chore: added stalled time window range

Signed-off-by: discord9 <discord9@163.com>

Cargo.lock

@@ -2336,7 +2336,6 @@ dependencies = [
  "num-traits",
  "once_cell",
  "paste",
- "promql",
  "s2",
  "serde",
  "serde_json",
@@ -5134,7 +5133,7 @@ dependencies = [
 [[package]]
 name = "greptime-proto"
 version = "0.1.0"
-source = "git+https://github.com/GreptimeTeam/greptime-proto.git?rev=454c52634c3bac27de10bf0d85d5533eed1cf03f#454c52634c3bac27de10bf0d85d5533eed1cf03f"
+source = "git+https://github.com/GreptimeTeam/greptime-proto.git?rev=52083925a15d741c259800a9a54eba3467939180#52083925a15d741c259800a9a54eba3467939180"
 dependencies = [
  "prost 0.13.5",
  "serde",

Cargo.toml

@@ -133,7 +133,7 @@ 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 = "454c52634c3bac27de10bf0d85d5533eed1cf03f" }
+greptime-proto = { git = "https://github.com/GreptimeTeam/greptime-proto.git", rev = "52083925a15d741c259800a9a54eba3467939180" }
 hex = "0.4"
 http = "1"
 humantime = "2.1"

config/config.md

@@ -232,6 +232,7 @@
 | `grpc.bind_addr` | String | `127.0.0.1:4001` | The address to bind the gRPC server. |
 | `grpc.server_addr` | String | `127.0.0.1:4001` | The address advertised to the metasrv, and used for connections from outside the host.<br/>If left empty or unset, the server will automatically use the IP address of the first network interface<br/>on the host, with the same port number as the one specified in `grpc.bind_addr`. |
 | `grpc.runtime_size` | Integer | `8` | The number of server worker threads. |
+| `grpc.flight_compression` | String | `arrow_ipc` | Compression mode for frontend side Arrow IPC service. Available options:<br/>- `none`: disable all compression<br/>- `transport`: only enable gRPC transport compression (zstd)<br/>- `arrow_ipc`: only enable Arrow IPC compression (lz4)<br/>- `all`: enable all compression. |
 | `grpc.tls` | -- | -- | gRPC server TLS options, see `mysql.tls` section. |
 | `grpc.tls.mode` | String | `disable` | TLS mode. |
 | `grpc.tls.cert_path` | String | Unset | Certificate file path. |
@@ -404,6 +405,7 @@
 | `grpc.runtime_size` | Integer | `8` | The number of server worker threads. |
 | `grpc.max_recv_message_size` | String | `512MB` | The maximum receive message size for gRPC server. |
 | `grpc.max_send_message_size` | String | `512MB` | The maximum send message size for gRPC server. |
+| `grpc.flight_compression` | String | `arrow_ipc` | Compression mode for datanode side Arrow IPC service. Available options:<br/>- `none`: disable all compression<br/>- `transport`: only enable gRPC transport compression (zstd)<br/>- `arrow_ipc`: only enable Arrow IPC compression (lz4)<br/>- `all`: enable all compression. |
 | `grpc.tls` | -- | -- | gRPC server TLS options, see `mysql.tls` section. |
 | `grpc.tls.mode` | String | `disable` | TLS mode. |
 | `grpc.tls.cert_path` | String | Unset | Certificate file path. |

config/datanode.example.toml

@@ -44,6 +44,13 @@ runtime_size = 8
 max_recv_message_size = "512MB"
 ## The maximum send message size for gRPC server.
 max_send_message_size = "512MB"
+## Compression mode for datanode side Arrow IPC service. Available options:
+## - `none`: disable all compression
+## - `transport`: only enable gRPC transport compression (zstd)
+## - `arrow_ipc`: only enable Arrow IPC compression (lz4)
+## - `all`: enable all compression.
+## Default to `none`
+flight_compression = "arrow_ipc"
 
 ## gRPC server TLS options, see `mysql.tls` section.
 [grpc.tls]

config/frontend.example.toml

@@ -54,6 +54,13 @@ bind_addr = "127.0.0.1:4001"
 server_addr = "127.0.0.1:4001"
 ## The number of server worker threads.
 runtime_size = 8
+## Compression mode for frontend side Arrow IPC service. Available options:
+## - `none`: disable all compression
+## - `transport`: only enable gRPC transport compression (zstd)
+## - `arrow_ipc`: only enable Arrow IPC compression (lz4)
+## - `all`: enable all compression.
+## Default to `none`
+flight_compression = "arrow_ipc"
 
 ## gRPC server TLS options, see `mysql.tls` section.
 [grpc.tls]

src/client/src/client.rs

@@ -162,12 +162,23 @@ impl Client {
             .as_bytes() as usize
     }
 
-    pub fn make_flight_client(&self) -> Result<FlightClient> {
+    pub fn make_flight_client(
+        &self,
+        send_compression: bool,
+        accept_compression: bool,
+    ) -> Result<FlightClient> {
         let (addr, channel) = self.find_channel()?;
 
-        let client = FlightServiceClient::new(channel)
+        let mut client = FlightServiceClient::new(channel)
             .max_decoding_message_size(self.max_grpc_recv_message_size())
             .max_encoding_message_size(self.max_grpc_send_message_size());
+        // todo(hl): support compression methods.
+        if send_compression {
+            client = client.send_compressed(CompressionEncoding::Zstd);
+        }
+        if accept_compression {
+            client = client.accept_compressed(CompressionEncoding::Zstd);
+        }
 
         Ok(FlightClient { addr, client })
     }

src/client/src/client_manager.rs

@@ -49,7 +49,16 @@ impl NodeManager for NodeClients {
     async fn datanode(&self, datanode: &Peer) -> DatanodeRef {
         let client = self.get_client(datanode).await;
 
-        Arc::new(RegionRequester::new(client))
+        let ChannelConfig {
+            send_compression,
+            accept_compression,
+            ..
+        } = self.channel_manager.config();
+        Arc::new(RegionRequester::new(
+            client,
+            *send_compression,
+            *accept_compression,
+        ))
     }
 
     async fn flownode(&self, flownode: &Peer) -> FlownodeRef {

src/client/src/database.rs

@@ -287,7 +287,7 @@ impl Database {
         let mut request = tonic::Request::new(request);
         Self::put_hints(request.metadata_mut(), hints)?;
 
-        let mut client = self.client.make_flight_client()?;
+        let mut client = self.client.make_flight_client(false, false)?;
 
         let response = client.mut_inner().do_get(request).await.or_else(|e| {
             let tonic_code = e.code();
@@ -409,7 +409,7 @@ impl Database {
             MetadataValue::from_str(db_to_put).context(InvalidTonicMetadataValueSnafu)?,
         );
 
-        let mut client = self.client.make_flight_client()?;
+        let mut client = self.client.make_flight_client(false, false)?;
         let response = client.mut_inner().do_put(request).await?;
         let response = response
             .into_inner()

src/client/src/flow.rs

@@ -12,7 +12,7 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
-use api::v1::flow::{FlowRequest, FlowResponse};
+use api::v1::flow::{DirtyWindowRequest, DirtyWindowRequests, FlowRequest, FlowResponse};
 use api::v1::region::InsertRequests;
 use common_error::ext::BoxedError;
 use common_meta::node_manager::Flownode;
@@ -44,6 +44,16 @@ impl Flownode for FlowRequester {
             .map_err(BoxedError::new)
             .context(common_meta::error::ExternalSnafu)
     }
+
+    async fn handle_mark_window_dirty(
+        &self,
+        req: DirtyWindowRequest,
+    ) -> common_meta::error::Result<FlowResponse> {
+        self.handle_mark_window_dirty(req)
+            .await
+            .map_err(BoxedError::new)
+            .context(common_meta::error::ExternalSnafu)
+    }
 }
 
 impl FlowRequester {
@@ -91,4 +101,20 @@ impl FlowRequester {
             .into_inner();
         Ok(response)
     }
+
+    async fn handle_mark_window_dirty(&self, req: DirtyWindowRequest) -> Result<FlowResponse> {
+        let (addr, mut client) = self.client.raw_flow_client()?;
+        let response = client
+            .handle_mark_dirty_time_window(DirtyWindowRequests {
+                requests: vec![req],
+            })
+            .await
+            .or_else(|e| {
+                let code = e.code();
+                let err: crate::error::Error = e.into();
+                Err(BoxedError::new(err)).context(FlowServerSnafu { addr, code })
+            })?
+            .into_inner();
+        Ok(response)
+    }
 }

src/client/src/region.rs

@@ -46,6 +46,8 @@ use crate::{metrics, Client, Error};
 #[derive(Debug)]
 pub struct RegionRequester {
     client: Client,
+    send_compression: bool,
+    accept_compression: bool,
 }
 
 #[async_trait]
@@ -89,12 +91,18 @@ impl Datanode for RegionRequester {
 }
 
 impl RegionRequester {
-    pub fn new(client: Client) -> Self {
+    pub fn new(client: Client, send_compression: bool, accept_compression: bool) -> Self {
-        Self { client }
+        Self {
+            client,
+            send_compression,
+            accept_compression,
+        }
     }
 
     pub async fn do_get_inner(&self, ticket: Ticket) -> Result<SendableRecordBatchStream> {
-        let mut flight_client = self.client.make_flight_client()?;
+        let mut flight_client = self
+            .client
+            .make_flight_client(self.send_compression, self.accept_compression)?;
         let response = flight_client
             .mut_inner()
             .do_get(ticket)

src/cmd/src/frontend.rs

@@ -364,12 +364,16 @@ impl StartCommand {
 
         // frontend to datanode need not timeout.
         // Some queries are expected to take long time.
-        let channel_config = ChannelConfig {
+        let mut channel_config = ChannelConfig {
             timeout: None,
             tcp_nodelay: opts.datanode.client.tcp_nodelay,
             connect_timeout: Some(opts.datanode.client.connect_timeout),
             ..Default::default()
         };
+        if opts.grpc.flight_compression.transport_compression() {
+            channel_config.accept_compression = true;
+            channel_config.send_compression = true;
+        }
         let client = NodeClients::new(channel_config);
 
         let instance = FrontendBuilder::new(

src/common/function/Cargo.toml

@@ -47,7 +47,6 @@ num = "0.4"
 num-traits = "0.2"
 once_cell.workspace = true
 paste.workspace = true
-promql.workspace = true
 s2 = { version = "0.0.12", optional = true }
 serde.workspace = true
 serde_json.workspace = true

src/common/function/src/adjust_flow.rs

@@ -0,0 +1,90 @@
+// Copyright 2023 Greptime Team
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+use common_macro::admin_fn;
+use common_query::error::{
+    InvalidFuncArgsSnafu, MissingFlowServiceHandlerSnafu, Result, UnsupportedInputDataTypeSnafu,
+};
+use common_query::prelude::Signature;
+use datafusion::logical_expr::Volatility;
+use datatypes::value::{Value, ValueRef};
+use session::context::QueryContextRef;
+use snafu::ensure;
+use store_api::storage::ConcreteDataType;
+
+use crate::handlers::FlowServiceHandlerRef;
+use crate::helper::parse_catalog_flow;
+
+fn adjust_signature() -> Signature {
+    Signature::exact(
+        vec![
+            ConcreteDataType::string_datatype(), // flow name
+            ConcreteDataType::uint64_datatype(), // min_run_interval in seconds
+            ConcreteDataType::uint64_datatype(), // max filter number per query
+        ],
+        Volatility::Immutable,
+    )
+}
+
+#[admin_fn(
+    name = AdjustFlowFunction,
+    display_name = adjust_flow,
+    sig_fn = adjust_signature,
+    ret = uint64
+)]
+pub(crate) async fn adjust_flow(
+    flow_service_handler: &FlowServiceHandlerRef,
+    query_ctx: &QueryContextRef,
+    params: &[ValueRef<'_>],
+) -> Result<Value> {
+    ensure!(
+        params.len() == 3,
+        InvalidFuncArgsSnafu {
+            err_msg: format!(
+                "The length of the args is not correct, expect 3, have: {}",
+                params.len()
+            ),
+        }
+    );
+
+    let (flow_name, min_run_interval, max_filter_num) = match (params[0], params[1], params[2]) {
+        (
+            ValueRef::String(flow_name),
+            ValueRef::UInt64(min_run_interval),
+            ValueRef::UInt64(max_filter_num),
+        ) => (flow_name, min_run_interval, max_filter_num),
+        _ => {
+            return UnsupportedInputDataTypeSnafu {
+                function: "adjust_flow",
+                datatypes: params.iter().map(|v| v.data_type()).collect::<Vec<_>>(),
+            }
+            .fail();
+        }
+    };
+
+    let (catalog_name, flow_name) = parse_catalog_flow(flow_name, query_ctx)?;
+
+    let res = flow_service_handler
+        .adjust(
+            &catalog_name,
+            &flow_name,
+            min_run_interval,
+            max_filter_num as usize,
+            query_ctx.clone(),
+        )
+        .await?;
+    let affected_rows = res.affected_rows;
+
+    Ok(Value::from(affected_rows))
+}

src/common/function/src/admin.rs

@@ -26,6 +26,7 @@ use flush_compact_table::{CompactTableFunction, FlushTableFunction};
 use migrate_region::MigrateRegionFunction;
 use remove_region_follower::RemoveRegionFollowerFunction;
 
