fix: ctx cancel should be the last step while stopping server (#31220)

issue: #31219

Signed-off-by: jaime <yun.zhang@zilliz.com>

internal/datacoord/server.go

@@ -1126,9 +1126,6 @@ func (s *Server) Stop() error {
 	s.cluster.Close()
 	logutil.Logger(s.ctx).Info("datacoord cluster stopped")
 
-	s.stopServerLoop()
-	logutil.Logger(s.ctx).Info("datacoord serverloop stopped")
-
 	if s.session != nil {
 		s.session.Stop()
 	}
@@ -1136,6 +1133,9 @@ func (s *Server) Stop() error {
 	if s.icSession != nil {
 		s.icSession.Stop()
 	}
+
+	s.stopServerLoop()
+	logutil.Logger(s.ctx).Info("datacoord serverloop stopped")
 	logutil.Logger(s.ctx).Warn("datacoord stop successful")
 
 	return nil

internal/datanode/data_node.go

@@ -426,8 +426,6 @@ func (node *DataNode) Stop() error {
 	node.stopOnce.Do(func() {
 		// https://github.com/milvus-io/milvus/issues/12282
 		node.UpdateStateCode(commonpb.StateCode_Abnormal)
-		// Delay the cancellation of ctx to ensure that the session is automatically recycled after closed the flow graph
-		node.cancel()
 
 		node.eventManager.CloseAll()
 
@@ -460,6 +458,7 @@ func (node *DataNode) Stop() error {
 			node.importManager.Close()
 		}
 
+		node.cancel()
 		node.stopWaiter.Wait()
 	})
 	return nil

internal/distributed/datacoord/service.go

@@ -61,7 +61,7 @@ type Server struct {
 
 	serverID atomic.Int64
 
-	wg        sync.WaitGroup
+	grpcWG    sync.WaitGroup
 	dataCoord types.DataCoordComponent
 
 	etcdCli *clientv3.Client
@@ -135,7 +135,7 @@ func (s *Server) init() error {
 
 func (s *Server) startGrpc() error {
 	Params := &paramtable.Get().DataCoordGrpcServerCfg
-	s.wg.Add(1)
+	s.grpcWG.Add(1)
 	go s.startGrpcLoop(Params.Port.GetAsInt())
 	// wait for grpc server loop start
 	err := <-s.grpcErrChan
@@ -144,7 +144,7 @@ func (s *Server) startGrpc() error {
 
 func (s *Server) startGrpcLoop(grpcPort int) {
 	defer logutil.LogPanic()
-	defer s.wg.Done()
+	defer s.grpcWG.Done()
 
 	Params := &paramtable.Get().DataCoordGrpcServerCfg
 	log.Debug("network port", zap.Int("port", grpcPort))
@@ -229,8 +229,6 @@ func (s *Server) Stop() (err error) {
 		logger.Info("Datacoord stopped", zap.Error(err))
 	}()
 
-	s.cancel()
-
 	if s.etcdCli != nil {
 		defer s.etcdCli.Close()
 	}
@@ -240,14 +238,16 @@ func (s *Server) Stop() (err error) {
 	if s.grpcServer != nil {
 		utils.GracefulStopGRPCServer(s.grpcServer)
 	}
+	s.grpcWG.Wait()
 
+	logger.Info("internal server[dataCoord] start to stop")
 	err = s.dataCoord.Stop()
 	if err != nil {
+		log.Error("failed to close dataCoord", zap.Error(err))
 		return err
 	}
 
-	s.wg.Wait()
-
+	s.cancel()
 	return nil
 }
 

internal/distributed/datanode/service.go

@@ -57,7 +57,7 @@ import (
 
 type Server struct {
 	datanode    types.DataNodeComponent
-	wg          sync.WaitGroup
+	grpcWG      sync.WaitGroup
 	grpcErrChan chan error
 	grpcServer  *grpc.Server
 	ctx         context.Context
@@ -97,7 +97,7 @@ func NewServer(ctx context.Context, factory dependency.Factory) (*Server, error)
 
 func (s *Server) startGrpc() error {
 	Params := &paramtable.Get().DataNodeGrpcServerCfg
-	s.wg.Add(1)
+	s.grpcWG.Add(1)
 	go s.startGrpcLoop(Params.Port.GetAsInt())
 	// wait for grpc server loop start
 	err := <-s.grpcErrChan
@@ -106,7 +106,7 @@ func (s *Server) startGrpc() error {
 
