Merge pull request #6071 from lichuang/metasrv_metrics

Feature: add meta service network metrics and http health checker

docs/doc/50-manage/00-metasrv/50-metasrv-metrics.md

@@ -21,6 +21,7 @@ These metrics describe the status of the `metasrv`. All these metrics are prefix
 | ----------------- | ------------------------------------------------- | ------- |
 | current_leader_id | Current leader id of cluster, 0 means no leader.  | IntGauge   |
 | is_leader         | Whether or not this node is current leader.       | Gauge   |
+| node_is_health    | Whether or not this node is health.               | IntGauge |
 | leader_changes    | Number of leader changes seen.                    | Counter |
 | applying_snapshot | Whether or not statemachine is applying snapshot. | Gauge   |
 | proposals_applied | Total number of consensus proposals applied.      | Gauge   |
@@ -32,6 +33,8 @@ These metrics describe the status of the `metasrv`. All these metrics are prefix
 
 `is_leader` indicate if this `metasrv` currently is the leader of cluster, and `leader_changes` show the total number of leader changes since start.If change leader too frequently, it will impact the performance of `metasrv`, also it signal that the cluster is unstable.
 
+If and only if the node state is `Follower` or `Leader` , `node_is_health` is 1, otherwise is 0.
+
 `proposals_applied` records the total number of applied write requests.
 
 `proposals_pending` indicates how many proposals are queued to commit currently.Rising pending proposals suggests there is a high client load or the member cannot commit proposals.
@@ -40,9 +43,9 @@ These metrics describe the status of the `metasrv`. All these metrics are prefix
 
 `watchers` show the total number of active watchers currently.
 
-### Network
+### Raft Network
 
-These metrics describe the network status of the `metasrv`. All these metrics are prefixed with `metasrv_network_`.
+These metrics describe the network status of raft nodes in the `metasrv`. All these metrics are prefixed with `metasrv_raft_network_`.
 
 | Name                    | Description                                       | Labels                            | Type          |
 | ----------------------- | ------------------------------------------------- | --------------------------------- | ------------- |
@@ -71,3 +74,15 @@ These metrics describe the network status of the `metasrv`. All these metrics ar
 `snapshot_recv_success` and `snapshot_recv_failures` indicates the success and fail number of receive snapshot.`snapshot_recv_inflights` indicate the inflight receiving snapshot, each time receive a snapshot, this field will increment by one, after receiving snapshot is done, this field will decrement by one.
 
 `snapshot_recv_seconds` indicate the total latency distributions of snapshot receives.
+
+### Meta Network
+
+These metrics describe the network status of meta service in the `metasrv`. All these metrics are prefixed with `metasrv_meta_network_`.
+
+| Name             | Description                                            | Type       |
+| ---------------- | ------------------------------------------------------ | ---------- |
+| meta_sent_bytes  | Total number of sent bytes to meta grpc client.        | IntCounter |
+| meta_recv_bytes  | Total number of recv bytes from meta grpc client.      | IntCounter |
+| meta_inflights   | Total number of inflight meta grpc requests.           | IntGauge   |
+| meta_req_success | Total number of success request from meta grpc client. | IntCounter |
+| meta_req_failed  | Total number of fail request from meta grpc client.    | IntCounter |

metasrv/src/api/grpc/grpc_service.rs

@@ -49,6 +49,9 @@ use tonic::Streaming;
 use crate::executor::ActionHandler;
 use crate::meta_service::meta_service_impl::GrpcStream;
 use crate::meta_service::MetaNode;
+use crate::metrics::add_meta_metrics_meta_request_inflights;
+use crate::metrics::incr_meta_metrics_meta_recv_bytes;
+use crate::metrics::incr_meta_metrics_meta_sent_bytes;
 use crate::version::from_digit_ver;
 use crate::version::to_digit_ver;
 use crate::version::METASRV_SEMVER;
@@ -147,22 +150,41 @@ impl MetaService for MetaServiceImpl {
         self.check_token(request.metadata())?;
         common_tracing::extract_remote_span_as_parent(&request);
 
