- a/src/query/src/datafusion/planner.rs
+++ b/src/query/src/datafusion/planner.rs
@@ -18,7 +18,9 @@ use std::sync::Arc;

 use arrow_schema::DataType;
 use catalog::table_source::DfTableSourceProvider;
-use common_function::aggr::{UddSketchState, UDDSKETCH_STATE_NAME};
+use common_function::aggr::{
+    HllState, UddSketchState, HLL_MERGE_NAME, HLL_NAME, UDDSKETCH_STATE_NAME,
+};
 use common_function::scalars::udf::create_udf;
 use common_query::logical_plan::create_aggregate_function;
 use datafusion::common::TableReference;
@@ -169,6 +171,12 @@ impl ContextProvider for DfContextProviderAdapter {
        if name == UDDSKETCH_STATE_NAME {
            return Some(Arc::new(UddSketchState::udf_impl()));
        }
+        if name == HLL_NAME {
+            return Some(Arc::new(HllState::state_udf_impl()));
+        }
+        if name == HLL_MERGE_NAME {
+            return Some(Arc::new(HllState::merge_udf_impl()));
+        }

        self.engine_state.aggregate_function(name).map_or_else(
            || self.session_state.aggregate_functions().get(name).cloned(),
--- a/src/query/src/promql/planner.rs
+++ b/src/query/src/promql/planner.rs
@@ -52,7 +52,7 @@ use promql::extension_plan::{
 use promql::functions::{
    AbsentOverTime, AvgOverTime, Changes, CountOverTime, Delta, Deriv, HoltWinters, IDelta,
    Increase, LastOverTime, MaxOverTime, MinOverTime, PredictLinear, PresentOverTime,
-    QuantileOverTime, Rate, Resets, StddevOverTime, StdvarOverTime, SumOverTime,
+    QuantileOverTime, Rate, Resets, Round, StddevOverTime, StdvarOverTime, SumOverTime,
 };
 use promql_parser::label::{MatchOp, Matcher, Matchers, METRIC_NAME};
 use promql_parser::parser::token::TokenType;
@@ -200,10 +200,9 @@ impl PromPlanner {
            PromExpr::Paren(ParenExpr { expr }) => {
                self.prom_expr_to_plan(expr, session_state).await?
            }
-            PromExpr::Subquery(SubqueryExpr { .. }) => UnsupportedExprSnafu {
-                name: "Prom Subquery",
+            PromExpr::Subquery(expr) => {
+                self.prom_subquery_expr_to_plan(session_state, expr).await?
            }
-            .fail()?,
            PromExpr::NumberLiteral(lit) => self.prom_number_lit_to_plan(lit)?,
            PromExpr::StringLiteral(lit) => self.prom_string_lit_to_plan(lit)?,
            PromExpr::VectorSelector(selector) => {
@@ -218,6 +217,48 @@ impl PromPlanner {
        Ok(res)
    }

+    async fn prom_subquery_expr_to_plan(
+        &mut self,
+        session_state: &SessionState,
+        subquery_expr: &SubqueryExpr,
+    ) -> Result<LogicalPlan> {
+        let SubqueryExpr {
+            expr, range, step, ..
+        } = subquery_expr;
+
+        let current_interval = self.ctx.interval;
+        if let Some(step) = step {
+            self.ctx.interval = step.as_millis() as _;
+        }
+        let current_start = self.ctx.start;
+        self.ctx.start -= range.as_millis() as i64 - self.ctx.interval;
+        let input = self.prom_expr_to_plan(expr, session_state).await?;
+        self.ctx.interval = current_interval;
+        self.ctx.start = current_start;
+
+        ensure!(!range.is_zero(), ZeroRangeSelectorSnafu);
+        let range_ms = range.as_millis() as _;
+        self.ctx.range = Some(range_ms);
+
+        let manipulate = RangeManipulate::new(
+            self.ctx.start,
+            self.ctx.end,
+            self.ctx.interval,
+            range_ms,
+            self.ctx
+                .time_index_column
+                .clone()
+                .expect("time index should be set in `setup_context`"),
+            self.ctx.field_columns.clone(),
+            input,
+        )
+        .context(DataFusionPlanningSnafu)?;
+
+        Ok(LogicalPlan::Extension(Extension {
+            node: Arc::new(manipulate),
+        }))
+    }
+
    async fn prom_aggr_expr_to_plan(
        &mut self,
        session_state: &SessionState,
@@ -441,6 +482,7 @@ impl PromPlanner {
                    // if left plan or right plan tag is empty, means case like `scalar(...) + host` or `host + scalar(...)`
                    // under this case we only join on time index
                    left_context.tag_columns.is_empty() || right_context.tag_columns.is_empty(),
+                    modifier,
                )?;
                let join_plan_schema = join_plan.schema().clone();

@@ -1468,6 +1510,20 @@ impl PromPlanner {

                ScalarFunc::GeneratedExpr
            }
+            "round" => {
+                let nearest = match other_input_exprs.pop_front() {
+                    Some(DfExpr::Literal(ScalarValue::Float64(Some(t)))) => t,
+                    Some(DfExpr::Literal(ScalarValue::Int64(Some(t)))) => t as f64,
+                    None => 0.0,
+                    other => UnexpectedPlanExprSnafu {
+                        desc: format!("expected f64 literal as t, but found {:?}", other),
+                    }
+                    .fail()?,
+                };
+
+                ScalarFunc::DataFusionUdf(Arc::new(Round::scalar_udf(nearest)))
+            }
+
            _ => {
                if let Some(f) = session_state.scalar_functions().get(func.name) {
                    ScalarFunc::DataFusionBuiltin(f.clone())
@@ -1674,7 +1730,7 @@ impl PromPlanner {
        ensure!(
            !src_labels.is_empty(),
            FunctionInvalidArgumentSnafu {
-                fn_name: "label_join",
+                fn_name: "label_join"
            }
        );

@@ -2121,24 +2177,49 @@ impl PromPlanner {
        left_time_index_column: Option<String>,
        right_time_index_column: Option<String>,
        only_join_time_index: bool,
+        modifier: &Option<BinModifier>,
    ) -> Result<LogicalPlan> {
        let mut left_tag_columns = if only_join_time_index {
-            vec![]
+            BTreeSet::new()
        } else {
            self.ctx
                .tag_columns
                .iter()
-                .map(Column::from_name)
-                .collect::<Vec<_>>()
+                .cloned()
+                .collect::<BTreeSet<_>>()
        };
        let mut right_tag_columns = left_tag_columns.clone();

+        // apply modifier
+        if let Some(modifier) = modifier {
+            // apply label modifier
+            if let Some(matching) = &modifier.matching {
+                match matching {
+                    // keeps columns mentioned in `on`
+                    LabelModifier::Include(on) => {
+                        let mask = on.labels.iter().cloned().collect::<BTreeSet<_>>();
+                        left_tag_columns = left_tag_columns.intersection(&mask).cloned().collect();
+                        right_tag_columns =
+                            right_tag_columns.intersection(&mask).cloned().collect();
+                    }
+                    // removes columns memtioned in `ignoring`
+                    LabelModifier::Exclude(ignoring) => {
+                        // doesn't check existence of label
+                        for label in &ignoring.labels {
+                            let _ = left_tag_columns.remove(label);
+                            let _ = right_tag_columns.remove(label);
+                        }
+                    }
+                }
+            }
+        }
+
        // push time index column if it exists
        if let (Some(left_time_index_column), Some(right_time_index_column)) =
            (left_time_index_column, right_time_index_column)
        {
-            left_tag_columns.push(Column::from_name(left_time_index_column));
-            right_tag_columns.push(Column::from_name(right_time_index_column));
+            left_tag_columns.insert(left_time_index_column);
+            right_tag_columns.insert(right_time_index_column);
        }

        let right = LogicalPlanBuilder::from(right)
@@ -2154,7 +2235,16 @@ impl PromPlanner {
            .join(
                right,
                JoinType::Inner,
-                (left_tag_columns, right_tag_columns),
+                (
+                    left_tag_columns
+                        .into_iter()
+                        .map(Column::from_name)
+                        .collect::<Vec<_>>(),
+                    right_tag_columns
+                        .into_iter()
+                        .map(Column::from_name)
+                        .collect::<Vec<_>>(),
+                ),
                None,
            )
            .context(DataFusionPlanningSnafu)?
@@ -3340,6 +3430,59 @@ mod test {
        indie_query_plan_compare(query, expected).await;
    }

