Merge pull request #5021 from wzshiming/fix/signal_repeatedly

Fix repeated sending signal

pkg/cri/server/container_stop.go

@@ -17,6 +17,7 @@
 package server
 
 import (
+	"sync/atomic"
 	"syscall"
 	"time"
 
@@ -129,10 +130,23 @@ func (c *criService) stopContainer(ctx context.Context, container containerstore
 		if err != nil {
 			return errors.Wrapf(err, "failed to parse stop signal %q", stopSignal)
 		}
+
+		var sswt bool
+		if container.IsStopSignaledWithTimeout == nil {
+			log.G(ctx).Infof("unable to ensure stop signal %v was not sent twice to container %v", sig, id)
+			sswt = true
+		} else {
+			sswt = atomic.CompareAndSwapUint32(container.IsStopSignaledWithTimeout, 0, 1)
+		}
+
+		if sswt {
 			log.G(ctx).Infof("Stop container %q with signal %v", id, sig)
 			if err = task.Kill(ctx, sig); err != nil && !errdefs.IsNotFound(err) {
 				return errors.Wrapf(err, "failed to stop container %q", id)
 			}
+		} else {
+			log.G(ctx).Infof("Skipping the sending of signal %v to container %q because a prior stop with timeout>0 request already sent the signal", sig, id)
+		}
 
 		sigTermCtx, sigTermCtxCancel := context.WithTimeout(ctx, timeout)
 		defer sigTermCtxCancel()

pkg/cri/store/container/container.go

@@ -42,6 +42,9 @@ type Container struct {
 	IO *cio.ContainerIO
 	// StopCh is used to propagate the stop information of the container.
 	*store.StopCh
+	// IsStopSignaledWithTimeout the default is 0, and it is set to 1 after sending
+	// the signal once to avoid repeated sending of the signal.
+	IsStopSignaledWithTimeout *uint32
 }
 
 // Opts sets specific information to newly created Container.
@@ -83,6 +86,7 @@ func NewContainer(metadata Metadata, opts ...Opts) (Container, error) {
 	c := Container{
 		Metadata:                  metadata,
 		StopCh:                    store.NewStopCh(),
+		IsStopSignaledWithTimeout: new(uint32),
 	}
 	for _, o := range opts {
 		if err := o(&c); err != nil {