Merge pull request #72259 from bsalamat/fix_nominated_node

pkg/scheduler/core/generic_scheduler.go

@@ -362,7 +362,7 @@ func (g *genericScheduler) processPreemptionWithExtenders(
 // worth the complexity, especially because we generally expect to have a very
 // small number of nominated pods per node.
 func (g *genericScheduler) getLowerPriorityNominatedPods(pod *v1.Pod, nodeName string) []*v1.Pod {
-	pods := g.schedulingQueue.WaitingPodsForNode(nodeName)
+	pods := g.schedulingQueue.NominatedPodsForNode(nodeName)
 
 	if len(pods) == 0 {
 		return nil
@@ -509,7 +509,7 @@ func addNominatedPods(pod *v1.Pod, meta predicates.PredicateMetadata,
 		// This may happen only in tests.
 		return false, meta, nodeInfo
 	}
-	nominatedPods := queue.WaitingPodsForNode(nodeInfo.Node().Name)
+	nominatedPods := queue.NominatedPodsForNode(nodeInfo.Node().Name)
 	if nominatedPods == nil || len(nominatedPods) == 0 {
 		return false, meta, nodeInfo
 	}

pkg/scheduler/internal/queue/scheduling_queue_test.go

@@ -105,8 +105,14 @@ func TestPriorityQueue_Add(t *testing.T) {
 	if err := q.Add(&highPriorityPod); err != nil {
 		t.Errorf("add failed: %v", err)
 	}
-	expectedNominatedPods := map[string][]*v1.Pod{
+	expectedNominatedPods := &nominatedPodMap{
+		nominatedPodToNode: map[types.UID]string{
+			medPriorityPod.UID:   "node1",
+			unschedulablePod.UID: "node1",
+		},
+		nominatedPods: map[string][]*v1.Pod{
 			"node1": {&medPriorityPod, &unschedulablePod},
+		},
 	}
 	if !reflect.DeepEqual(q.nominatedPods, expectedNominatedPods) {
 		t.Errorf("Unexpected nominated map after adding pods. Expected: %v, got: %v", expectedNominatedPods, q.nominatedPods)
@@ -120,8 +126,8 @@ func TestPriorityQueue_Add(t *testing.T) {
 	if p, err := q.Pop(); err != nil || p != &unschedulablePod {
 		t.Errorf("Expected: %v after Pop, but got: %v", unschedulablePod.Name, p.Name)
 	}
-	if len(q.nominatedPods["node1"]) != 2 {
-		t.Errorf("Expected medPriorityPod and unschedulablePod to be still present in nomindatePods: %v", q.nominatedPods["node1"])
+	if len(q.nominatedPods.nominatedPods["node1"]) != 2 {
+		t.Errorf("Expected medPriorityPod and unschedulablePod to be still present in nomindatePods: %v", q.nominatedPods.nominatedPods["node1"])
 	}
 }
 