+use crate::adjust_flow::AdjustFlowFunction;
 use crate::flush_flow::FlushFlowFunction;
 use crate::function_registry::FunctionRegistry;
 
@@ -43,5 +44,6 @@ impl AdminFunction {
         registry.register_async(Arc::new(FlushTableFunction));
         registry.register_async(Arc::new(CompactTableFunction));
         registry.register_async(Arc::new(FlushFlowFunction));
+        registry.register_async(Arc::new(AdjustFlowFunction));
     }
 }

src/common/function/src/flush_flow.rs

@@ -12,21 +12,19 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
-use common_error::ext::BoxedError;
 use common_macro::admin_fn;
 use common_query::error::{
-    ExecuteSnafu, InvalidFuncArgsSnafu, MissingFlowServiceHandlerSnafu, Result,
+    InvalidFuncArgsSnafu, MissingFlowServiceHandlerSnafu, Result, UnsupportedInputDataTypeSnafu,
-    UnsupportedInputDataTypeSnafu,
 };
 use common_query::prelude::Signature;
 use datafusion::logical_expr::Volatility;
 use datatypes::value::{Value, ValueRef};
 use session::context::QueryContextRef;
-use snafu::{ensure, ResultExt};
+use snafu::ensure;
-use sql::parser::ParserContext;
 use store_api::storage::ConcreteDataType;
 
 use crate::handlers::FlowServiceHandlerRef;
+use crate::helper::parse_catalog_flow;
 
 fn flush_signature() -> Signature {
     Signature::uniform(
@@ -47,20 +45,6 @@ pub(crate) async fn flush_flow(
     query_ctx: &QueryContextRef,
     params: &[ValueRef<'_>],
 ) -> Result<Value> {
-    let (catalog_name, flow_name) = parse_flush_flow(params, query_ctx)?;
-
-    let res = flow_service_handler
-        .flush(&catalog_name, &flow_name, query_ctx.clone())
-        .await?;
-    let affected_rows = res.affected_rows;
-
-    Ok(Value::from(affected_rows))
-}
-
-fn parse_flush_flow(
-    params: &[ValueRef<'_>],
-    query_ctx: &QueryContextRef,
-) -> Result<(String, String)> {
     ensure!(
         params.len() == 1,
         InvalidFuncArgsSnafu {
@@ -70,7 +54,6 @@ fn parse_flush_flow(
             ),
         }
     );
-
     let ValueRef::String(flow_name) = params[0] else {
         return UnsupportedInputDataTypeSnafu {
             function: "flush_flow",
@@ -78,27 +61,14 @@ fn parse_flush_flow(
         }
         .fail();
     };
-    let obj_name = ParserContext::parse_table_name(flow_name, query_ctx.sql_dialect())
+    let (catalog_name, flow_name) = parse_catalog_flow(flow_name, query_ctx)?;
-        .map_err(BoxedError::new)
-        .context(ExecuteSnafu)?;
 
-    let (catalog_name, flow_name) = match &obj_name.0[..] {
+    let res = flow_service_handler
-        [flow_name] => (
+        .flush(&catalog_name, &flow_name, query_ctx.clone())
-            query_ctx.current_catalog().to_string(),
+        .await?;
-            flow_name.value.clone(),
+    let affected_rows = res.affected_rows;
-        ),
+
-        [catalog, flow_name] => (catalog.value.clone(), flow_name.value.clone()),
+    Ok(Value::from(affected_rows))
-        _ => {
-            return InvalidFuncArgsSnafu {
-                err_msg: format!(
-                    "expect flow name to be <catalog>.<flow-name> or <flow-name>, actual: {}",
-                    obj_name
-                ),
-            }
-            .fail()
-        }
-    };
-    Ok((catalog_name, flow_name))
 }
 
 #[cfg(test)]
@@ -154,10 +124,7 @@ mod test {
             ("catalog.flow_name", ("catalog", "flow_name")),
         ];
         for (input, expected) in testcases.iter() {
-            let args = vec![*input];
+            let result = parse_catalog_flow(input, &QueryContext::arc()).unwrap();
-            let args = args.into_iter().map(ValueRef::String).collect::<Vec<_>>();
-
-            let result = parse_flush_flow(&args, &QueryContext::arc()).unwrap();
             assert_eq!(*expected, (result.0.as_str(), result.1.as_str()));
         }
     }

src/common/function/src/function_registry.rs

@@ -39,7 +39,7 @@ use crate::system::SystemFunction;
 
 #[derive(Default)]
 pub struct FunctionRegistry {
-    scalar_functions: RwLock<HashMap<String, ScalarFunctionFactory>>,
+    functions: RwLock<HashMap<String, ScalarFunctionFactory>>,
     async_functions: RwLock<HashMap<String, AsyncFunctionRef>>,
     aggregate_functions: RwLock<HashMap<String, AggregateUDF>>,
 }
@@ -48,7 +48,7 @@ impl FunctionRegistry {
     pub fn register(&self, func: impl Into<ScalarFunctionFactory>) {
         let func = func.into();
         let _ = self
-            .scalar_functions
+            .functions
             .write()
             .unwrap()
             .insert(func.name().to_string(), func);
@@ -87,17 +87,13 @@ impl FunctionRegistry {
             .collect()
     }
 
-    pub fn get_scalar_function(&self, name: &str) -> Option<ScalarFunctionFactory> {
+    #[cfg(test)]
-        self.scalar_functions.read().unwrap().get(name).cloned()
+    pub fn get_function(&self, name: &str) -> Option<ScalarFunctionFactory> {
+        self.functions.read().unwrap().get(name).cloned()
     }
 
     pub fn scalar_functions(&self) -> Vec<ScalarFunctionFactory> {
-        self.scalar_functions
+        self.functions.read().unwrap().values().cloned().collect()
-            .read()
-            .unwrap()
-            .values()
-            .cloned()
-            .collect()
     }
 
     pub fn aggregate_functions(&self) -> Vec<AggregateUDF> {
@@ -148,11 +144,6 @@ pub static FUNCTION_REGISTRY: Lazy<Arc<FunctionRegistry>> = Lazy::new(|| {
     // Approximate functions
     ApproximateFunction::register(&function_registry);
 
-    // PromQL aggregate functions
-    for aggr in promql::functions::aggr_funcs() {
-        function_registry.register_aggr(aggr);
-    }
-
     Arc::new(function_registry)
 });
 
@@ -165,10 +156,10 @@ mod tests {
     fn test_function_registry() {
         let registry = FunctionRegistry::default();
 
-        assert!(registry.get_scalar_function("test_and").is_none());
+        assert!(registry.get_function("test_and").is_none());
         assert!(registry.scalar_functions().is_empty());
         registry.register_scalar(TestAndFunction);
-        let _ = registry.get_scalar_function("test_and").unwrap();
+        let _ = registry.get_function("test_and").unwrap();
         assert_eq!(1, registry.scalar_functions().len());
     }
 }

src/common/function/src/handlers.rs

@@ -87,6 +87,15 @@ pub trait FlowServiceHandler: Send + Sync {
         flow: &str,
         ctx: QueryContextRef,
     ) -> Result<api::v1::flow::FlowResponse>;
+
+    async fn adjust(
+        &self,
+        catalog: &str,
+        flow: &str,
+        min_run_interval_secs: u64,
+        max_filter_num_per_query: usize,
+        ctx: QueryContextRef,
+    ) -> Result<api::v1::flow::FlowResponse>;
 }
 
 pub type TableMutationHandlerRef = Arc<dyn TableMutationHandler>;

src/common/function/src/helper.rs

@@ -12,12 +12,15 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
-use common_query::error::{InvalidInputTypeSnafu, Result};
+use common_error::ext::BoxedError;
+use common_query::error::{ExecuteSnafu, InvalidFuncArgsSnafu, InvalidInputTypeSnafu, Result};
 use common_query::prelude::{Signature, TypeSignature, Volatility};
 use datatypes::prelude::ConcreteDataType;
 use datatypes::types::cast::cast;
 use datatypes::value::ValueRef;
+use session::context::QueryContextRef;
 use snafu::ResultExt;
+use sql::parser::ParserContext;
 
 /// Create a function signature with oneof signatures of interleaving two arguments.
 pub fn one_of_sigs2(args1: Vec<ConcreteDataType>, args2: Vec<ConcreteDataType>) -> Signature {
@@ -43,3 +46,30 @@ pub fn cast_u64(value: &ValueRef) -> Result<Option<u64>> {
         })
         .map(|v| v.as_u64())
 }
+
+pub fn parse_catalog_flow(
+    flow_name: &str,
+    query_ctx: &QueryContextRef,
+) -> Result<(String, String)> {
+    let obj_name = ParserContext::parse_table_name(flow_name, query_ctx.sql_dialect())
+        .map_err(BoxedError::new)
+        .context(ExecuteSnafu)?;
+
+    let (catalog_name, flow_name) = match &obj_name.0[..] {
+        [flow_name] => (
+            query_ctx.current_catalog().to_string(),
+            flow_name.value.clone(),
+        ),
+        [catalog, flow_name] => (catalog.value.clone(), flow_name.value.clone()),
+        _ => {
+            return InvalidFuncArgsSnafu {
+                err_msg: format!(
+                    "expect flow name to be <catalog>.<flow-name> or <flow-name>, actual: {}",
+                    obj_name
+                ),
+            }
+            .fail()
+        }
+    };
+    Ok((catalog_name, flow_name))
+}

src/common/function/src/lib.rs

@@ -15,6 +15,7 @@
 #![feature(let_chains)]
 #![feature(try_blocks)]
 
+mod adjust_flow;
 mod admin;
 mod flush_flow;
 mod macros;

src/common/function/src/state.rs

@@ -148,6 +148,17 @@ impl FunctionState {
             ) -> Result<api::v1::flow::FlowResponse> {
                 todo!()
             }
+
+            async fn adjust(
+                &self,
+                _catalog: &str,
+                _flow: &str,
+                _min_run_interval_secs: u64,
+                _max_filter_num_per_query: usize,
+                _ctx: QueryContextRef,
+            ) -> Result<api::v1::flow::FlowResponse> {
+                todo!()
+            }
         }
 
         Self {

src/common/grpc/src/channel_manager.rs

@@ -296,6 +296,8 @@ pub struct ChannelConfig {
     pub max_recv_message_size: ReadableSize,
     // Max gRPC sending(encoding) message size
     pub max_send_message_size: ReadableSize,
+    pub send_compression: bool,
+    pub accept_compression: bool,
 }
 
 impl Default for ChannelConfig {
@@ -316,6 +318,8 @@ impl Default for ChannelConfig {
             client_tls: None,
             max_recv_message_size: DEFAULT_MAX_GRPC_RECV_MESSAGE_SIZE,
             max_send_message_size: DEFAULT_MAX_GRPC_SEND_MESSAGE_SIZE,
+            send_compression: false,
+            accept_compression: false,
         }
     }
 }
@@ -566,6 +570,8 @@ mod tests {
                 client_tls: None,
                 max_recv_message_size: DEFAULT_MAX_GRPC_RECV_MESSAGE_SIZE,
                 max_send_message_size: DEFAULT_MAX_GRPC_SEND_MESSAGE_SIZE,
+                send_compression: false,
+                accept_compression: false,
             },
             default_cfg
         );
@@ -610,6 +616,8 @@ mod tests {
                 }),
                 max_recv_message_size: DEFAULT_MAX_GRPC_RECV_MESSAGE_SIZE,
                 max_send_message_size: DEFAULT_MAX_GRPC_SEND_MESSAGE_SIZE,
+                send_compression: false,
+                accept_compression: false,
             },
             cfg
         );

src/common/grpc/src/flight.rs

@@ -64,6 +64,7 @@ impl Default for FlightEncoder {
 }
 
 impl FlightEncoder {
+    /// Creates new [FlightEncoder] with compression disabled.
     pub fn with_compression_disabled() -> Self {
         let write_options = writer::IpcWriteOptions::default()
             .try_with_compression(None)

src/common/meta/src/node_manager.rs

@@ -15,7 +15,7 @@
 use std::sync::Arc;
 
 use api::region::RegionResponse;
-use api::v1::flow::{FlowRequest, FlowResponse};
+use api::v1::flow::{DirtyWindowRequest, FlowRequest, FlowResponse};
 use api::v1::region::{InsertRequests, RegionRequest};
 pub use common_base::AffectedRows;
 use common_query::request::QueryRequest;
@@ -42,6 +42,9 @@ pub trait Flownode: Send + Sync {
     async fn handle(&self, request: FlowRequest) -> Result<FlowResponse>;
 
     async fn handle_inserts(&self, request: InsertRequests) -> Result<FlowResponse>;
+
+    /// Handles requests to mark time window as dirty.
+    async fn handle_mark_window_dirty(&self, req: DirtyWindowRequest) -> Result<FlowResponse>;
 }
 
 pub type FlownodeRef = Arc<dyn Flownode>;

src/common/meta/src/test_util.rs

@@ -15,7 +15,7 @@
 use std::sync::Arc;
 
 use api::region::RegionResponse;
-use api::v1::flow::{FlowRequest, FlowResponse};
+use api::v1::flow::{DirtyWindowRequest, FlowRequest, FlowResponse};
 use api::v1::region::{InsertRequests, RegionRequest};
 pub use common_base::AffectedRows;
 use common_query::request::QueryRequest;
@@ -67,6 +67,14 @@ pub trait MockFlownodeHandler: Sync + Send + Clone {
     ) -> Result<FlowResponse> {
         unimplemented!()
     }
+
+    async fn handle_mark_window_dirty(
+        &self,
+        _peer: &Peer,
+        _req: DirtyWindowRequest,
+    ) -> Result<FlowResponse> {
+        unimplemented!()
+    }
 }
 
