risingwave_rpc_client/

meta_client.rs

// Copyright 2025 RisingWave Labs
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
//     http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.

use std::collections::{BTreeMap, HashMap, HashSet};
use std::fmt::{Debug, Display};
use std::sync::Arc;
use std::sync::atomic::AtomicBool;
use std::sync::atomic::Ordering::Relaxed;
use std::thread;
use std::time::{Duration, SystemTime};

use anyhow::{Context, anyhow};
use async_trait::async_trait;
use cluster_limit_service_client::ClusterLimitServiceClient;
use either::Either;
use futures::stream::BoxStream;
use list_rate_limits_response::RateLimitInfo;
use lru::LruCache;
use replace_job_plan::ReplaceJob;
use risingwave_common::RW_VERSION;
use risingwave_common::catalog::{
    AlterDatabaseParam, FunctionId, IndexId, ObjectId, SecretId, TableId,
};
use risingwave_common::config::{MAX_CONNECTION_WINDOW_SIZE, MetaConfig};
use risingwave_common::hash::WorkerSlotMapping;
use risingwave_common::id::{
    ConnectionId, DatabaseId, JobId, SchemaId, SinkId, SubscriptionId, ViewId, WorkerId,
};
use risingwave_common::monitor::EndpointExt;
use risingwave_common::system_param::reader::SystemParamsReader;
use risingwave_common::telemetry::report::TelemetryInfoFetcher;
use risingwave_common::util::addr::HostAddr;
use risingwave_common::util::meta_addr::MetaAddressStrategy;
use risingwave_common::util::resource_util::cpu::total_cpu_available;
use risingwave_common::util::resource_util::hostname;
use risingwave_common::util::resource_util::memory::system_memory_available_bytes;
use risingwave_error::bail;
use risingwave_error::tonic::ErrorIsFromTonicServerImpl;
use risingwave_hummock_sdk::version::{HummockVersion, HummockVersionDelta};
use risingwave_hummock_sdk::{
    CompactionGroupId, HummockEpoch, HummockVersionId, ObjectIdRange, SyncResult,
};
use risingwave_pb::backup_service::backup_service_client::BackupServiceClient;
use risingwave_pb::backup_service::*;
use risingwave_pb::catalog::{
    Connection, PbComment, PbDatabase, PbFunction, PbIndex, PbSchema, PbSink, PbSource,
    PbSubscription, PbTable, PbView, Table,
};
use risingwave_pb::cloud_service::cloud_service_client::CloudServiceClient;
use risingwave_pb::cloud_service::*;
use risingwave_pb::common::worker_node::Property;
use risingwave_pb::common::{HostAddress, OptionalUint32, OptionalUint64, WorkerNode, WorkerType};
use risingwave_pb::connector_service::sink_coordination_service_client::SinkCoordinationServiceClient;
use risingwave_pb::ddl_service::alter_owner_request::Object;
use risingwave_pb::ddl_service::create_iceberg_table_request::{PbSinkJobInfo, PbTableJobInfo};
use risingwave_pb::ddl_service::create_materialized_view_request::PbBackfillType;
use risingwave_pb::ddl_service::ddl_service_client::DdlServiceClient;
use risingwave_pb::ddl_service::*;
use risingwave_pb::hummock::compact_task::TaskStatus;
use risingwave_pb::hummock::get_compaction_score_response::PickerInfo;
use risingwave_pb::hummock::hummock_manager_service_client::HummockManagerServiceClient;
use risingwave_pb::hummock::rise_ctl_update_compaction_config_request::mutable_config::MutableConfig;
use risingwave_pb::hummock::subscribe_compaction_event_request::Register;
use risingwave_pb::hummock::write_limits::WriteLimit;
use risingwave_pb::hummock::*;
use risingwave_pb::iceberg_compaction::subscribe_iceberg_compaction_event_request::Register as IcebergRegister;
use risingwave_pb::iceberg_compaction::{
    SubscribeIcebergCompactionEventRequest, SubscribeIcebergCompactionEventResponse,
    subscribe_iceberg_compaction_event_request,
};
use risingwave_pb::id::{ActorId, FragmentId, SourceId};
use risingwave_pb::meta::alter_connector_props_request::{
    AlterConnectorPropsObject, AlterIcebergTableIds, ExtraOptions,
};
use risingwave_pb::meta::cancel_creating_jobs_request::PbJobs;
use risingwave_pb::meta::cluster_service_client::ClusterServiceClient;
use risingwave_pb::meta::event_log_service_client::EventLogServiceClient;
use risingwave_pb::meta::heartbeat_service_client::HeartbeatServiceClient;
use risingwave_pb::meta::hosted_iceberg_catalog_service_client::HostedIcebergCatalogServiceClient;
use risingwave_pb::meta::list_actor_splits_response::ActorSplit;
use risingwave_pb::meta::list_actor_states_response::ActorState;
use risingwave_pb::meta::list_cdc_progress_response::PbCdcProgress;
use risingwave_pb::meta::list_iceberg_tables_response::IcebergTable;
use risingwave_pb::meta::list_object_dependencies_response::PbObjectDependencies;
use risingwave_pb::meta::list_streaming_job_states_response::StreamingJobState;
use risingwave_pb::meta::list_table_fragments_response::TableFragmentInfo;
use risingwave_pb::meta::meta_member_service_client::MetaMemberServiceClient;
use risingwave_pb::meta::notification_service_client::NotificationServiceClient;
use risingwave_pb::meta::scale_service_client::ScaleServiceClient;
use risingwave_pb::meta::serving_service_client::ServingServiceClient;
use risingwave_pb::meta::session_param_service_client::SessionParamServiceClient;
use risingwave_pb::meta::stream_manager_service_client::StreamManagerServiceClient;
use risingwave_pb::meta::system_params_service_client::SystemParamsServiceClient;
use risingwave_pb::meta::telemetry_info_service_client::TelemetryInfoServiceClient;
use risingwave_pb::meta::update_worker_node_schedulability_request::Schedulability;
use risingwave_pb::meta::{FragmentDistribution, *};
use risingwave_pb::monitor_service::monitor_service_client::MonitorServiceClient;
use risingwave_pb::monitor_service::stack_trace_request::ActorTracesFormat;
use risingwave_pb::monitor_service::{StackTraceRequest, StackTraceResponse};
use risingwave_pb::secret::PbSecretRef;
use risingwave_pb::stream_plan::StreamFragmentGraph;
use risingwave_pb::user::alter_default_privilege_request::Operation as AlterDefaultPrivilegeOperation;
use risingwave_pb::user::update_user_request::UpdateField;
use risingwave_pb::user::user_service_client::UserServiceClient;
use risingwave_pb::user::*;
use thiserror_ext::AsReport;
use tokio::sync::mpsc::{Receiver, UnboundedSender, unbounded_channel};
use tokio::sync::oneshot::Sender;
use tokio::sync::{RwLock, mpsc, oneshot};
use tokio::task::JoinHandle;
use tokio::time::{self};
use tokio_retry::strategy::{ExponentialBackoff, jitter};
use tokio_stream::wrappers::UnboundedReceiverStream;
use tonic::transport::Endpoint;
use tonic::{Code, Request, Streaming};

use crate::channel::{Channel, WrappedChannelExt};
use crate::error::{Result, RpcError};
use crate::hummock_meta_client::{
    CompactionEventItem, HummockMetaClient, HummockMetaClientChangeLogInfo,
    IcebergCompactionEventItem,
};
use crate::meta_rpc_client_method_impl;

/// Client to meta server. Cloning the instance is lightweight.
#[derive(Clone, Debug)]
pub struct MetaClient {
    worker_id: WorkerId,
    worker_type: WorkerType,
    host_addr: HostAddr,
    inner: GrpcMetaClient,
    meta_config: Arc<MetaConfig>,
    cluster_id: String,
    shutting_down: Arc<AtomicBool>,
}

impl MetaClient {
    pub fn worker_id(&self) -> WorkerId {
        self.worker_id
    }

    pub fn host_addr(&self) -> &HostAddr {
        &self.host_addr
    }

    pub fn worker_type(&self) -> WorkerType {
        self.worker_type
    }

    pub fn cluster_id(&self) -> &str {
        &self.cluster_id
    }

    /// Subscribe to notification from meta.
    pub async fn subscribe(
        &self,
        subscribe_type: SubscribeType,
    ) -> Result<Streaming<SubscribeResponse>> {
        let request = SubscribeRequest {
            subscribe_type: subscribe_type as i32,
            host: Some(self.host_addr.to_protobuf()),
            worker_id: self.worker_id(),
        };

        let retry_strategy = GrpcMetaClient::retry_strategy_to_bound(
            Duration::from_secs(self.meta_config.max_heartbeat_interval_secs as u64),
            true,
        );

        tokio_retry::Retry::spawn(retry_strategy, || async {
            let request = request.clone();
            self.inner.subscribe(request).await
        })
        .await
    }

    pub async fn create_connection(
        &self,
        connection_name: String,
        database_id: DatabaseId,
        schema_id: SchemaId,
        owner_id: u32,
        req: create_connection_request::Payload,
    ) -> Result<WaitVersion> {
        let request = CreateConnectionRequest {
            name: connection_name,
            database_id,
            schema_id,
            owner_id,
            payload: Some(req),
        };
        let resp = self.inner.create_connection(request).await?;
        Ok(resp
            .version
            .ok_or_else(|| anyhow!("wait version not set"))?)
    }

    pub async fn create_secret(
        &self,
        secret_name: String,
        database_id: DatabaseId,
        schema_id: SchemaId,
        owner_id: u32,
        value: Vec<u8>,
    ) -> Result<WaitVersion> {
        let request = CreateSecretRequest {
            name: secret_name,
            database_id,
            schema_id,
            owner_id,
            value,
        };
        let resp = self.inner.create_secret(request).await?;
        Ok(resp
            .version
            .ok_or_else(|| anyhow!("wait version not set"))?)
    }

    pub async fn list_connections(&self, _name: Option<&str>) -> Result<Vec<Connection>> {
        let request = ListConnectionsRequest {};
        let resp = self.inner.list_connections(request).await?;
        Ok(resp.connections)
    }

    pub async fn drop_connection(
        &self,
        connection_id: ConnectionId,
        cascade: bool,
    ) -> Result<WaitVersion> {
        let request = DropConnectionRequest {
            connection_id,
            cascade,
        };
        let resp = self.inner.drop_connection(request).await?;
        Ok(resp
            .version
            .ok_or_else(|| anyhow!("wait version not set"))?)
    }

    pub async fn drop_secret(&self, secret_id: SecretId) -> Result<WaitVersion> {
        let request = DropSecretRequest { secret_id };
        let resp = self.inner.drop_secret(request).await?;
        Ok(resp
            .version
            .ok_or_else(|| anyhow!("wait version not set"))?)
    }

    /// Register the current node to the cluster and set the corresponding worker id.
    ///
    /// Retry if there's connection issue with the meta node. Exit the process if the registration fails.
    pub async fn register_new(
        addr_strategy: MetaAddressStrategy,
        worker_type: WorkerType,
        addr: &HostAddr,
        property: Property,
        meta_config: Arc<MetaConfig>,
    ) -> (Self, SystemParamsReader) {
        let ret =
            Self::register_new_inner(addr_strategy, worker_type, addr, property, meta_config).await;

        match ret {
            Ok(ret) => ret,
            Err(err) => {
                tracing::error!(error = %err.as_report(), "failed to register worker, exiting...");
                std::process::exit(1);
            }
        }
    }

    async fn register_new_inner(
        addr_strategy: MetaAddressStrategy,
        worker_type: WorkerType,
        addr: &HostAddr,
        property: Property,
        meta_config: Arc<MetaConfig>,
    ) -> Result<(Self, SystemParamsReader)> {
        tracing::info!("register meta client using strategy: {}", addr_strategy);