 // startGrpcLoop starts the grep loop of datanode component.
 func (s *Server) startGrpcLoop(grpcPort int) {
-	defer s.wg.Done()
+	defer s.grpcWG.Done()
 	Params := &paramtable.Get().DataNodeGrpcServerCfg
 	kaep := keepalive.EnforcementPolicy{
 		MinTime:             5 * time.Second, // If a client pings more than once every 5 seconds, terminate the connection
@@ -202,24 +202,26 @@ func (s *Server) Run() error {
 func (s *Server) Stop() (err error) {
 	Params := &paramtable.Get().DataNodeGrpcServerCfg
 	logger := log.With(zap.String("address", Params.GetAddress()))
-	logger.Info("Datanode stopping")
+	logger.Info("datanode stopping")
 	defer func() {
-		logger.Info("Datanode stopped", zap.Error(err))
+		logger.Info("datanode stopped", zap.Error(err))
 	}()
 
-	s.cancel()
 	if s.etcdCli != nil {
 		defer s.etcdCli.Close()
 	}
 	if s.grpcServer != nil {
 		utils.GracefulStopGRPCServer(s.grpcServer)
 	}
+	s.grpcWG.Wait()
 
+	logger.Info("internal server[datanode] start to stop")
 	err = s.datanode.Stop()
 	if err != nil {
+		log.Error("failed to close datanode", zap.Error(err))
 		return err
 	}
-	s.wg.Wait()
+	s.cancel()
 	return nil
 }
 

internal/distributed/datanode/service_test.go

@@ -405,11 +405,4 @@ func Test_Run(t *testing.T) {
 
 	err = server.Run()
 	assert.Error(t, err)
-
-	server.datanode = &MockDataNode{
-		stopErr: errors.New("error"),
-	}
-
-	err = server.Stop()
-	assert.Error(t, err)
 }

internal/distributed/indexnode/service.go

@@ -61,7 +61,7 @@ type Server struct {
 
 	loopCtx    context.Context
 	loopCancel func()
-	loopWg     sync.WaitGroup
+	grpcWG     sync.WaitGroup
 
 	etcdCli *clientv3.Client
 }
@@ -81,7 +81,7 @@ func (s *Server) Run() error {
 
 // startGrpcLoop starts the grep loop of IndexNode component.
 func (s *Server) startGrpcLoop(grpcPort int) {
-	defer s.loopWg.Done()
+	defer s.grpcWG.Done()
 
 	Params := &paramtable.Get().IndexNodeGrpcServerCfg
 	log.Debug("IndexNode", zap.String("network address", Params.GetAddress()), zap.Int("network port: ", grpcPort))
@@ -159,7 +159,7 @@ func (s *Server) init() error {
 		}
 	}()
 
-	s.loopWg.Add(1)
+	s.grpcWG.Add(1)
 	go s.startGrpcLoop(Params.Port.GetAsInt())
 	// wait for grpc server loop start
 	err = <-s.grpcErrChan
@@ -220,17 +220,21 @@ func (s *Server) Stop() (err error) {
 	}()
 
 	if s.indexnode != nil {
-		s.indexnode.Stop()
+		err := s.indexnode.Stop()
+		if err != nil {
+			log.Error("failed to close indexnode", zap.Error(err))
+			return err
+		}
 	}
-	s.loopCancel()
 	if s.etcdCli != nil {
 		defer s.etcdCli.Close()
 	}
 	if s.grpcServer != nil {
 		utils.GracefulStopGRPCServer(s.grpcServer)
 	}
-	s.loopWg.Wait()
+	s.grpcWG.Wait()
 