 /// A mock struct implements [NodeManager] only implement the `datanode` method.
@@ -134,6 +142,10 @@ impl<T: MockFlownodeHandler> Flownode for MockNode<T> {
     async fn handle_inserts(&self, requests: InsertRequests) -> Result<FlowResponse> {
         self.handler.handle_inserts(&self.peer, requests).await
     }
+
+    async fn handle_mark_window_dirty(&self, req: DirtyWindowRequest) -> Result<FlowResponse> {
+        self.handler.handle_mark_window_dirty(&self.peer, req).await
+    }
 }
 
 #[async_trait::async_trait]

src/datanode/src/datanode.rs

@@ -372,6 +372,7 @@ impl DatanodeBuilder {
             opts.max_concurrent_queries,
             //TODO: revaluate the hardcoded timeout on the next version of datanode concurrency limiter.
             Duration::from_millis(100),
+            opts.grpc.flight_compression,
         );
 
         let object_store_manager = Self::build_object_store_manager(&opts.storage).await?;

src/datanode/src/region_server.rs

@@ -50,6 +50,7 @@ use query::QueryEngineRef;
 use servers::error::{self as servers_error, ExecuteGrpcRequestSnafu, Result as ServerResult};
 use servers::grpc::flight::{FlightCraft, FlightRecordBatchStream, TonicStream};
 use servers::grpc::region_server::RegionServerHandler;
+use servers::grpc::FlightCompression;
 use session::context::{QueryContextBuilder, QueryContextRef};
 use snafu::{ensure, OptionExt, ResultExt};
 use store_api::metric_engine_consts::{
@@ -80,6 +81,7 @@ use crate::event_listener::RegionServerEventListenerRef;
 #[derive(Clone)]
 pub struct RegionServer {
     inner: Arc<RegionServerInner>,
+    flight_compression: FlightCompression,
 }
 
 pub struct RegionStat {
@@ -93,6 +95,7 @@ impl RegionServer {
         query_engine: QueryEngineRef,
         runtime: Runtime,
         event_listener: RegionServerEventListenerRef,
+        flight_compression: FlightCompression,
     ) -> Self {
         Self::with_table_provider(
             query_engine,
@@ -101,6 +104,7 @@ impl RegionServer {
             Arc::new(DummyTableProviderFactory),
             0,
             Duration::from_millis(0),
+            flight_compression,
         )
     }
 
@@ -111,6 +115,7 @@ impl RegionServer {
         table_provider_factory: TableProviderFactoryRef,
         max_concurrent_queries: usize,
         concurrent_query_limiter_timeout: Duration,
+        flight_compression: FlightCompression,
     ) -> Self {
         Self {
             inner: Arc::new(RegionServerInner::new(
@@ -123,6 +128,7 @@ impl RegionServer {
                     concurrent_query_limiter_timeout,
                 ),
             )),
+            flight_compression,
         }
     }
 
@@ -536,7 +542,11 @@ impl FlightCraft for RegionServer {
             .trace(tracing_context.attach(info_span!("RegionServer::handle_read")))
             .await?;
 
-        let stream = Box::pin(FlightRecordBatchStream::new(result, tracing_context));
+        let stream = Box::pin(FlightRecordBatchStream::new(
+            result,
+            tracing_context,
+            self.flight_compression,
+        ));
         Ok(Response::new(stream))
     }
 }

src/datanode/src/tests.rs

@@ -28,6 +28,7 @@ use query::dataframe::DataFrame;
 use query::planner::LogicalPlanner;
 use query::query_engine::{DescribeResult, QueryEngineState};
 use query::{QueryEngine, QueryEngineContext};
+use servers::grpc::FlightCompression;
 use session::context::QueryContextRef;
 use store_api::metadata::RegionMetadataRef;
 use store_api::region_engine::{
@@ -97,6 +98,7 @@ pub fn mock_region_server() -> RegionServer {
         Arc::new(MockQueryEngine),
         Runtime::builder().build().unwrap(),
         Box::new(NoopRegionServerEventListener),
+        FlightCompression::default(),
     )
 }
 

src/flow/Cargo.toml

@@ -61,6 +61,7 @@ prost.workspace = true
 query.workspace = true
 rand.workspace = true
 serde.workspace = true
+serde_json.workspace = true
 servers.workspace = true
 session.workspace = true
 smallvec.workspace = true

src/flow/src/adapter.rs

@@ -316,7 +316,7 @@ impl StreamingEngine {
             );
 
             METRIC_FLOW_ROWS
-                .with_label_values(&["out"])
+                .with_label_values(&["out-streaming"])
                 .inc_by(total_rows as u64);
 
             let now = self.tick_manager.tick();

src/flow/src/adapter/flownode_impl.rs

@@ -18,7 +18,7 @@ use std::sync::atomic::AtomicBool;
 use std::sync::Arc;
 
 use api::v1::flow::{
-    flow_request, CreateRequest, DropRequest, FlowRequest, FlowResponse, FlushFlow,
+    flow_request, AdjustFlow, CreateRequest, DropRequest, FlowRequest, FlowResponse, FlushFlow,
 };
 use api::v1::region::InsertRequests;
 use catalog::CatalogManager;
@@ -31,7 +31,9 @@ use common_runtime::JoinHandle;
 use common_telemetry::{error, info, trace, warn};
 use datatypes::value::Value;
 use futures::TryStreamExt;
+use greptime_proto::v1::flow::DirtyWindowRequest;
 use itertools::Itertools;
+use serde::{Deserialize, Serialize};
 use session::context::QueryContextBuilder;
 use snafu::{ensure, IntoError, OptionExt, ResultExt};
 use store_api::storage::{RegionId, TableId};
@@ -46,7 +48,7 @@ use crate::error::{
     IllegalCheckTaskStateSnafu, InsertIntoFlowSnafu, InternalSnafu, JoinTaskSnafu, ListFlowsSnafu,
     NoAvailableFrontendSnafu, SyncCheckTaskSnafu, UnexpectedSnafu,
 };
-use crate::metrics::METRIC_FLOW_TASK_COUNT;
+use crate::metrics::{METRIC_FLOW_ROWS, METRIC_FLOW_TASK_COUNT};
 use crate::repr::{self, DiffRow};
 use crate::{Error, FlowId};
 
@@ -689,6 +691,9 @@ impl FlowEngine for FlowDualEngine {
         let mut to_stream_engine = Vec::with_capacity(request.requests.len());
         let mut to_batch_engine = request.requests;
 
+        let mut batching_row_cnt = 0;
+        let mut streaming_row_cnt = 0;
+
         {
             // not locking this, or recover flows will be starved when also handling flow inserts
             let src_table2flow = self.src_table2flow.read().await;
@@ -698,9 +703,11 @@ impl FlowEngine for FlowDualEngine {
                 let is_in_stream = src_table2flow.in_stream(table_id);
                 let is_in_batch = src_table2flow.in_batch(table_id);
                 if is_in_stream {
+                    streaming_row_cnt += req.rows.as_ref().map(|rs| rs.rows.len()).unwrap_or(0);
                     to_stream_engine.push(req.clone());
                 }
                 if is_in_batch {
+                    batching_row_cnt += req.rows.as_ref().map(|rs| rs.rows.len()).unwrap_or(0);
                     return true;
                 }
                 if !is_in_batch && !is_in_stream {
@@ -713,6 +720,14 @@ impl FlowEngine for FlowDualEngine {
             // can't use drop due to https://github.com/rust-lang/rust/pull/128846
         }
 
+        METRIC_FLOW_ROWS
+            .with_label_values(&["in-streaming"])
+            .inc_by(streaming_row_cnt as u64);
+
+        METRIC_FLOW_ROWS
+            .with_label_values(&["in-batching"])
+            .inc_by(batching_row_cnt as u64);
+
         let streaming_engine = self.streaming_engine.clone();
         let stream_handler: JoinHandle<Result<(), Error>> =
             common_runtime::spawn_global(async move {
@@ -809,6 +824,25 @@ impl common_meta::node_manager::Flownode for FlowDualEngine {
                     ..Default::default()
                 })
             }
+            Some(flow_request::Body::Adjust(AdjustFlow { flow_id, options })) => {
+                #[derive(Debug, Serialize, Deserialize)]
+                struct Options {
+                    min_run_interval_secs: u64,
+                    max_filter_num_per_query: usize,
+                }
+                let options: Options = serde_json::from_str(&options).with_context(|_| {
+                    common_meta::error::DeserializeFromJsonSnafu { input: options }
+                })?;
+                self.batching_engine
+                    .adjust_flow(
+                        flow_id.unwrap().id as u64,
+                        options.min_run_interval_secs,
+                        options.max_filter_num_per_query,
+                    )
+                    .await
+                    .map_err(to_meta_err(snafu::location!()))?;
+                Ok(Default::default())
+            }
             other => common_meta::error::InvalidFlowRequestBodySnafu { body: other }.fail(),
         }
     }
@@ -819,6 +853,11 @@ impl common_meta::node_manager::Flownode for FlowDualEngine {
             .map(|_| Default::default())
             .map_err(to_meta_err(snafu::location!()))
     }
+
+    async fn handle_mark_window_dirty(&self, _req: DirtyWindowRequest) -> MetaResult<FlowResponse> {
+        // todo: implement
+        unimplemented!()
+    }
 }
 
 /// return a function to convert `crate::error::Error` to `common_meta::error::Error`
@@ -926,6 +965,11 @@ impl common_meta::node_manager::Flownode for StreamingEngine {
             .map(|_| Default::default())
             .map_err(to_meta_err(snafu::location!()))
     }
+
+    async fn handle_mark_window_dirty(&self, _req: DirtyWindowRequest) -> MetaResult<FlowResponse> {
+        // todo: implement
+        unimplemented!()
+    }
 }
 
 impl FlowEngine for StreamingEngine {

src/flow/src/batching_mode/engine.rs

@@ -17,6 +17,7 @@
 use std::collections::{BTreeMap, HashMap};
 use std::sync::Arc;
 
+use api::v1::flow::{DirtyWindowRequests, FlowResponse};
 use catalog::CatalogManagerRef;
 use common_error::ext::BoxedError;
 use common_meta::ddl::create_flow::FlowType;
@@ -29,8 +30,7 @@ use common_telemetry::{debug, info};
 use common_time::TimeToLive;
 use query::QueryEngineRef;
 use snafu::{ensure, OptionExt, ResultExt};
-use store_api::storage::RegionId;
+use store_api::storage::{RegionId, TableId};
-use table::metadata::TableId;
 use tokio::sync::{oneshot, RwLock};
 
 use crate::batching_mode::frontend_client::FrontendClient;
@@ -42,6 +42,7 @@ use crate::error::{
     ExternalSnafu, FlowAlreadyExistSnafu, FlowNotFoundSnafu, TableNotFoundMetaSnafu,
     UnexpectedSnafu, UnsupportedSnafu,
 };
+use crate::metrics::METRIC_FLOW_BATCHING_ENGINE_BULK_MARK_TIME_WINDOW_RANGE;
 use crate::{CreateFlowArgs, Error, FlowId, TableName};
 
 /// Batching mode Engine, responsible for driving all the batching mode tasks
@@ -77,6 +78,122 @@ impl BatchingEngine {
         }
     }
 
