Merge pull request #40330 from janetkuo/kill-failed-daemon-pods

pkg/controller/daemon/daemoncontroller.go

@@ -461,26 +461,42 @@ func (dsc *DaemonSetsController) manage(ds *extensions.DaemonSet) error {
 		return fmt.Errorf("couldn't get list of nodes when syncing daemon set %#v: %v", ds, err)
 	}
 	var nodesNeedingDaemonPods, podsToDelete []string
+	var failedPodsObserved int
 	for _, node := range nodeList.Items {
 		_, shouldSchedule, shouldContinueRunning, err := dsc.nodeShouldRunDaemonPod(&node, ds)
 		if err != nil {
 			continue
 		}
 
-		daemonPods, isRunning := nodeToDaemonPods[node.Name]
+		daemonPods, exists := nodeToDaemonPods[node.Name]
 
 		switch {
-		case shouldSchedule && !isRunning:
+		case shouldSchedule && !exists:
 			// If daemon pod is supposed to be running on node, but isn't, create daemon pod.
 			nodesNeedingDaemonPods = append(nodesNeedingDaemonPods, node.Name)
-		case shouldContinueRunning && len(daemonPods) > 1:
+		case shouldContinueRunning:
+			// If a daemon pod failed, delete it
+			// If there's no daemon pods left on this node, we will create it in the next sync loop
+			var daemonPodsRunning []*v1.Pod
+			for i := range daemonPods {
+				pod := daemonPods[i]
+				if pod.Status.Phase == v1.PodFailed {
+					glog.V(2).Infof("Found failed daemon pod %s/%s on node %s, will try to kill it", pod.Namespace, node.Name, pod.Name)
+					podsToDelete = append(podsToDelete, pod.Name)
+					failedPodsObserved++
+				} else {
+					daemonPodsRunning = append(daemonPodsRunning, pod)
+				}
+			}
 			// If daemon pod is supposed to be running on node, but more than 1 daemon pod is running, delete the excess daemon pods.
-			// Sort the daemon pods by creation time, so the the oldest is preserved.
-			sort.Sort(podByCreationTimestamp(daemonPods))
-			for i := 1; i < len(daemonPods); i++ {
-				podsToDelete = append(podsToDelete, daemonPods[i].Name)
+			// Sort the daemon pods by creation time, so the oldest is preserved.
+			if len(daemonPodsRunning) > 1 {
+				sort.Sort(podByCreationTimestamp(daemonPodsRunning))
+				for i := 1; i < len(daemonPodsRunning); i++ {
+					podsToDelete = append(podsToDelete, daemonPods[i].Name)
+				}
 			}
-		case !shouldContinueRunning && isRunning:
+		case !shouldContinueRunning && exists:
 			// If daemon pod isn't supposed to run on node, but it is, delete all daemon pods on node.
 			for i := range daemonPods {
 				podsToDelete = append(podsToDelete, daemonPods[i].Name)
@@ -547,6 +563,10 @@ func (dsc *DaemonSetsController) manage(ds *extensions.DaemonSet) error {
 	for err := range errCh {
 		errors = append(errors, err)
 	}
+	// Throw an error when the daemon pods fail, to use ratelimiter to prevent kill-recreate hot loop
+	if failedPodsObserved > 0 {
+		errors = append(errors, fmt.Errorf("deleted %d failed pods of DaemonSet %s/%s", failedPodsObserved, ds.Namespace, ds.Name))
+	}
 	return utilerrors.NewAggregate(errors)
 }
 
@@ -774,7 +794,7 @@ func (dsc *DaemonSetsController) nodeShouldRunDaemonPod(node *v1.Node, ds *exten
 				predicates.ErrTaintsTolerationsNotMatch:
 				return false, false, false, fmt.Errorf("unexpected reason: GeneralPredicates should not return reason %s", reason.GetReason())
 			default:
-				glog.V(4).Infof("unknownd predicate failure reason: %s", reason.GetReason())
+				glog.V(4).Infof("unknown predicate failure reason: %s", reason.GetReason())
 				wantToRun, shouldSchedule, shouldContinueRunning = false, false, false
 				emitEvent = true
 			}

pkg/controller/daemon/daemoncontroller_test.go

@@ -138,6 +138,14 @@ func addPods(podStore cache.Store, nodeName string, label map[string]string, num
 	}
 }
 
+func addFailedPods(podStore cache.Store, nodeName string, label map[string]string, number int) {
+	for i := 0; i < number; i++ {
+		pod := newPod(fmt.Sprintf("%s-", nodeName), nodeName, label)
+		pod.Status = v1.PodStatus{Phase: v1.PodFailed}
+		podStore.Add(pod)
+	}
+}
+
 func newTestController(initialObjects ...runtime.Object) (*DaemonSetsController, *controller.FakePodControl, *fake.Clientset) {
 	clientset := fake.NewSimpleClientset(initialObjects...)
 	informerFactory := informers.NewSharedInformerFactory(clientset, nil, controller.NoResyncPeriodFunc())
@@ -653,6 +661,31 @@ func TestObservedGeneration(t *testing.T) {
 	}
 }
 
+// DaemonSet controller should kill all failed pods and create at most 1 pod on every node.
+func TestDaemonKillFailedPods(t *testing.T) {
+	tests := []struct {
+		numFailedPods, numNormalPods, expectedCreates, expectedDeletes int
+		test                                                           string
+	}{
+		{numFailedPods: 0, numNormalPods: 1, expectedCreates: 0, expectedDeletes: 0, test: "normal (do nothing)"},
+		{numFailedPods: 0, numNormalPods: 0, expectedCreates: 1, expectedDeletes: 0, test: "no pods (create 1)"},
+		{numFailedPods: 1, numNormalPods: 0, expectedCreates: 0, expectedDeletes: 1, test: "1 failed pod (kill 1), 0 normal pod (create 0; will create in the next sync)"},
+		{numFailedPods: 1, numNormalPods: 3, expectedCreates: 0, expectedDeletes: 3, test: "1 failed pod (kill 1), 3 normal pods (kill 2)"},
+		{numFailedPods: 2, numNormalPods: 1, expectedCreates: 0, expectedDeletes: 2, test: "2 failed pods (kill 2), 1 normal pod"},
+	}
+
+	for _, test := range tests {
+		t.Logf("test case: %s\n", test.test)
+		manager, podControl, _ := newTestController()
+		addNodes(manager.nodeStore.Store, 0, 1, nil)
+		addFailedPods(manager.podStore.Indexer, "node-0", simpleDaemonSetLabel, test.numFailedPods)
+		addPods(manager.podStore.Indexer, "node-0", simpleDaemonSetLabel, test.numNormalPods)
+		ds := newDaemonSet("foo")
+		manager.dsStore.Add(ds)
+		syncAndValidateDaemonSets(t, manager, ds, podControl, test.expectedCreates, test.expectedDeletes)
+	}
+}
+
 func TestNodeShouldRunDaemonPod(t *testing.T) {
 	cases := []struct {
 		podsOnNode                                       []*v1.Pod