+	s.loopCancel()
 	return nil
 }
 

internal/distributed/proxy/service.go

@@ -745,8 +745,10 @@ func (s *Server) Stop() (err error) {
 
 	s.wg.Wait()
 
+	logger.Info("internal server[proxy] start to stop")
 	err = s.proxy.Stop()
 	if err != nil {
+		log.Error("failed to close proxy", zap.Error(err))
 		return err
 	}
 

internal/distributed/querycoord/service.go

@@ -57,7 +57,7 @@ import (
 
 // Server is the grpc server of QueryCoord.
 type Server struct {
-	wg         sync.WaitGroup
+	grpcWG     sync.WaitGroup
 	loopCtx    context.Context
 	loopCancel context.CancelFunc
 	grpcServer *grpc.Server
@@ -147,7 +147,7 @@ func (s *Server) init() error {
 		log.Info("Connected to tikv. Using tikv as metadata storage.")
 	}
 
-	s.wg.Add(1)
+	s.grpcWG.Add(1)
 	go s.startGrpcLoop(rpcParams.Port.GetAsInt())
 	// wait for grpc server loop start
 	err = <-s.grpcErrChan
@@ -204,7 +204,7 @@ func (s *Server) init() error {
 }
 
 func (s *Server) startGrpcLoop(grpcPort int) {
-	defer s.wg.Done()
+	defer s.grpcWG.Done()
 	Params := &paramtable.Get().QueryCoordGrpcServerCfg
 	kaep := keepalive.EnforcementPolicy{
 		MinTime:             5 * time.Second, // If a client pings more than once every 5 seconds, terminate the connection
@@ -283,11 +283,18 @@ func (s *Server) Stop() (err error) {
 	if s.etcdCli != nil {
 		defer s.etcdCli.Close()
 	}
-	s.loopCancel()
+
 	if s.grpcServer != nil {
 		utils.GracefulStopGRPCServer(s.grpcServer)
 	}
-	return s.queryCoord.Stop()
+	s.grpcWG.Wait()
+
+	logger.Info("internal server[queryCoord] start to stop")
+	if err := s.queryCoord.Stop(); err != nil {
+		log.Error("failed to close queryCoord", zap.Error(err))
+	}
+	s.loopCancel()
+	return nil
 }
 
 // SetRootCoord sets root coordinator's client

internal/distributed/querynode/service.go

@@ -58,7 +58,7 @@ type UniqueID = typeutil.UniqueID
 // Server is the grpc server of QueryNode.
 type Server struct {
 	querynode   types.QueryNodeComponent
-	wg          sync.WaitGroup
+	grpcWG      sync.WaitGroup
 	ctx         context.Context
 	cancel      context.CancelFunc
 	grpcErrChan chan error
@@ -118,7 +118,7 @@ func (s *Server) init() error {
 	s.SetEtcdClient(etcdCli)
 	s.querynode.SetAddress(Params.GetAddress())
 	log.Debug("QueryNode connect to etcd successfully")
-	s.wg.Add(1)
+	s.grpcWG.Add(1)
 	go s.startGrpcLoop(Params.Port.GetAsInt())
 	// wait for grpc server loop start
 	err = <-s.grpcErrChan
@@ -152,7 +152,7 @@ func (s *Server) start() error {
 
 // startGrpcLoop starts the grpc loop of QueryNode component.
 func (s *Server) startGrpcLoop(grpcPort int) {
-	defer s.wg.Done()
+	defer s.grpcWG.Done()
 	Params := &paramtable.Get().QueryNodeGrpcServerCfg
 	kaep := keepalive.EnforcementPolicy{
 		MinTime:             5 * time.Second, // If a client pings more than once every 5 seconds, terminate the connection
@@ -245,19 +245,22 @@ func (s *Server) Stop() (err error) {
 		logger.Info("QueryNode stopped", zap.Error(err))
 	}()
 