+    #[tokio::test]
+    async fn test_hash_join() {
+        let mut eval_stmt = EvalStmt {
+            expr: PromExpr::NumberLiteral(NumberLiteral { val: 1.0 }),
+            start: UNIX_EPOCH,
+            end: UNIX_EPOCH
+                .checked_add(Duration::from_secs(100_000))
+                .unwrap(),
+            interval: Duration::from_secs(5),
+            lookback_delta: Duration::from_secs(1),
+        };
+
+        let case = r#"http_server_requests_seconds_sum{uri="/accounts/login"} / ignoring(kubernetes_pod_name,kubernetes_namespace) http_server_requests_seconds_count{uri="/accounts/login"}"#;
+
+        let prom_expr = parser::parse(case).unwrap();
+        eval_stmt.expr = prom_expr;
+        let table_provider = build_test_table_provider_with_fields(
+            &[
+                (
+                    DEFAULT_SCHEMA_NAME.to_string(),
+                    "http_server_requests_seconds_sum".to_string(),
+                ),
+                (
+                    DEFAULT_SCHEMA_NAME.to_string(),
+                    "http_server_requests_seconds_count".to_string(),
+                ),
+            ],
+            &["uri", "kubernetes_namespace", "kubernetes_pod_name"],
+        )
+        .await;
+        // Should be ok
+        let plan = PromPlanner::stmt_to_plan(table_provider, &eval_stmt, &build_session_state())
+            .await
+            .unwrap();
+        let expected = r#"Projection: http_server_requests_seconds_count.uri, http_server_requests_seconds_count.kubernetes_namespace, http_server_requests_seconds_count.kubernetes_pod_name, http_server_requests_seconds_count.greptime_timestamp, http_server_requests_seconds_sum.greptime_value / http_server_requests_seconds_count.greptime_value AS http_server_requests_seconds_sum.greptime_value / http_server_requests_seconds_count.greptime_value
+  Inner Join: http_server_requests_seconds_sum.greptime_timestamp = http_server_requests_seconds_count.greptime_timestamp, http_server_requests_seconds_sum.uri = http_server_requests_seconds_count.uri
+    SubqueryAlias: http_server_requests_seconds_sum
+      PromInstantManipulate: range=[0..100000000], lookback=[1000], interval=[5000], time index=[greptime_timestamp]
+        PromSeriesNormalize: offset=[0], time index=[greptime_timestamp], filter NaN: [false]
+          PromSeriesDivide: tags=["uri", "kubernetes_namespace", "kubernetes_pod_name"]
+            Sort: http_server_requests_seconds_sum.uri DESC NULLS LAST, http_server_requests_seconds_sum.kubernetes_namespace DESC NULLS LAST, http_server_requests_seconds_sum.kubernetes_pod_name DESC NULLS LAST, http_server_requests_seconds_sum.greptime_timestamp DESC NULLS LAST
+              Filter: http_server_requests_seconds_sum.uri = Utf8("/accounts/login") AND http_server_requests_seconds_sum.greptime_timestamp >= TimestampMillisecond(-1000, None) AND http_server_requests_seconds_sum.greptime_timestamp <= TimestampMillisecond(100001000, None)
+                TableScan: http_server_requests_seconds_sum
+    SubqueryAlias: http_server_requests_seconds_count
+      PromInstantManipulate: range=[0..100000000], lookback=[1000], interval=[5000], time index=[greptime_timestamp]
+        PromSeriesNormalize: offset=[0], time index=[greptime_timestamp], filter NaN: [false]
+          PromSeriesDivide: tags=["uri", "kubernetes_namespace", "kubernetes_pod_name"]
+            Sort: http_server_requests_seconds_count.uri DESC NULLS LAST, http_server_requests_seconds_count.kubernetes_namespace DESC NULLS LAST, http_server_requests_seconds_count.kubernetes_pod_name DESC NULLS LAST, http_server_requests_seconds_count.greptime_timestamp DESC NULLS LAST
+              Filter: http_server_requests_seconds_count.uri = Utf8("/accounts/login") AND http_server_requests_seconds_count.greptime_timestamp >= TimestampMillisecond(-1000, None) AND http_server_requests_seconds_count.greptime_timestamp <= TimestampMillisecond(100001000, None)
+                TableScan: http_server_requests_seconds_count"#;
+        assert_eq!(plan.to_string(), expected);
+    }
+
    #[tokio::test]
    async fn test_nested_histogram_quantile() {
        let mut eval_stmt = EvalStmt {
--- a/src/query/src/query_engine/default_serializer.rs
+++ b/src/query/src/query_engine/default_serializer.rs
@@ -15,7 +15,7 @@
 use std::sync::Arc;

 use common_error::ext::BoxedError;
-use common_function::aggr::UddSketchState;
+use common_function::aggr::{HllState, UddSketchState};
 use common_function::function_registry::FUNCTION_REGISTRY;
 use common_function::scalars::udf::create_udf;
 use common_query::error::RegisterUdfSnafu;
@@ -127,6 +127,8 @@ impl SubstraitPlanDecoder for DefaultPlanDecoder {
                .register_udf(udf)
                .context(RegisterUdfSnafu { name: func.name() })?;
            let _ = session_state.register_udaf(Arc::new(UddSketchState::udf_impl()));
+            let _ = session_state.register_udaf(Arc::new(HllState::state_udf_impl()));
+            let _ = session_state.register_udaf(Arc::new(HllState::merge_udf_impl()));
        }
        let logical_plan = DFLogicalSubstraitConvertor
            .decode(message, session_state)
--- a/src/query/src/sql/show_create_table.rs
+++ b/src/query/src/sql/show_create_table.rs
@@ -327,7 +327,7 @@ CREATE TABLE IF NOT EXISTS "system_metrics" (
  "host" STRING NULL INVERTED INDEX,
  "cpu" DOUBLE NULL,
  "disk" FLOAT NULL,
-  "msg" STRING NULL FULLTEXT WITH(analyzer = 'English', case_sensitive = 'false'),
+  "msg" STRING NULL FULLTEXT INDEX WITH(analyzer = 'English', case_sensitive = 'false'),
  "ts" TIMESTAMP(3) NOT NULL DEFAULT current_timestamp(),
  TIME INDEX ("ts"),
  PRIMARY KEY ("id", "host")
--- a/src/servers/Cargo.toml
+++ b/src/servers/Cargo.toml
@@ -65,6 +65,7 @@ http-body = "1"
 humantime.workspace = true
 humantime-serde.workspace = true
 hyper = { workspace = true, features = ["full"] }
+indexmap = "2.7"
 influxdb_line_protocol = { git = "https://github.com/evenyag/influxdb_iox", branch = "feat/line-protocol" }
 itertools.workspace = true
 jsonb.workspace = true
--- a/src/servers/dashboard/VERSION
+++ b/src/servers/dashboard/VERSION
@@ -1 +1 @@
-v0.7.10
+v0.7.11
--- a/src/servers/src/http/result/prometheus_resp.rs
+++ b/src/servers/src/http/result/prometheus_resp.rs
@@ -13,7 +13,7 @@
 // limitations under the License.