        // Retry until reaching `max_heartbeat_interval_secs`
        let retry_strategy = GrpcMetaClient::retry_strategy_to_bound(
            Duration::from_secs(meta_config.max_heartbeat_interval_secs as u64),
            true,
        );

        if property.is_unschedulable {
            tracing::warn!("worker {:?} registered as unschedulable", addr.clone());
        }
        let init_result: Result<_> = tokio_retry::RetryIf::spawn(
            retry_strategy,
            || async {
                let grpc_meta_client =
                    GrpcMetaClient::new(&addr_strategy, meta_config.clone()).await?;

                let add_worker_resp = grpc_meta_client
                    .add_worker_node(AddWorkerNodeRequest {
                        worker_type: worker_type as i32,
                        host: Some(addr.to_protobuf()),
                        property: Some(property.clone()),
                        resource: Some(risingwave_pb::common::worker_node::Resource {
                            rw_version: RW_VERSION.to_owned(),
                            total_memory_bytes: system_memory_available_bytes() as _,
                            total_cpu_cores: total_cpu_available() as _,
                            hostname: hostname(),
                        }),
                    })
                    .await
                    .context("failed to add worker node")?;

                let system_params_resp = grpc_meta_client
                    .get_system_params(GetSystemParamsRequest {})
                    .await
                    .context("failed to get initial system params")?;

                Ok((add_worker_resp, system_params_resp, grpc_meta_client))
            },
            // Only retry if there's any transient connection issue.
            // If the error is from our implementation or business, do not retry it.
            |e: &RpcError| !e.is_from_tonic_server_impl(),
        )
        .await;

        let (add_worker_resp, system_params_resp, grpc_meta_client) = init_result?;
        let worker_id = add_worker_resp
            .node_id
            .expect("AddWorkerNodeResponse::node_id is empty");

        let meta_client = Self {
            worker_id,
            worker_type,
            host_addr: addr.clone(),
            inner: grpc_meta_client,
            meta_config: meta_config.clone(),
            cluster_id: add_worker_resp.cluster_id,
            shutting_down: Arc::new(false.into()),
        };

        static REPORT_PANIC: std::sync::Once = std::sync::Once::new();
        REPORT_PANIC.call_once(|| {
            let meta_client_clone = meta_client.clone();
            std::panic::update_hook(move |default_hook, info| {
                // Try to report panic event to meta node.
                meta_client_clone.try_add_panic_event_blocking(info, None);
                default_hook(info);
            });
        });

        Ok((meta_client, system_params_resp.params.unwrap().into()))
    }

    /// Activate the current node in cluster to confirm it's ready to serve.
    pub async fn activate(&self, addr: &HostAddr) -> Result<()> {
        let request = ActivateWorkerNodeRequest {
            host: Some(addr.to_protobuf()),
            node_id: self.worker_id,
        };
        let retry_strategy = GrpcMetaClient::retry_strategy_to_bound(
            Duration::from_secs(self.meta_config.max_heartbeat_interval_secs as u64),
            true,
        );
        tokio_retry::Retry::spawn(retry_strategy, || async {
            let request = request.clone();
            self.inner.activate_worker_node(request).await
        })
        .await?;

        Ok(())
    }

    /// Send heartbeat signal to meta service.
    pub async fn send_heartbeat(&self, node_id: WorkerId) -> Result<()> {
        let request = HeartbeatRequest { node_id };
        let resp = self.inner.heartbeat(request).await?;
        if let Some(status) = resp.status
            && status.code() == risingwave_pb::common::status::Code::UnknownWorker
        {
            // Ignore the error if we're already shutting down.
            // Otherwise, exit the process.
            if !self.shutting_down.load(Relaxed) {
                tracing::error!(message = status.message, "worker expired");
                std::process::exit(1);
            }
        }
        Ok(())
    }

    pub async fn create_database(&self, db: PbDatabase) -> Result<WaitVersion> {
        let request = CreateDatabaseRequest { db: Some(db) };
        let resp = self.inner.create_database(request).await?;
        // TODO: handle error in `resp.status` here
        Ok(resp
            .version
            .ok_or_else(|| anyhow!("wait version not set"))?)
    }

    pub async fn create_schema(&self, schema: PbSchema) -> Result<WaitVersion> {
        let request = CreateSchemaRequest {
            schema: Some(schema),
        };
        let resp = self.inner.create_schema(request).await?;
        // TODO: handle error in `resp.status` here
        Ok(resp
            .version
            .ok_or_else(|| anyhow!("wait version not set"))?)
    }

    pub async fn create_materialized_view(
        &self,
        table: PbTable,
        graph: StreamFragmentGraph,
        dependencies: HashSet<ObjectId>,
        specific_resource_group: Option<String>,
        if_not_exists: bool,
    ) -> Result<WaitVersion> {
        let request = CreateMaterializedViewRequest {
            materialized_view: Some(table),
            fragment_graph: Some(graph),
            backfill: PbBackfillType::Regular as _,
            dependencies: dependencies.into_iter().collect(),
            specific_resource_group,
            if_not_exists,
        };
        let resp = self.inner.create_materialized_view(request).await?;
        // TODO: handle error in `resp.status` here
        Ok(resp
            .version
            .ok_or_else(|| anyhow!("wait version not set"))?)
    }

    pub async fn drop_materialized_view(
        &self,
        table_id: TableId,
        cascade: bool,
    ) -> Result<WaitVersion> {
        let request = DropMaterializedViewRequest { table_id, cascade };

        let resp = self.inner.drop_materialized_view(request).await?;
        Ok(resp
            .version
            .ok_or_else(|| anyhow!("wait version not set"))?)
    }

    pub async fn create_source(
        &self,
        source: PbSource,
        graph: Option<StreamFragmentGraph>,
        if_not_exists: bool,
    ) -> Result<WaitVersion> {
        let request = CreateSourceRequest {
            source: Some(source),
            fragment_graph: graph,
            if_not_exists,
        };

        let resp = self.inner.create_source(request).await?;
        Ok(resp
            .version
            .ok_or_else(|| anyhow!("wait version not set"))?)
    }

    pub async fn create_sink(
        &self,
        sink: PbSink,
        graph: StreamFragmentGraph,
        dependencies: HashSet<ObjectId>,
        if_not_exists: bool,
    ) -> Result<WaitVersion> {
        let request = CreateSinkRequest {
            sink: Some(sink),
            fragment_graph: Some(graph),
            dependencies: dependencies.into_iter().collect(),
            if_not_exists,
        };

        let resp = self.inner.create_sink(request).await?;
        Ok(resp
            .version
            .ok_or_else(|| anyhow!("wait version not set"))?)
    }

    pub async fn create_subscription(&self, subscription: PbSubscription) -> Result<WaitVersion> {
        let request = CreateSubscriptionRequest {
            subscription: Some(subscription),
        };

        let resp = self.inner.create_subscription(request).await?;
        Ok(resp
            .version
            .ok_or_else(|| anyhow!("wait version not set"))?)
    }

    pub async fn create_function(&self, function: PbFunction) -> Result<WaitVersion> {
        let request = CreateFunctionRequest {
            function: Some(function),
        };
        let resp = self.inner.create_function(request).await?;
        Ok(resp
            .version
            .ok_or_else(|| anyhow!("wait version not set"))?)
    }

    pub async fn create_table(
        &self,
        source: Option<PbSource>,
        table: PbTable,
        graph: StreamFragmentGraph,
        job_type: PbTableJobType,
        if_not_exists: bool,
        dependencies: HashSet<ObjectId>,
    ) -> Result<WaitVersion> {
        let request = CreateTableRequest {
            materialized_view: Some(table),
            fragment_graph: Some(graph),
            source,
            job_type: job_type as _,
            if_not_exists,
            dependencies: dependencies.into_iter().collect(),
        };
        let resp = self.inner.create_table(request).await?;
        // TODO: handle error in `resp.status` here
        Ok(resp
            .version
            .ok_or_else(|| anyhow!("wait version not set"))?)
    }

    pub async fn comment_on(&self, comment: PbComment) -> Result<WaitVersion> {
        let request = CommentOnRequest {
            comment: Some(comment),
        };
        let resp = self.inner.comment_on(request).await?;
        Ok(resp
            .version
            .ok_or_else(|| anyhow!("wait version not set"))?)
    }

    pub async fn alter_name(
        &self,
        object: alter_name_request::Object,
        name: &str,
    ) -> Result<WaitVersion> {
        let request = AlterNameRequest {
            object: Some(object),
            new_name: name.to_owned(),
        };
        let resp = self.inner.alter_name(request).await?;
        Ok(resp
            .version
            .ok_or_else(|| anyhow!("wait version not set"))?)
    }

    pub async fn alter_owner(&self, object: Object, owner_id: u32) -> Result<WaitVersion> {
        let request = AlterOwnerRequest {
            object: Some(object),
            owner_id,
        };
        let resp = self.inner.alter_owner(request).await?;
        Ok(resp
            .version
            .ok_or_else(|| anyhow!("wait version not set"))?)
    }

    pub async fn alter_database_param(
        &self,
        database_id: DatabaseId,
        param: AlterDatabaseParam,
    ) -> Result<WaitVersion> {
        let request = match param {
            AlterDatabaseParam::BarrierIntervalMs(barrier_interval_ms) => {
                let barrier_interval_ms = OptionalUint32 {
                    value: barrier_interval_ms,
                };
                AlterDatabaseParamRequest {
                    database_id,
                    param: Some(alter_database_param_request::Param::BarrierIntervalMs(
                        barrier_interval_ms,
                    )),
                }
            }
            AlterDatabaseParam::CheckpointFrequency(checkpoint_frequency) => {
                let checkpoint_frequency = OptionalUint64 {
                    value: checkpoint_frequency,
                };
                AlterDatabaseParamRequest {
                    database_id,
                    param: Some(alter_database_param_request::Param::CheckpointFrequency(
                        checkpoint_frequency,
                    )),
                }
            }
        };
        let resp = self.inner.alter_database_param(request).await?;
        Ok(resp
            .version
            .ok_or_else(|| anyhow!("wait version not set"))?)
    }

    pub async fn alter_set_schema(
        &self,
        object: alter_set_schema_request::Object,
        new_schema_id: SchemaId,
    ) -> Result<WaitVersion> {
        let request = AlterSetSchemaRequest {
            new_schema_id,
            object: Some(object),
        };
        let resp = self.inner.alter_set_schema(request).await?;
        Ok(resp
            .version
            .ok_or_else(|| anyhow!("wait version not set"))?)
    }

    pub async fn alter_source(&self, source: PbSource) -> Result<WaitVersion> {
        let request = AlterSourceRequest {
            source: Some(source),
        };
        let resp = self.inner.alter_source(request).await?;
        Ok(resp
            .version
            .ok_or_else(|| anyhow!("wait version not set"))?)
    }

    pub async fn alter_parallelism(
        &self,
        job_id: JobId,
        parallelism: PbTableParallelism,
        deferred: bool,
    ) -> Result<()> {
        let request = AlterParallelismRequest {
            table_id: job_id,
            parallelism: Some(parallelism),
            deferred,
        };

        self.inner.alter_parallelism(request).await?;
        Ok(())
    }

    pub async fn alter_fragment_parallelism(
        &self,
        fragment_ids: Vec<FragmentId>,
        parallelism: Option<PbTableParallelism>,
    ) -> Result<()> {
        let request = AlterFragmentParallelismRequest {
            fragment_ids,
            parallelism,
        };

        self.inner.alter_fragment_parallelism(request).await?;
        Ok(())
    }

    pub async fn alter_cdc_table_backfill_parallelism(
        &self,
        table_id: JobId,
        parallelism: PbTableParallelism,
    ) -> Result<()> {
        let request = AlterCdcTableBackfillParallelismRequest {
            table_id,
            parallelism: Some(parallelism),
        };
        self.inner
            .alter_cdc_table_backfill_parallelism(request)
            .await?;
        Ok(())
    }

    pub async fn alter_resource_group(
        &self,
        table_id: TableId,
        resource_group: Option<String>,
        deferred: bool,
    ) -> Result<()> {
        let request = AlterResourceGroupRequest {
            table_id,
            resource_group,
            deferred,
        };