+    pub async fn handle_mark_dirty_time_window(
+        &self,
+        reqs: DirtyWindowRequests,
+    ) -> Result<FlowResponse, Error> {
+        let table_info_mgr = self.table_meta.table_info_manager();
+
+        let mut group_by_table_id: HashMap<u32, Vec<_>> = HashMap::new();
+        for r in reqs.requests {
+            let tid = TableId::from(r.table_id);
+            let entry = group_by_table_id.entry(tid).or_default();
+            entry.extend(r.dirty_time_ranges);
+        }
+        let tids = group_by_table_id.keys().cloned().collect::<Vec<TableId>>();
+        let table_infos =
+            table_info_mgr
+                .batch_get(&tids)
+                .await
+                .with_context(|_| TableNotFoundMetaSnafu {
+                    msg: format!("Failed to get table info for table ids: {:?}", tids),
+                })?;
+
+        let group_by_table_name = group_by_table_id
+            .into_iter()
+            .filter_map(|(id, rows)| {
+                let table_name = table_infos.get(&id).map(|info| info.table_name());
+                let Some(table_name) = table_name else {
+                    warn!("Failed to get table infos for table id: {:?}", id);
+                    return None;
+                };
+                let table_name = [
+                    table_name.catalog_name,
+                    table_name.schema_name,
+                    table_name.table_name,
+                ];
+                let schema = &table_infos.get(&id).unwrap().table_info.meta.schema;
+                let time_index_unit = schema.column_schemas[schema.timestamp_index.unwrap()]
+                    .data_type
+                    .as_timestamp()
+                    .unwrap()
+                    .unit();
+                Some((table_name, (rows, time_index_unit)))
+            })
+            .collect::<HashMap<_, _>>();
+
+        let group_by_table_name = Arc::new(group_by_table_name);
+
+        let mut handles = Vec::new();
+        let tasks = self.tasks.read().await;
+
+        for (_flow_id, task) in tasks.iter() {
+            let src_table_names = &task.config.source_table_names;
+
+            if src_table_names
+                .iter()
+                .all(|name| !group_by_table_name.contains_key(name))
+            {
+                continue;
+            }
+
+            let group_by_table_name = group_by_table_name.clone();
+            let task = task.clone();
+
+            let handle: JoinHandle<Result<(), Error>> = tokio::spawn(async move {
+                let src_table_names = &task.config.source_table_names;
+                let mut all_dirty_windows = vec![];
+                for src_table_name in src_table_names {
+                    if let Some((window_ranges, unit)) = group_by_table_name.get(src_table_name) {
+                        let Some(expr) = &task.config.time_window_expr else {
+                            continue;
+                        };
+                        for window in window_ranges {
+                            let align_start = expr
+                                .eval(common_time::Timestamp::new(window.start_value, *unit))?
+                                .0
+                                .context(UnexpectedSnafu {
+                                    reason: "Failed to eval start value",
+                                })?;
+
+                            let align_end = expr
+                                .eval(common_time::Timestamp::new(window.end_value, *unit))?
+                                .1
+                                .context(UnexpectedSnafu {
+                                    reason: "Failed to eval end value",
+                                })?;
+                            all_dirty_windows.push((align_start, align_end));
+                        }
+                    }
+                }
+                let mut state = task.state.write().unwrap();
+                let flow_id_label = task.config.flow_id.to_string();
+                for (s, e) in all_dirty_windows {
+                    METRIC_FLOW_BATCHING_ENGINE_BULK_MARK_TIME_WINDOW_RANGE
+                        .with_label_values(&[&flow_id_label])
+                        .observe(e.sub(&s).unwrap_or_default().num_seconds() as f64);
+                    state.dirty_time_windows.add_window(s, Some(e));
+                }
+                Ok(())
+            });
+            handles.push(handle);
+        }
+        drop(tasks);
+        for handle in handles {
+            match handle.await {
+                Err(e) => {
+                    warn!("Failed to handle inserts: {e}");
+                }
+                Ok(Ok(())) => (),
+                Ok(Err(e)) => {
+                    warn!("Failed to handle inserts: {e}");
+                }
+            }
+        }
+
+        Ok(Default::default())
+    }
+
     pub async fn handle_inserts_inner(
         &self,
         request: api::v1::region::InsertRequests,
@@ -388,6 +505,20 @@ impl BatchingEngine {
     pub async fn flow_exist_inner(&self, flow_id: FlowId) -> bool {
         self.tasks.read().await.contains_key(&flow_id)
     }
+
+    pub async fn adjust_flow(
+        &self,
+        flow_id: FlowId,
+        min_run_interval_secs: u64,
+        max_filter_num_per_query: usize,
+    ) -> Result<(), Error> {
+        let task = self.tasks.read().await.get(&flow_id).cloned();
+        let task = task.with_context(|| FlowNotFoundSnafu { id: flow_id })?;
+        debug!("Adjusting flow {flow_id} with min_run_interval_secs={} and max_filter_num_per_query={}", min_run_interval_secs, max_filter_num_per_query);
+        task.adjust(min_run_interval_secs, max_filter_num_per_query);
+
+        Ok(())
+    }
 }
 
 impl FlowEngine for BatchingEngine {

src/flow/src/batching_mode/frontend_client.rs

@@ -14,8 +14,9 @@
 
 //! Frontend client to run flow as batching task which is time-window-aware normal query triggered every tick set by user
 
-use std::sync::{Arc, Weak};
+use std::collections::HashMap;
-use std::time::SystemTime;
+use std::sync::{Arc, Mutex, Weak};
+use std::time::{Duration, Instant, SystemTime};
 
 use api::v1::greptime_request::Request;
 use api::v1::CreateTableExpr;
@@ -26,20 +27,21 @@ use common_meta::cluster::{NodeInfo, NodeInfoKey, Role};
 use common_meta::peer::Peer;
 use common_meta::rpc::store::RangeRequest;
 use common_query::Output;
-use common_telemetry::warn;
+use common_telemetry::{debug, warn};
+use itertools::Itertools;
 use meta_client::client::MetaClient;
-use rand::rng;
-use rand::seq::SliceRandom;
 use servers::query_handler::grpc::GrpcQueryHandler;
 use session::context::{QueryContextBuilder, QueryContextRef};
 use snafu::{OptionExt, ResultExt};
 
+use crate::batching_mode::task::BatchingTask;
 use crate::batching_mode::{
     DEFAULT_BATCHING_ENGINE_QUERY_TIMEOUT, FRONTEND_ACTIVITY_TIMEOUT, GRPC_CONN_TIMEOUT,
     GRPC_MAX_RETRIES,
 };
 use crate::error::{ExternalSnafu, InvalidRequestSnafu, NoAvailableFrontendSnafu, UnexpectedSnafu};
-use crate::{Error, FlowAuthHeader};
+use crate::metrics::METRIC_FLOW_BATCHING_ENGINE_GUESS_FE_LOAD;
+use crate::{Error, FlowAuthHeader, FlowId};
 
 /// Just like [`GrpcQueryHandler`] but use BoxedError
 ///
@@ -74,6 +76,105 @@ impl<
 
 type HandlerMutable = Arc<std::sync::Mutex<Option<Weak<dyn GrpcQueryHandlerWithBoxedError>>>>;
 
+/// Statistics about running query on this frontend from flownode
+#[derive(Debug, Default, Clone)]
+struct FrontendStat {
+    /// The query for flow id has been running since this timestamp
+    since: HashMap<FlowId, Instant>,
+    /// The average query time for each flow id
+    /// This is used to calculate the average query time for each flow id
+    past_query_avg: HashMap<FlowId, (usize, Duration)>,
+}
+
+#[derive(Debug, Default, Clone)]
+pub struct FrontendStats {
+    /// The statistics for each flow id
+    stats: Arc<Mutex<HashMap<String, FrontendStat>>>,
+}
+
+impl FrontendStats {
+    pub fn observe(&self, frontend_addr: &str, flow_id: FlowId) -> FrontendStatsGuard {
+        let mut stats = self.stats.lock().expect("Failed to lock frontend stats");
+        let stat = stats.entry(frontend_addr.to_string()).or_default();
+        stat.since.insert(flow_id, Instant::now());
+
+        FrontendStatsGuard {
+            stats: self.stats.clone(),
+            frontend_addr: frontend_addr.to_string(),
+            cur: flow_id,
+        }
+    }
+
+    /// return frontend addrs sorted by load, from lightest to heaviest
+    /// The load is calculated as the total average query time for each flow id plus running query's total running time elapsed
+    pub fn sort_by_load(&self) -> Vec<String> {
+        let stats = self.stats.lock().expect("Failed to lock frontend stats");
+        let fe_load_factor = stats
+            .iter()
+            .map(|(node_addr, stat)| {
+                // total expected avg running time for all currently running queries
+                let total_expect_avg_run_time = stat
+                    .since
+                    .keys()
+                    .map(|f| {
+                        let (count, total_duration) =
+                            stat.past_query_avg.get(f).unwrap_or(&(0, Duration::ZERO));
+                        if *count == 0 {
+                            0.0
+                        } else {
+                            total_duration.as_secs_f64() / *count as f64
+                        }
+                    })
+                    .sum::<f64>();
+                let total_cur_running_time = stat
+                    .since
+                    .values()
+                    .map(|since| since.elapsed().as_secs_f64())
+                    .sum::<f64>();
+                (
+                    node_addr.to_string(),
+                    total_expect_avg_run_time + total_cur_running_time,
+                )
+            })
+            .sorted_by(|(_, load_a), (_, load_b)| {
+                load_a
+                    .partial_cmp(load_b)
+                    .unwrap_or(std::cmp::Ordering::Equal)
+            })
+            .collect::<Vec<_>>();
+        debug!("Frontend load factor: {:?}", fe_load_factor);
+        for (node_addr, load) in &fe_load_factor {
+            METRIC_FLOW_BATCHING_ENGINE_GUESS_FE_LOAD
+                .with_label_values(&[&node_addr.to_string()])
+                .observe(*load);
+        }
+        fe_load_factor
+            .into_iter()
+            .map(|(addr, _)| addr)
+            .collect::<Vec<_>>()
+    }
+}
+
+pub struct FrontendStatsGuard {
+    stats: Arc<Mutex<HashMap<String, FrontendStat>>>,
+    frontend_addr: String,
+    cur: FlowId,
+}
+
+impl Drop for FrontendStatsGuard {
+    fn drop(&mut self) {
+        let mut stats = self.stats.lock().expect("Failed to lock frontend stats");
+        if let Some(stat) = stats.get_mut(&self.frontend_addr) {
+            if let Some(since) = stat.since.remove(&self.cur) {
+                let elapsed = since.elapsed();
+                let (count, total_duration) = stat.past_query_avg.entry(self.cur).or_default();
+                *count += 1;
+                *total_duration += elapsed;
+            }
+        }
+    }
+}
+
 /// A simple frontend client able to execute sql using grpc protocol
 ///
 /// This is for computation-heavy query which need to offload computation to frontend, lifting the load from flownode
@@ -83,6 +184,7 @@ pub enum FrontendClient {
         meta_client: Arc<MetaClient>,
         chnl_mgr: ChannelManager,
         auth: Option<FlowAuthHeader>,
+        fe_stats: FrontendStats,
     },
     Standalone {
         /// for the sake of simplicity still use grpc even in standalone mode
@@ -114,6 +216,7 @@ impl FrontendClient {
                 ChannelManager::with_config(cfg)
             },
             auth,
+            fe_stats: Default::default(),
         }
     }
 
@@ -192,6 +295,7 @@ impl FrontendClient {
             meta_client: _,
             chnl_mgr,
             auth,
+            fe_stats,
         } = self
         else {
             return UnexpectedSnafu {
@@ -208,8 +312,21 @@ impl FrontendClient {
                 .duration_since(SystemTime::UNIX_EPOCH)
                 .unwrap()
                 .as_millis() as i64;
-            // shuffle the frontends to avoid always pick the same one
+            let node_addrs_by_load = fe_stats.sort_by_load();
-            frontends.shuffle(&mut rng());
+            // index+1 to load order asc, so that the lightest node has load 1 and non-existent node has load 0
+            let addr2load = node_addrs_by_load
+                .iter()
+                .enumerate()
+                .map(|(i, id)| (id.clone(), i + 1))
+                .collect::<HashMap<_, _>>();
+            // sort frontends by load, from lightest to heaviest
+            frontends.sort_by(|(_, a), (_, b)| {
+                // if not even in stats, treat as 0 load since never been queried
+                let load_a = addr2load.get(&a.peer.addr).unwrap_or(&0);
+                let load_b = addr2load.get(&b.peer.addr).unwrap_or(&0);
+                load_a.cmp(load_b)
+            });
+            debug!("Frontend nodes sorted by load: {:?}", frontends);
 
             // found node with maximum last_activity_ts
             for (_, node_info) in frontends
@@ -257,6 +374,7 @@ impl FrontendClient {
         create: CreateTableExpr,
         catalog: &str,
         schema: &str,
+        task: Option<&BatchingTask>,
     ) -> Result<u32, Error> {
         self.handle(
             Request::Ddl(api::v1::DdlRequest {
@@ -265,6 +383,7 @@ impl FrontendClient {
             catalog,
             schema,
             &mut None,
+            task,
         )
         .await
     }
@@ -276,15 +395,19 @@ impl FrontendClient {
         catalog: &str,
         schema: &str,
         peer_desc: &mut Option<PeerDesc>,
+        task: Option<&BatchingTask>,
     ) -> Result<u32, Error> {
         match self {
-            FrontendClient::Distributed { .. } => {
+            FrontendClient::Distributed { fe_stats, .. } => {
                 let db = self.get_random_active_frontend(catalog, schema).await?;
 
                 *peer_desc = Some(PeerDesc::Dist {
                     peer: db.peer.clone(),
                 });
 
+                let flow_id = task.map(|t| t.config.flow_id).unwrap_or_default();
+                let _guard = fe_stats.observe(&db.peer.addr, flow_id);
+
                 db.database
                     .handle_with_retry(req.clone(), GRPC_MAX_RETRIES)
                     .await

src/flow/src/batching_mode/state.rs

@@ -31,7 +31,9 @@ use crate::batching_mode::time_window::TimeWindowExpr;
 use crate::batching_mode::MIN_REFRESH_DURATION;
 use crate::error::{DatatypesSnafu, InternalSnafu, TimeSnafu, UnexpectedSnafu};
 use crate::metrics::{
-    METRIC_FLOW_BATCHING_ENGINE_QUERY_TIME_RANGE, METRIC_FLOW_BATCHING_ENGINE_QUERY_WINDOW_CNT,
+    METRIC_FLOW_BATCHING_ENGINE_QUERY_WINDOW_CNT, METRIC_FLOW_BATCHING_ENGINE_QUERY_WINDOW_SIZE,
+    METRIC_FLOW_BATCHING_ENGINE_STALLED_QUERY_WINDOW_CNT,
+    METRIC_FLOW_BATCHING_ENGINE_STALLED_WINDOW_SIZE,
 };
 use crate::{Error, FlowId};
 