 //! prom supply the prometheus HTTP API Server compliance
-use std::collections::{BTreeMap, HashMap};
+use std::collections::HashMap;

 use axum::http::HeaderValue;
 use axum::response::{IntoResponse, Response};
@@ -25,6 +25,7 @@ use common_recordbatch::RecordBatches;
 use datatypes::prelude::ConcreteDataType;
 use datatypes::scalars::ScalarVector;
 use datatypes::vectors::{Float64Vector, StringVector, TimestampMillisecondVector};
+use indexmap::IndexMap;
 use promql_parser::label::METRIC_NAME;
 use promql_parser::parser::value::ValueType;
 use serde::{Deserialize, Serialize};
@@ -229,7 +230,9 @@ impl PrometheusJsonResponse {
        })?;

        let metric_name = (METRIC_NAME, metric_name.as_str());
-        let mut buffer = BTreeMap::<Vec<(&str, &str)>, Vec<(f64, String)>>::new();
+        // Preserves the order of output tags.
+        // Tag order matters, e.g., after sorc and sort_desc, the output order must be kept.
+        let mut buffer = IndexMap::<Vec<(&str, &str)>, Vec<(f64, String)>>::new();

        let schema = batches.schema();
        for batch in batches.iter() {
--- a/src/sql/src/statements/create.rs
+++ b/src/sql/src/statements/create.rs
@@ -156,9 +156,9 @@ impl Display for Column {
        if let Some(fulltext_options) = &self.extensions.fulltext_index_options {
            if !fulltext_options.is_empty() {
                let options = fulltext_options.kv_pairs();
-                write!(f, " FULLTEXT WITH({})", format_list_comma!(options))?;
+                write!(f, " FULLTEXT INDEX WITH({})", format_list_comma!(options))?;
            } else {
-                write!(f, " FULLTEXT")?;
+                write!(f, " FULLTEXT INDEX")?;
            }
        }

--- a/src/store-api/src/region_request.rs
+++ b/src/store-api/src/region_request.rs
@@ -222,7 +222,12 @@ fn make_region_creates(creates: CreateRequests) -> Result<Vec<(RegionId, RegionR

 fn parse_region_drop(drop: DropRequest) -> Result<(RegionId, RegionDropRequest)> {
    let region_id = drop.region_id.into();
-    Ok((region_id, RegionDropRequest {}))
+    Ok((
+        region_id,
+        RegionDropRequest {
+            fast_path: drop.fast_path,
+        },
+    ))
 }

 fn make_region_drop(drop: DropRequest) -> Result<Vec<(RegionId, RegionRequest)>> {
@@ -397,8 +402,10 @@ impl RegionCreateRequest {
    }
 }

-#[derive(Debug, Clone, Default)]
-pub struct RegionDropRequest {}
+#[derive(Debug, Clone)]
+pub struct RegionDropRequest {
+    pub fast_path: bool,
+}

 /// Open region request.
 #[derive(Debug, Clone)]
--- a/tests-integration/src/standalone.rs
+++ b/tests-integration/src/standalone.rs
@@ -40,7 +40,7 @@ use common_procedure::options::ProcedureConfig;
 use common_procedure::ProcedureManagerRef;
 use common_wal::config::{DatanodeWalConfig, MetasrvWalConfig};
 use datanode::datanode::DatanodeBuilder;
-use flow::FlownodeBuilder;
+use flow::{FlownodeBuilder, FrontendClient};
 use frontend::instance::builder::FrontendBuilder;
 use frontend::instance::{FrontendInstance, Instance, StandaloneDatanodeManager};
 use meta_srv::metasrv::{FLOW_ID_SEQ, TABLE_ID_SEQ};
@@ -164,12 +164,15 @@ impl GreptimeDbStandaloneBuilder {
            Some(procedure_manager.clone()),
        );

+        let fe_server_addr = opts.frontend_options().grpc.bind_addr.clone();
+        let frontend_client = FrontendClient::from_static_grpc_addr(fe_server_addr);
        let flow_builder = FlownodeBuilder::new(
            Default::default(),
            plugins.clone(),
            table_metadata_manager.clone(),
            catalog_manager.clone(),
            flow_metadata_manager.clone(),
+            Arc::new(frontend_client),
        );
        let flownode = Arc::new(flow_builder.build().await.unwrap());

--- a/tests/cases/standalone/common/aggregate/hll.result
+++ b/tests/cases/standalone/common/aggregate/hll.result
@@ -0,0 +1,84 @@
+CREATE TABLE test_hll (
+    `id` INT PRIMARY KEY,
+    `value` STRING,
+    `ts` timestamp time index default now()
+);
+
+Affected Rows: 0
+
+INSERT INTO test_hll (`id`, `value`) VALUES
+    (1, "a"),
+    (2, "b"),
+    (5, "e"),
+    (6, "f"),
+    (7, "g"),
+    (8, "h"),
+    (9, "i"),
+    (10, "j"),
+    (11, "i"),
+    (12, "j"),
+    (13, "i"),
+    (14, "n"),
+    (15, "o");
+
+Affected Rows: 13
+
+select hll_count(hll(`value`)) from test_hll;
+
+--------------------------------+
+| hll_count(hll(test_hll.value)) |
+--------------------------------+
+| 10                             |
+--------------------------------+
+
+INSERT INTO test_hll (`id`, `value`) VALUES
+    (16, "b"),
+    (17, "i"),
+    (18, "j"),
+    (19, "s"),
+    (20, "t");
+
+Affected Rows: 5
+
+select hll_count(hll(`value`)) from test_hll;
+
+--------------------------------+
+| hll_count(hll(test_hll.value)) |
+--------------------------------+
+| 12                             |
+--------------------------------+
+
+create table test_hll_merge (
+    `id` INT PRIMARY KEY,
+    `state` BINARY,
+    `ts` timestamp time index default now()
+);
+
+Affected Rows: 0
+
+insert into test_hll_merge (`id`, `state`)
+select 1, hll(`value`) from test_hll;
+
+Affected Rows: 1
+
+insert into test_hll_merge (`id`, `state`)
+select 2, hll(`value`) from test_hll;
+
+Affected Rows: 1
+
+select hll_count(hll_merge(`state`)) from test_hll_merge;
+
+--------------------------------------------+
+| hll_count(hll_merge(test_hll_merge.state)) |
+--------------------------------------------+
+| 12                                         |
+--------------------------------------------+
+
+drop table test_hll;
+
+Affected Rows: 0
+
+drop table test_hll_merge;
+
+Affected Rows: 0
+
--- a/tests/cases/standalone/common/aggregate/hll.sql
+++ b/tests/cases/standalone/common/aggregate/hll.sql
@@ -0,0 +1,49 @@
+CREATE TABLE test_hll (
+    `id` INT PRIMARY KEY,
+    `value` STRING,
+    `ts` timestamp time index default now()
+);
+
+INSERT INTO test_hll (`id`, `value`) VALUES
+    (1, "a"),
+    (2, "b"),
+    (5, "e"),
+    (6, "f"),
+    (7, "g"),
+    (8, "h"),
+    (9, "i"),
+    (10, "j"),
+    (11, "i"),
+    (12, "j"),
+    (13, "i"),
+    (14, "n"),
+    (15, "o");
+
+select hll_count(hll(`value`)) from test_hll;
+
+INSERT INTO test_hll (`id`, `value`) VALUES
+    (16, "b"),
+    (17, "i"),
+    (18, "j"),
+    (19, "s"),
+    (20, "t");
+
+select hll_count(hll(`value`)) from test_hll;
+
+create table test_hll_merge (
+    `id` INT PRIMARY KEY,
+    `state` BINARY,
+    `ts` timestamp time index default now()
+);
+
+insert into test_hll_merge (`id`, `state`)
+select 1, hll(`value`) from test_hll;
+
+insert into test_hll_merge (`id`, `state`)
+select 2, hll(`value`) from test_hll;
+
+select hll_count(hll_merge(`state`)) from test_hll_merge;
+
+drop table test_hll;
+
+drop table test_hll_merge;
--- a/tests/cases/standalone/common/alter/change_col_fulltext_options.result
+++ b/tests/cases/standalone/common/alter/change_col_fulltext_options.result
@@ -79,20 +79,20 @@ SELECT * FROM test WHERE MATCHES(message, 'hello') ORDER BY message;
 -- SQLNESS ARG restart=true
 SHOW CREATE TABLE test;