        self.inner.alter_resource_group(request).await?;
        Ok(())
    }

    pub async fn alter_swap_rename(
        &self,
        object: alter_swap_rename_request::Object,
    ) -> Result<WaitVersion> {
        let request = AlterSwapRenameRequest {
            object: Some(object),
        };
        let resp = self.inner.alter_swap_rename(request).await?;
        Ok(resp
            .version
            .ok_or_else(|| anyhow!("wait version not set"))?)
    }

    pub async fn alter_secret(
        &self,
        secret_id: SecretId,
        secret_name: String,
        database_id: DatabaseId,
        schema_id: SchemaId,
        owner_id: u32,
        value: Vec<u8>,
    ) -> Result<WaitVersion> {
        let request = AlterSecretRequest {
            secret_id,
            name: secret_name,
            database_id,
            schema_id,
            owner_id,
            value,
        };
        let resp = self.inner.alter_secret(request).await?;
        Ok(resp
            .version
            .ok_or_else(|| anyhow!("wait version not set"))?)
    }

    pub async fn replace_job(
        &self,
        graph: StreamFragmentGraph,
        replace_job: ReplaceJob,
    ) -> Result<WaitVersion> {
        let request = ReplaceJobPlanRequest {
            plan: Some(ReplaceJobPlan {
                fragment_graph: Some(graph),
                replace_job: Some(replace_job),
            }),
        };
        let resp = self.inner.replace_job_plan(request).await?;
        // TODO: handle error in `resp.status` here
        Ok(resp
            .version
            .ok_or_else(|| anyhow!("wait version not set"))?)
    }

    pub async fn auto_schema_change(&self, schema_change: SchemaChangeEnvelope) -> Result<()> {
        let request = AutoSchemaChangeRequest {
            schema_change: Some(schema_change),
        };
        let _ = self.inner.auto_schema_change(request).await?;
        Ok(())
    }

    pub async fn create_view(
        &self,
        view: PbView,
        dependencies: HashSet<ObjectId>,
    ) -> Result<WaitVersion> {
        let request = CreateViewRequest {
            view: Some(view),
            dependencies: dependencies.into_iter().collect(),
        };
        let resp = self.inner.create_view(request).await?;
        // TODO: handle error in `resp.status` here
        Ok(resp
            .version
            .ok_or_else(|| anyhow!("wait version not set"))?)
    }

    pub async fn create_index(
        &self,
        index: PbIndex,
        table: PbTable,
        graph: StreamFragmentGraph,
        if_not_exists: bool,
    ) -> Result<WaitVersion> {
        let request = CreateIndexRequest {
            index: Some(index),
            index_table: Some(table),
            fragment_graph: Some(graph),
            if_not_exists,
        };
        let resp = self.inner.create_index(request).await?;
        // TODO: handle error in `resp.status` here
        Ok(resp
            .version
            .ok_or_else(|| anyhow!("wait version not set"))?)
    }

    pub async fn drop_table(
        &self,
        source_id: Option<u32>,
        table_id: TableId,
        cascade: bool,
    ) -> Result<WaitVersion> {
        let request = DropTableRequest {
            source_id: source_id.map(risingwave_pb::ddl_service::drop_table_request::SourceId::Id),
            table_id,
            cascade,
        };

        let resp = self.inner.drop_table(request).await?;
        Ok(resp
            .version
            .ok_or_else(|| anyhow!("wait version not set"))?)
    }

    pub async fn compact_iceberg_table(&self, sink_id: SinkId) -> Result<u64> {
        let request = CompactIcebergTableRequest { sink_id };
        let resp = self.inner.compact_iceberg_table(request).await?;
        Ok(resp.task_id)
    }

    pub async fn expire_iceberg_table_snapshots(&self, sink_id: SinkId) -> Result<()> {
        let request = ExpireIcebergTableSnapshotsRequest { sink_id };
        let _resp = self.inner.expire_iceberg_table_snapshots(request).await?;
        Ok(())
    }

    pub async fn drop_view(&self, view_id: ViewId, cascade: bool) -> Result<WaitVersion> {
        let request = DropViewRequest { view_id, cascade };
        let resp = self.inner.drop_view(request).await?;
        Ok(resp
            .version
            .ok_or_else(|| anyhow!("wait version not set"))?)
    }

    pub async fn drop_source(&self, source_id: SourceId, cascade: bool) -> Result<WaitVersion> {
        let request = DropSourceRequest { source_id, cascade };
        let resp = self.inner.drop_source(request).await?;
        Ok(resp
            .version
            .ok_or_else(|| anyhow!("wait version not set"))?)
    }

    pub async fn drop_sink(&self, sink_id: SinkId, cascade: bool) -> Result<WaitVersion> {
        let request = DropSinkRequest { sink_id, cascade };
        let resp = self.inner.drop_sink(request).await?;
        Ok(resp
            .version
            .ok_or_else(|| anyhow!("wait version not set"))?)
    }

    pub async fn drop_subscription(
        &self,
        subscription_id: SubscriptionId,
        cascade: bool,
    ) -> Result<WaitVersion> {
        let request = DropSubscriptionRequest {
            subscription_id,
            cascade,
        };
        let resp = self.inner.drop_subscription(request).await?;
        Ok(resp
            .version
            .ok_or_else(|| anyhow!("wait version not set"))?)
    }

    pub async fn drop_index(&self, index_id: IndexId, cascade: bool) -> Result<WaitVersion> {
        let request = DropIndexRequest { index_id, cascade };
        let resp = self.inner.drop_index(request).await?;
        Ok(resp
            .version
            .ok_or_else(|| anyhow!("wait version not set"))?)
    }

    pub async fn drop_function(
        &self,
        function_id: FunctionId,
        cascade: bool,
    ) -> Result<WaitVersion> {
        let request = DropFunctionRequest {
            function_id,
            cascade,
        };
        let resp = self.inner.drop_function(request).await?;
        Ok(resp
            .version
            .ok_or_else(|| anyhow!("wait version not set"))?)
    }

    pub async fn drop_database(&self, database_id: DatabaseId) -> Result<WaitVersion> {
        let request = DropDatabaseRequest { database_id };
        let resp = self.inner.drop_database(request).await?;
        Ok(resp
            .version
            .ok_or_else(|| anyhow!("wait version not set"))?)
    }

    pub async fn drop_schema(&self, schema_id: SchemaId, cascade: bool) -> Result<WaitVersion> {
        let request = DropSchemaRequest { schema_id, cascade };
        let resp = self.inner.drop_schema(request).await?;
        Ok(resp
            .version
            .ok_or_else(|| anyhow!("wait version not set"))?)
    }

    // TODO: using UserInfoVersion instead as return type.
    pub async fn create_user(&self, user: UserInfo) -> Result<u64> {
        let request = CreateUserRequest { user: Some(user) };
        let resp = self.inner.create_user(request).await?;
        Ok(resp.version)
    }

    pub async fn drop_user(&self, user_id: u32) -> Result<u64> {
        let request = DropUserRequest { user_id };
        let resp = self.inner.drop_user(request).await?;
        Ok(resp.version)
    }

    pub async fn update_user(
        &self,
        user: UserInfo,
        update_fields: Vec<UpdateField>,
    ) -> Result<u64> {
        let request = UpdateUserRequest {
            user: Some(user),
            update_fields: update_fields
                .into_iter()
                .map(|field| field as i32)
                .collect::<Vec<_>>(),
        };
        let resp = self.inner.update_user(request).await?;
        Ok(resp.version)
    }

    pub async fn grant_privilege(
        &self,
        user_ids: Vec<u32>,
        privileges: Vec<GrantPrivilege>,
        with_grant_option: bool,
        granted_by: u32,
    ) -> Result<u64> {
        let request = GrantPrivilegeRequest {
            user_ids,
            privileges,
            with_grant_option,
            granted_by,
        };
        let resp = self.inner.grant_privilege(request).await?;
        Ok(resp.version)
    }

    pub async fn revoke_privilege(
        &self,
        user_ids: Vec<u32>,
        privileges: Vec<GrantPrivilege>,
        granted_by: u32,
        revoke_by: u32,
        revoke_grant_option: bool,
        cascade: bool,
    ) -> Result<u64> {
        let request = RevokePrivilegeRequest {
            user_ids,
            privileges,
            granted_by,
            revoke_by,
            revoke_grant_option,
            cascade,
        };
        let resp = self.inner.revoke_privilege(request).await?;
        Ok(resp.version)
    }

    pub async fn alter_default_privilege(
        &self,
        user_ids: Vec<u32>,
        database_id: DatabaseId,
        schema_ids: Vec<SchemaId>,
        operation: AlterDefaultPrivilegeOperation,
        granted_by: u32,
    ) -> Result<()> {
        let request = AlterDefaultPrivilegeRequest {
            user_ids,
            database_id,
            schema_ids,
            operation: Some(operation),
            granted_by,
        };
        self.inner.alter_default_privilege(request).await?;
        Ok(())
    }

    /// Unregister the current node from the cluster.
    pub async fn unregister(&self) -> Result<()> {
        let request = DeleteWorkerNodeRequest {
            host: Some(self.host_addr.to_protobuf()),
        };
        self.inner.delete_worker_node(request).await?;
        self.shutting_down.store(true, Relaxed);
        Ok(())
    }

    /// Try to unregister the current worker from the cluster with best effort. Log the result.
    pub async fn try_unregister(&self) {
        match self.unregister().await {
            Ok(_) => {
                tracing::info!(
                    worker_id = %self.worker_id(),
                    "successfully unregistered from meta service",
                )
            }
            Err(e) => {
                tracing::warn!(
                    error = %e.as_report(),
                    worker_id = %self.worker_id(),
                    "failed to unregister from meta service",
                );
            }
        }
    }

    pub async fn update_schedulability(
        &self,
        worker_ids: &[WorkerId],
        schedulability: Schedulability,
    ) -> Result<UpdateWorkerNodeSchedulabilityResponse> {
        let request = UpdateWorkerNodeSchedulabilityRequest {
            worker_ids: worker_ids.to_vec(),
            schedulability: schedulability.into(),
        };
        let resp = self
            .inner
            .update_worker_node_schedulability(request)
            .await?;
        Ok(resp)
    }

    pub async fn list_worker_nodes(
        &self,
        worker_type: Option<WorkerType>,
    ) -> Result<Vec<WorkerNode>> {
        let request = ListAllNodesRequest {
            worker_type: worker_type.map(Into::into),
            include_starting_nodes: true,
        };
        let resp = self.inner.list_all_nodes(request).await?;
        Ok(resp.nodes)
    }