@@ -52,6 +54,11 @@ pub struct TaskState {
     pub(crate) shutdown_rx: oneshot::Receiver<()>,
     /// Task handle
     pub(crate) task_handle: Option<tokio::task::JoinHandle<()>>,
+
+    /// min run interval in seconds
+    pub(crate) min_run_interval: Option<u64>,
+    /// max filter number per query
+    pub(crate) max_filter_num: Option<usize>,
 }
 impl TaskState {
     pub fn new(query_ctx: QueryContextRef, shutdown_rx: oneshot::Receiver<()>) -> Self {
@@ -63,6 +70,8 @@ impl TaskState {
             exec_state: ExecState::Idle,
             shutdown_rx,
             task_handle: None,
+            min_run_interval: None,
+            max_filter_num: None,
         }
     }
 
@@ -86,24 +95,21 @@ impl TaskState {
     /// TODO: Make this behavior configurable.
     pub fn get_next_start_query_time(
         &self,
-        flow_id: FlowId,
+        _flow_id: FlowId,
-        time_window_size: &Option<Duration>,
+        _time_window_size: &Option<Duration>,
         max_timeout: Option<Duration>,
     ) -> Instant {
-        let last_duration = max_timeout
+        let next_duration = max_timeout
             .unwrap_or(self.last_query_duration)
             .min(self.last_query_duration)
-            .max(MIN_REFRESH_DURATION);
+            .max(
-
+                self.min_run_interval
-        let next_duration = time_window_size
+                    .map(Duration::from_secs)
-            .map(|t| {
+                    .unwrap_or(MIN_REFRESH_DURATION),
-                let half = t / 2;
+            );
-                half.max(last_duration)
-            })
-            .unwrap_or(last_duration);
 
         // if have dirty time window, execute immediately to clean dirty time window
-        if self.dirty_time_windows.windows.is_empty() {
+        /*if self.dirty_time_windows.windows.is_empty() {
             self.last_update_time + next_duration
         } else {
             debug!(
@@ -113,7 +119,10 @@ impl TaskState {
                 self.dirty_time_windows.windows
             );
             Instant::now()
-        }
+        }*/
+
+        // wait for next duration anyway
+        self.last_update_time + next_duration
     }
 }
 
@@ -206,47 +215,63 @@ impl DirtyTimeWindows {
 
         // get the first `window_cnt` time windows
         let max_time_range = window_size * window_cnt as i32;
-        let nth = {
-            let mut cur_time_range = chrono::Duration::zero();
-            let mut nth_key = None;
-            for (idx, (start, end)) in self.windows.iter().enumerate() {
-                // if time range is too long, stop
-                if cur_time_range > max_time_range {
-                    nth_key = Some(*start);
-                    break;
-                }
 
-                // if we have enough time windows, stop
+        let mut to_be_query = BTreeMap::new();
-                if idx >= window_cnt {
+        let mut new_windows = self.windows.clone();
-                    nth_key = Some(*start);
+        let mut cur_time_range = chrono::Duration::zero();
-                    break;
+        for (idx, (start, end)) in self.windows.iter().enumerate() {
-                }
+            let first_end = start
+                .add_duration(window_size.to_std().unwrap())
+                .context(TimeSnafu)?;
+            let end = end.unwrap_or(first_end);
 
-                if let Some(end) = end {
+            // if time range is too long, stop
-                    if let Some(x) = end.sub(start) {
+            if cur_time_range >= max_time_range {
-                        cur_time_range += x;
+                break;
-                    }
-                }
             }
 
-            nth_key
+            // if we have enough time windows, stop
-        };
+            if idx >= window_cnt {
-        let first_nth = {
+                break;
-            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)
             }
-        };
+
+            if let Some(x) = end.sub(start) {
+                if cur_time_range + x <= max_time_range {
+                    to_be_query.insert(*start, Some(end));
+                    new_windows.remove(start);
+                    cur_time_range += x;
+                } else {
+                    // too large a window, split it
+                    // split at window_size * times
+                    let surplus = max_time_range - cur_time_range;
+                    let times = surplus.num_seconds() / window_size.num_seconds();
+
+                    let split_offset = window_size * times as i32;
+                    let split_at = start
+                        .add_duration(split_offset.to_std().unwrap())
+                        .context(TimeSnafu)?;
+                    to_be_query.insert(*start, Some(split_at));
+
+                    // remove the original window
+                    new_windows.remove(start);
+                    new_windows.insert(split_at, Some(end));
+                    cur_time_range += split_offset;
+                    break;
+                }
+            }
+        }
+
+        self.windows = new_windows;
 
         METRIC_FLOW_BATCHING_ENGINE_QUERY_WINDOW_CNT
             .with_label_values(&[flow_id.to_string().as_str()])
-            .observe(first_nth.len() as f64);
+            .observe(to_be_query.len() as f64);
 
-        let full_time_range = first_nth
+        METRIC_FLOW_BATCHING_ENGINE_STALLED_QUERY_WINDOW_CNT
+            .with_label_values(&[flow_id.to_string().as_str()])
+            .observe(self.windows.len() as f64);
+
+        let full_time_range = to_be_query
             .iter()
             .fold(chrono::Duration::zero(), |acc, (start, end)| {
                 if let Some(end) = end {
@@ -256,12 +281,27 @@ impl DirtyTimeWindows {
                 }
             })
             .num_seconds() as f64;
-        METRIC_FLOW_BATCHING_ENGINE_QUERY_TIME_RANGE
+        METRIC_FLOW_BATCHING_ENGINE_QUERY_WINDOW_SIZE
             .with_label_values(&[flow_id.to_string().as_str()])
             .observe(full_time_range);
 
+        let stalled_time_range =
+            self.windows
+                .iter()
+                .fold(chrono::Duration::zero(), |acc, (start, end)| {
+                    if let Some(end) = end {
+                        acc + end.sub(start).unwrap_or(chrono::Duration::zero())
+                    } else {
+                        acc
+                    }
+                });
+
+        METRIC_FLOW_BATCHING_ENGINE_STALLED_WINDOW_SIZE
+            .with_label_values(&[flow_id.to_string().as_str()])
+            .observe(stalled_time_range.num_seconds() as f64);
+
         let mut expr_lst = vec![];
-        for (start, end) in first_nth.into_iter() {
+        for (start, end) in to_be_query.into_iter() {
             // align using time window exprs
             let (start, end) = if let Some(ctx) = task_ctx {
                 let Some(time_window_expr) = &ctx.config.time_window_expr else {
@@ -495,6 +535,64 @@ mod test {
                     "((ts >= CAST('1970-01-01 00:00:00' AS TIMESTAMP)) AND (ts < CAST('1970-01-01 00:00:21' AS TIMESTAMP)))",
                 )
             ),
+            // split range
+            (
+                Vec::from_iter((0..20).map(|i|Timestamp::new_second(i*3)).chain(std::iter::once(
+                    Timestamp::new_second(60 + 3 * (DirtyTimeWindows::MERGE_DIST as i64 + 1)),
+                ))),
+                (chrono::Duration::seconds(3), None),
+                BTreeMap::from([
+                (
+                    Timestamp::new_second(0),
+                    Some(Timestamp::new_second(
+                        60
+                    )),
+                ),
+                (
+                    Timestamp::new_second(60 + 3 * (DirtyTimeWindows::MERGE_DIST as i64 + 1)),
+                    Some(Timestamp::new_second(
+                        60 + 3 * (DirtyTimeWindows::MERGE_DIST as i64 + 1) + 3
+                    )),
+                )]),
+                Some(
+                    "((ts >= CAST('1970-01-01 00:00:00' AS TIMESTAMP)) AND (ts < CAST('1970-01-01 00:01:00' AS TIMESTAMP)))",
+                )
+            ),
+            // split 2 min into 1 min
+            (
+                Vec::from_iter((0..40).map(|i|Timestamp::new_second(i*3))),
+                (chrono::Duration::seconds(3), None),
+                BTreeMap::from([
+                (
+                    Timestamp::new_second(0),
+                    Some(Timestamp::new_second(
+                        40 * 3
+                    )),
+                )]),
+                Some(
+                    "((ts >= CAST('1970-01-01 00:00:00' AS TIMESTAMP)) AND (ts < CAST('1970-01-01 00:01:00' AS TIMESTAMP)))",
+                )
+            ),
+            // split 3s + 1min into 3s + 57s
+            (
+                Vec::from_iter(std::iter::once(Timestamp::new_second(0)).chain((0..40).map(|i|Timestamp::new_second(20+i*3)))),
+                (chrono::Duration::seconds(3), None),
+                BTreeMap::from([
+                (
+                    Timestamp::new_second(0),
+                    Some(Timestamp::new_second(
+                        3
+                    )),
+                ),(
+                    Timestamp::new_second(20),
+                    Some(Timestamp::new_second(
+                        140
+                    )),
+                )]),
+                Some(
+                    "(((ts >= CAST('1970-01-01 00:00:00' AS TIMESTAMP)) AND (ts < CAST('1970-01-01 00:00:03' AS TIMESTAMP))) OR ((ts >= CAST('1970-01-01 00:00:20' AS TIMESTAMP)) AND (ts < CAST('1970-01-01 00:01:17' AS TIMESTAMP))))",
+                )
+            ),
             // expired
             (
                 vec![
@@ -511,6 +609,8 @@ mod test {
                 None
             ),
         ];
+        // let len = testcases.len();
+        // let testcases = testcases[(len - 2)..(len - 1)].to_vec();
         for (lower_bounds, (window_size, expire_lower_bound), expected, expected_filter_expr) in
             testcases
         {

src/flow/src/batching_mode/task.rs

@@ -61,7 +61,9 @@ use crate::error::{
     SubstraitEncodeLogicalPlanSnafu, UnexpectedSnafu,
 };
 use crate::metrics::{
-    METRIC_FLOW_BATCHING_ENGINE_QUERY_TIME, METRIC_FLOW_BATCHING_ENGINE_SLOW_QUERY,
+    METRIC_FLOW_BATCHING_ENGINE_ERROR_CNT, METRIC_FLOW_BATCHING_ENGINE_QUERY_TIME,
+    METRIC_FLOW_BATCHING_ENGINE_SLOW_QUERY, METRIC_FLOW_BATCHING_ENGINE_START_QUERY_CNT,
+    METRIC_FLOW_ROWS,
 };
 use crate::{Error, FlowId};
 
@@ -144,6 +146,12 @@ impl BatchingTask {
         })
     }
 
+    pub fn adjust(&self, min_run_interval_secs: u64, max_filter_num_per_query: usize) {
+        let mut state = self.state.write().unwrap();
+        state.min_run_interval = Some(min_run_interval_secs);
+        state.max_filter_num = Some(max_filter_num_per_query);
+    }
+
     /// mark time window range (now - expire_after, now) as dirty (or (0, now) if expire_after not set)
     ///
     /// useful for flush_flow to flush dirty time windows range
@@ -280,7 +288,7 @@ impl BatchingTask {
         let catalog = &self.config.sink_table_name[0];
         let schema = &self.config.sink_table_name[1];
         frontend_client
-            .create(expr.clone(), catalog, schema)
+            .create(expr.clone(), catalog, schema, Some(self))
             .await?;
         Ok(())
     }
@@ -361,7 +369,7 @@ impl BatchingTask {
             };
 
             frontend_client
-                .handle(req, catalog, schema, &mut peer_desc)
+                .handle(req, catalog, schema, &mut peer_desc, Some(self))
                 .await
         };
 
@@ -371,6 +379,9 @@ impl BatchingTask {
                 "Flow {flow_id} executed, affected_rows: {affected_rows:?}, elapsed: {:?}",
                 elapsed
             );
+            METRIC_FLOW_ROWS
+                .with_label_values(&[format!("{}-out-batching", flow_id).as_str()])
+                .inc_by(*affected_rows as _);
         } else if let Err(err) = &res {
             warn!(
                 "Failed to execute Flow {flow_id} on frontend {:?}, result: {err:?}, elapsed: {:?} with query: {}",
@@ -410,6 +421,7 @@ impl BatchingTask {
         engine: QueryEngineRef,
         frontend_client: Arc<FrontendClient>,
     ) {
+        let flow_id_str = self.config.flow_id.to_string();
         loop {
             // first check if shutdown signal is received
             // if so, break the loop
@@ -427,6 +439,9 @@ impl BatchingTask {
                     Err(TryRecvError::Empty) => (),
                 }
             }
+            METRIC_FLOW_BATCHING_ENGINE_START_QUERY_CNT
+                .with_label_values(&[&flow_id_str])
+                .inc();
 
             let new_query = match self.gen_insert_plan(&engine).await {
                 Ok(new_query) => new_query,
@@ -473,6 +488,9 @@ impl BatchingTask {
                 }
                 // TODO(discord9): this error should have better place to go, but for now just print error, also more context is needed
                 Err(err) => {
+                    METRIC_FLOW_BATCHING_ENGINE_ERROR_CNT
+                        .with_label_values(&[&flow_id_str])
+                        .inc();
                     match new_query {
                         Some(query) => {
                             common_telemetry::error!(err; "Failed to execute query for flow={} with query: {query}", self.config.flow_id)
@@ -580,19 +598,20 @@ impl BatchingTask {
                 ),
             })?;
 