-+-------+---------------------------------------------------------------------------------------+
-| Table | Create Table                                                                          |
-+-------+---------------------------------------------------------------------------------------+
-| test  | CREATE TABLE IF NOT EXISTS "test" (                                                   |
-|       |   "message" STRING NULL FULLTEXT WITH(analyzer = 'Chinese', case_sensitive = 'true'), |
-|       |   "time" TIMESTAMP(3) NOT NULL,                                                       |
-|       |   TIME INDEX ("time")                                                                 |
-|       | )                                                                                     |
-|       |                                                                                       |
-|       | ENGINE=mito                                                                           |
-|       | WITH(                                                                                 |
-|       |   append_mode = 'true'                                                                |
-|       | )                                                                                     |
-+-------+---------------------------------------------------------------------------------------+
+-------+---------------------------------------------------------------------------------------------+
+| Table | Create Table                                                                                |
+-------+---------------------------------------------------------------------------------------------+
+| test  | CREATE TABLE IF NOT EXISTS "test" (                                                         |
+|       |   "message" STRING NULL FULLTEXT INDEX WITH(analyzer = 'Chinese', case_sensitive = 'true'), |
+|       |   "time" TIMESTAMP(3) NOT NULL,                                                             |
+|       |   TIME INDEX ("time")                                                                       |
+|       | )                                                                                           |
+|       |                                                                                             |
+|       | ENGINE=mito                                                                                 |
+|       | WITH(                                                                                       |
+|       |   append_mode = 'true'                                                                      |
+|       | )                                                                                           |
+-------+---------------------------------------------------------------------------------------------+

 SHOW INDEX FROM test;

@@ -138,20 +138,20 @@ Affected Rows: 0

 SHOW CREATE TABLE test;

-+-------+---------------------------------------------------------------------------------------+
-| Table | Create Table                                                                          |
-+-------+---------------------------------------------------------------------------------------+
-| test  | CREATE TABLE IF NOT EXISTS "test" (                                                   |
-|       |   "message" STRING NULL FULLTEXT WITH(analyzer = 'Chinese', case_sensitive = 'true'), |
-|       |   "time" TIMESTAMP(3) NOT NULL,                                                       |
-|       |   TIME INDEX ("time")                                                                 |
-|       | )                                                                                     |
-|       |                                                                                       |
-|       | ENGINE=mito                                                                           |
-|       | WITH(                                                                                 |
-|       |   append_mode = 'true'                                                                |
-|       | )                                                                                     |
-+-------+---------------------------------------------------------------------------------------+
+-------+---------------------------------------------------------------------------------------------+
+| Table | Create Table                                                                                |
+-------+---------------------------------------------------------------------------------------------+
+| test  | CREATE TABLE IF NOT EXISTS "test" (                                                         |
+|       |   "message" STRING NULL FULLTEXT INDEX WITH(analyzer = 'Chinese', case_sensitive = 'true'), |
+|       |   "time" TIMESTAMP(3) NOT NULL,                                                             |
+|       |   TIME INDEX ("time")                                                                       |
+|       | )                                                                                           |
+|       |                                                                                             |
+|       | ENGINE=mito                                                                                 |
+|       | WITH(                                                                                       |
+|       |   append_mode = 'true'                                                                      |
+|       | )                                                                                           |
+-------+---------------------------------------------------------------------------------------------+

 SHOW INDEX FROM test;

--- a/tests/cases/standalone/common/create/create_with_fulltext.result
+++ b/tests/cases/standalone/common/create/create_with_fulltext.result
@@ -7,18 +7,18 @@ Affected Rows: 0

 SHOW CREATE TABLE log;

-+-------+------------------------------------------------------------------------------------+
-| Table | Create Table                                                                       |
-+-------+------------------------------------------------------------------------------------+
-| log   | CREATE TABLE IF NOT EXISTS "log" (                                                 |
-|       |   "ts" TIMESTAMP(3) NOT NULL,                                                      |
-|       |   "msg" STRING NULL FULLTEXT WITH(analyzer = 'English', case_sensitive = 'false'), |
-|       |   TIME INDEX ("ts")                                                                |
-|       | )                                                                                  |
-|       |                                                                                    |
-|       | ENGINE=mito                                                                        |
-|       |                                                                                    |
-+-------+------------------------------------------------------------------------------------+
+-------+------------------------------------------------------------------------------------------+
+| Table | Create Table                                                                             |
+-------+------------------------------------------------------------------------------------------+
+| log   | CREATE TABLE IF NOT EXISTS "log" (                                                       |
+|       |   "ts" TIMESTAMP(3) NOT NULL,                                                            |
+|       |   "msg" STRING NULL FULLTEXT INDEX WITH(analyzer = 'English', case_sensitive = 'false'), |
+|       |   TIME INDEX ("ts")                                                                      |
+|       | )                                                                                        |
+|       |                                                                                          |
+|       | ENGINE=mito                                                                              |
+|       |                                                                                          |
+-------+------------------------------------------------------------------------------------------+

 DROP TABLE log;

@@ -33,18 +33,18 @@ Affected Rows: 0

 SHOW CREATE TABLE log_with_opts;

-+---------------+-----------------------------------------------------------------------------------+
-| Table         | Create Table                                                                      |
-+---------------+-----------------------------------------------------------------------------------+
-| log_with_opts | CREATE TABLE IF NOT EXISTS "log_with_opts" (                                      |
-|               |   "ts" TIMESTAMP(3) NOT NULL,                                                     |
-|               |   "msg" STRING NULL FULLTEXT WITH(analyzer = 'English', case_sensitive = 'true'), |
-|               |   TIME INDEX ("ts")                                                               |
-|               | )                                                                                 |
-|               |                                                                                   |
-|               | ENGINE=mito                                                                       |
-|               |                                                                                   |
-+---------------+-----------------------------------------------------------------------------------+
+---------------+-----------------------------------------------------------------------------------------+
+| Table         | Create Table                                                                            |
+---------------+-----------------------------------------------------------------------------------------+
+| log_with_opts | CREATE TABLE IF NOT EXISTS "log_with_opts" (                                            |
+|               |   "ts" TIMESTAMP(3) NOT NULL,                                                           |
+|               |   "msg" STRING NULL FULLTEXT INDEX WITH(analyzer = 'English', case_sensitive = 'true'), |
+|               |   TIME INDEX ("ts")                                                                     |
+|               | )                                                                                       |
+|               |                                                                                         |
+|               | ENGINE=mito                                                                             |
+|               |                                                                                         |
+---------------+-----------------------------------------------------------------------------------------+

