Set the pod phase based on pod restart policy when provider failed (#361)

Update the resource manager to include the deleting pods in the GetPods function

manager/resource.go

@@ -86,16 +86,14 @@ func (rm *ResourceManager) SetPods(pods *v1.PodList) {
 	rm.Lock()
 	defer rm.Unlock()
 
-	rm.pods = make(map[string]*v1.Pod, len(pods.Items))
-	rm.configMapRef = make(map[string]int64, 0)
-	rm.secretRef = make(map[string]int64, 0)
-	rm.configMaps = make(map[string]*v1.ConfigMap, len(pods.Items))
-	rm.secrets = make(map[string]*v1.Secret, len(pods.Items))
-
 	for k, p := range pods.Items {
-		rm.pods[rm.getStoreKey(p.Namespace, p.Name)] = &pods.Items[k]
+		podKey := rm.getStoreKey(p.Namespace, p.Name)
-
+		if p.DeletionTimestamp != nil {
-		rm.incrementRefCounters(&p)
+			rm.deletingPods[podKey] = &pods.Items[k]
+		} else {
+			rm.pods[podKey] = &pods.Items[k]
+			rm.incrementRefCounters(&p)
+		}
 	}
 }
 
@@ -173,10 +171,13 @@ func (rm *ResourceManager) GetPods() []*v1.Pod {
 	rm.RLock()
 	defer rm.RUnlock()
 
-	pods := make([]*v1.Pod, 0, len(rm.pods))
+	pods := make([]*v1.Pod, 0, len(rm.pods)+len(rm.deletingPods))
 	for _, p := range rm.pods {
 		pods = append(pods, p)
 	}
+	for _, p := range rm.deletingPods {
+		pods = append(pods, p)
+	}
 
 	return pods
 }

vkubelet/vkubelet.go

@@ -349,7 +349,11 @@ func (s *Server) reconcile(ctx context.Context) {
 			}
 		}
 
-		if pod.DeletionTimestamp == nil && pod.Status.Phase != corev1.PodFailed && providerPod == nil {
+		if providerPod == nil &&
+		   pod.DeletionTimestamp == nil &&
+		   pod.Status.Phase != corev1.PodSucceeded &&
+		   pod.Status.Phase != corev1.PodFailed &&
+		   pod.Status.Reason != PodStatusReason_ProviderFailed {
 			logger.Debug("Creating pod")
 			if err := s.createPod(ctx, pod); err != nil {
 				logger.WithError(err).Error("Error creating pod")
@@ -378,8 +382,13 @@ func (s *Server) createPod(ctx context.Context, pod *corev1.Pod) error {
 	logger := log.G(ctx).WithField("pod", pod.Name)
 
 	if origErr := s.provider.CreatePod(ctx, pod); origErr != nil {
+		podPhase := corev1.PodPending
+		if pod.Spec.RestartPolicy == corev1.RestartPolicyNever {
+			podPhase = corev1.PodFailed
+		}
+
 		pod.ResourceVersion = "" // Blank out resource version to prevent object has been modified error
-		pod.Status.Phase = corev1.PodFailed
+		pod.Status.Phase = podPhase
 		pod.Status.Reason = PodStatusReason_ProviderFailed
 		pod.Status.Message = origErr.Error()
 
@@ -426,7 +435,9 @@ func (s *Server) updatePodStatuses(ctx context.Context) {
 	// Update all the pods with the provider status.
 	pods := s.resourceManager.GetPods()
 	for _, pod := range pods {
-		if pod.Status.Phase == corev1.PodSucceeded || (pod.Status.Phase == corev1.PodFailed && pod.Status.Reason == PodStatusReason_ProviderFailed) {
+		if pod.Status.Phase == corev1.PodSucceeded ||
+		   pod.Status.Phase == corev1.PodFailed ||
+		   pod.Status.Reason == PodStatusReason_ProviderFailed {
 			continue
 		}