@@ -131,8 +137,14 @@ func TestPriorityQueue_AddIfNotPresent(t *testing.T) {
 	q.AddIfNotPresent(&highPriNominatedPod) // Must not add anything.
 	q.AddIfNotPresent(&medPriorityPod)
 	q.AddIfNotPresent(&unschedulablePod)
-	expectedNominatedPods := map[string][]*v1.Pod{
+	expectedNominatedPods := &nominatedPodMap{
+		nominatedPodToNode: map[types.UID]string{
+			medPriorityPod.UID:   "node1",
+			unschedulablePod.UID: "node1",
+		},
+		nominatedPods: map[string][]*v1.Pod{
 			"node1": {&medPriorityPod, &unschedulablePod},
+		},
 	}
 	if !reflect.DeepEqual(q.nominatedPods, expectedNominatedPods) {
 		t.Errorf("Unexpected nominated map after adding pods. Expected: %v, got: %v", expectedNominatedPods, q.nominatedPods)
@@ -143,8 +155,8 @@ func TestPriorityQueue_AddIfNotPresent(t *testing.T) {
 	if p, err := q.Pop(); err != nil || p != &unschedulablePod {
 		t.Errorf("Expected: %v after Pop, but got: %v", unschedulablePod.Name, p.Name)
 	}
-	if len(q.nominatedPods["node1"]) != 2 {
-		t.Errorf("Expected medPriorityPod and unschedulablePod to be still present in nomindatePods: %v", q.nominatedPods["node1"])
+	if len(q.nominatedPods.nominatedPods["node1"]) != 2 {
+		t.Errorf("Expected medPriorityPod and unschedulablePod to be still present in nomindatePods: %v", q.nominatedPods.nominatedPods["node1"])
 	}
 	if q.unschedulableQ.get(&highPriNominatedPod) != &highPriNominatedPod {
 		t.Errorf("Pod %v was not found in the unschedulableQ.", highPriNominatedPod.Name)
@@ -157,8 +169,15 @@ func TestPriorityQueue_AddUnschedulableIfNotPresent(t *testing.T) {
 	q.AddUnschedulableIfNotPresent(&highPriNominatedPod) // Must not add anything.
 	q.AddUnschedulableIfNotPresent(&medPriorityPod)      // This should go to activeQ.
 	q.AddUnschedulableIfNotPresent(&unschedulablePod)
-	expectedNominatedPods := map[string][]*v1.Pod{
+	expectedNominatedPods := &nominatedPodMap{
+		nominatedPodToNode: map[types.UID]string{
+			medPriorityPod.UID:      "node1",
+			unschedulablePod.UID:    "node1",
+			highPriNominatedPod.UID: "node1",
+		},
+		nominatedPods: map[string][]*v1.Pod{
 			"node1": {&highPriNominatedPod, &medPriorityPod, &unschedulablePod},
+		},
 	}
 	if !reflect.DeepEqual(q.nominatedPods, expectedNominatedPods) {
 		t.Errorf("Unexpected nominated map after adding pods. Expected: %v, got: %v", expectedNominatedPods, q.nominatedPods)
@@ -169,7 +188,7 @@ func TestPriorityQueue_AddUnschedulableIfNotPresent(t *testing.T) {
 	if p, err := q.Pop(); err != nil || p != &medPriorityPod {
 		t.Errorf("Expected: %v after Pop, but got: %v", medPriorityPod.Name, p.Name)
 	}
-	if len(q.nominatedPods) != 1 {
+	if len(q.nominatedPods.nominatedPods) != 1 {
 		t.Errorf("Expected nomindatePods to have one element: %v", q.nominatedPods)
 	}
 	if q.unschedulableQ.get(&unschedulablePod) != &unschedulablePod {
@@ -186,8 +205,8 @@ func TestPriorityQueue_Pop(t *testing.T) {
 		if p, err := q.Pop(); err != nil || p != &medPriorityPod {
 			t.Errorf("Expected: %v after Pop, but got: %v", medPriorityPod.Name, p.Name)
 		}
-		if len(q.nominatedPods["node1"]) != 1 {
-			t.Errorf("Expected medPriorityPod to be present in nomindatePods: %v", q.nominatedPods["node1"])
+		if len(q.nominatedPods.nominatedPods["node1"]) != 1 {
+			t.Errorf("Expected medPriorityPod to be present in nomindatePods: %v", q.nominatedPods.nominatedPods["node1"])
 		}
 	}()
 	q.Add(&medPriorityPod)
@@ -200,7 +219,7 @@ func TestPriorityQueue_Update(t *testing.T) {
 	if _, exists, _ := q.activeQ.Get(&highPriorityPod); !exists {
 		t.Errorf("Expected %v to be added to activeQ.", highPriorityPod.Name)
 	}
-	if len(q.nominatedPods) != 0 {
+	if len(q.nominatedPods.nominatedPods) != 0 {
 		t.Errorf("Expected nomindatePods to be empty: %v", q.nominatedPods)
 	}
 	// Update highPriorityPod and add a nominatedNodeName to it.
@@ -208,7 +227,7 @@ func TestPriorityQueue_Update(t *testing.T) {
 	if q.activeQ.Len() != 1 {
 		t.Error("Expected only one item in activeQ.")
 	}
-	if len(q.nominatedPods) != 1 {
+	if len(q.nominatedPods.nominatedPods) != 1 {
 		t.Errorf("Expected one item in nomindatePods map: %v", q.nominatedPods)
 	}
 	// Updating an unschedulable pod which is not in any of the two queues, should
@@ -243,13 +262,13 @@ func TestPriorityQueue_Delete(t *testing.T) {
 	if _, exists, _ := q.activeQ.Get(&highPriNominatedPod); exists {
 		t.Errorf("Didn't expect %v to be in activeQ.", highPriorityPod.Name)
 	}
-	if len(q.nominatedPods) != 1 {
-		t.Errorf("Expected nomindatePods to have only 'unschedulablePod': %v", q.nominatedPods)
+	if len(q.nominatedPods.nominatedPods) != 1 {
+		t.Errorf("Expected nomindatePods to have only 'unschedulablePod': %v", q.nominatedPods.nominatedPods)
 	}
 	if err := q.Delete(&unschedulablePod); err != nil {
 		t.Errorf("delete failed: %v", err)
 	}
-	if len(q.nominatedPods) != 0 {
+	if len(q.nominatedPods.nominatedPods) != 0 {
 		t.Errorf("Expected nomindatePods to be empty: %v", q.nominatedPods)
 	}
 }
@@ -321,7 +340,7 @@ func TestPriorityQueue_AssignedPodAdded(t *testing.T) {
 	}
 }
 