 DROP TABLE log_with_opts;

@@ -60,19 +60,19 @@ Affected Rows: 0

 SHOW CREATE TABLE log_multi_fulltext_cols;

-+-------------------------+-------------------------------------------------------------------------------------+
-| Table                   | Create Table                                                                        |
-+-------------------------+-------------------------------------------------------------------------------------+
-| log_multi_fulltext_cols | CREATE TABLE IF NOT EXISTS "log_multi_fulltext_cols" (                              |
-|                         |   "ts" TIMESTAMP(3) NOT NULL,                                                       |
-|                         |   "msg" STRING NULL FULLTEXT WITH(analyzer = 'English', case_sensitive = 'false'),  |
-|                         |   "msg2" STRING NULL FULLTEXT WITH(analyzer = 'English', case_sensitive = 'false'), |
-|                         |   TIME INDEX ("ts")                                                                 |
-|                         | )                                                                                   |
-|                         |                                                                                     |
-|                         | ENGINE=mito                                                                         |
-|                         |                                                                                     |
-+-------------------------+-------------------------------------------------------------------------------------+
+-------------------------+-------------------------------------------------------------------------------------------+
+| Table                   | Create Table                                                                              |
+-------------------------+-------------------------------------------------------------------------------------------+
+| log_multi_fulltext_cols | CREATE TABLE IF NOT EXISTS "log_multi_fulltext_cols" (                                    |
+|                         |   "ts" TIMESTAMP(3) NOT NULL,                                                             |
+|                         |   "msg" STRING NULL FULLTEXT INDEX WITH(analyzer = 'English', case_sensitive = 'false'),  |
+|                         |   "msg2" STRING NULL FULLTEXT INDEX WITH(analyzer = 'English', case_sensitive = 'false'), |
+|                         |   TIME INDEX ("ts")                                                                       |
+|                         | )                                                                                         |
+|                         |                                                                                           |
+|                         | ENGINE=mito                                                                               |
+|                         |                                                                                           |
+-------------------------+-------------------------------------------------------------------------------------------+

 DROP TABLE log_multi_fulltext_cols;

--- a/tests/cases/standalone/common/promql/round_fn.result
+++ b/tests/cases/standalone/common/promql/round_fn.result
@@ -0,0 +1,81 @@
+create table cache_hit (
+    ts timestamp time index,
+    job string,
+    greptime_value double,
+    primary key (job)
+);
+
+Affected Rows: 0
+
+insert into cache_hit values
+    (3000, "read", 123.45),
+    (3000, "write", 234.567),
+    (4000, "read", 345.678),
+    (4000, "write", 456.789);
+
+Affected Rows: 4
+
+-- SQLNESS SORT_RESULT 3 1
+tql eval (3, 4, '1s') round(cache_hit, 0.01);
+
+---------------------+----------------------------+-------+
+| ts                  | prom_round(greptime_value) | job   |
+---------------------+----------------------------+-------+
+| 1970-01-01T00:00:03 | 123.45                     | read  |
+| 1970-01-01T00:00:03 | 234.57                     | write |
+| 1970-01-01T00:00:04 | 345.68                     | read  |
+| 1970-01-01T00:00:04 | 456.79                     | write |
+---------------------+----------------------------+-------+
+
+-- SQLNESS SORT_RESULT 3 1
+tql eval (3, 4, '1s') round(cache_hit, 0.1);
+
+---------------------+----------------------------+-------+
+| ts                  | prom_round(greptime_value) | job   |
+---------------------+----------------------------+-------+
+| 1970-01-01T00:00:03 | 123.5                      | read  |
+| 1970-01-01T00:00:03 | 234.60000000000002         | write |
+| 1970-01-01T00:00:04 | 345.70000000000005         | read  |
+| 1970-01-01T00:00:04 | 456.8                      | write |
+---------------------+----------------------------+-------+
+
+-- SQLNESS SORT_RESULT 3 1
+tql eval (3, 4, '1s') round(cache_hit, 1.0);
+
+---------------------+----------------------------+-------+
+| ts                  | prom_round(greptime_value) | job   |
+---------------------+----------------------------+-------+
+| 1970-01-01T00:00:03 | 123.0                      | read  |
+| 1970-01-01T00:00:03 | 235.0                      | write |
+| 1970-01-01T00:00:04 | 346.0                      | read  |
+| 1970-01-01T00:00:04 | 457.0                      | write |
+---------------------+----------------------------+-------+
+
+-- SQLNESS SORT_RESULT 3 1
+tql eval (3, 4, '1s') round(cache_hit);
+
+---------------------+----------------------------+-------+
+| ts                  | prom_round(greptime_value) | job   |
+---------------------+----------------------------+-------+
+| 1970-01-01T00:00:03 | 123.0                      | read  |
+| 1970-01-01T00:00:03 | 235.0                      | write |
+| 1970-01-01T00:00:04 | 346.0                      | read  |
+| 1970-01-01T00:00:04 | 457.0                      | write |
+---------------------+----------------------------+-------+
+
+-- SQLNESS SORT_RESULT 3 1
+tql eval (3, 4, '1s') round(cache_hit, 10.0);
+
+---------------------+----------------------------+-------+
+| ts                  | prom_round(greptime_value) | job   |
+---------------------+----------------------------+-------+
+| 1970-01-01T00:00:03 | 120.0                      | read  |
+| 1970-01-01T00:00:03 | 230.0                      | write |
+| 1970-01-01T00:00:04 | 350.0                      | read  |
+| 1970-01-01T00:00:04 | 460.0                      | write |
+---------------------+----------------------------+-------+
+
+drop table cache_hit;
+
+Affected Rows: 0
+
--- a/tests/cases/standalone/common/promql/round_fn.sql
+++ b/tests/cases/standalone/common/promql/round_fn.sql
@@ -0,0 +1,30 @@
+
+create table cache_hit (
+    ts timestamp time index,
+    job string,
+    greptime_value double,
+    primary key (job)
+);
+
+insert into cache_hit values
+    (3000, "read", 123.45),
+    (3000, "write", 234.567),
+    (4000, "read", 345.678),
+    (4000, "write", 456.789);
+
+-- SQLNESS SORT_RESULT 3 1
+tql eval (3, 4, '1s') round(cache_hit, 0.01);
+
+-- SQLNESS SORT_RESULT 3 1
+tql eval (3, 4, '1s') round(cache_hit, 0.1);
+
+-- SQLNESS SORT_RESULT 3 1
+tql eval (3, 4, '1s') round(cache_hit, 1.0);
+
+-- SQLNESS SORT_RESULT 3 1
+tql eval (3, 4, '1s') round(cache_hit);
+
+-- SQLNESS SORT_RESULT 3 1
+tql eval (3, 4, '1s') round(cache_hit, 10.0);
+
+drop table cache_hit;
--- a/tests/cases/standalone/common/promql/set_operation.result
+++ b/tests/cases/standalone/common/promql/set_operation.result
@@ -638,3 +638,78 @@ drop table cache_miss;