    /// Starts a heartbeat worker.
    pub fn start_heartbeat_loop(
        meta_client: MetaClient,
        min_interval: Duration,
    ) -> (JoinHandle<()>, Sender<()>) {
        let (shutdown_tx, mut shutdown_rx) = tokio::sync::oneshot::channel();
        let join_handle = tokio::spawn(async move {
            let mut min_interval_ticker = tokio::time::interval(min_interval);
            loop {
                tokio::select! {
                    biased;
                    // Shutdown
                    _ = &mut shutdown_rx => {
                        tracing::info!("Heartbeat loop is stopped");
                        return;
                    }
                    // Wait for interval
                    _ = min_interval_ticker.tick() => {},
                }
                tracing::debug!(target: "events::meta::client_heartbeat", "heartbeat");
                match tokio::time::timeout(
                    // TODO: decide better min_interval for timeout
                    min_interval * 3,
                    meta_client.send_heartbeat(meta_client.worker_id()),
                )
                .await
                {
                    Ok(Ok(_)) => {}
                    Ok(Err(err)) => {
                        tracing::warn!(error = %err.as_report(), "Failed to send_heartbeat");
                    }
                    Err(_) => {
                        tracing::warn!("Failed to send_heartbeat: timeout");
                    }
                }
            }
        });
        (join_handle, shutdown_tx)
    }

    pub async fn risectl_list_state_tables(&self) -> Result<Vec<PbTable>> {
        let request = RisectlListStateTablesRequest {};
        let resp = self.inner.risectl_list_state_tables(request).await?;
        Ok(resp.tables)
    }

    pub async fn flush(&self, database_id: DatabaseId) -> Result<HummockVersionId> {
        let request = FlushRequest { database_id };
        let resp = self.inner.flush(request).await?;
        Ok(HummockVersionId::new(resp.hummock_version_id))
    }

    pub async fn wait(&self) -> Result<()> {
        let request = WaitRequest {};
        self.inner.wait(request).await?;
        Ok(())
    }

    pub async fn recover(&self) -> Result<()> {
        let request = RecoverRequest {};
        self.inner.recover(request).await?;
        Ok(())
    }

    pub async fn cancel_creating_jobs(&self, jobs: PbJobs) -> Result<Vec<u32>> {
        let request = CancelCreatingJobsRequest { jobs: Some(jobs) };
        let resp = self.inner.cancel_creating_jobs(request).await?;
        Ok(resp.canceled_jobs)
    }

    pub async fn list_table_fragments(
        &self,
        job_ids: &[JobId],
    ) -> Result<HashMap<JobId, TableFragmentInfo>> {
        let request = ListTableFragmentsRequest {
            table_ids: job_ids.to_vec(),
        };
        let resp = self.inner.list_table_fragments(request).await?;
        Ok(resp.table_fragments)
    }

    pub async fn list_streaming_job_states(&self) -> Result<Vec<StreamingJobState>> {
        let resp = self
            .inner
            .list_streaming_job_states(ListStreamingJobStatesRequest {})
            .await?;
        Ok(resp.states)
    }

    pub async fn list_fragment_distributions(&self) -> Result<Vec<FragmentDistribution>> {
        let resp = self
            .inner
            .list_fragment_distribution(ListFragmentDistributionRequest {})
            .await?;
        Ok(resp.distributions)
    }

    pub async fn list_creating_fragment_distribution(&self) -> Result<Vec<FragmentDistribution>> {
        let resp = self
            .inner
            .list_creating_fragment_distribution(ListCreatingFragmentDistributionRequest {})
            .await?;
        Ok(resp.distributions)
    }

    pub async fn get_fragment_by_id(
        &self,
        fragment_id: FragmentId,
    ) -> Result<Option<FragmentDistribution>> {
        let resp = self
            .inner
            .get_fragment_by_id(GetFragmentByIdRequest { fragment_id })
            .await?;
        Ok(resp.distribution)
    }

    pub async fn get_fragment_vnodes(
        &self,
        fragment_id: FragmentId,
    ) -> Result<Vec<(ActorId, Vec<u32>)>> {
        let resp = self
            .inner
            .get_fragment_vnodes(GetFragmentVnodesRequest { fragment_id })
            .await?;
        Ok(resp
            .actor_vnodes
            .into_iter()
            .map(|actor| (actor.actor_id, actor.vnode_indices))
            .collect())
    }

    pub async fn get_actor_vnodes(&self, actor_id: ActorId) -> Result<Vec<u32>> {
        let resp = self
            .inner
            .get_actor_vnodes(GetActorVnodesRequest { actor_id })
            .await?;
        Ok(resp.vnode_indices)
    }

    pub async fn list_actor_states(&self) -> Result<Vec<ActorState>> {
        let resp = self
            .inner
            .list_actor_states(ListActorStatesRequest {})
            .await?;
        Ok(resp.states)
    }

    pub async fn list_actor_splits(&self) -> Result<Vec<ActorSplit>> {
        let resp = self
            .inner
            .list_actor_splits(ListActorSplitsRequest {})
            .await?;

        Ok(resp.actor_splits)
    }

    pub async fn list_object_dependencies(&self) -> Result<Vec<PbObjectDependencies>> {
        let resp = self
            .inner
            .list_object_dependencies(ListObjectDependenciesRequest {})
            .await?;
        Ok(resp.dependencies)
    }

    pub async fn pause(&self) -> Result<PauseResponse> {
        let request = PauseRequest {};
        let resp = self.inner.pause(request).await?;
        Ok(resp)
    }

    pub async fn resume(&self) -> Result<ResumeResponse> {
        let request = ResumeRequest {};
        let resp = self.inner.resume(request).await?;
        Ok(resp)
    }

    pub async fn apply_throttle(
        &self,
        kind: PbThrottleTarget,
        id: u32,
        rate: Option<u32>,
    ) -> Result<ApplyThrottleResponse> {
        let request = ApplyThrottleRequest {
            kind: kind as i32,
            id,
            rate,
        };
        let resp = self.inner.apply_throttle(request).await?;
        Ok(resp)
    }

    pub async fn get_cluster_recovery_status(&self) -> Result<RecoveryStatus> {
        let resp = self
            .inner
            .get_cluster_recovery_status(GetClusterRecoveryStatusRequest {})
            .await?;
        Ok(resp.get_status().unwrap())
    }

    pub async fn get_cluster_info(&self) -> Result<GetClusterInfoResponse> {
        let request = GetClusterInfoRequest {};
        let resp = self.inner.get_cluster_info(request).await?;
        Ok(resp)
    }

    pub async fn reschedule(
        &self,
        worker_reschedules: HashMap<u32, PbWorkerReschedule>,
        revision: u64,
        resolve_no_shuffle_upstream: bool,
    ) -> Result<(bool, u64)> {
        let request = RescheduleRequest {
            revision,
            resolve_no_shuffle_upstream,
            worker_reschedules,
        };
        let resp = self.inner.reschedule(request).await?;
        Ok((resp.success, resp.revision))
    }

    pub async fn risectl_get_pinned_versions_summary(
        &self,
    ) -> Result<RiseCtlGetPinnedVersionsSummaryResponse> {
        let request = RiseCtlGetPinnedVersionsSummaryRequest {};
        self.inner
            .rise_ctl_get_pinned_versions_summary(request)
            .await
    }

    pub async fn risectl_get_checkpoint_hummock_version(
        &self,
    ) -> Result<RiseCtlGetCheckpointVersionResponse> {
        let request = RiseCtlGetCheckpointVersionRequest {};
        self.inner.rise_ctl_get_checkpoint_version(request).await
    }

    pub async fn risectl_pause_hummock_version_checkpoint(
        &self,
    ) -> Result<RiseCtlPauseVersionCheckpointResponse> {
        let request = RiseCtlPauseVersionCheckpointRequest {};
        self.inner.rise_ctl_pause_version_checkpoint(request).await
    }

    pub async fn risectl_resume_hummock_version_checkpoint(
        &self,
    ) -> Result<RiseCtlResumeVersionCheckpointResponse> {
        let request = RiseCtlResumeVersionCheckpointRequest {};
        self.inner.rise_ctl_resume_version_checkpoint(request).await
    }

    pub async fn init_metadata_for_replay(
        &self,
        tables: Vec<PbTable>,
        compaction_groups: Vec<CompactionGroupInfo>,
    ) -> Result<()> {
        let req = InitMetadataForReplayRequest {
            tables,
            compaction_groups,
        };
        let _resp = self.inner.init_metadata_for_replay(req).await?;
        Ok(())
    }

    pub async fn replay_version_delta(
        &self,
        version_delta: HummockVersionDelta,
    ) -> Result<(HummockVersion, Vec<CompactionGroupId>)> {
        let req = ReplayVersionDeltaRequest {
            version_delta: Some(version_delta.into()),
        };
        let resp = self.inner.replay_version_delta(req).await?;
        Ok((
            HummockVersion::from_rpc_protobuf(&resp.version.unwrap()),
            resp.modified_compaction_groups,
        ))
    }

    pub async fn list_version_deltas(
        &self,
        start_id: HummockVersionId,
        num_limit: u32,
        committed_epoch_limit: HummockEpoch,
    ) -> Result<Vec<HummockVersionDelta>> {
        let req = ListVersionDeltasRequest {
            start_id: start_id.to_u64(),
            num_limit,
            committed_epoch_limit,
        };
        Ok(self
            .inner
            .list_version_deltas(req)
            .await?
            .version_deltas
            .unwrap()
            .version_deltas
            .iter()
            .map(HummockVersionDelta::from_rpc_protobuf)
            .collect())
    }

    pub async fn trigger_compaction_deterministic(
        &self,
        version_id: HummockVersionId,
        compaction_groups: Vec<CompactionGroupId>,
    ) -> Result<()> {
        let req = TriggerCompactionDeterministicRequest {
            version_id: version_id.to_u64(),
            compaction_groups,
        };
        self.inner.trigger_compaction_deterministic(req).await?;
        Ok(())
    }

    pub async fn disable_commit_epoch(&self) -> Result<HummockVersion> {
        let req = DisableCommitEpochRequest {};
        Ok(HummockVersion::from_rpc_protobuf(
            &self
                .inner
                .disable_commit_epoch(req)
                .await?
                .current_version
                .unwrap(),
        ))
    }

    pub async fn get_assigned_compact_task_num(&self) -> Result<usize> {
        let req = GetAssignedCompactTaskNumRequest {};
        let resp = self.inner.get_assigned_compact_task_num(req).await?;
        Ok(resp.num_tasks as usize)
    }

    pub async fn risectl_list_compaction_group(&self) -> Result<Vec<CompactionGroupInfo>> {
        let req = RiseCtlListCompactionGroupRequest {};
        let resp = self.inner.rise_ctl_list_compaction_group(req).await?;
        Ok(resp.compaction_groups)
    }

    pub async fn risectl_update_compaction_config(
        &self,
        compaction_groups: &[CompactionGroupId],
        configs: &[MutableConfig],
    ) -> Result<()> {
        let req = RiseCtlUpdateCompactionConfigRequest {
            compaction_group_ids: compaction_groups.to_vec(),
            configs: configs
                .iter()
                .map(
                    |c| rise_ctl_update_compaction_config_request::MutableConfig {
                        mutable_config: Some(c.clone()),
                    },
                )
                .collect(),
        };
        let _resp = self.inner.rise_ctl_update_compaction_config(req).await?;
        Ok(())
    }

    pub async fn backup_meta(&self, remarks: Option<String>) -> Result<u64> {
        let req = BackupMetaRequest { remarks };
        let resp = self.inner.backup_meta(req).await?;
        Ok(resp.job_id)
    }

    pub async fn get_backup_job_status(&self, job_id: u64) -> Result<(BackupJobStatus, String)> {
        let req = GetBackupJobStatusRequest { job_id };
        let resp = self.inner.get_backup_job_status(req).await?;
        Ok((resp.job_status(), resp.message))
    }

    pub async fn delete_meta_snapshot(&self, snapshot_ids: &[u64]) -> Result<()> {
        let req = DeleteMetaSnapshotRequest {
            snapshot_ids: snapshot_ids.to_vec(),
        };
        let _resp = self.inner.delete_meta_snapshot(req).await?;
        Ok(())
    }

    pub async fn get_meta_snapshot_manifest(&self) -> Result<MetaSnapshotManifest> {
        let req = GetMetaSnapshotManifestRequest {};
        let resp = self.inner.get_meta_snapshot_manifest(req).await?;
        Ok(resp.manifest.expect("should exist"))
    }

    pub async fn get_telemetry_info(&self) -> Result<TelemetryInfoResponse> {
        let req = GetTelemetryInfoRequest {};
        let resp = self.inner.get_telemetry_info(req).await?;
        Ok(resp)
    }

    pub async fn get_meta_store_endpoint(&self) -> Result<String> {
        let req = GetMetaStoreInfoRequest {};
        let resp = self.inner.get_meta_store_info(req).await?;
        Ok(resp.meta_store_endpoint)
    }

    pub async fn alter_sink_props(
        &self,
        sink_id: SinkId,
        changed_props: BTreeMap<String, String>,
        changed_secret_refs: BTreeMap<String, PbSecretRef>,
        connector_conn_ref: Option<ConnectionId>,
    ) -> Result<()> {
        let req = AlterConnectorPropsRequest {
            object_id: sink_id.as_raw_id(),
            changed_props: changed_props.into_iter().collect(),
            changed_secret_refs: changed_secret_refs.into_iter().collect(),
            connector_conn_ref,
            object_type: AlterConnectorPropsObject::Sink as i32,
            extra_options: None,
        };
        let _resp = self.inner.alter_connector_props(req).await?;
        Ok(())
    }