+        incr_meta_metrics_meta_recv_bytes(request.get_ref().encoded_len() as u64);
+
         let action: MetaGrpcWriteReq = request.try_into()?;
+
+        add_meta_metrics_meta_request_inflights(1);
+
         tracing::info!("Receive write_action: {:?}", action);
 
         let body = self.action_handler.execute_write(action).await;
+
+        add_meta_metrics_meta_request_inflights(-1);
+
+        incr_meta_metrics_meta_sent_bytes(body.encoded_len() as u64);
+
         Ok(Response::new(body))
     }
 
     async fn read_msg(&self, request: Request<RaftRequest>) -> Result<Response<RaftReply>, Status> {
         self.check_token(request.metadata())?;
         common_tracing::extract_remote_span_as_parent(&request);
 
+        incr_meta_metrics_meta_recv_bytes(request.get_ref().encoded_len() as u64);
+
         let action: MetaGrpcReadReq = request.try_into()?;
+
+        add_meta_metrics_meta_request_inflights(1);
+
         tracing::info!("Receive read_action: {:?}", action);
 
         let res = self.action_handler.execute_read(action).await;
 
+        add_meta_metrics_meta_request_inflights(-1);
+
+        incr_meta_metrics_meta_sent_bytes(res.encoded_len() as u64);
+
         Ok(Response::new(res))
     }
 
@@ -210,13 +232,20 @@ impl MetaService for MetaServiceImpl {
         request: Request<TxnRequest>,
     ) -> Result<Response<TxnReply>, Status> {
         self.check_token(request.metadata())?;
+        incr_meta_metrics_meta_recv_bytes(request.get_ref().encoded_len() as u64);
+        add_meta_metrics_meta_request_inflights(1);
+
         common_tracing::extract_remote_span_as_parent(&request);
 
         let request = request.into_inner();
 
         tracing::info!("Receive txn_request: {:?}", request);
 
         let body = self.action_handler.execute_txn(request).await;
+        add_meta_metrics_meta_request_inflights(-1);
+
+        incr_meta_metrics_meta_sent_bytes(body.encoded_len() as u64);
+
         Ok(Response::new(body))
     }
 
@@ -228,7 +257,11 @@ impl MetaService for MetaServiceImpl {
         let members = meta_node.get_meta_addrs().await.map_err(|e| {
             Status::internal(format!("Cannot get metasrv member list, error: {:?}", e))
         })?;
-        Ok(Response::new(MemberListReply { data: members }))
+
+        let resp = MemberListReply { data: members };
+        incr_meta_metrics_meta_sent_bytes(resp.encoded_len() as u64);
+
+        Ok(Response::new(resp))
     }
 }
 

metasrv/src/api/http_service.rs

@@ -20,13 +20,18 @@ use common_base::base::Stoppable;
 use common_exception::Result;
 use common_tracing::tracing;
 use poem::get;
+use poem::http::StatusCode;
 use poem::listener::RustlsConfig;
+use poem::web::Json;
 use poem::Endpoint;
 use poem::EndpointExt;
+use poem::IntoResponse;
+use poem::Response;
 use poem::Route;
 
 use crate::configs::Config;
 use crate::meta_service::MetaNode;
+use crate::metrics::get_meta_metrics_node_is_health;
 