 Affected Rows: 0

+create table cache_hit_with_null_label (
+    ts timestamp time index,
+    job string,
+    null_label string null,
+    greptime_value double,
+    primary key (job, null_label)
+);
+
+Affected Rows: 0
+
+create table cache_miss_with_null_label (
+    ts timestamp time index,
+    job string,
+    null_label string null,
+    greptime_value double,
+    primary key (job, null_label)
+);
+
+Affected Rows: 0
+
+insert into cache_hit_with_null_label values
+    (3000, "read", null, 1.0),
+    (3000, "write", null, 2.0),
+    (4000, "read", null, 3.0),
+    (4000, "write", null, 4.0);
+
+Affected Rows: 4
+
+insert into cache_miss_with_null_label values
+    (3000, "read", null, 1.0),
+    (3000, "write", null, 2.0),
+    (4000, "read", null, 1.0),
+    (4000, "write", null, 2.0);
+
+Affected Rows: 4
+
+-- SQLNESS SORT_RESULT 3 1
+-- null!=null, so it will returns the empty set.
+tql eval (3, 4, '1s') cache_hit_with_null_label / (cache_miss_with_null_label + cache_hit_with_null_label);
+
++
++
+
+-- SQLNESS SORT_RESULT 3 1
+tql eval (3, 4, '1s') cache_hit_with_null_label / ignoring(null_label) (cache_miss_with_null_label + ignoring(null_label) cache_hit_with_null_label);
+
+-------+------------+---------------------+---------------------------------------------------------------------------------------------------------------+
+| job   | null_label | ts                  | lhs.greptime_value / rhs.cache_miss_with_null_label.greptime_value + cache_hit_with_null_label.greptime_value |
+-------+------------+---------------------+---------------------------------------------------------------------------------------------------------------+
+| read  |            | 1970-01-01T00:00:03 | 0.5                                                                                                           |
+| read  |            | 1970-01-01T00:00:04 | 0.75                                                                                                          |
+| write |            | 1970-01-01T00:00:03 | 0.5                                                                                                           |
+| write |            | 1970-01-01T00:00:04 | 0.6666666666666666                                                                                            |
+-------+------------+---------------------+---------------------------------------------------------------------------------------------------------------+
+
+-- SQLNESS SORT_RESULT 3 1
+tql eval (3, 4, '1s') cache_hit_with_null_label / on(job) (cache_miss_with_null_label + on(job) cache_hit_with_null_label);
+
+-------+------------+---------------------+---------------------------------------------------------------------------------------------------------------+
+| job   | null_label | ts                  | lhs.greptime_value / rhs.cache_miss_with_null_label.greptime_value + cache_hit_with_null_label.greptime_value |
+-------+------------+---------------------+---------------------------------------------------------------------------------------------------------------+
+| read  |            | 1970-01-01T00:00:03 | 0.5                                                                                                           |
+| read  |            | 1970-01-01T00:00:04 | 0.75                                                                                                          |
+| write |            | 1970-01-01T00:00:03 | 0.5                                                                                                           |
+| write |            | 1970-01-01T00:00:04 | 0.6666666666666666                                                                                            |
+-------+------------+---------------------+---------------------------------------------------------------------------------------------------------------+
+
+drop table cache_hit_with_null_label;
+
+Affected Rows: 0
+
+drop table cache_miss_with_null_label;
+
+Affected Rows: 0
+
--- a/tests/cases/standalone/common/promql/set_operation.sql
+++ b/tests/cases/standalone/common/promql/set_operation.sql
@@ -295,3 +295,45 @@ tql eval (3, 4, '1s') cache_hit / (cache_miss + cache_hit);
 drop table cache_hit;

 drop table cache_miss;
+
+create table cache_hit_with_null_label (
+    ts timestamp time index,
+    job string,
+    null_label string null,
+    greptime_value double,
+    primary key (job, null_label)
+);
+
+create table cache_miss_with_null_label (
+    ts timestamp time index,
+    job string,
+    null_label string null,
+    greptime_value double,
+    primary key (job, null_label)
+);
+
+insert into cache_hit_with_null_label values
+    (3000, "read", null, 1.0),
+    (3000, "write", null, 2.0),
+    (4000, "read", null, 3.0),
+    (4000, "write", null, 4.0);
+
+insert into cache_miss_with_null_label values
+    (3000, "read", null, 1.0),
+    (3000, "write", null, 2.0),
+    (4000, "read", null, 1.0),
+    (4000, "write", null, 2.0);
+
+-- SQLNESS SORT_RESULT 3 1
+-- null!=null, so it will returns the empty set.
+tql eval (3, 4, '1s') cache_hit_with_null_label / (cache_miss_with_null_label + cache_hit_with_null_label);
+
+-- SQLNESS SORT_RESULT 3 1
+tql eval (3, 4, '1s') cache_hit_with_null_label / ignoring(null_label) (cache_miss_with_null_label + ignoring(null_label) cache_hit_with_null_label);
+
+-- SQLNESS SORT_RESULT 3 1
+tql eval (3, 4, '1s') cache_hit_with_null_label / on(job) (cache_miss_with_null_label + on(job) cache_hit_with_null_label);
+
+drop table cache_hit_with_null_label;
+
+drop table cache_miss_with_null_label;
--- a/tests/cases/standalone/common/promql/simple_histogram.result
+++ b/tests/cases/standalone/common/promql/simple_histogram.result
@@ -295,3 +295,40 @@ drop table histogram3_bucket;

 Affected Rows: 0

+-- test with invalid data (unaligned buckets)
+create table histogram4_bucket (
+    ts timestamp time index,
+    le string,
+    s string,
+    val double,
+    primary key (s, le),
+);
+
+Affected Rows: 0
+
+insert into histogram4_bucket values
+    (2900000, "0.1", "a", 0),
+    (2900000, "1", "a", 10),
+    (2900000, "5", "a", 20),
+    (2900000, "+Inf", "a", 150),
+    (3000000, "0.1", "a", 50),
+    (3000000, "1", "a", 70),
+    (3000000, "5", "a", 120),
+    -- INF here is missing
+;
+
+Affected Rows: 7
+
+tql eval (2900, 3000, '100s') histogram_quantile(0.9, histogram4_bucket);
+
+---------------------+---+-----+
+| ts                  | s | val |
+---------------------+---+-----+
+| 1970-01-01T00:48:20 | a | 5.0 |
+| 1970-01-01T00:50:00 | a | 5.0 |
+---------------------+---+-----+
+
+drop table histogram4_bucket;
+
+Affected Rows: 0
+
--- a/tests/cases/standalone/common/promql/simple_histogram.sql
+++ b/tests/cases/standalone/common/promql/simple_histogram.sql
@@ -163,3 +163,27 @@ insert into histogram3_bucket values
 tql eval (3000, 3005, '3s') histogram_quantile(0.5, sum by(le, s) (rate(histogram3_bucket[5m])));