    pub async fn alter_iceberg_table_props(
        &self,
        table_id: TableId,
        sink_id: SinkId,
        source_id: SourceId,
        changed_props: BTreeMap<String, String>,
        changed_secret_refs: BTreeMap<String, PbSecretRef>,
        connector_conn_ref: Option<ConnectionId>,
    ) -> Result<()> {
        let req = AlterConnectorPropsRequest {
            object_id: table_id.as_raw_id(),
            changed_props: changed_props.into_iter().collect(),
            changed_secret_refs: changed_secret_refs.into_iter().collect(),
            connector_conn_ref,
            object_type: AlterConnectorPropsObject::IcebergTable as i32,
            extra_options: Some(ExtraOptions::AlterIcebergTableIds(AlterIcebergTableIds {
                sink_id,
                source_id,
            })),
        };
        let _resp = self.inner.alter_connector_props(req).await?;
        Ok(())
    }

    pub async fn alter_source_connector_props(
        &self,
        source_id: SourceId,
        changed_props: BTreeMap<String, String>,
        changed_secret_refs: BTreeMap<String, PbSecretRef>,
        connector_conn_ref: Option<ConnectionId>,
    ) -> Result<()> {
        let req = AlterConnectorPropsRequest {
            object_id: source_id.as_raw_id(),
            changed_props: changed_props.into_iter().collect(),
            changed_secret_refs: changed_secret_refs.into_iter().collect(),
            connector_conn_ref,
            object_type: AlterConnectorPropsObject::Source as i32,
            extra_options: None,
        };
        let _resp = self.inner.alter_connector_props(req).await?;
        Ok(())
    }

    pub async fn set_system_param(
        &self,
        param: String,
        value: Option<String>,
    ) -> Result<Option<SystemParamsReader>> {
        let req = SetSystemParamRequest { param, value };
        let resp = self.inner.set_system_param(req).await?;
        Ok(resp.params.map(SystemParamsReader::from))
    }

    pub async fn get_session_params(&self) -> Result<String> {
        let req = GetSessionParamsRequest {};
        let resp = self.inner.get_session_params(req).await?;
        Ok(resp.params)
    }

    pub async fn set_session_param(&self, param: String, value: Option<String>) -> Result<String> {
        let req = SetSessionParamRequest { param, value };
        let resp = self.inner.set_session_param(req).await?;
        Ok(resp.param)
    }

    pub async fn get_ddl_progress(&self) -> Result<Vec<DdlProgress>> {
        let req = GetDdlProgressRequest {};
        let resp = self.inner.get_ddl_progress(req).await?;
        Ok(resp.ddl_progress)
    }

    pub async fn split_compaction_group(
        &self,
        group_id: CompactionGroupId,
        table_ids_to_new_group: &[TableId],
        partition_vnode_count: u32,
    ) -> Result<CompactionGroupId> {
        let req = SplitCompactionGroupRequest {
            group_id,
            table_ids: table_ids_to_new_group.to_vec(),
            partition_vnode_count,
        };
        let resp = self.inner.split_compaction_group(req).await?;
        Ok(resp.new_group_id)
    }

    pub async fn get_tables(
        &self,
        table_ids: Vec<TableId>,
        include_dropped_tables: bool,
    ) -> Result<HashMap<TableId, Table>> {
        let req = GetTablesRequest {
            table_ids,
            include_dropped_tables,
        };
        let resp = self.inner.get_tables(req).await?;
        Ok(resp.tables)
    }

    pub async fn list_serving_vnode_mappings(
        &self,
    ) -> Result<HashMap<FragmentId, (u32, WorkerSlotMapping)>> {
        let req = GetServingVnodeMappingsRequest {};
        let resp = self.inner.get_serving_vnode_mappings(req).await?;
        let mappings = resp
            .worker_slot_mappings
            .into_iter()
            .map(|p| {
                (
                    p.fragment_id,
                    (
                        resp.fragment_to_table
                            .get(&p.fragment_id)
                            .cloned()
                            .unwrap_or(0),
                        WorkerSlotMapping::from_protobuf(p.mapping.as_ref().unwrap()),
                    ),
                )
            })
            .collect();
        Ok(mappings)
    }

    pub async fn risectl_list_compaction_status(
        &self,
    ) -> Result<(
        Vec<CompactStatus>,
        Vec<CompactTaskAssignment>,
        Vec<CompactTaskProgress>,
    )> {
        let req = RiseCtlListCompactionStatusRequest {};
        let resp = self.inner.rise_ctl_list_compaction_status(req).await?;
        Ok((
            resp.compaction_statuses,
            resp.task_assignment,
            resp.task_progress,
        ))
    }

    pub async fn get_compaction_score(
        &self,
        compaction_group_id: CompactionGroupId,
    ) -> Result<Vec<PickerInfo>> {
        let req = GetCompactionScoreRequest {
            compaction_group_id,
        };
        let resp = self.inner.get_compaction_score(req).await?;
        Ok(resp.scores)
    }

    pub async fn risectl_rebuild_table_stats(&self) -> Result<()> {
        let req = RiseCtlRebuildTableStatsRequest {};
        let _resp = self.inner.rise_ctl_rebuild_table_stats(req).await?;
        Ok(())
    }

    pub async fn list_branched_object(&self) -> Result<Vec<BranchedObject>> {
        let req = ListBranchedObjectRequest {};
        let resp = self.inner.list_branched_object(req).await?;
        Ok(resp.branched_objects)
    }

    pub async fn list_active_write_limit(&self) -> Result<HashMap<u64, WriteLimit>> {
        let req = ListActiveWriteLimitRequest {};
        let resp = self.inner.list_active_write_limit(req).await?;
        Ok(resp.write_limits)
    }

    pub async fn list_hummock_meta_config(&self) -> Result<HashMap<String, String>> {
        let req = ListHummockMetaConfigRequest {};
        let resp = self.inner.list_hummock_meta_config(req).await?;
        Ok(resp.configs)
    }

    pub async fn delete_worker_node(&self, worker: HostAddress) -> Result<()> {
        let _resp = self
            .inner
            .delete_worker_node(DeleteWorkerNodeRequest { host: Some(worker) })
            .await?;

        Ok(())
    }

    pub async fn rw_cloud_validate_source(
        &self,
        source_type: SourceType,
        source_config: HashMap<String, String>,
    ) -> Result<RwCloudValidateSourceResponse> {
        let req = RwCloudValidateSourceRequest {
            source_type: source_type.into(),
            source_config,
        };
        let resp = self.inner.rw_cloud_validate_source(req).await?;
        Ok(resp)
    }

    pub async fn sink_coordinate_client(&self) -> SinkCoordinationRpcClient {
        self.inner.core.read().await.sink_coordinate_client.clone()
    }

    pub async fn list_compact_task_assignment(&self) -> Result<Vec<CompactTaskAssignment>> {
        let req = ListCompactTaskAssignmentRequest {};
        let resp = self.inner.list_compact_task_assignment(req).await?;
        Ok(resp.task_assignment)
    }

    pub async fn list_event_log(&self) -> Result<Vec<EventLog>> {
        let req = ListEventLogRequest::default();
        let resp = self.inner.list_event_log(req).await?;
        Ok(resp.event_logs)
    }

    pub async fn list_compact_task_progress(&self) -> Result<Vec<CompactTaskProgress>> {
        let req = ListCompactTaskProgressRequest {};
        let resp = self.inner.list_compact_task_progress(req).await?;
        Ok(resp.task_progress)
    }

    #[cfg(madsim)]
    pub fn try_add_panic_event_blocking(
        &self,
        panic_info: impl Display,
        timeout_millis: Option<u64>,
    ) {
    }

    /// If `timeout_millis` is None, default is used.
    #[cfg(not(madsim))]
    pub fn try_add_panic_event_blocking(
        &self,
        panic_info: impl Display,
        timeout_millis: Option<u64>,
    ) {
        let event = event_log::EventWorkerNodePanic {
            worker_id: self.worker_id,
            worker_type: self.worker_type.into(),
            host_addr: Some(self.host_addr.to_protobuf()),
            panic_info: format!("{panic_info}"),
        };
        let grpc_meta_client = self.inner.clone();
        let _ = thread::spawn(move || {
            let rt = tokio::runtime::Builder::new_current_thread()
                .enable_all()
                .build()
                .unwrap();
            let req = AddEventLogRequest {
                event: Some(add_event_log_request::Event::WorkerNodePanic(event)),
            };
            rt.block_on(async {
                let _ = tokio::time::timeout(
                    Duration::from_millis(timeout_millis.unwrap_or(1000)),
                    grpc_meta_client.add_event_log(req),
                )
                .await;
            });
        })
        .join();
    }

    pub async fn add_sink_fail_evet(
        &self,
        sink_id: SinkId,
        sink_name: String,
        connector: String,
        error: String,
    ) -> Result<()> {
        let event = event_log::EventSinkFail {
            sink_id,
            sink_name,
            connector,
            error,
        };
        let req = AddEventLogRequest {
            event: Some(add_event_log_request::Event::SinkFail(event)),
        };
        self.inner.add_event_log(req).await?;
        Ok(())
    }

    pub async fn add_cdc_auto_schema_change_fail_event(
        &self,
        source_id: SourceId,
        table_name: String,
        cdc_table_id: String,
        upstream_ddl: String,
        fail_info: String,
    ) -> Result<()> {
        let event = event_log::EventAutoSchemaChangeFail {
            table_id: source_id.as_cdc_table_id(),
            table_name,
            cdc_table_id,
            upstream_ddl,
            fail_info,
        };
        let req = AddEventLogRequest {
            event: Some(add_event_log_request::Event::AutoSchemaChangeFail(event)),
        };
        self.inner.add_event_log(req).await?;
        Ok(())
    }

    pub async fn cancel_compact_task(&self, task_id: u64, task_status: TaskStatus) -> Result<bool> {
        let req = CancelCompactTaskRequest {
            task_id,
            task_status: task_status as _,
        };
        let resp = self.inner.cancel_compact_task(req).await?;
        Ok(resp.ret)
    }

    pub async fn get_version_by_epoch(
        &self,
        epoch: HummockEpoch,
        table_id: TableId,
    ) -> Result<PbHummockVersion> {
        let req = GetVersionByEpochRequest { epoch, table_id };
        let resp = self.inner.get_version_by_epoch(req).await?;
        Ok(resp.version.unwrap())
    }

    pub async fn get_cluster_limits(
        &self,
    ) -> Result<Vec<risingwave_common::util::cluster_limit::ClusterLimit>> {
        let req = GetClusterLimitsRequest {};
        let resp = self.inner.get_cluster_limits(req).await?;
        Ok(resp.active_limits.into_iter().map(|l| l.into()).collect())
    }

    pub async fn merge_compaction_group(
        &self,
        left_group_id: CompactionGroupId,
        right_group_id: CompactionGroupId,
    ) -> Result<()> {
        let req = MergeCompactionGroupRequest {
            left_group_id,
            right_group_id,
        };
        self.inner.merge_compaction_group(req).await?;
        Ok(())
    }

    /// List all rate limits for sources and backfills
    pub async fn list_rate_limits(&self) -> Result<Vec<RateLimitInfo>> {
        let request = ListRateLimitsRequest {};
        let resp = self.inner.list_rate_limits(request).await?;
        Ok(resp.rate_limits)
    }

    pub async fn list_cdc_progress(&self) -> Result<HashMap<JobId, PbCdcProgress>> {
        let request = ListCdcProgressRequest {};
        let resp = self.inner.list_cdc_progress(request).await?;
        Ok(resp.cdc_progress)
    }

    pub async fn create_iceberg_table(
        &self,
        table_job_info: PbTableJobInfo,
        sink_job_info: PbSinkJobInfo,
        iceberg_source: PbSource,
        if_not_exists: bool,
    ) -> Result<WaitVersion> {
        let request = CreateIcebergTableRequest {
            table_info: Some(table_job_info),
            sink_info: Some(sink_job_info),
            iceberg_source: Some(iceberg_source),
            if_not_exists,
        };

        let resp = self.inner.create_iceberg_table(request).await?;
        Ok(resp
            .version
            .ok_or_else(|| anyhow!("wait version not set"))?)
    }

    pub async fn list_hosted_iceberg_tables(&self) -> Result<Vec<IcebergTable>> {
        let request = ListIcebergTablesRequest {};
        let resp = self.inner.list_iceberg_tables(request).await?;
        Ok(resp.iceberg_tables)
    }

