fix(client): synchronize graceful shutdown duration (#5442)

client/service.go

@@ -22,6 +22,7 @@ import (
 	"net/http"
 	"os"
 	"sync"
+	"sync/atomic"
 	"time"
 
 	"github.com/fatedier/golib/crypto"
@@ -109,6 +110,9 @@ func setServiceOptionsDefault(options *ServiceOptions) error {
 // Service is the client service that connects to frps and provides proxy services.
 type Service struct {
 	ctlMu sync.RWMutex
+	// Stores gracefulShutdownDuration independently from ctlMu, because the
+	// graceful shutdown wait may hold ctlMu for an arbitrary duration.
+	gracefulShutdownDuration atomic.Int64
 	// manager control connection with server
 	ctl *Control
 	// Uniq id got from frps, it will be attached to loginMsg.
@@ -150,7 +154,6 @@ type Service struct {
 	ctx context.Context
 	// call cancel to stop service
 	cancel context.CancelCauseFunc
-	gracefulShutdownDuration time.Duration
 
 	connectorCreator func(context.Context, *v1.ClientCommonConfig) Connector
 	handleWorkConnCb func(*v1.ProxyBaseConfig, net.Conn, *msg.StartWorkConn) bool
@@ -412,7 +415,7 @@ func (svr *Service) Close() {
 }
 
 func (svr *Service) GracefulClose(d time.Duration) {
-	svr.gracefulShutdownDuration = d
+	svr.gracefulShutdownDuration.Store(int64(d))
 	svr.cancel(nil)
 }
 
@@ -429,7 +432,8 @@ func (svr *Service) stop() {
 	svr.ctlMu.Lock()
 	defer svr.ctlMu.Unlock()
 	if svr.ctl != nil {
-		svr.ctl.GracefulClose(svr.gracefulShutdownDuration)
+		d := time.Duration(svr.gracefulShutdownDuration.Load())
+		svr.ctl.GracefulClose(d)
 		svr.ctl = nil
 	}
 	if svr.webServer != nil {

client/service_shutdown_test.go

@@ -0,0 +1,95 @@
+package client
+
+import (
+	"context"
+	"net"
+	"sync"
+	"testing"
+	"time"
+
+	"github.com/fatedier/frp/client/proxy"
+	"github.com/fatedier/frp/client/visitor"
+	v1 "github.com/fatedier/frp/pkg/config/v1"
+	"github.com/fatedier/frp/pkg/msg"
+)
+
+type gracefulCloseTestConnector struct {
+	conn net.Conn
+}
+
+func (*gracefulCloseTestConnector) Connect() (*msg.Conn, error) { return nil, net.ErrClosed }
+func (c *gracefulCloseTestConnector) Close() error              { return c.conn.Close() }
+
+func newGracefulCloseTestService() *Service {
+	ctx := context.Background()
+	common := &v1.ClientCommonConfig{}
+	serverConn, clientConn := net.Pipe()
+	ctl := &Control{
+		ctx: ctx,
+		sessionCtx: &SessionContext{
+			Common:    common,
+			RunID:     "graceful-close-race",
+			Conn:      msg.NewConn(clientConn, msg.NewV1ReadWriter(clientConn)),
+			Connector: &gracefulCloseTestConnector{conn: serverConn},
+		},
+		doneCh: make(chan struct{}),
+	}
+	ctl.pm = proxy.NewManager(ctx, common, nil, nil, nil)
+	ctl.vm = visitor.NewManager(ctx, "graceful-close-race", common, nil, nil, nil)
+	return &Service{ctl: ctl, cancel: context.CancelCauseFunc(func(error) {})}
+}
+
+func TestGracefulCloseAndStopSynchronizeDuration(t *testing.T) {
+	for i := range 10000 {
+		svr := newGracefulCloseTestService()
+		start := make(chan struct{})
+		var wg sync.WaitGroup
+		wg.Add(2)
+		go func() {
+			defer wg.Done()
+			<-start
+			svr.GracefulClose(time.Duration(i))
+		}()
+		go func() {
+			defer wg.Done()
+			<-start
+			svr.stop()
+		}()
+		close(start)
+		wg.Wait()
+	}
+}
+
+func TestGracefulCloseDoesNotBlockDuringStop(t *testing.T) {
+	const gracefulDuration = 200 * time.Millisecond
+
+	svr := newGracefulCloseTestService()
+	svr.GracefulClose(gracefulDuration)
+	stopDone := make(chan struct{})
+	go func() {
+		svr.stop()
+		close(stopDone)
+	}()
+	defer func() {
+		select {
+		case <-stopDone:
+		case <-time.After(time.Second):
+			t.Error("stop did not finish")
+		}
+	}()
+
+	deadline := time.Now().Add(time.Second)
+	for svr.ctlMu.TryLock() {
+		svr.ctlMu.Unlock()
+		if time.Now().After(deadline) {
+			t.Fatal("stop did not acquire ctlMu")
+		}
+		time.Sleep(time.Millisecond)
+	}
+
+	start := time.Now()
+	svr.GracefulClose(0)
+	if elapsed := time.Since(start); elapsed >= gracefulDuration/2 {
+		t.Fatalf("GracefulClose blocked for %v while stop was waiting", elapsed)
+	}
+}