-        let expr = self
+        let expr = {
-            .state
+            let mut state = self.state.write().unwrap();
-            .write()
+            let max_window_cnt = state
-            .unwrap()
+                .max_filter_num
-            .dirty_time_windows
+                .unwrap_or(DirtyTimeWindows::MAX_FILTER_NUM);
-            .gen_filter_exprs(
+            state.dirty_time_windows.gen_filter_exprs(
                 &col_name,
                 Some(l),
                 window_size,
-                DirtyTimeWindows::MAX_FILTER_NUM,
+                max_window_cnt,
                 self.config.flow_id,
                 Some(self),
-            )?;
+            )?
+        };
 
         debug!(
             "Flow id={:?}, Generated filter expr: {:?}",

src/flow/src/metrics.rs

@@ -42,6 +42,14 @@ lazy_static! {
         vec![60., 2. * 60., 3. * 60., 5. * 60., 10. * 60.]
     )
     .unwrap();
+    pub static ref METRIC_FLOW_BATCHING_ENGINE_STALLED_QUERY_WINDOW_CNT: HistogramVec =
+        register_histogram_vec!(
+            "greptime_flow_batching_engine_stalled_query_window_cnt",
+            "flow batching engine stalled query time window count",
+            &["flow_id"],
+            vec![0.0, 5., 10., 20., 40.]
+        )
+        .unwrap();
     pub static ref METRIC_FLOW_BATCHING_ENGINE_QUERY_WINDOW_CNT: HistogramVec =
         register_histogram_vec!(
             "greptime_flow_batching_engine_query_window_cnt",
@@ -50,19 +58,57 @@ lazy_static! {
             vec![0.0, 5., 10., 20., 40.]
         )
         .unwrap();
-    pub static ref METRIC_FLOW_BATCHING_ENGINE_QUERY_TIME_RANGE: HistogramVec =
+    pub static ref METRIC_FLOW_BATCHING_ENGINE_QUERY_WINDOW_SIZE: HistogramVec =
         register_histogram_vec!(
-            "greptime_flow_batching_engine_query_time_range_secs",
+            "greptime_flow_batching_engine_query_window_size_secs",
-            "flow batching engine query time range(seconds)",
+            "flow batching engine query window size(seconds)",
             &["flow_id"],
             vec![60., 4. * 60., 16. * 60., 64. * 60., 256. * 60.]
         )
         .unwrap();
+    pub static ref METRIC_FLOW_BATCHING_ENGINE_STALLED_WINDOW_SIZE: HistogramVec =
+        register_histogram_vec!(
+            "greptime_flow_batching_engine_stalled_window_size_secs",
+            "flow batching engine stalled window size(seconds)",
+            &["flow_id"],
+            vec![60., 4. * 60., 16. * 60., 64. * 60., 256. * 60.]
+        )
+        .unwrap();
+    pub static ref METRIC_FLOW_BATCHING_ENGINE_BULK_MARK_TIME_WINDOW_RANGE: HistogramVec =
+        register_histogram_vec!(
+            "greptime_flow_batching_engine_bulk_mark_time_window_range_secs",
+            "flow batching engine query time window range marked by bulk memtable in seconds",
+            &["flow_id"],
+            vec![0.0, 60., 4. * 60., 16. * 60., 64. * 60., 256. * 60.]
+        )
+        .unwrap();
+    pub static ref METRIC_FLOW_BATCHING_ENGINE_START_QUERY_CNT: IntCounterVec =
+        register_int_counter_vec!(
+            "greptime_flow_batching_start_query_count",
+            "flow batching engine started query count",
+            &["flow_id"],
+        )
+        .unwrap();
+    pub static ref METRIC_FLOW_BATCHING_ENGINE_ERROR_CNT: IntCounterVec =
+        register_int_counter_vec!(
+            "greptime_flow_batching_error_count",
+            "flow batching engine error count per flow id",
+            &["flow_id"]
+        )
+        .unwrap();
+    pub static ref METRIC_FLOW_BATCHING_ENGINE_GUESS_FE_LOAD: HistogramVec =
+        register_histogram_vec!(
+            "greptime_flow_batching_engine_guess_fe_load",
+            "flow batching engine guessed frontend load",
+            &["fe_addr"],
+            vec![60., 4. * 60., 16. * 60., 64. * 60., 256. * 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!(
         "greptime_flow_processed_rows",
-        "Count of rows flowing through the system",
+        "Count of rows flowing through the system.",
         &["direction"]
     )
     .unwrap();

src/flow/src/server.rs

@@ -17,6 +17,7 @@
 use std::net::SocketAddr;
 use std::sync::Arc;
 
+use api::v1::flow::DirtyWindowRequests;
 use api::v1::{RowDeleteRequests, RowInsertRequests};
 use cache::{TABLE_FLOWNODE_SET_CACHE_NAME, TABLE_ROUTE_CACHE_NAME};
 use catalog::CatalogManagerRef;
@@ -136,6 +137,18 @@ impl flow_server::Flow for FlowService {
             .map(Response::new)
             .map_err(to_status_with_last_err)
     }
+
+    async fn handle_mark_dirty_time_window(
+        &self,
+        reqs: Request<DirtyWindowRequests>,
+    ) -> Result<Response<FlowResponse>, Status> {
+        self.dual_engine
+            .batching_engine()
+            .handle_mark_dirty_time_window(reqs.into_inner())
+            .await
+            .map(Response::new)
+            .map_err(to_status_with_last_err)
+    }
 }
 
 #[derive(Clone)]

src/frontend/src/instance/grpc.rs

@@ -35,8 +35,8 @@ use servers::query_handler::grpc::GrpcQueryHandler;
 use servers::query_handler::sql::SqlQueryHandler;
 use session::context::QueryContextRef;
 use snafu::{ensure, OptionExt, ResultExt};
-use table::metadata::TableId;
 use table::table_name::TableName;
+use table::TableRef;
 
 use crate::error::{
     CatalogSnafu, DataFusionSnafu, Error, InFlightWriteBytesExceededSnafu,
@@ -235,34 +235,33 @@ impl GrpcQueryHandler for Instance {
 
     async fn put_record_batch(
         &self,
-        table: &TableName,
+        table_name: &TableName,
-        table_id: &mut Option<TableId>,
+        table_ref: &mut Option<TableRef>,
         decoder: &mut FlightDecoder,
         data: FlightData,
     ) -> Result<AffectedRows> {
-        let table_id = if let Some(table_id) = table_id {
+        let table = if let Some(table) = table_ref {
-            *table_id
+            table.clone()
         } else {
             let table = self
                 .catalog_manager()
                 .table(
-                    &table.catalog_name,
+                    &table_name.catalog_name,
-                    &table.schema_name,
+                    &table_name.schema_name,
-                    &table.table_name,
+                    &table_name.table_name,
                     None,
                 )
                 .await
                 .context(CatalogSnafu)?
                 .with_context(|| TableNotFoundSnafu {
-                    table_name: table.to_string(),
+                    table_name: table_name.to_string(),
                 })?;
-            let id = table.table_info().table_id();
+            *table_ref = Some(table.clone());
-            *table_id = Some(id);
+            table
-            id
         };
 
         self.inserter
-            .handle_bulk_insert(table_id, decoder, data)
+            .handle_bulk_insert(table, decoder, data)
             .await
             .context(TableOperationSnafu)
     }

src/frontend/src/server.rs

@@ -154,6 +154,7 @@ where
             ServerGrpcQueryHandlerAdapter::arc(self.instance.clone()),
             user_provider.clone(),
             runtime,
+            opts.grpc.flight_compression,
         );
 
         let grpc_server = builder

src/operator/src/bulk_insert.rs

@@ -12,18 +12,28 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
+use std::collections::HashSet;
+
 use ahash::{HashMap, HashMapExt};
+use api::v1::flow::{DirtyWindowRequest, WindowRange};
 use api::v1::region::{
     bulk_insert_request, region_request, BulkInsertRequest, RegionRequest, RegionRequestHeader,
 };
 use api::v1::ArrowIpc;
+use arrow::array::{
+    Array, TimestampMicrosecondArray, TimestampMillisecondArray, TimestampNanosecondArray,
+    TimestampSecondArray,
+};
+use arrow::datatypes::{DataType, Int64Type, TimeUnit};
+use arrow::record_batch::RecordBatch;
 use common_base::AffectedRows;
 use common_grpc::flight::{FlightDecoder, FlightEncoder, FlightMessage};
 use common_grpc::FlightData;
+use common_telemetry::error;
 use common_telemetry::tracing_context::TracingContext;
-use snafu::ResultExt;
+use snafu::{OptionExt, ResultExt};
-use store_api::storage::RegionId;
+use store_api::storage::{RegionId, TableId};
-use table::metadata::TableId;
+use table::TableRef;
 
 use crate::insert::Inserter;
 use crate::{error, metrics};
@@ -32,10 +42,11 @@ impl Inserter {
     /// Handle bulk insert request.
     pub async fn handle_bulk_insert(
         &self,
-        table_id: TableId,
+        table: TableRef,
         decoder: &mut FlightDecoder,
         data: FlightData,
     ) -> error::Result<AffectedRows> {
+        let table_id = table.table_info().table_id();
         let decode_timer = metrics::HANDLE_BULK_INSERT_ELAPSED
             .with_label_values(&["decode_request"])
             .start_timer();
@@ -48,6 +59,20 @@ impl Inserter {
             return Ok(0);
         };
         decode_timer.observe_duration();
+        if let Some((min, max)) = compute_timestamp_range(
+            &record_batch,
+            &table
+                .table_info()
+                .meta
+                .schema
+                .timestamp_column()
+                .as_ref()
+                .unwrap()
+                .name,
+        )? {
+            // notify flownode to update dirty time windows.
+            self.update_flow_dirty_window(table_id, min, max);
+        }
         metrics::BULK_REQUEST_MESSAGE_SIZE.observe(body_size as f64);
         metrics::BULK_REQUEST_ROWS
             .with_label_values(&["raw"])
@@ -216,4 +241,88 @@ impl Inserter {
         crate::metrics::DIST_INGEST_ROW_COUNT.inc_by(rows_inserted as u64);
         Ok(rows_inserted)
     }
+
+    fn update_flow_dirty_window(&self, table_id: TableId, min: i64, max: i64) {
+        let table_flownode_set_cache = self.table_flownode_set_cache.clone();
+        let node_manager = self.node_manager.clone();
+        common_runtime::spawn_global(async move {
+            let result = table_flownode_set_cache
+                .get(table_id)
+                .await
+                .context(error::RequestInsertsSnafu);
+            let flownodes = match result {
+                Ok(flownodes) => flownodes.unwrap_or_default(),
+                Err(e) => {
+                    error!(e; "Failed to get flownodes for table id: {}", table_id);
+                    return;
+                }
+            };
+
+            let peers: HashSet<_> = flownodes.values().cloned().collect();
+            for peer in peers {
+                let node_manager = node_manager.clone();
+                common_runtime::spawn_global(async move {
+                    if let Err(e) = node_manager
+                        .flownode(&peer)
+                        .await
+                        .handle_mark_window_dirty(DirtyWindowRequest {
+                            table_id,
+                            dirty_time_ranges: vec![WindowRange {
+                                start_value: min,
+                                end_value: max,
+                            }],
+                        })
+                        .await
+                        .context(error::RequestInsertsSnafu)
+                    {
+                        error!(e; "Failed to mark time window as dirty, table: {}, min: {}, max: {}", table_id, min, max);
+                    }
+                });
+            }
+        });
+    }
+}
+
+/// Calculate the timestamp range of record batch. Return `None` if record batch is empty.
+fn compute_timestamp_range(
+    rb: &RecordBatch,
+    timestamp_index_name: &str,
+) -> error::Result<Option<(i64, i64)>> {
+    let ts_col = rb
+        .column_by_name(timestamp_index_name)
+        .context(error::ColumnNotFoundSnafu {
+            msg: timestamp_index_name,
+        })?;
+    if rb.num_rows() == 0 {
+        return Ok(None);
+    }
+    let primitive = match ts_col.data_type() {
+        DataType::Timestamp(unit, _) => match unit {
+            TimeUnit::Second => ts_col
+                .as_any()
+                .downcast_ref::<TimestampSecondArray>()
+                .unwrap()
+                .reinterpret_cast::<Int64Type>(),
+            TimeUnit::Millisecond => ts_col
+                .as_any()
+                .downcast_ref::<TimestampMillisecondArray>()
+                .unwrap()
+                .reinterpret_cast::<Int64Type>(),
+            TimeUnit::Microsecond => ts_col
+                .as_any()
+                .downcast_ref::<TimestampMicrosecondArray>()
+                .unwrap()
+                .reinterpret_cast::<Int64Type>(),
+            TimeUnit::Nanosecond => ts_col
+                .as_any()
+                .downcast_ref::<TimestampNanosecondArray>()
+                .unwrap()
+                .reinterpret_cast::<Int64Type>(),
+        },
+        t => {
+            return error::InvalidTimeIndexTypeSnafu { ty: t.clone() }.fail();
+        }
+    };
+
+    Ok(arrow::compute::min(&primitive).zip(arrow::compute::max(&primitive)))
 }

src/operator/src/error.rs

@@ -837,6 +837,13 @@ pub enum Error {
         #[snafu(implicit)]
         location: Location,
     },
+
+    #[snafu(display("Invalid time index type: {}", ty))]
+    InvalidTimeIndexType {
+        ty: arrow::datatypes::DataType,
+        #[snafu(implicit)]
+        location: Location,
+    },
 }
 
 pub type Result<T> = std::result::Result<T, Error>;
@@ -964,6 +971,7 @@ impl ErrorExt for Error {
             Error::ColumnOptions { source, .. } => source.status_code(),
             Error::DecodeFlightData { source, .. } => source.status_code(),
             Error::ComputeArrow { .. } => StatusCode::Internal,
+            Error::InvalidTimeIndexType { .. } => StatusCode::InvalidArguments,
         }
     }
 

src/operator/src/flow.rs

@@ -12,7 +12,7 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
-use api::v1::flow::FlowRequestHeader;
+use api::v1::flow::{AdjustFlow, FlowRequestHeader};
 use async_trait::async_trait;
 use common_error::ext::BoxedError;
 use common_function::handlers::FlowServiceHandler;
@@ -22,6 +22,7 @@ use common_query::error::Result;
 use common_telemetry::tracing_context::TracingContext;
 use futures::stream::FuturesUnordered;
 use futures::StreamExt;
+use serde_json::json;
 use session::context::QueryContextRef;
 use snafu::{OptionExt, ResultExt};
 
