Merge pull request #126543 from gjkim42/restart-failed-to-start-container

Restart the init container to not be stuck in created state

pkg/kubelet/kuberuntime/kuberuntime_container.go

@@ -1084,7 +1084,12 @@ func (m *kubeGenericRuntimeManager) computeInitContainerActions(pod *v1.Pod, pod
 
 		switch status.State {
 		case kubecontainer.ContainerStateCreated:
-			// nothing to do but wait for it to start
+			// The main sync loop should have created and started the container
+			// in one step. If the init container is in the 'created' state,
+			// it is likely that the container runtime failed to start it. To
+			// prevent the container from getting stuck in the 'created' state,
+			// restart it.
+			changes.InitContainersToStart = append(changes.InitContainersToStart, i)
 
 		case kubecontainer.ContainerStateRunning:
 			if !types.IsRestartableInitContainer(container) {

pkg/kubelet/kuberuntime/kuberuntime_manager_test.go

@@ -1171,6 +1171,17 @@ func TestComputePodActions(t *testing.T) {
 				ContainersToStart: []int{1},
 			},
 		},
+		"Restart the container if the container is in created state": {
+			mutatePodFn: func(pod *v1.Pod) { pod.Spec.RestartPolicy = v1.RestartPolicyNever },
+			mutateStatusFn: func(status *kubecontainer.PodStatus) {
+				status.ContainerStatuses[1].State = kubecontainer.ContainerStateCreated
+			},
+			actions: podActions{
+				SandboxID:         baseStatus.SandboxStatuses[0].Id,
+				ContainersToKill:  map[kubecontainer.ContainerID]containerToKillInfo{},
+				ContainersToStart: []int{1},
+			},
+		},
 	} {
 		pod, status := makeBasePodAndStatus()
 		if test.mutatePodFn != nil {
@@ -1545,12 +1556,17 @@ func TestComputePodActionsWithRestartableInitContainers(t *testing.T) {
 				ContainersToKill:      getKillMapWithInitContainers(basePod, baseStatus, []int{}),
 			},
 		},
-		"initialization in progress; do nothing": {
+		"an init container is stuck in the created state; restart it": {
 			mutatePodFn: func(pod *v1.Pod) { pod.Spec.RestartPolicy = v1.RestartPolicyAlways },
 			mutateStatusFn: func(pod *v1.Pod, status *kubecontainer.PodStatus) {
 				status.ContainerStatuses[2].State = kubecontainer.ContainerStateCreated
 			},
-			actions: noAction,
+			actions: podActions{
+				SandboxID:             baseStatus.SandboxStatuses[0].Id,
+				InitContainersToStart: []int{2},
+				ContainersToStart:     []int{},
+				ContainersToKill:      getKillMapWithInitContainers(basePod, baseStatus, []int{}),
+			},
 		},
 		"restartable init container has started; start the next": {
 			mutatePodFn: func(pod *v1.Pod) { pod.Spec.RestartPolicy = v1.RestartPolicyAlways },