 pub struct HttpService {
     cfg: Config,
@@ -46,7 +51,7 @@ impl HttpService {
     fn build_router(&self) -> impl Endpoint {
         #[cfg_attr(not(feature = "memory-profiling"), allow(unused_mut))]
         let mut route = Route::new()
-            .at("/v1/health", get(super::http::v1::health::health_handler))
+            .at("/v1/health", get(health_handler))
             .at("/v1/config", get(super::http::v1::config::config_handler))
             .at(
                 "/v1/cluster/nodes",
@@ -129,3 +134,14 @@ impl Stoppable for HttpService {
         self.shutdown_handler.stop(force).await
     }
 }
+
+#[poem::handler]
+pub async fn health_handler() -> Response {
+    if !get_meta_metrics_node_is_health() {
+        return StatusCode::SERVICE_UNAVAILABLE.into_response();
+    }
+    Json(super::http::v1::health::HealthCheckResponse {
+        status: super::http::v1::health::HealthCheckStatus::Pass,
+    })
+    .into_response()
+}

metasrv/src/executor/action_handler.rs

@@ -24,6 +24,7 @@ use common_meta_types::TxnReply;
 use common_meta_types::TxnRequest;
 
 use crate::meta_service::MetaNode;
+use crate::metrics::incr_meta_metrics_meta_request_result;
 
 pub struct ActionHandler {
     /// The raft-based meta data entry.
@@ -48,6 +49,7 @@ impl ActionHandler {
         match action {
             MetaGrpcWriteReq::UpsertKV(a) => {
                 let r = self.meta_node.upsert_kv(a).await;
+                incr_meta_metrics_meta_request_result(r.is_ok());
                 RaftReply::from(r)
             }
         }
@@ -59,25 +61,32 @@ impl ActionHandler {
         match action {
             MetaGrpcReadReq::GetKV(a) => {
                 let r = self.meta_node.get_kv(&a.key).await;
+                incr_meta_metrics_meta_request_result(r.is_ok());
                 RaftReply::from(r)
             }
             MetaGrpcReadReq::MGetKV(a) => {
                 let r = self.meta_node.mget_kv(&a.keys).await;
+                incr_meta_metrics_meta_request_result(r.is_ok());
                 RaftReply::from(r)
             }
             MetaGrpcReadReq::ListKV(a) => {
                 let r = self.meta_node.prefix_list_kv(&a.prefix).await;
+                incr_meta_metrics_meta_request_result(r.is_ok());
                 RaftReply::from(r)
             }
             MetaGrpcReadReq::PrefixListKV(a) => {
                 let r = self.meta_node.prefix_list_kv(&a.0).await;
+                incr_meta_metrics_meta_request_result(r.is_ok());
                 RaftReply::from(r)
             }
         }
     }
 
     pub async fn execute_txn(&self, req: TxnRequest) -> TxnReply {
-        match self.meta_node.transaction(req).await {
+        let ret = self.meta_node.transaction(req).await;
+        incr_meta_metrics_meta_request_result(ret.is_ok());
+
+        match ret {
             Ok(resp) => resp,
             Err(err) => TxnReply {
                 success: false,

metasrv/src/meta_service/raftmeta.rs

@@ -53,6 +53,7 @@ use openraft::Config;
 use openraft::Raft;
 use openraft::RaftMetrics;
 use openraft::SnapshotPolicy;
+use openraft::State;
 use tonic::Status;
 
 use crate::meta_service::meta_leader::MetaLeader;
@@ -63,6 +64,7 @@ use crate::metrics::incr_meta_metrics_leader_change;
 use crate::metrics::incr_meta_metrics_read_failed;
 use crate::metrics::set_meta_metrics_current_leader;
 use crate::metrics::set_meta_metrics_is_leader;
+use crate::metrics::set_meta_metrics_node_is_health;
 use crate::metrics::set_meta_metrics_proposals_applied;
 use crate::network::Network;
 use crate::store::MetaRaftStore;
@@ -379,6 +381,11 @@ impl MetaNode {
                         };
                         if changed.is_ok() {
                             let mm = metrics_rx.borrow().clone();
+
+                            set_meta_metrics_node_is_health(
+                                mm.state == State::Follower || mm.state == State::Leader,
+                            );
+
                             if let Some(cur) = mm.current_leader {
                                 // if current leader has changed?
                                 if let Some(leader) = current_leader {