@@ -57,9 +58,96 @@ impl FlowServiceHandler for FlowServiceOperator {
     ) -> Result<api::v1::flow::FlowResponse> {
         self.flush_inner(catalog, flow, ctx).await
     }
+
+    async fn adjust(
+        &self,
+        catalog: &str,
+        flow: &str,
+        min_run_interval_secs: u64,
+        max_filter_num_per_query: usize,
+        ctx: QueryContextRef,
+    ) -> Result<api::v1::flow::FlowResponse> {
+        self.adjust_inner(
+            catalog,
+            flow,
+            min_run_interval_secs,
+            max_filter_num_per_query,
+            ctx,
+        )
+        .await
+    }
 }
 
 impl FlowServiceOperator {
+    async fn adjust_inner(
+        &self,
+        catalog: &str,
+        flow: &str,
+        min_run_interval_secs: u64,
+        max_filter_num_per_query: usize,
+        ctx: QueryContextRef,
+    ) -> Result<api::v1::flow::FlowResponse> {
+        let id = self
+            .flow_metadata_manager
+            .flow_name_manager()
+            .get(catalog, flow)
+            .await
+            .map_err(BoxedError::new)
+            .context(common_query::error::ExecuteSnafu)?
+            .context(common_meta::error::FlowNotFoundSnafu {
+                flow_name: format!("{}.{}", catalog, flow),
+            })
+            .map_err(BoxedError::new)
+            .context(common_query::error::ExecuteSnafu)?
+            .flow_id();
+
+        let all_flownode_peers = self
+            .flow_metadata_manager
+            .flow_route_manager()
+            .routes(id)
+            .await
+            .map_err(BoxedError::new)
+            .context(common_query::error::ExecuteSnafu)?;
+
+        // order of flownodes doesn't matter here
+        let all_flow_nodes = FuturesUnordered::from_iter(
+            all_flownode_peers
+                .iter()
+                .map(|(_key, peer)| self.node_manager.flownode(peer.peer())),
+        )
+        .collect::<Vec<_>>()
+        .await;
+
+        // TODO(discord9): use proper type for flow options
+        let options = json!({
+            "min_run_interval_secs": min_run_interval_secs,
+            "max_filter_num_per_query": max_filter_num_per_query,
+        });
+
+        for node in all_flow_nodes {
+            let _res = {
+                use api::v1::flow::{flow_request, FlowRequest};
+                let flush_req = FlowRequest {
+                    header: Some(FlowRequestHeader {
+                        tracing_context: TracingContext::from_current_span().to_w3c(),
+                        query_context: Some(
+                            common_meta::rpc::ddl::QueryContext::from(ctx.clone()).into(),
+                        ),
+                    }),
+                    body: Some(flow_request::Body::Adjust(AdjustFlow {
+                        flow_id: Some(api::v1::FlowId { id }),
+                        options: options.to_string(),
+                    })),
+                };
+                node.handle(flush_req)
+                    .await
+                    .map_err(BoxedError::new)
+                    .context(common_query::error::ExecuteSnafu)?
+            };
+        }
+        Ok(Default::default())
+    }
+
     /// Flush the flownodes according to the flow id.
     async fn flush_inner(
         &self,

src/operator/src/insert.rs

@@ -78,7 +78,7 @@ pub struct Inserter {
     catalog_manager: CatalogManagerRef,
     pub(crate) partition_manager: PartitionRuleManagerRef,
     pub(crate) node_manager: NodeManagerRef,
-    table_flownode_set_cache: TableFlownodeSetCacheRef,
+    pub(crate) table_flownode_set_cache: TableFlownodeSetCacheRef,
 }
 
 pub type InserterRef = Arc<Inserter>;

src/promql/src/functions.rs

@@ -34,7 +34,6 @@ pub use changes::Changes;
 use datafusion::arrow::array::{ArrayRef, Float64Array, TimestampMillisecondArray};
 use datafusion::error::DataFusionError;
 use datafusion::physical_plan::ColumnarValue;
-use datafusion_expr::{AggregateUDF, ScalarUDF};
 pub use deriv::Deriv;
 pub use extrapolate_rate::{Delta, Increase, Rate};
 pub use holt_winters::HoltWinters;
@@ -45,39 +44,6 @@ pub use quantile_aggr::{quantile_udaf, QUANTILE_NAME};
 pub use resets::Resets;
 pub use round::Round;
 
-/// Range functions for PromQL.
-pub fn range_funcs() -> Vec<ScalarUDF> {
-    vec![
-        IDelta::<false>::scalar_udf(),
-        IDelta::<true>::scalar_udf(),
-        Rate::scalar_udf(),
-        Increase::scalar_udf(),
-        Delta::scalar_udf(),
-        Resets::scalar_udf(),
-        Changes::scalar_udf(),
-        Deriv::scalar_udf(),
-        Round::scalar_udf(),
-        AvgOverTime::scalar_udf(),
-        MinOverTime::scalar_udf(),
-        MaxOverTime::scalar_udf(),
-        SumOverTime::scalar_udf(),
-        CountOverTime::scalar_udf(),
-        LastOverTime::scalar_udf(),
-        AbsentOverTime::scalar_udf(),
-        PresentOverTime::scalar_udf(),
-        StddevOverTime::scalar_udf(),
-        StdvarOverTime::scalar_udf(),
-        QuantileOverTime::scalar_udf(),
-        PredictLinear::scalar_udf(),
-        HoltWinters::scalar_udf(),
-    ]
-}
-
-/// Aggregate functions for PromQL.
-pub fn aggr_funcs() -> Vec<AggregateUDF> {
-    vec![quantile_udaf()]
-}
-
 /// Extracts an array from a `ColumnarValue`.
 ///
 /// If the `ColumnarValue` is a scalar, it converts it to an array of size 1.

src/promql/src/functions/quantile_aggr.rs

@@ -40,8 +40,8 @@ pub struct QuantileAccumulator {
 
 /// Create a quantile `AggregateUDF` for PromQL quantile operator,
 /// which calculates φ-quantile (0 ≤ φ ≤ 1) over dimensions
-pub fn quantile_udaf() -> AggregateUDF {
+pub fn quantile_udaf() -> Arc<AggregateUDF> {
-    create_udaf(
+    Arc::new(create_udaf(
         QUANTILE_NAME,
         // Input type: (φ, values)
         vec![DataType::Float64, DataType::Float64],
@@ -63,7 +63,7 @@ pub fn quantile_udaf() -> AggregateUDF {
             )]
             .into(),
         )]),
-    )
+    ))
 }
 
 impl QuantileAccumulator {

src/query/src/promql/planner.rs

@@ -1948,7 +1948,7 @@ impl PromPlanner {
             token::T_QUANTILE => {
                 let q = Self::get_param_value_as_f64(op, param)?;
                 non_col_args.push(lit(q));
-                Arc::new(quantile_udaf())
+                quantile_udaf()
             }
             token::T_AVG => avg_udaf(),
             token::T_COUNT_VALUES | token::T_COUNT => count_udaf(),

src/query/src/query_engine/default_serializer.rs

@@ -28,6 +28,11 @@ use datafusion::execution::{FunctionRegistry, SessionStateBuilder};
 use datafusion::logical_expr::LogicalPlan;
 use datafusion_expr::UserDefinedLogicalNode;
 use greptime_proto::substrait_extension::MergeScan as PbMergeScan;
+use promql::functions::{
+    quantile_udaf, AbsentOverTime, AvgOverTime, Changes, CountOverTime, Delta, Deriv, IDelta,
+    Increase, LastOverTime, MaxOverTime, MinOverTime, PresentOverTime, Rate, Resets, Round,
+    StddevOverTime, StdvarOverTime, SumOverTime,
+};
 use prost::Message;
 use session::context::QueryContextRef;
 use snafu::ResultExt;
@@ -112,15 +117,12 @@ impl SubstraitPlanDecoder for DefaultPlanDecoder {
         let mut session_state = SessionStateBuilder::new_from_existing(self.session_state.clone())
             .with_catalog_list(catalog_list)
             .build();
-
         // Substrait decoder will look up the UDFs in SessionState, so we need to register them
         // Note: the query context must be passed to set the timezone
         // We MUST register the UDFs after we build the session state, otherwise the UDFs will be lost
         // if they have the same name as the default UDFs or their alias.
         // e.g. The default UDF `to_char()` has an alias `date_format()`, if we register a UDF with the name `date_format()`
         // before we build the session state, the UDF will be lost.
-
-        // Scalar functions
         for func in FUNCTION_REGISTRY.scalar_functions() {
             let udf = func.provide(FunctionContext {
                 query_ctx: self.query_ctx.clone(),
@@ -131,15 +133,6 @@ impl SubstraitPlanDecoder for DefaultPlanDecoder {
                 .context(RegisterUdfSnafu { name: func.name() })?;
         }
 
-        // PromQL range functions
-        for func in promql::functions::range_funcs() {
-            let name = func.name().to_string();
-            session_state
-                .register_udf(Arc::new(func))
-                .context(RegisterUdfSnafu { name })?;
-        }
-
-        // Aggregate functions
         for func in FUNCTION_REGISTRY.aggregate_functions() {
             let name = func.name().to_string();
             session_state
@@ -147,6 +140,29 @@ impl SubstraitPlanDecoder for DefaultPlanDecoder {
                 .context(RegisterUdfSnafu { name })?;
         }
 
+        let _ = session_state.register_udaf(quantile_udaf());
+
+        let _ = session_state.register_udf(Arc::new(IDelta::<false>::scalar_udf()));
+        let _ = session_state.register_udf(Arc::new(IDelta::<true>::scalar_udf()));
+        let _ = session_state.register_udf(Arc::new(Rate::scalar_udf()));
+        let _ = session_state.register_udf(Arc::new(Increase::scalar_udf()));
+        let _ = session_state.register_udf(Arc::new(Delta::scalar_udf()));
+        let _ = session_state.register_udf(Arc::new(Resets::scalar_udf()));
+        let _ = session_state.register_udf(Arc::new(Changes::scalar_udf()));
+        let _ = session_state.register_udf(Arc::new(Deriv::scalar_udf()));
+        let _ = session_state.register_udf(Arc::new(Round::scalar_udf()));
+        let _ = session_state.register_udf(Arc::new(AvgOverTime::scalar_udf()));
+        let _ = session_state.register_udf(Arc::new(MinOverTime::scalar_udf()));
+        let _ = session_state.register_udf(Arc::new(MaxOverTime::scalar_udf()));
+        let _ = session_state.register_udf(Arc::new(SumOverTime::scalar_udf()));
+        let _ = session_state.register_udf(Arc::new(CountOverTime::scalar_udf()));
+        let _ = session_state.register_udf(Arc::new(LastOverTime::scalar_udf()));
+        let _ = session_state.register_udf(Arc::new(AbsentOverTime::scalar_udf()));
+        let _ = session_state.register_udf(Arc::new(PresentOverTime::scalar_udf()));
+        let _ = session_state.register_udf(Arc::new(StddevOverTime::scalar_udf()));
+        let _ = session_state.register_udf(Arc::new(StdvarOverTime::scalar_udf()));
+        // TODO(ruihang): add quantile_over_time, predict_linear, holt_winters, round
+
         let logical_plan = DFLogicalSubstraitConvertor
             .decode(message, session_state)
             .await

src/servers/src/grpc.rs

@@ -66,6 +66,8 @@ pub struct GrpcOptions {
     pub max_recv_message_size: ReadableSize,
     /// Max gRPC sending(encoding) message size
     pub max_send_message_size: ReadableSize,
+    /// Compression mode in Arrow Flight service.
+    pub flight_compression: FlightCompression,
     pub runtime_size: usize,
     #[serde(default = "Default::default")]
     pub tls: TlsOption,
@@ -114,6 +116,7 @@ impl Default for GrpcOptions {
             server_addr: String::new(),
             max_recv_message_size: DEFAULT_MAX_GRPC_RECV_MESSAGE_SIZE,
             max_send_message_size: DEFAULT_MAX_GRPC_SEND_MESSAGE_SIZE,
+            flight_compression: FlightCompression::ArrowIpc,
             runtime_size: 8,
             tls: TlsOption::default(),
         }
@@ -132,6 +135,30 @@ impl GrpcOptions {
     }
 }
 