    /// Get the await tree of all nodes in the cluster.
    pub async fn get_cluster_stack_trace(
        &self,
        actor_traces_format: ActorTracesFormat,
    ) -> Result<StackTraceResponse> {
        let request = StackTraceRequest {
            actor_traces_format: actor_traces_format as i32,
        };
        let resp = self.inner.stack_trace(request).await?;
        Ok(resp)
    }

    pub async fn set_sync_log_store_aligned(&self, job_id: JobId, aligned: bool) -> Result<()> {
        let request = SetSyncLogStoreAlignedRequest { job_id, aligned };
        self.inner.set_sync_log_store_aligned(request).await?;
        Ok(())
    }

    pub async fn refresh(&self, request: RefreshRequest) -> Result<RefreshResponse> {
        self.inner.refresh(request).await
    }

    pub async fn list_unmigrated_tables(&self) -> Result<HashMap<TableId, String>> {
        let request = ListUnmigratedTablesRequest {};
        let resp = self.inner.list_unmigrated_tables(request).await?;
        Ok(resp
            .tables
            .into_iter()
            .map(|table| (table.table_id, table.table_name))
            .collect())
    }
}

#[async_trait]
impl HummockMetaClient for MetaClient {
    async fn unpin_version_before(&self, unpin_version_before: HummockVersionId) -> Result<()> {
        let req = UnpinVersionBeforeRequest {
            context_id: self.worker_id(),
            unpin_version_before: unpin_version_before.to_u64(),
        };
        self.inner.unpin_version_before(req).await?;
        Ok(())
    }

    async fn get_current_version(&self) -> Result<HummockVersion> {
        let req = GetCurrentVersionRequest::default();
        Ok(HummockVersion::from_rpc_protobuf(
            &self
                .inner
                .get_current_version(req)
                .await?
                .current_version
                .unwrap(),
        ))
    }

    async fn get_new_object_ids(&self, number: u32) -> Result<ObjectIdRange> {
        let resp = self
            .inner
            .get_new_object_ids(GetNewObjectIdsRequest { number })
            .await?;
        Ok(ObjectIdRange::new(resp.start_id, resp.end_id))
    }

    async fn commit_epoch_with_change_log(
        &self,
        _epoch: HummockEpoch,
        _sync_result: SyncResult,
        _change_log_info: Option<HummockMetaClientChangeLogInfo>,
    ) -> Result<()> {
        panic!("Only meta service can commit_epoch in production.")
    }

    async fn trigger_manual_compaction(
        &self,
        compaction_group_id: u64,
        table_id: JobId,
        level: u32,
        sst_ids: Vec<u64>,
    ) -> Result<()> {
        // TODO: support key_range parameter
        let req = TriggerManualCompactionRequest {
            compaction_group_id,
            table_id,
            // if table_id not exist, manual_compaction will include all the sst
            // without check internal_table_id
            level,
            sst_ids,
            ..Default::default()
        };

        self.inner.trigger_manual_compaction(req).await?;
        Ok(())
    }

    async fn trigger_full_gc(
        &self,
        sst_retention_time_sec: u64,
        prefix: Option<String>,
    ) -> Result<()> {
        self.inner
            .trigger_full_gc(TriggerFullGcRequest {
                sst_retention_time_sec,
                prefix,
            })
            .await?;
        Ok(())
    }

    async fn subscribe_compaction_event(
        &self,
    ) -> Result<(
        UnboundedSender<SubscribeCompactionEventRequest>,
        BoxStream<'static, CompactionEventItem>,
    )> {
        let (request_sender, request_receiver) =
            unbounded_channel::<SubscribeCompactionEventRequest>();
        request_sender
            .send(SubscribeCompactionEventRequest {
                event: Some(subscribe_compaction_event_request::Event::Register(
                    Register {
                        context_id: self.worker_id(),
                    },
                )),
                create_at: SystemTime::now()
                    .duration_since(SystemTime::UNIX_EPOCH)
                    .expect("Clock may have gone backwards")
                    .as_millis() as u64,
            })
            .context("Failed to subscribe compaction event")?;

        let stream = self
            .inner
            .subscribe_compaction_event(Request::new(UnboundedReceiverStream::new(
                request_receiver,
            )))
            .await?;

        Ok((request_sender, Box::pin(stream)))
    }

    async fn get_version_by_epoch(
        &self,
        epoch: HummockEpoch,
        table_id: TableId,
    ) -> Result<PbHummockVersion> {
        self.get_version_by_epoch(epoch, table_id).await
    }

    async fn subscribe_iceberg_compaction_event(
        &self,
    ) -> Result<(
        UnboundedSender<SubscribeIcebergCompactionEventRequest>,
        BoxStream<'static, IcebergCompactionEventItem>,
    )> {
        let (request_sender, request_receiver) =
            unbounded_channel::<SubscribeIcebergCompactionEventRequest>();
        request_sender
            .send(SubscribeIcebergCompactionEventRequest {
                event: Some(subscribe_iceberg_compaction_event_request::Event::Register(
                    IcebergRegister {
                        context_id: self.worker_id(),
                    },
                )),
                create_at: SystemTime::now()
                    .duration_since(SystemTime::UNIX_EPOCH)
                    .expect("Clock may have gone backwards")
                    .as_millis() as u64,
            })
            .context("Failed to subscribe compaction event")?;

        let stream = self
            .inner
            .subscribe_iceberg_compaction_event(Request::new(UnboundedReceiverStream::new(
                request_receiver,
            )))
            .await?;

        Ok((request_sender, Box::pin(stream)))
    }
}

#[async_trait]
impl TelemetryInfoFetcher for MetaClient {
    async fn fetch_telemetry_info(&self) -> std::result::Result<Option<String>, String> {
        let resp = self
            .get_telemetry_info()
            .await
            .map_err(|e| e.to_report_string())?;
        let tracking_id = resp.get_tracking_id().ok();
        Ok(tracking_id.map(|id| id.to_owned()))
    }
}

pub type SinkCoordinationRpcClient = SinkCoordinationServiceClient<Channel>;

#[derive(Debug, Clone)]
struct GrpcMetaClientCore {
    cluster_client: ClusterServiceClient<Channel>,
    meta_member_client: MetaMemberServiceClient<Channel>,
    heartbeat_client: HeartbeatServiceClient<Channel>,
    ddl_client: DdlServiceClient<Channel>,
    hummock_client: HummockManagerServiceClient<Channel>,
    notification_client: NotificationServiceClient<Channel>,
    stream_client: StreamManagerServiceClient<Channel>,
    user_client: UserServiceClient<Channel>,
    scale_client: ScaleServiceClient<Channel>,
    backup_client: BackupServiceClient<Channel>,
    telemetry_client: TelemetryInfoServiceClient<Channel>,
    system_params_client: SystemParamsServiceClient<Channel>,
    session_params_client: SessionParamServiceClient<Channel>,
    serving_client: ServingServiceClient<Channel>,
    cloud_client: CloudServiceClient<Channel>,
    sink_coordinate_client: SinkCoordinationRpcClient,
    event_log_client: EventLogServiceClient<Channel>,
    cluster_limit_client: ClusterLimitServiceClient<Channel>,
    hosted_iceberg_catalog_service_client: HostedIcebergCatalogServiceClient<Channel>,
    monitor_client: MonitorServiceClient<Channel>,
}

impl GrpcMetaClientCore {
    pub(crate) fn new(channel: Channel) -> Self {
        let cluster_client = ClusterServiceClient::new(channel.clone());
        let meta_member_client = MetaMemberClient::new(channel.clone());
        let heartbeat_client = HeartbeatServiceClient::new(channel.clone());
        let ddl_client =
            DdlServiceClient::new(channel.clone()).max_decoding_message_size(usize::MAX);
        let hummock_client =
            HummockManagerServiceClient::new(channel.clone()).max_decoding_message_size(usize::MAX);
        let notification_client =
            NotificationServiceClient::new(channel.clone()).max_decoding_message_size(usize::MAX);
        let stream_client =
            StreamManagerServiceClient::new(channel.clone()).max_decoding_message_size(usize::MAX);
        let user_client = UserServiceClient::new(channel.clone());
        let scale_client =
            ScaleServiceClient::new(channel.clone()).max_decoding_message_size(usize::MAX);
        let backup_client = BackupServiceClient::new(channel.clone());
        let telemetry_client =
            TelemetryInfoServiceClient::new(channel.clone()).max_decoding_message_size(usize::MAX);
        let system_params_client = SystemParamsServiceClient::new(channel.clone());
        let session_params_client = SessionParamServiceClient::new(channel.clone());
        let serving_client = ServingServiceClient::new(channel.clone());
        let cloud_client = CloudServiceClient::new(channel.clone());
        let sink_coordinate_client = SinkCoordinationServiceClient::new(channel.clone())
            .max_decoding_message_size(usize::MAX);
        let event_log_client = EventLogServiceClient::new(channel.clone());
        let cluster_limit_client = ClusterLimitServiceClient::new(channel.clone());
        let hosted_iceberg_catalog_service_client =
            HostedIcebergCatalogServiceClient::new(channel.clone());
        let monitor_client = MonitorServiceClient::new(channel);

        GrpcMetaClientCore {
            cluster_client,
            meta_member_client,
            heartbeat_client,
            ddl_client,
            hummock_client,
            notification_client,
            stream_client,
            user_client,
            scale_client,
            backup_client,
            telemetry_client,
            system_params_client,
            session_params_client,
            serving_client,
            cloud_client,
            sink_coordinate_client,
            event_log_client,
            cluster_limit_client,
            hosted_iceberg_catalog_service_client,
            monitor_client,
        }
    }
}

/// Client to meta server. Cloning the instance is lightweight.
///
/// It is a wrapper of tonic client. See [`crate::meta_rpc_client_method_impl`].
#[derive(Debug, Clone)]
struct GrpcMetaClient {
    member_monitor_event_sender: mpsc::Sender<Sender<Result<()>>>,
    core: Arc<RwLock<GrpcMetaClientCore>>,
}

type MetaMemberClient = MetaMemberServiceClient<Channel>;

struct MetaMemberGroup {
    members: LruCache<http::Uri, Option<MetaMemberClient>>,
}

struct MetaMemberManagement {
    core_ref: Arc<RwLock<GrpcMetaClientCore>>,
    members: Either<MetaMemberClient, MetaMemberGroup>,
    current_leader: http::Uri,
    meta_config: Arc<MetaConfig>,
}

impl MetaMemberManagement {
    const META_MEMBER_REFRESH_PERIOD: Duration = Duration::from_secs(5);

    fn host_address_to_uri(addr: HostAddress) -> http::Uri {
        format!("http://{}:{}", addr.host, addr.port)
            .parse()
            .unwrap()
    }

    async fn recreate_core(&self, channel: Channel) {
        let mut core = self.core_ref.write().await;
        *core = GrpcMetaClientCore::new(channel);
    }

    async fn refresh_members(&mut self) -> Result<()> {
        let leader_addr = match self.members.as_mut() {
            Either::Left(client) => {
                let resp = client
                    .to_owned()
                    .members(MembersRequest {})
                    .await
                    .map_err(RpcError::from_meta_status)?;
                let resp = resp.into_inner();
                resp.members.into_iter().find(|member| member.is_leader)
            }
            Either::Right(member_group) => {
                let mut fetched_members = None;

                for (addr, client) in &mut member_group.members {
                    let members: Result<_> = try {
                        let mut client = match client {
                            Some(cached_client) => cached_client.to_owned(),
                            None => {
                                let endpoint = GrpcMetaClient::addr_to_endpoint(addr.clone());
                                let channel = GrpcMetaClient::connect_to_endpoint(endpoint)
                                    .await
                                    .context("failed to create client")?;
                                let new_client: MetaMemberClient =
                                    MetaMemberServiceClient::new(channel);
                                *client = Some(new_client.clone());