 drop table histogram3_bucket;
+
+-- test with invalid data (unaligned buckets)
+create table histogram4_bucket (
+    ts timestamp time index,
+    le string,
+    s string,
+    val double,
+    primary key (s, le),
+);
+
+insert into histogram4_bucket values
+    (2900000, "0.1", "a", 0),
+    (2900000, "1", "a", 10),
+    (2900000, "5", "a", 20),
+    (2900000, "+Inf", "a", 150),
+    (3000000, "0.1", "a", 50),
+    (3000000, "1", "a", 70),
+    (3000000, "5", "a", 120),
+    -- INF here is missing
+;
+
+tql eval (2900, 3000, '100s') histogram_quantile(0.9, histogram4_bucket);
+
+drop table histogram4_bucket;
--- a/tests/cases/standalone/common/promql/subquery.result
+++ b/tests/cases/standalone/common/promql/subquery.result
@@ -0,0 +1,65 @@
+create table metric_total (
+    ts timestamp time index,
+    val double,
+);
+
+Affected Rows: 0
+
+insert into metric_total values
+    (0, 1),
+    (10000, 2);
+
+Affected Rows: 2
+
+tql eval (10, 10, '1s') sum_over_time(metric_total[50s:10s]);
+
+---------------------+----------------------------------+
+| ts                  | prom_sum_over_time(ts_range,val) |
+---------------------+----------------------------------+
+| 1970-01-01T00:00:10 | 3.0                              |
+---------------------+----------------------------------+
+
+tql eval (10, 10, '1s') sum_over_time(metric_total[50s:5s]);
+
+---------------------+----------------------------------+
+| ts                  | prom_sum_over_time(ts_range,val) |
+---------------------+----------------------------------+
+| 1970-01-01T00:00:10 | 4.0                              |
+---------------------+----------------------------------+
+
+tql eval (300, 300, '1s') sum_over_time(metric_total[50s:10s]);
+
+---------------------+----------------------------------+
+| ts                  | prom_sum_over_time(ts_range,val) |
+---------------------+----------------------------------+
+| 1970-01-01T00:05:00 | 10.0                             |
+---------------------+----------------------------------+
+
+tql eval (359, 359, '1s') sum_over_time(metric_total[60s:10s]);
+
+---------------------+----------------------------------+
+| ts                  | prom_sum_over_time(ts_range,val) |
+---------------------+----------------------------------+
+| 1970-01-01T00:05:59 | 2.0                              |
+---------------------+----------------------------------+
+
+tql eval (10, 10, '1s') rate(metric_total[20s:10s]);
+
+---------------------+----------------------------+
+| ts                  | prom_rate(ts_range,val,ts) |
+---------------------+----------------------------+
+| 1970-01-01T00:00:10 | 0.1                        |
+---------------------+----------------------------+
+
+tql eval (20, 20, '1s') rate(metric_total[20s:5s]);
+
+---------------------+----------------------------+
+| ts                  | prom_rate(ts_range,val,ts) |
+---------------------+----------------------------+
+| 1970-01-01T00:00:20 | 0.06666666666666667        |
+---------------------+----------------------------+
+
+drop table metric_total;
+
+Affected Rows: 0
+
--- a/tests/cases/standalone/common/promql/subquery.sql
+++ b/tests/cases/standalone/common/promql/subquery.sql
@@ -0,0 +1,22 @@
+create table metric_total (
+    ts timestamp time index,
+    val double,
+);
+
+insert into metric_total values
+    (0, 1),
+    (10000, 2);
+
+tql eval (10, 10, '1s') sum_over_time(metric_total[50s:10s]);
+
+tql eval (10, 10, '1s') sum_over_time(metric_total[50s:5s]);
+
+tql eval (300, 300, '1s') sum_over_time(metric_total[50s:10s]);
+
+tql eval (359, 359, '1s') sum_over_time(metric_total[60s:10s]);
+
+tql eval (10, 10, '1s') rate(metric_total[20s:10s]);
+
+tql eval (20, 20, '1s') rate(metric_total[20s:5s]);
+
+drop table metric_total;
--- a/tests/cases/standalone/common/show/show_create.result
+++ b/tests/cases/standalone/common/show/show_create.result
@@ -373,20 +373,20 @@ Affected Rows: 0

 show create table test_column_constrain_composite_indexes;

-+-----------------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------+
-| Table                                   | Create Table                                                                                                                                                  |
-+-----------------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------+
-| test_column_constrain_composite_indexes | CREATE TABLE IF NOT EXISTS "test_column_constrain_composite_indexes" (                                                                                        |
-|                                         |   "id" INT NULL SKIPPING INDEX WITH(granularity = '10240', type = 'BLOOM') INVERTED INDEX,                                                                    |
-|                                         |   "host" STRING NULL FULLTEXT WITH(analyzer = 'English', case_sensitive = 'false') SKIPPING INDEX WITH(granularity = '10240', type = 'BLOOM') INVERTED INDEX, |
-|                                         |   "ts" TIMESTAMP(3) NOT NULL,                                                                                                                                 |
-|                                         |   TIME INDEX ("ts"),                                                                                                                                          |
-|                                         |   PRIMARY KEY ("host")                                                                                                                                        |
-|                                         | )                                                                                                                                                             |
-|                                         |                                                                                                                                                               |
-|                                         | ENGINE=mito                                                                                                                                                   |
-|                                         |                                                                                                                                                               |
-+-----------------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------+
+-----------------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
+| Table                                   | Create Table                                                                                                                                                        |
+-----------------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+
+| test_column_constrain_composite_indexes | CREATE TABLE IF NOT EXISTS "test_column_constrain_composite_indexes" (                                                                                              |
+|                                         |   "id" INT NULL SKIPPING INDEX WITH(granularity = '10240', type = 'BLOOM') INVERTED INDEX,                                                                          |
+|                                         |   "host" STRING NULL FULLTEXT INDEX WITH(analyzer = 'English', case_sensitive = 'false') SKIPPING INDEX WITH(granularity = '10240', type = 'BLOOM') INVERTED INDEX, |
+|                                         |   "ts" TIMESTAMP(3) NOT NULL,                                                                                                                                       |
+|                                         |   TIME INDEX ("ts"),                                                                                                                                                |
+|                                         |   PRIMARY KEY ("host")                                                                                                                                              |
+|                                         | )                                                                                                                                                                   |
+|                                         |                                                                                                                                                                     |
+|                                         | ENGINE=mito                                                                                                                                                         |
+|                                         |                                                                                                                                                                     |
+-----------------------------------------+---------------------------------------------------------------------------------------------------------------------------------------------------------------------+