+	logger.Info("internal server[querynode] start to stop")
 	err = s.querynode.Stop()
 	if err != nil {
+		log.Error("failed to close querynode", zap.Error(err))
 		return err
 	}
 	if s.etcdCli != nil {
 		defer s.etcdCli.Close()
 	}
 
-	s.cancel()
 	if s.grpcServer != nil {
 		utils.GracefulStopGRPCServer(s.grpcServer)
 	}
-	s.wg.Wait()
+	s.grpcWG.Wait()
+
+	s.cancel()
 	return nil
 }
 

internal/distributed/rootcoord/service.go

@@ -61,7 +61,7 @@ type Server struct {
 	grpcServer  *grpc.Server
 	grpcErrChan chan error
 
-	wg sync.WaitGroup
+	grpcWG sync.WaitGroup
 
 	ctx    context.Context
 	cancel context.CancelFunc
@@ -234,7 +234,7 @@ func (s *Server) init() error {
 }
 
 func (s *Server) startGrpc(port int) error {
-	s.wg.Add(1)
+	s.grpcWG.Add(1)
 	go s.startGrpcLoop(port)
 	// wait for grpc server loop start
 	err := <-s.grpcErrChan
@@ -242,7 +242,7 @@ func (s *Server) startGrpc(port int) error {
 }
 
 func (s *Server) startGrpcLoop(port int) {
-	defer s.wg.Done()
+	defer s.grpcWG.Done()
 	Params := &paramtable.Get().RootCoordGrpcServerCfg
 	kaep := keepalive.EnforcementPolicy{
 		MinTime:             5 * time.Second, // If a client pings more than once every 5 seconds, terminate the connection
@@ -330,11 +330,10 @@ func (s *Server) Stop() (err error) {
 		defer s.tikvCli.Close()
 	}
 
-	s.cancel()
 	if s.grpcServer != nil {
 		utils.GracefulStopGRPCServer(s.grpcServer)
 	}
-	s.wg.Wait()
+	s.grpcWG.Wait()
 
 	if s.dataCoord != nil {
 		if err := s.dataCoord.Close(); err != nil {
@@ -347,10 +346,13 @@ func (s *Server) Stop() (err error) {
 		}
 	}
 	if s.rootCoord != nil {
+		logger.Info("internal server[rootCoord] start to stop")
 		if err := s.rootCoord.Stop(); err != nil {
-			log.Error("Failed to close close rootCoord", zap.Error(err))
+			log.Error("Failed to close rootCoord", zap.Error(err))
 		}
 	}
+
+	s.cancel()
 	return nil
 }
 

internal/indexnode/indexnode.go

@@ -257,7 +257,6 @@ func (i *IndexNode) Stop() error {
 				task.cancel()
 			}
 		}
-		i.loopCancel()
 		if i.sched != nil {
 			i.sched.Close()
 		}
@@ -266,6 +265,7 @@ func (i *IndexNode) Stop() error {
 		}
 
 		i.CloseSegcore()
+		i.loopCancel()
 		log.Info("Index node stopped.")
 	})
 	return nil

internal/querycoordv2/server.go

@@ -451,11 +451,6 @@ func (s *Server) startServerLoop() {
 }
 
 func (s *Server) Stop() error {
-	// stop the components from outside to inside,
-	// to make the dependencies stopped working properly,
-	// cancel the server context first to stop receiving requests
-	s.cancel()
-
 	// FOLLOW the dependence graph:
 	// job scheduler -> checker controller -> task scheduler -> dist controller -> cluster -> session
 	// observers -> dist controller
@@ -503,6 +498,7 @@ func (s *Server) Stop() error {
 		s.session.Stop()
 	}
 
