Add an extra loop when the pod watcher is closed unexpectedly. (#226)

vkubelet/vkubelet.go

@@ -192,20 +192,24 @@ func (s *Server) registerNode() error {
 // Run starts the server, registers it with Kubernetes and begins watching/reconciling the cluster.
 // Run will block until Stop is called or a SIGINT or SIGTERM signal is received.
 func (s *Server) Run() error {
+	shouldStop := false
+
 	sig := make(chan os.Signal, 1)
 	signal.Notify(sig, syscall.SIGINT, syscall.SIGTERM)
 	go func() {
 		<-sig
+		shouldStop = true
 		s.Stop()
 	}()
 
+	for {
 		opts := metav1.ListOptions{
 			FieldSelector: fields.OneTermEqualSelector("spec.nodeName", s.nodeName).String(),
 		}
 
 		pods, err := s.k8sClient.CoreV1().Pods(s.namespace).List(opts)
 		if err != nil {
-		log.Fatal(err)
+			log.Fatal("Failed to list pods", err)
 		}
 		s.resourceManager.SetPods(pods)
 		s.reconcile()
@@ -213,16 +217,24 @@ func (s *Server) Run() error {
 		opts.ResourceVersion = pods.ResourceVersion
 		s.podWatcher, err = s.k8sClient.CoreV1().Pods(s.namespace).Watch(opts)
 		if err != nil {
-		log.Fatal(err)
+			log.Fatal("Failed to watch pods", err)
 		}
 
+		loop:
 		for {
 			select {
 			case ev, ok := <-s.podWatcher.ResultChan():
 				if !ok {
+					if shouldStop {
+						log.Println("Pod watcher is stopped.")
 						return nil
 					}
 
+					log.Println("Pod watcher connection is closed unexpectedly.")
+					break loop
+				}
+
+				log.Println("Pod watcher event is received:", ev.Type)
 				switch ev.Type {
 				case watch.Added:
 					s.resourceManager.AddPod(ev.Object.(*corev1.Pod))
@@ -234,6 +246,9 @@ func (s *Server) Run() error {
 				s.reconcile()
 			}
 		}
+
+		time.Sleep(5 * time.Second)
+	}
 }
 
 // Stop shutsdown the server.