 drop table test_column_constrain_composite_indexes;
Author	SHA1	Message	Date
discord9	7f7b974e8a	fix: conn pool leak & placeholder feature so ci can compile	2025-04-10 15:01:07 +08:00
discord9	4875ace0d0	fix: placeholder feature so ci can compile	2025-04-08 14:37:55 +08:00
discord9	a847d96649	fix: time window filter expr use OR	2025-04-07 16:50:17 +08:00
discord9	23a0a54e18	fix: convert timestamp unit too	2025-04-07 16:50:17 +08:00
discord9	78eb8b53f6	fix: quote&more info when time window too many chore: even more warning fix: filter first warn later	2025-04-07 16:50:17 +08:00
discord9	2455f39e8e	fix: subquery&cte time window expr	2025-04-07 16:46:53 +08:00
discord9	7fe0074202	refactor: even finer&limit time window num	2025-04-07 16:46:53 +08:00
discord9	e16bc203d0	feat: basic time window aware	2025-04-07 16:46:53 +08:00
discord9	9a3c26bb0a	metrics: better bucket&longer timeout	2025-04-07 16:46:53 +08:00
discord9	e1ff398c32	fix: timeout	2025-04-07 16:46:53 +08:00
discord9	780e3000de	fix: heartbeat&expire_after unit	2025-04-07 16:46:53 +08:00
discord9	2b5ddf8427	feat: time window in df plan WIP test: found out time window expr chore: pub tests: also unparsed tests: rm dup code feat: frontend client for recording rule fix: bound edgecase WIP WIP feat: rule engine feat: add init options& tmp rerounte to rule fix: dist client get fix: also not handle mirror write in flownode chore: clippy	2025-04-07 16:46:47 +08:00
Weny Xu	904d560175	feat(promql-planner): introduce vector matching binary operation (#5578 ) * feat(promql-planner): support vector matching for binary operation * test: add sqlness tests	2025-02-27 07:39:19 +00:00
Lei, HUANG	765d1277ee	fix(metasrv): clean expired nodes in memory (#5592 ) * fix/frontend-node-state: Refactor NodeInfoKey and Context Handling in Meta Server • Removed unused cluster_id from NodeInfoKey struct. • Updated HeartbeatHandlerGroup to return Context alongside HeartbeatResponse. • Added current_node_info to Context for tracking node information. • Implemented on_node_disconnect in Context to handle node disconnection events, specifically for Frontend roles. • Adjusted register_pusher function to return PusherId directly. • Updated tests to accommodate changes in Context structure. * fix/frontend-node-state: Refactor Heartbeat Handler Context Management Refactored the HeartbeatHandlerGroup::handle method to use a mutable reference for Context instead of passing it by value. This change simplifies the context management by eliminating the need to return the context with the response. Updated the Metasrv implementation to align with this new context handling approach, improving code clarity and reducing unnecessary context cloning. * revert: clean cluster info on disconnect * fix/frontend-node-state: Add Frontend Expiry Listener and Update NodeInfoKey Conversion • Introduced FrontendExpiryListener to manage the expiration of frontend nodes, including its integration with leadership change notifications. • Modified NodeInfoKey conversion to use references, enhancing efficiency and consistency across the codebase. • Updated collect_cluster_info_handler and metasrv to incorporate the new listener and conversion changes. • Added frontend_expiry module to the project structure for better organization and maintainability. * chore: add config for node expiry * add some doc * fix: clippy * fix/frontend-node-state: ### Refactor Node Expiry Handling - Configuration Update: Removed `node_expiry_tick` from `metasrv.example.toml` and `MetasrvOptions` in `metasrv.rs`. - Module Renaming: Renamed `frontend_expiry.rs` to `node_expiry_listener.rs` and updated references in `lib.rs`. - Code Refactoring: Replaced `FrontendExpiryListener` with `NodeExpiryListener` in `node_expiry_listener.rs` and `metasrv.rs`, removing the tick interval and adjusting logic to use a fixed 60-second interval for node expiry checks. * fix/frontend-node-state: Improve logging in `node_expiry_listener.rs` - Enhanced warning message to include peer information when an unrecognized node info key is encountered in `node_expiry_listener.rs`. * docs: update config docs * fix/frontend-node-state: Refactor Context Handling in Heartbeat Services - Updated `HeartbeatHandlerGroup` in `handler.rs` to pass `Context` by value instead of by mutable reference, allowing for more flexible context management. - Modified `Metasrv` implementation in `heartbeat.rs` to clone `Context` when passing to `handle` method, ensuring thread safety and consistency in asynchronous operations.	2025-02-27 06:16:36 +00:00
discord9	ccf42a9d97	fix: flow heartbeat retry (#5600 ) * fix: flow heartbeat retry * fix?: not sure if fixed * chore: per review	2025-02-27 03:58:21 +00:00
Weny Xu	71e2fb895f	feat: introduce `prom_round` fn (#5604 ) * feat: introduce `prom_round` fn * test: add sqlness tests	2025-02-27 03:30:15 +00:00
Ruihang Xia	c9671fd669	feat(promql): implement subquery (#5606 ) * feat: initial implement for promql subquery Signed-off-by: Ruihang Xia <waynestxia@gmail.com> * impl and test Signed-off-by: Ruihang Xia <waynestxia@gmail.com> * refactor Signed-off-by: Ruihang Xia <waynestxia@gmail.com> * fix clippy Signed-off-by: Ruihang Xia <waynestxia@gmail.com> --------- Signed-off-by: Ruihang Xia <waynestxia@gmail.com>	2025-02-27 03:28:04 +00:00
Ruihang Xia	b5efc75aab	feat(promql): ignore invalid input in histogram plan (#5607 ) Signed-off-by: Ruihang Xia <waynestxia@gmail.com>	2025-02-27 03:18:20 +00:00
Weny Xu	c1d18d9980	fix(prom): preserve the order of series in `PromQueryResult` (#5601 ) fix(prom): keep the order of tags	2025-02-26 13:40:09 +00:00
Lei, HUANG	5d9faaaf39	fix(metasrv): reject ddl when metasrv is follower (#5599 ) * fix/reject-ddl-in-follower-metasrv: Add leader check and logging for gRPC requests in `procedure.rs` - Implemented leader verification for `query_procedure_state`, `ddl`, and `procedure_details` gRPC requests in `procedure.rs`. - Added logging with `warn` for requests reaching a non-leader node. - Introduced `ResponseHeader` and `Error::is_not_leader()` to handle non-leader responses. * fix/reject-ddl-in-follower-metasrv: Improve leader address handling in `heartbeat.rs` - Refactor leader address retrieval by renaming `leader` to `leader_addr` for clarity. - Update `make_client` function to use a reference to `leader_addr`. - Enhance logging to include the leader address in the success message for creating a heartbeat stream. * fmt * fix/reject-ddl-in-follower-metasrv: Enhance Leader Check in `procedure.rs` - Updated the leader verification logic in `procedure.rs` to return a failed `MigrateRegionResponse` when the server is not the leader. - Added logging to warn when a migrate request is received by a non-leader server.	2025-02-26 08:10:40 +00:00
ZonaHe	538875abee	feat: update dashboard to v0.7.11 (#5597 ) Co-authored-by: sunchanglong <sunchanglong@users.noreply.github.com>	2025-02-26 07:57:59 +00:00
jeremyhi	5ed09c4584	fix: all heartbeat channel need to check leader (#5593 )	2025-02-25 10:45:30 +00:00
Yingwen	3f6a41eac5	fix: update show create table output for fulltext index (#5591 ) * fix: update full index syntax in show create table * test: update fulltext sqlness result	2025-02-25 09:36:27 +00:00
yihong	ff0dcf12c5	perf: close issue 4974 by do not delete columns when drop logical region about 100 times faster (#5561 ) * perf: do not delete columns when drop logical region in drop database Signed-off-by: yihong0618 <zouzou0208@gmail.com> * fix: make ci happy Signed-off-by: yihong0618 <zouzou0208@gmail.com> * fix: address review comments Signed-off-by: yihong0618 <zouzou0208@gmail.com> * fix: address some comments Signed-off-by: yihong0618 <zouzou0208@gmail.com> * fix: drop stupid comments by copilot Signed-off-by: yihong0618 <zouzou0208@gmail.com> * chore: minor refactor * chore: minor refactor * chore: update grpetime-proto --------- Signed-off-by: yihong0618 <zouzou0208@gmail.com> Co-authored-by: WenyXu <wenymedia@gmail.com>	2025-02-25 09:00:49 +00:00
Yingwen	5b1fca825a	fix: remove cached and uploaded files on failure (#5590 )	2025-02-25 08:51:37 +00:00
Ruihang Xia	7bd108e2be	feat: impl `hll_state`, `hll_merge` and `hll_calc` for incremental distinct counting (#5579 ) * basic impl Signed-off-by: Ruihang Xia <waynestxia@gmail.com> * more tests Signed-off-by: Ruihang Xia <waynestxia@gmail.com> * sqlness test Signed-off-by: Ruihang Xia <waynestxia@gmail.com> * fix clippy Signed-off-by: Ruihang Xia <waynestxia@gmail.com> * update with more test and logs Signed-off-by: Ruihang Xia <waynestxia@gmail.com> * impl Signed-off-by: Ruihang Xia <waynestxia@gmail.com> * impl merge fn Signed-off-by: Ruihang Xia <waynestxia@gmail.com> * rename function names Signed-off-by: Ruihang Xia <waynestxia@gmail.com> --------- Signed-off-by: Ruihang Xia <waynestxia@gmail.com>	2025-02-24 19:07:37 +00:00