+	s.cancel()
 	s.wg.Wait()
 	log.Info("QueryCoord stop successfully")
 	return nil

internal/querynodev2/server.go

@@ -476,8 +476,7 @@ func (node *QueryNode) Stop() error {
 		if node.pipelineManager != nil {
 			node.pipelineManager.Close()
 		}
-		// Delay the cancellation of ctx to ensure that the session is automatically recycled after closed the pipeline
-		node.cancel()
+
 		if node.session != nil {
 			node.session.Stop()
 		}
@@ -489,6 +488,9 @@ func (node *QueryNode) Stop() error {
 		}
 
 		node.CloseSegcore()
+
+		// Delay the cancellation of ctx to ensure that the session is automatically recycled after closed the pipeline
+		node.cancel()
 	})
 	return nil
 }

internal/rootcoord/root_coord.go

@@ -772,7 +772,7 @@ func (c *Core) revokeSession() {
 	if c.session != nil {
 		// wait at most one second to revoke
 		c.session.Stop()
-		log.Info("revoke rootcoord session")
+		log.Info("rootcoord session stop")
 	}
 }
 
@@ -784,12 +784,13 @@ func (c *Core) Stop() error {
 	if c.proxyWatcher != nil {
 		c.proxyWatcher.Stop()
 	}
-	c.cancelIfNotNil()
 	if c.quotaCenter != nil {
 		c.quotaCenter.stop()
 	}
-	c.wg.Wait()
+
 	c.revokeSession()
+	c.cancelIfNotNil()
+	c.wg.Wait()
 	return nil
 }
 

internal/util/sessionutil/session_util.go

@@ -151,6 +151,8 @@ type Session struct {
 	sessionTTL        int64
 	sessionRetryTimes int64
 	reuseNodeID       bool
+
+	isStopped atomic.Bool // set to true if stop method is invoked
 }
 
 type SessionOption func(session *Session)
@@ -239,6 +241,7 @@ func NewSessionWithEtcd(ctx context.Context, metaRoot string, client *clientv3.C
 		sessionTTL:        paramtable.Get().CommonCfg.SessionTTL.GetAsInt64(),
 		sessionRetryTimes: paramtable.Get().CommonCfg.SessionRetryTimes.GetAsInt64(),
 		reuseNodeID:       true,
+		isStopped:         *atomic.NewBool(false),
 	}
 
 	// integration test create cluster with different nodeId in one process
@@ -861,7 +864,8 @@ func (s *Session) LivenessCheck(ctx context.Context, callback func()) {
 		if callback != nil {
 			// before exit liveness check, callback to exit the session owner
 			defer func() {
-				if ctx.Err() == nil {
+				// the callback method will not be invoked if session is stopped.
+				if ctx.Err() == nil && !s.isStopped.Load() {
 					go callback()
 				}
 			}()
@@ -940,6 +944,7 @@ func (s *Session) deleteSession() bool {
 }
 
 func (s *Session) Stop() {
+	s.isStopped.Store(true)
 	s.Revoke(time.Second)
 	s.cancelKeepAlive()
 	s.deleteSession()
@@ -951,17 +956,28 @@ func (s *Session) Revoke(timeout time.Duration) {
 	if s == nil {
 		return
 	}
+	log.Info("start to revoke session", zap.String("sessionKey", s.activeKey))
 	if s.etcdCli == nil || s.LeaseID == nil {
+		log.Warn("skip remove session",
+			zap.String("sessionKey", s.activeKey),
+			zap.Bool("etcdCliIsNil", s.etcdCli == nil),
+			zap.Bool("LeaseIDIsNil", s.LeaseID == nil),
+		)
 		return
 	}
 	if s.Disconnected() {
+		log.Warn("skip remove session, connection is disconnected", zap.String("sessionKey", s.activeKey))
 		return
 	}
 	// can NOT use s.ctx, it may be Done here
 	ctx, cancel := context.WithTimeout(context.Background(), timeout)
 	defer cancel()
 	// ignores resp & error, just do best effort to revoke
-	_, _ = s.etcdCli.Revoke(ctx, *s.LeaseID)
+	_, err := s.etcdCli.Revoke(ctx, *s.LeaseID)
+	if err != nil {
+		log.Warn("failed to revoke session", zap.String("sessionKey", s.activeKey), zap.Error(err))
+	}
+	log.Info("revoke session successfully", zap.String("sessionKey", s.activeKey))
 }
 
 // UpdateRegistered update the state of registered.