                                new_client
                            }
                        };

                        let resp = client
                            .members(MembersRequest {})
                            .await
                            .context("failed to fetch members")?;

                        resp.into_inner().members
                    };

                    let fetched = members.is_ok();
                    fetched_members = Some(members);
                    if fetched {
                        break;
                    }
                }

                let members = fetched_members
                    .context("no member available in the list")?
                    .context("could not refresh members")?;

                // find new leader
                let mut leader = None;
                for member in members {
                    if member.is_leader {
                        leader = Some(member.clone());
                    }

                    let addr = Self::host_address_to_uri(member.address.unwrap());
                    // We don't clean any expired addrs here to deal with some extreme situations.
                    if !member_group.members.contains(&addr) {
                        tracing::info!("new meta member joined: {}", addr);
                        member_group.members.put(addr, None);
                    }
                }

                leader
            }
        };

        if let Some(leader) = leader_addr {
            let discovered_leader = Self::host_address_to_uri(leader.address.unwrap());

            if discovered_leader != self.current_leader {
                tracing::info!("new meta leader {} discovered", discovered_leader);

                let retry_strategy = GrpcMetaClient::retry_strategy_to_bound(
                    Duration::from_secs(self.meta_config.meta_leader_lease_secs),
                    false,
                );

                let channel = tokio_retry::Retry::spawn(retry_strategy, || async {
                    let endpoint = GrpcMetaClient::addr_to_endpoint(discovered_leader.clone());
                    GrpcMetaClient::connect_to_endpoint(endpoint).await
                })
                .await?;

                self.recreate_core(channel).await;
                self.current_leader = discovered_leader;
            }
        }

        Ok(())
    }
}

impl GrpcMetaClient {
    // See `Endpoint::http2_keep_alive_interval`
    const ENDPOINT_KEEP_ALIVE_INTERVAL_SEC: u64 = 60;
    // See `Endpoint::keep_alive_timeout`
    const ENDPOINT_KEEP_ALIVE_TIMEOUT_SEC: u64 = 60;
    // Retry base interval in ms for connecting to meta server.
    const INIT_RETRY_BASE_INTERVAL_MS: u64 = 10;
    // Max retry times for connecting to meta server.
    const INIT_RETRY_MAX_INTERVAL_MS: u64 = 2000;

    fn start_meta_member_monitor(
        &self,
        init_leader_addr: http::Uri,
        members: Either<MetaMemberClient, MetaMemberGroup>,
        force_refresh_receiver: Receiver<Sender<Result<()>>>,
        meta_config: Arc<MetaConfig>,
    ) -> Result<()> {
        let core_ref: Arc<RwLock<GrpcMetaClientCore>> = self.core.clone();
        let current_leader = init_leader_addr;

        let enable_period_tick = matches!(members, Either::Right(_));

        let member_management = MetaMemberManagement {
            core_ref,
            members,
            current_leader,
            meta_config,
        };

        let mut force_refresh_receiver = force_refresh_receiver;

        tokio::spawn(async move {
            let mut member_management = member_management;
            let mut ticker = time::interval(MetaMemberManagement::META_MEMBER_REFRESH_PERIOD);

            loop {
                let event: Option<Sender<Result<()>>> = if enable_period_tick {
                    tokio::select! {
                        _ = ticker.tick() => None,
                        result_sender = force_refresh_receiver.recv() => {
                            if result_sender.is_none() {
                                break;
                            }

                            result_sender
                        },
                    }
                } else {
                    let result_sender = force_refresh_receiver.recv().await;

                    if result_sender.is_none() {
                        break;
                    }

                    result_sender
                };

                let tick_result = member_management.refresh_members().await;
                if let Err(e) = tick_result.as_ref() {
                    tracing::warn!(error = %e.as_report(),  "refresh meta member client failed");
                }

                if let Some(sender) = event {
                    // ignore resp
                    let _resp = sender.send(tick_result);
                }
            }
        });

        Ok(())
    }

    async fn force_refresh_leader(&self) -> Result<()> {
        let (sender, receiver) = oneshot::channel();

        self.member_monitor_event_sender
            .send(sender)
            .await
            .map_err(|e| anyhow!(e))?;

        receiver.await.map_err(|e| anyhow!(e))?
    }

    /// Connect to the meta server from `addrs`.
    pub async fn new(strategy: &MetaAddressStrategy, config: Arc<MetaConfig>) -> Result<Self> {
        let (channel, addr) = match strategy {
            MetaAddressStrategy::LoadBalance(addr) => {
                Self::try_build_rpc_channel(vec![addr.clone()]).await
            }
            MetaAddressStrategy::List(addrs) => Self::try_build_rpc_channel(addrs.clone()).await,
        }?;
        let (force_refresh_sender, force_refresh_receiver) = mpsc::channel(1);
        let client = GrpcMetaClient {
            member_monitor_event_sender: force_refresh_sender,
            core: Arc::new(RwLock::new(GrpcMetaClientCore::new(channel))),
        };

        let meta_member_client = client.core.read().await.meta_member_client.clone();
        let members = match strategy {
            MetaAddressStrategy::LoadBalance(_) => Either::Left(meta_member_client),
            MetaAddressStrategy::List(addrs) => {
                let mut members = LruCache::new(20.try_into().unwrap());
                for addr in addrs {
                    members.put(addr.clone(), None);
                }
                members.put(addr.clone(), Some(meta_member_client));

                Either::Right(MetaMemberGroup { members })
            }
        };

        client.start_meta_member_monitor(addr, members, force_refresh_receiver, config.clone())?;

        client.force_refresh_leader().await?;

        Ok(client)
    }

    fn addr_to_endpoint(addr: http::Uri) -> Endpoint {
        Endpoint::from(addr).initial_connection_window_size(MAX_CONNECTION_WINDOW_SIZE)
    }

    pub(crate) async fn try_build_rpc_channel(
        addrs: impl IntoIterator<Item = http::Uri>,
    ) -> Result<(Channel, http::Uri)> {
        let endpoints: Vec<_> = addrs
            .into_iter()
            .map(|addr| (Self::addr_to_endpoint(addr.clone()), addr))
            .collect();

        let mut last_error = None;

        for (endpoint, addr) in endpoints {
            match Self::connect_to_endpoint(endpoint).await {
                Ok(channel) => {
                    tracing::info!("Connect to meta server {} successfully", addr);
                    return Ok((channel, addr));
                }
                Err(e) => {
                    tracing::warn!(
                        error = %e.as_report(),
                        "Failed to connect to meta server {}, trying again",
                        addr,
                    );
                    last_error = Some(e);
                }
            }
        }

        if let Some(last_error) = last_error {
            Err(anyhow::anyhow!(last_error)
                .context("failed to connect to all meta servers")
                .into())
        } else {
            bail!("no meta server address provided")
        }
    }

    async fn connect_to_endpoint(endpoint: Endpoint) -> Result<Channel> {
        let channel = endpoint
            .http2_keep_alive_interval(Duration::from_secs(Self::ENDPOINT_KEEP_ALIVE_INTERVAL_SEC))
            .keep_alive_timeout(Duration::from_secs(Self::ENDPOINT_KEEP_ALIVE_TIMEOUT_SEC))
            .connect_timeout(Duration::from_secs(5))
            .monitored_connect("grpc-meta-client", Default::default())
            .await?
            .wrapped();

        Ok(channel)
    }

    pub(crate) fn retry_strategy_to_bound(
        high_bound: Duration,
        exceed: bool,
    ) -> impl Iterator<Item = Duration> {
        let iter = ExponentialBackoff::from_millis(Self::INIT_RETRY_BASE_INTERVAL_MS)
            .max_delay(Duration::from_millis(Self::INIT_RETRY_MAX_INTERVAL_MS))
            .map(jitter);

        let mut sum = Duration::default();

        iter.take_while(move |duration| {
            sum += *duration;

            if exceed {
                sum < high_bound + *duration
            } else {
                sum < high_bound
            }
        })
    }
}

