Delete only unscheduled pods if node doesn't exist anymore.

Author: krzysztof-jastrzebski

pkg/controller/daemon/daemon_controller.go

@@ -960,10 +960,10 @@ func (dsc *DaemonSetsController) manage(ds *apps.DaemonSet, nodeList []*v1.Node,
 		failedPodsObserved += failedPodsObservedOnNode
 	}
 
-	// Remove pods assigned to not existing nodes when daemonset pods are scheduled by default scheduler.
+	// Remove unscheduled pods assigned to not existing nodes when daemonset pods are scheduled by scheduler.
 	// If node doesn't exist then pods are never scheduled and can't be deleted by PodGCController.
 	if utilfeature.DefaultFeatureGate.Enabled(features.ScheduleDaemonSetPods) {
-		podsToDelete = append(podsToDelete, getPodsWithoutNode(nodeList, nodeToDaemonPods)...)
+		podsToDelete = append(podsToDelete, getUnscheduledPodsWithoutNode(nodeList, nodeToDaemonPods)...)
 	}
 
 	// Label new pods using the hash label value of the current history when creating them
@@ -1527,8 +1527,9 @@ func failedPodsBackoffKey(ds *apps.DaemonSet, nodeName string) string {
 	return fmt.Sprintf("%s/%d/%s", ds.UID, ds.Status.ObservedGeneration, nodeName)
 }
 
-// getPodsWithoutNode returns list of pods assigned to not existing nodes.
-func getPodsWithoutNode(runningNodesList []*v1.Node, nodeToDaemonPods map[string][]*v1.Pod) []string {
+// getUnscheduledPodsWithoutNode returns list of unscheduled pods assigned to not existing nodes.
+// Returned pods can't be deleted by PodGCController so they should be deleted by DaemonSetController.
+func getUnscheduledPodsWithoutNode(runningNodesList []*v1.Node, nodeToDaemonPods map[string][]*v1.Pod) []string {
 	var results []string
 	isNodeRunning := make(map[string]bool)
 	for _, node := range runningNodesList {
@@ -1537,7 +1538,9 @@ func getPodsWithoutNode(runningNodesList []*v1.Node, nodeToDaemonPods map[string
 	for n, pods := range nodeToDaemonPods {
 		if !isNodeRunning[n] {
 			for _, pod := range pods {
-				results = append(results, pod.Name)
+				if len(pod.Spec.NodeName) == 0 {
+					results = append(results, pod.Name)
+				}
 			}
 		}
 	}

pkg/controller/daemon/daemon_controller_test.go

@@ -2447,7 +2447,7 @@ func TestDeleteNoDaemonPod(t *testing.T) {
 	}
 }
 
-func TestDeletePodForNotExistingNode(t *testing.T) {
+func TestDeleteUnscheduledPodForNotExistingNode(t *testing.T) {
 	for _, f := range []bool{true, false} {
 		defer utilfeaturetesting.SetFeatureGateDuringTest(t, utilfeature.DefaultFeatureGate, features.ScheduleDaemonSetPods, f)()
 		for _, strategy := range updateStrategies() {
@@ -2461,6 +2461,26 @@ func TestDeletePodForNotExistingNode(t *testing.T) {
 			addNodes(manager.nodeStore, 0, 1, nil)
 			addPods(manager.podStore, "node-0", simpleDaemonSetLabel, ds, 1)
 			addPods(manager.podStore, "node-1", simpleDaemonSetLabel, ds, 1)
+
+			podScheduledUsingAffinity := newPod("pod1-node-3", "", simpleDaemonSetLabel, ds)
+			podScheduledUsingAffinity.Spec.Affinity = &v1.Affinity{
+				NodeAffinity: &v1.NodeAffinity{
+					RequiredDuringSchedulingIgnoredDuringExecution: &v1.NodeSelector{
+						NodeSelectorTerms: []v1.NodeSelectorTerm{
+							{
+								MatchFields: []v1.NodeSelectorRequirement{
+									{
+										Key:      schedulerapi.NodeFieldSelectorKeyNodeName,
+										Operator: v1.NodeSelectorOpIn,
+										Values:   []string{"node-2"},
+									},
+								},
+							},
+						},
+					},
+				},
+			}
+			manager.podStore.Add(podScheduledUsingAffinity)
 			if f {
 				syncAndValidateDaemonSets(t, manager, ds, podControl, 0, 1, 0)
 			} else {