Merge pull request kubernetes#133243 from yuanwang04/restart-rules

ContainerRestartRules feature gate should work with probes

Author: k8s-ci-robot

pkg/kubelet/kuberuntime/kuberuntime_container.go

@@ -1222,7 +1222,12 @@ func (m *kubeGenericRuntimeManager) computeInitContainerActions(ctx context.Cont
 
 				restartOnFailure := restartOnFailure
 				if utilfeature.DefaultFeatureGate.Enabled(features.ContainerRestartRules) {
-					restartOnFailure = kubecontainer.ShouldContainerBeRestarted(container, pod, podStatus)
+					// Only container-level restart policy is used. The container-level restart
+					// rules are not evaluated because the container might not have exited, so
+					// there is no exit code on which the rules can be used.
+					if container.RestartPolicy != nil {
+						restartOnFailure = *container.RestartPolicy != v1.ContainerRestartPolicyNever
+					}
 				}
 				if !restartOnFailure {
 					changes.KillPod = true

pkg/kubelet/kuberuntime/kuberuntime_manager.go

@@ -589,16 +589,6 @@ func containerChanged(container *v1.Container, containerStatus *kubecontainer.St
 }
 
 func shouldRestartOnFailure(pod *v1.Pod) bool {
-	// With feature ContainerRestartRules enabled, the pod should be restarted
-	// on failure if any of its containers have container-level restart policy
-	// that is restartable.
-	if utilfeature.DefaultFeatureGate.Enabled(features.ContainerRestartRules) {
-		for _, c := range pod.Spec.Containers {
-			if podutil.IsContainerRestartable(pod.Spec, c) {
-				return true
-			}
-		}
-	}
 	return pod.Spec.RestartPolicy != v1.RestartPolicyNever
 }
 
@@ -1147,7 +1137,11 @@ func (m *kubeGenericRuntimeManager) computePodActions(ctx context.Context, pod *
 		var reason containerKillReason
 		restart := shouldRestartOnFailure(pod)
 		if utilfeature.DefaultFeatureGate.Enabled(features.ContainerRestartRules) {
-			restart = kubecontainer.ShouldContainerBeRestarted(&container, pod, podStatus)
+			// For probe failures, use container-level restart policy only. Container-level restart
+			// rules are not evaluated because the container is still running.
+			if container.RestartPolicy != nil {
+				restart = *container.RestartPolicy != v1.ContainerRestartPolicyNever
+			}
 		}
 		if _, _, changed := containerChanged(&container, containerStatus); changed {
 			message = fmt.Sprintf("Container %s definition changed", container.Name)

pkg/kubelet/kuberuntime/kuberuntime_manager_test.go

@@ -2110,6 +2110,11 @@ func TestComputePodActionsWithInitAndEphemeralContainers(t *testing.T) {
 }
 
 func TestComputePodActionsWithContainerRestartRules(t *testing.T) {
+	// Make sure existing test cases pass with feature enabled
+	featuregatetesting.SetFeatureGateDuringTest(t, utilfeature.DefaultFeatureGate, features.ContainerRestartRules, true)
+	TestComputePodActions(t)
+	TestComputePodActionsWithInitContainers(t)
+
 	var (
 		containerRestartPolicyAlways    = v1.ContainerRestartPolicyAlways
 		containerRestartPolicyOnFailure = v1.ContainerRestartPolicyOnFailure
@@ -2231,7 +2236,6 @@ func TestComputePodActionsWithContainerRestartRules(t *testing.T) {
 			},
 		},
 	} {
-		featuregatetesting.SetFeatureGateDuringTest(t, utilfeature.DefaultFeatureGate, features.ContainerRestartRules, true)
 		pod, status := makeBasePodAndStatus()
 		if test.mutatePodFn != nil {
 			test.mutatePodFn(pod)

test/e2e/node/pods.go

@@ -718,7 +718,7 @@ var _ = SIGDescribe("Pods Extended (pod generation)", feature.PodObservedGenerat
 	})
 })
 
-var _ = SIGDescribe("Pod Extended (container restart policy)", feature.ContainerRestartRules, framework.WithFeatureGate(features.ContainerRestartRules), func() {
+var _ = SIGDescribe("Pod Extended (container restart policy)", framework.WithFeatureGate(features.ContainerRestartRules), func() {
 	f := framework.NewDefaultFramework("pods")
 	f.NamespacePodSecurityLevel = admissionapi.LevelBaseline