macro_rules! for_all_meta_rpc {
    ($macro:ident) => {
        $macro! {
             { cluster_client, add_worker_node, AddWorkerNodeRequest, AddWorkerNodeResponse }
            ,{ cluster_client, activate_worker_node, ActivateWorkerNodeRequest, ActivateWorkerNodeResponse }
            ,{ cluster_client, delete_worker_node, DeleteWorkerNodeRequest, DeleteWorkerNodeResponse }
            ,{ cluster_client, update_worker_node_schedulability, UpdateWorkerNodeSchedulabilityRequest, UpdateWorkerNodeSchedulabilityResponse }
            ,{ cluster_client, list_all_nodes, ListAllNodesRequest, ListAllNodesResponse }
            ,{ cluster_client, get_cluster_recovery_status, GetClusterRecoveryStatusRequest, GetClusterRecoveryStatusResponse }
            ,{ cluster_client, get_meta_store_info, GetMetaStoreInfoRequest, GetMetaStoreInfoResponse }
            ,{ heartbeat_client, heartbeat, HeartbeatRequest, HeartbeatResponse }
            ,{ stream_client, flush, FlushRequest, FlushResponse }
            ,{ stream_client, pause, PauseRequest, PauseResponse }
            ,{ stream_client, resume, ResumeRequest, ResumeResponse }
            ,{ stream_client, apply_throttle, ApplyThrottleRequest, ApplyThrottleResponse }
            ,{ stream_client, cancel_creating_jobs, CancelCreatingJobsRequest, CancelCreatingJobsResponse }
            ,{ stream_client, list_table_fragments, ListTableFragmentsRequest, ListTableFragmentsResponse }
            ,{ stream_client, list_streaming_job_states, ListStreamingJobStatesRequest, ListStreamingJobStatesResponse }
            ,{ stream_client, list_fragment_distribution, ListFragmentDistributionRequest, ListFragmentDistributionResponse }
            ,{ stream_client, list_creating_fragment_distribution, ListCreatingFragmentDistributionRequest, ListCreatingFragmentDistributionResponse }
            ,{ stream_client, list_actor_states, ListActorStatesRequest, ListActorStatesResponse }
            ,{ stream_client, list_actor_splits, ListActorSplitsRequest, ListActorSplitsResponse }
            ,{ stream_client, list_object_dependencies, ListObjectDependenciesRequest, ListObjectDependenciesResponse }
            ,{ stream_client, recover, RecoverRequest, RecoverResponse }
            ,{ stream_client, list_rate_limits, ListRateLimitsRequest, ListRateLimitsResponse }
            ,{ stream_client, list_cdc_progress, ListCdcProgressRequest, ListCdcProgressResponse }
            ,{ stream_client, alter_connector_props, AlterConnectorPropsRequest, AlterConnectorPropsResponse }
            ,{ stream_client, get_fragment_by_id, GetFragmentByIdRequest, GetFragmentByIdResponse }
            ,{ stream_client, get_fragment_vnodes, GetFragmentVnodesRequest, GetFragmentVnodesResponse }
            ,{ stream_client, get_actor_vnodes, GetActorVnodesRequest, GetActorVnodesResponse }
            ,{ stream_client, set_sync_log_store_aligned, SetSyncLogStoreAlignedRequest, SetSyncLogStoreAlignedResponse }
            ,{ stream_client, refresh, RefreshRequest, RefreshResponse }
            ,{ stream_client, list_unmigrated_tables, ListUnmigratedTablesRequest, ListUnmigratedTablesResponse }
            ,{ ddl_client, create_table, CreateTableRequest, CreateTableResponse }
            ,{ ddl_client, alter_name, AlterNameRequest, AlterNameResponse }
            ,{ ddl_client, alter_owner, AlterOwnerRequest, AlterOwnerResponse }
            ,{ ddl_client, alter_set_schema, AlterSetSchemaRequest, AlterSetSchemaResponse }
            ,{ ddl_client, alter_parallelism, AlterParallelismRequest, AlterParallelismResponse }
            ,{ ddl_client, alter_fragment_parallelism, AlterFragmentParallelismRequest, AlterFragmentParallelismResponse }
            ,{ ddl_client, alter_cdc_table_backfill_parallelism, AlterCdcTableBackfillParallelismRequest, AlterCdcTableBackfillParallelismResponse }
            ,{ ddl_client, alter_resource_group, AlterResourceGroupRequest, AlterResourceGroupResponse }
            ,{ ddl_client, alter_database_param, AlterDatabaseParamRequest, AlterDatabaseParamResponse }
            ,{ ddl_client, create_materialized_view, CreateMaterializedViewRequest, CreateMaterializedViewResponse }
            ,{ ddl_client, create_view, CreateViewRequest, CreateViewResponse }
            ,{ ddl_client, create_source, CreateSourceRequest, CreateSourceResponse }
            ,{ ddl_client, create_sink, CreateSinkRequest, CreateSinkResponse }
            ,{ ddl_client, create_subscription, CreateSubscriptionRequest, CreateSubscriptionResponse }
            ,{ ddl_client, create_schema, CreateSchemaRequest, CreateSchemaResponse }
            ,{ ddl_client, create_database, CreateDatabaseRequest, CreateDatabaseResponse }
            ,{ ddl_client, create_secret, CreateSecretRequest, CreateSecretResponse }
            ,{ ddl_client, create_index, CreateIndexRequest, CreateIndexResponse }
            ,{ ddl_client, create_function, CreateFunctionRequest, CreateFunctionResponse }
            ,{ ddl_client, drop_table, DropTableRequest, DropTableResponse }
            ,{ ddl_client, drop_materialized_view, DropMaterializedViewRequest, DropMaterializedViewResponse }
            ,{ ddl_client, drop_view, DropViewRequest, DropViewResponse }
            ,{ ddl_client, drop_source, DropSourceRequest, DropSourceResponse }
            ,{ ddl_client, drop_secret, DropSecretRequest, DropSecretResponse}
            ,{ ddl_client, drop_sink, DropSinkRequest, DropSinkResponse }
            ,{ ddl_client, drop_subscription, DropSubscriptionRequest, DropSubscriptionResponse }
            ,{ ddl_client, drop_database, DropDatabaseRequest, DropDatabaseResponse }
            ,{ ddl_client, drop_schema, DropSchemaRequest, DropSchemaResponse }
            ,{ ddl_client, drop_index, DropIndexRequest, DropIndexResponse }
            ,{ ddl_client, drop_function, DropFunctionRequest, DropFunctionResponse }
            ,{ ddl_client, replace_job_plan, ReplaceJobPlanRequest, ReplaceJobPlanResponse }
            ,{ ddl_client, alter_source, AlterSourceRequest, AlterSourceResponse }
            ,{ ddl_client, risectl_list_state_tables, RisectlListStateTablesRequest, RisectlListStateTablesResponse }
            ,{ ddl_client, get_ddl_progress, GetDdlProgressRequest, GetDdlProgressResponse }
            ,{ ddl_client, create_connection, CreateConnectionRequest, CreateConnectionResponse }
            ,{ ddl_client, list_connections, ListConnectionsRequest, ListConnectionsResponse }
            ,{ ddl_client, drop_connection, DropConnectionRequest, DropConnectionResponse }
            ,{ ddl_client, comment_on, CommentOnRequest, CommentOnResponse }
            ,{ ddl_client, get_tables, GetTablesRequest, GetTablesResponse }
            ,{ ddl_client, wait, WaitRequest, WaitResponse }
            ,{ ddl_client, auto_schema_change, AutoSchemaChangeRequest, AutoSchemaChangeResponse }
            ,{ ddl_client, alter_swap_rename, AlterSwapRenameRequest, AlterSwapRenameResponse }
            ,{ ddl_client, alter_secret, AlterSecretRequest, AlterSecretResponse }
            ,{ ddl_client, compact_iceberg_table, CompactIcebergTableRequest, CompactIcebergTableResponse }
            ,{ ddl_client, expire_iceberg_table_snapshots, ExpireIcebergTableSnapshotsRequest, ExpireIcebergTableSnapshotsResponse }
            ,{ ddl_client, create_iceberg_table, CreateIcebergTableRequest, CreateIcebergTableResponse }
            ,{ hummock_client, unpin_version_before, UnpinVersionBeforeRequest, UnpinVersionBeforeResponse }
            ,{ hummock_client, get_current_version, GetCurrentVersionRequest, GetCurrentVersionResponse }
            ,{ hummock_client, replay_version_delta, ReplayVersionDeltaRequest, ReplayVersionDeltaResponse }
            ,{ hummock_client, list_version_deltas, ListVersionDeltasRequest, ListVersionDeltasResponse }
            ,{ hummock_client, get_assigned_compact_task_num, GetAssignedCompactTaskNumRequest, GetAssignedCompactTaskNumResponse }
            ,{ hummock_client, trigger_compaction_deterministic, TriggerCompactionDeterministicRequest, TriggerCompactionDeterministicResponse }
            ,{ hummock_client, disable_commit_epoch, DisableCommitEpochRequest, DisableCommitEpochResponse }
            ,{ hummock_client, get_new_object_ids, GetNewObjectIdsRequest, GetNewObjectIdsResponse }
            ,{ hummock_client, trigger_manual_compaction, TriggerManualCompactionRequest, TriggerManualCompactionResponse }
            ,{ hummock_client, trigger_full_gc, TriggerFullGcRequest, TriggerFullGcResponse }
            ,{ hummock_client, rise_ctl_get_pinned_versions_summary, RiseCtlGetPinnedVersionsSummaryRequest, RiseCtlGetPinnedVersionsSummaryResponse }
            ,{ hummock_client, rise_ctl_list_compaction_group, RiseCtlListCompactionGroupRequest, RiseCtlListCompactionGroupResponse }
            ,{ hummock_client, rise_ctl_update_compaction_config, RiseCtlUpdateCompactionConfigRequest, RiseCtlUpdateCompactionConfigResponse }
            ,{ hummock_client, rise_ctl_get_checkpoint_version, RiseCtlGetCheckpointVersionRequest, RiseCtlGetCheckpointVersionResponse }
            ,{ hummock_client, rise_ctl_pause_version_checkpoint, RiseCtlPauseVersionCheckpointRequest, RiseCtlPauseVersionCheckpointResponse }
            ,{ hummock_client, rise_ctl_resume_version_checkpoint, RiseCtlResumeVersionCheckpointRequest, RiseCtlResumeVersionCheckpointResponse }
            ,{ hummock_client, init_metadata_for_replay, InitMetadataForReplayRequest, InitMetadataForReplayResponse }
            ,{ hummock_client, split_compaction_group, SplitCompactionGroupRequest, SplitCompactionGroupResponse }
            ,{ hummock_client, rise_ctl_list_compaction_status, RiseCtlListCompactionStatusRequest, RiseCtlListCompactionStatusResponse }
            ,{ hummock_client, get_compaction_score, GetCompactionScoreRequest, GetCompactionScoreResponse }
            ,{ hummock_client, rise_ctl_rebuild_table_stats, RiseCtlRebuildTableStatsRequest, RiseCtlRebuildTableStatsResponse }
            ,{ hummock_client, subscribe_compaction_event, impl tonic::IntoStreamingRequest<Message = SubscribeCompactionEventRequest>, Streaming<SubscribeCompactionEventResponse> }
            ,{ hummock_client, subscribe_iceberg_compaction_event, impl tonic::IntoStreamingRequest<Message = SubscribeIcebergCompactionEventRequest>, Streaming<SubscribeIcebergCompactionEventResponse> }
            ,{ hummock_client, list_branched_object, ListBranchedObjectRequest, ListBranchedObjectResponse }
            ,{ hummock_client, list_active_write_limit, ListActiveWriteLimitRequest, ListActiveWriteLimitResponse }
            ,{ hummock_client, list_hummock_meta_config, ListHummockMetaConfigRequest, ListHummockMetaConfigResponse }
            ,{ hummock_client, list_compact_task_assignment, ListCompactTaskAssignmentRequest, ListCompactTaskAssignmentResponse }
            ,{ hummock_client, list_compact_task_progress, ListCompactTaskProgressRequest, ListCompactTaskProgressResponse }
            ,{ hummock_client, cancel_compact_task, CancelCompactTaskRequest, CancelCompactTaskResponse}
            ,{ hummock_client, get_version_by_epoch, GetVersionByEpochRequest, GetVersionByEpochResponse }
            ,{ hummock_client, merge_compaction_group, MergeCompactionGroupRequest, MergeCompactionGroupResponse }
            ,{ user_client, create_user, CreateUserRequest, CreateUserResponse }
            ,{ user_client, update_user, UpdateUserRequest, UpdateUserResponse }
            ,{ user_client, drop_user, DropUserRequest, DropUserResponse }
            ,{ user_client, grant_privilege, GrantPrivilegeRequest, GrantPrivilegeResponse }
            ,{ user_client, revoke_privilege, RevokePrivilegeRequest, RevokePrivilegeResponse }
            ,{ user_client, alter_default_privilege, AlterDefaultPrivilegeRequest, AlterDefaultPrivilegeResponse }
            ,{ scale_client, get_cluster_info, GetClusterInfoRequest, GetClusterInfoResponse }
            ,{ scale_client, reschedule, RescheduleRequest, RescheduleResponse }
            ,{ notification_client, subscribe, SubscribeRequest, Streaming<SubscribeResponse> }
            ,{ backup_client, backup_meta, BackupMetaRequest, BackupMetaResponse }
            ,{ backup_client, get_backup_job_status, GetBackupJobStatusRequest, GetBackupJobStatusResponse }
            ,{ backup_client, delete_meta_snapshot, DeleteMetaSnapshotRequest, DeleteMetaSnapshotResponse}
            ,{ backup_client, get_meta_snapshot_manifest, GetMetaSnapshotManifestRequest, GetMetaSnapshotManifestResponse}
            ,{ telemetry_client, get_telemetry_info, GetTelemetryInfoRequest, TelemetryInfoResponse}
            ,{ system_params_client, get_system_params, GetSystemParamsRequest, GetSystemParamsResponse }
            ,{ system_params_client, set_system_param, SetSystemParamRequest, SetSystemParamResponse }
            ,{ session_params_client, get_session_params, GetSessionParamsRequest, GetSessionParamsResponse }
            ,{ session_params_client, set_session_param, SetSessionParamRequest, SetSessionParamResponse }
            ,{ serving_client, get_serving_vnode_mappings, GetServingVnodeMappingsRequest, GetServingVnodeMappingsResponse }
            ,{ cloud_client, rw_cloud_validate_source, RwCloudValidateSourceRequest, RwCloudValidateSourceResponse }
            ,{ event_log_client, list_event_log, ListEventLogRequest, ListEventLogResponse }
            ,{ event_log_client, add_event_log, AddEventLogRequest, AddEventLogResponse }
            ,{ cluster_limit_client, get_cluster_limits, GetClusterLimitsRequest, GetClusterLimitsResponse }
            ,{ hosted_iceberg_catalog_service_client, list_iceberg_tables, ListIcebergTablesRequest, ListIcebergTablesResponse }
            ,{ monitor_client, stack_trace, StackTraceRequest, StackTraceResponse }
        }
    };
}

impl GrpcMetaClient {
    async fn refresh_client_if_needed(&self, code: Code) {
        if matches!(
            code,
            Code::Unknown | Code::Unimplemented | Code::Unavailable
        ) {
            tracing::debug!("matching tonic code {}", code);
            let (result_sender, result_receiver) = oneshot::channel();
            if self
                .member_monitor_event_sender
                .try_send(result_sender)
                .is_ok()
            {
                if let Ok(Err(e)) = result_receiver.await {
                    tracing::warn!(error = %e.as_report(), "force refresh meta client failed");
                }
            } else {
                tracing::debug!("skipping the current refresh, somewhere else is already doing it")
            }
        }
    }
}

impl GrpcMetaClient {
    for_all_meta_rpc! { meta_rpc_client_method_impl }
}