+#[derive(Debug, Serialize, Deserialize, Clone, Copy, PartialEq, Eq, Default)]
+#[serde(rename_all = "snake_case")]
+pub enum FlightCompression {
+    /// Disable all compression in Arrow Flight service.
+    #[default]
+    None,
+    /// Enable only transport layer compression (zstd).
+    Transport,
+    /// Enable only payload compression (lz4)
+    ArrowIpc,
+    /// Enable all compression.
+    All,
+}
+
+impl FlightCompression {
+    pub fn transport_compression(&self) -> bool {
+        self == &FlightCompression::Transport || self == &FlightCompression::All
+    }
+
+    pub fn arrow_compression(&self) -> bool {
+        self == &FlightCompression::ArrowIpc || self == &FlightCompression::All
+    }
+}
+
 pub struct GrpcServer {
     // states
     shutdown_tx: Mutex<Option<Sender<()>>>,

src/servers/src/grpc/flight.rs

@@ -45,7 +45,7 @@ use tonic::{Request, Response, Status, Streaming};
 use crate::error::{InvalidParameterSnafu, ParseJsonSnafu, Result, ToJsonSnafu};
 pub use crate::grpc::flight::stream::FlightRecordBatchStream;
 use crate::grpc::greptime_handler::{get_request_type, GreptimeRequestHandler};
-use crate::grpc::TonicResult;
+use crate::grpc::{FlightCompression, TonicResult};
 use crate::http::header::constants::GREPTIME_DB_HEADER_NAME;
 use crate::http::AUTHORIZATION_HEADER;
 use crate::{error, hint_headers};
@@ -195,9 +195,14 @@ impl FlightCraft for GreptimeRequestHandler {
             protocol = "grpc",
             request_type = get_request_type(&request)
         );
+        let flight_compression = self.flight_compression;
         async {
             let output = self.handle_request(request, hints).await?;
-            let stream = to_flight_data_stream(output, TracingContext::from_current_span());
+            let stream = to_flight_data_stream(
+                output,
+                TracingContext::from_current_span(),
+                flight_compression,
+            );
             Ok(Response::new(stream))
         }
         .trace(span)
@@ -365,14 +370,16 @@ impl Stream for PutRecordBatchRequestStream {
 fn to_flight_data_stream(
     output: Output,
     tracing_context: TracingContext,
+    flight_compression: FlightCompression,
 ) -> TonicStream<FlightData> {
     match output.data {
         OutputData::Stream(stream) => {
-            let stream = FlightRecordBatchStream::new(stream, tracing_context);
+            let stream = FlightRecordBatchStream::new(stream, tracing_context, flight_compression);
             Box::pin(stream) as _
         }
         OutputData::RecordBatches(x) => {
-            let stream = FlightRecordBatchStream::new(x.as_stream(), tracing_context);
+            let stream =
+                FlightRecordBatchStream::new(x.as_stream(), tracing_context, flight_compression);
             Box::pin(stream) as _
         }
         OutputData::AffectedRows(rows) => {

src/servers/src/grpc/flight/stream.rs

@@ -30,6 +30,7 @@ use tokio::task::JoinHandle;
 
 use crate::error;
 use crate::grpc::flight::TonicResult;
+use crate::grpc::FlightCompression;
 
 #[pin_project(PinnedDrop)]
 pub struct FlightRecordBatchStream {
@@ -41,18 +42,27 @@ pub struct FlightRecordBatchStream {
 }
 
 impl FlightRecordBatchStream {
-    pub fn new(recordbatches: SendableRecordBatchStream, tracing_context: TracingContext) -> Self {
+    pub fn new(
+        recordbatches: SendableRecordBatchStream,
+        tracing_context: TracingContext,
+        compression: FlightCompression,
+    ) -> Self {
         let (tx, rx) = mpsc::channel::<TonicResult<FlightMessage>>(1);
         let join_handle = common_runtime::spawn_global(async move {
             Self::flight_data_stream(recordbatches, tx)
                 .trace(tracing_context.attach(info_span!("flight_data_stream")))
                 .await
         });
+        let encoder = if compression.arrow_compression() {
+            FlightEncoder::default()
+        } else {
+            FlightEncoder::with_compression_disabled()
+        };
         Self {
             rx,
             join_handle,
             done: false,
-            encoder: FlightEncoder::with_compression_disabled(),
+            encoder,
         }
     }
 
@@ -161,7 +171,11 @@ mod test {
         let recordbatches = RecordBatches::try_new(schema.clone(), vec![recordbatch.clone()])
             .unwrap()
             .as_stream();
-        let mut stream = FlightRecordBatchStream::new(recordbatches, TracingContext::default());
+        let mut stream = FlightRecordBatchStream::new(
+            recordbatches,
+            TracingContext::default(),
+            FlightCompression::default(),
+        );
 
         let mut raw_data = Vec::with_capacity(2);
         raw_data.push(stream.next().await.unwrap().unwrap());

src/servers/src/grpc/greptime_handler.rs

@@ -40,7 +40,7 @@ use futures_util::StreamExt;
 use session::context::{QueryContext, QueryContextBuilder, QueryContextRef};
 use session::hints::READ_PREFERENCE_HINT;
 use snafu::{OptionExt, ResultExt};
-use table::metadata::TableId;
+use table::TableRef;
 use tokio::sync::mpsc;
 
 use crate::error::Error::UnsupportedAuthScheme;
@@ -49,7 +49,7 @@ use crate::error::{
     JoinTaskSnafu, NotFoundAuthHeaderSnafu, Result, UnknownHintSnafu,
 };
 use crate::grpc::flight::{PutRecordBatchRequest, PutRecordBatchRequestStream};
-use crate::grpc::TonicResult;
+use crate::grpc::{FlightCompression, TonicResult};
 use crate::metrics;
 use crate::metrics::{METRIC_AUTH_FAILURE, METRIC_SERVER_GRPC_DB_REQUEST_TIMER};
 use crate::query_handler::grpc::ServerGrpcQueryHandlerRef;
@@ -59,6 +59,7 @@ pub struct GreptimeRequestHandler {
     handler: ServerGrpcQueryHandlerRef,
     user_provider: Option<UserProviderRef>,
     runtime: Option<Runtime>,
+    pub(crate) flight_compression: FlightCompression,
 }
 
 impl GreptimeRequestHandler {
@@ -66,11 +67,13 @@ impl GreptimeRequestHandler {
         handler: ServerGrpcQueryHandlerRef,
         user_provider: Option<UserProviderRef>,
         runtime: Option<Runtime>,
+        flight_compression: FlightCompression,
     ) -> Self {
         Self {
             handler,
             user_provider,
             runtime,
+            flight_compression,
         }
     }
 
@@ -146,8 +149,8 @@ impl GreptimeRequestHandler {
             .clone()
             .unwrap_or_else(common_runtime::global_runtime);
         runtime.spawn(async move {
-            // Cached table id
+            // Cached table ref
-            let mut table_id: Option<TableId> = None;
+            let mut table_ref: Option<TableRef> = None;
 
             let mut decoder = FlightDecoder::default();
             while let Some(request) = stream.next().await {
@@ -166,7 +169,7 @@ impl GreptimeRequestHandler {
 
                 let timer = metrics::GRPC_BULK_INSERT_ELAPSED.start_timer();
                 let result = handler
-                    .put_record_batch(&table_name, &mut table_id, &mut decoder, data)
+                    .put_record_batch(&table_name, &mut table_ref, &mut decoder, data)
                     .await
                     .inspect_err(|e| error!(e; "Failed to handle flight record batches"));
                 timer.observe_duration();

src/servers/src/query_handler/grpc.rs

@@ -23,8 +23,8 @@ use common_grpc::flight::FlightDecoder;
 use common_query::Output;
 use session::context::QueryContextRef;
 use snafu::ResultExt;
-use table::metadata::TableId;
 use table::table_name::TableName;
+use table::TableRef;
 
 use crate::error::{self, Result};
 
@@ -45,8 +45,8 @@ pub trait GrpcQueryHandler {
 
     async fn put_record_batch(
         &self,
-        table: &TableName,
+        table_name: &TableName,
-        table_id: &mut Option<TableId>,
+        table_ref: &mut Option<TableRef>,
         decoder: &mut FlightDecoder,
         flight_data: FlightData,
     ) -> std::result::Result<AffectedRows, Self::Error>;
@@ -77,13 +77,13 @@ where
 
     async fn put_record_batch(
         &self,
-        table: &TableName,
+        table_name: &TableName,
-        table_id: &mut Option<TableId>,
+        table_ref: &mut Option<TableRef>,
         decoder: &mut FlightDecoder,
         data: FlightData,
     ) -> Result<AffectedRows> {
         self.0
-            .put_record_batch(table, table_id, decoder, data)
+            .put_record_batch(table_name, table_ref, decoder, data)
             .await
             .map_err(BoxedError::new)
             .context(error::ExecuteGrpcRequestSnafu)

src/servers/tests/mod.rs

@@ -34,7 +34,6 @@ use servers::query_handler::sql::{ServerSqlQueryHandlerRef, SqlQueryHandler};
 use session::context::QueryContextRef;
 use snafu::ensure;
 use sql::statements::statement::Statement;
-use table::metadata::TableId;
 use table::table_name::TableName;
 use table::TableRef;
 
@@ -160,15 +159,11 @@ impl GrpcQueryHandler for DummyInstance {
 
     async fn put_record_batch(
         &self,
-        table: &TableName,
+        _table_name: &TableName,
-        table_id: &mut Option<TableId>,
+        _table_ref: &mut Option<TableRef>,
-        decoder: &mut FlightDecoder,
+        _decoder: &mut FlightDecoder,
-        data: FlightData,
+        _data: FlightData,
     ) -> std::result::Result<AffectedRows, Self::Error> {
-        let _ = table;
-        let _ = data;
-        let _ = table_id;
-        let _ = decoder;
         unimplemented!()
     }
 }

tests-integration/src/grpc/flight.rs

@@ -34,7 +34,7 @@ mod test {
     use itertools::Itertools;
     use servers::grpc::builder::GrpcServerBuilder;
     use servers::grpc::greptime_handler::GreptimeRequestHandler;
-    use servers::grpc::GrpcServerConfig;
+    use servers::grpc::{FlightCompression, GrpcServerConfig};
     use servers::query_handler::grpc::ServerGrpcQueryHandlerAdapter;
     use servers::server::Server;
 
@@ -94,6 +94,7 @@ mod test {
             )
             .ok(),
             Some(runtime.clone()),
+            FlightCompression::default(),
         );
         let mut grpc_server = GrpcServerBuilder::new(GrpcServerConfig::default(), runtime)
             .flight_handler(Arc::new(greptime_request_handler))
@@ -139,8 +140,7 @@ mod test {
         let schema = record_batches[0].schema.arrow_schema().clone();
 
         let stream = futures::stream::once(async move {
-            let mut schema_data =
+            let mut schema_data = FlightEncoder::default().encode(FlightMessage::Schema(schema));
-                FlightEncoder::with_compression_disabled().encode(FlightMessage::Schema(schema));
             let metadata = DoPutMetadata::new(0);
             schema_data.app_metadata = serde_json::to_vec(&metadata).unwrap().into();
             // first message in "DoPut" stream should carry table name in flight descriptor
@@ -155,7 +155,7 @@ mod test {
             tokio_stream::iter(record_batches)
                 .enumerate()
                 .map(|(i, x)| {
-                    let mut encoder = FlightEncoder::with_compression_disabled();
+                    let mut encoder = FlightEncoder::default();
                     let message = FlightMessage::RecordBatch(x.into_df_record_batch());
                     let mut data = encoder.encode(message);
                     let metadata = DoPutMetadata::new((i + 1) as i64);

tests-integration/src/test_util.rs

@@ -42,7 +42,7 @@ use object_store::test_util::TempFolder;
 use object_store::ObjectStore;
 use servers::grpc::builder::GrpcServerBuilder;
 use servers::grpc::greptime_handler::GreptimeRequestHandler;
-use servers::grpc::{GrpcOptions, GrpcServer, GrpcServerConfig};
+use servers::grpc::{FlightCompression, GrpcOptions, GrpcServer, GrpcServerConfig};
 use servers::http::{HttpOptions, HttpServerBuilder, PromValidationMode};
 use servers::metrics_handler::MetricsHandler;
 use servers::mysql::server::{MysqlServer, MysqlSpawnConfig, MysqlSpawnRef};
@@ -585,6 +585,7 @@ pub async fn setup_grpc_server_with(
         ServerGrpcQueryHandlerAdapter::arc(fe_instance_ref.clone()),
         user_provider.clone(),
         Some(runtime.clone()),
+        FlightCompression::default(),
     );
 
     let flight_handler = Arc::new(greptime_request_handler.clone());

tests-integration/tests/http.rs

@@ -1025,6 +1025,7 @@ bind_addr = "127.0.0.1:4001"
 server_addr = "127.0.0.1:4001"
 max_recv_message_size = "512MiB"
 max_send_message_size = "512MiB"
+flight_compression = "arrow_ipc"
 runtime_size = 8
 
 [grpc.tls]