-func TestPriorityQueue_WaitingPodsForNode(t *testing.T) {
+func TestPriorityQueue_NominatedPodsForNode(t *testing.T) {
 	q := NewPriorityQueue(nil)
 	q.Add(&medPriorityPod)
 	q.Add(&unschedulablePod)
@@ -330,10 +349,10 @@ func TestPriorityQueue_WaitingPodsForNode(t *testing.T) {
 		t.Errorf("Expected: %v after Pop, but got: %v", highPriorityPod.Name, p.Name)
 	}
 	expectedList := []*v1.Pod{&medPriorityPod, &unschedulablePod}
-	if !reflect.DeepEqual(expectedList, q.WaitingPodsForNode("node1")) {
+	if !reflect.DeepEqual(expectedList, q.NominatedPodsForNode("node1")) {
 		t.Error("Unexpected list of nominated Pods for node.")
 	}
-	if q.WaitingPodsForNode("node2") != nil {
+	if q.NominatedPodsForNode("node2") != nil {
 		t.Error("Expected list of nominated Pods for node2 to be empty.")
 	}
 }
@@ -354,6 +373,75 @@ func TestPriorityQueue_PendingPods(t *testing.T) {
 	}
 }
 
+func TestPriorityQueue_UpdateNominatedPodForNode(t *testing.T) {
+	q := NewPriorityQueue(nil)
+	if err := q.Add(&medPriorityPod); err != nil {
+		t.Errorf("add failed: %v", err)
+	}
+	// Update unschedulablePod on a different node than specified in the pod.
+	q.UpdateNominatedPodForNode(&unschedulablePod, "node5")
+
+	// Update nominated node name of a pod on a node that is not specified in the pod object.
+	q.UpdateNominatedPodForNode(&highPriorityPod, "node2")
+	expectedNominatedPods := &nominatedPodMap{
+		nominatedPodToNode: map[types.UID]string{
+			medPriorityPod.UID:   "node1",
+			highPriorityPod.UID:  "node2",
+			unschedulablePod.UID: "node5",
+		},
+		nominatedPods: map[string][]*v1.Pod{
+			"node1": {&medPriorityPod},
+			"node2": {&highPriorityPod},
+			"node5": {&unschedulablePod},
+		},
+	}
+	if !reflect.DeepEqual(q.nominatedPods, expectedNominatedPods) {
+		t.Errorf("Unexpected nominated map after adding pods. Expected: %v, got: %v", expectedNominatedPods, q.nominatedPods)
+	}
+	if p, err := q.Pop(); err != nil || p != &medPriorityPod {
+		t.Errorf("Expected: %v after Pop, but got: %v", medPriorityPod.Name, p.Name)
+	}
+	// List of nominated pods shouldn't change after popping them from the queue.
+	if !reflect.DeepEqual(q.nominatedPods, expectedNominatedPods) {
+		t.Errorf("Unexpected nominated map after popping pods. Expected: %v, got: %v", expectedNominatedPods, q.nominatedPods)
+	}
+	// Update one of the nominated pods that doesn't have nominatedNodeName in the
+	// pod object. It should be updated correctly.
+	q.UpdateNominatedPodForNode(&highPriorityPod, "node4")
+	expectedNominatedPods = &nominatedPodMap{
+		nominatedPodToNode: map[types.UID]string{
+			medPriorityPod.UID:   "node1",
+			highPriorityPod.UID:  "node4",
+			unschedulablePod.UID: "node5",
+		},
+		nominatedPods: map[string][]*v1.Pod{
+			"node1": {&medPriorityPod},
+			"node4": {&highPriorityPod},
+			"node5": {&unschedulablePod},
+		},
+	}
+	if !reflect.DeepEqual(q.nominatedPods, expectedNominatedPods) {
+		t.Errorf("Unexpected nominated map after updating pods. Expected: %v, got: %v", expectedNominatedPods, q.nominatedPods)
+	}
+
+	// Delete a nominated pod that doesn't have nominatedNodeName in the pod
+	// object. It should be deleted.
+	q.DeleteNominatedPodIfExists(&highPriorityPod)
+	expectedNominatedPods = &nominatedPodMap{
+		nominatedPodToNode: map[types.UID]string{
+			medPriorityPod.UID:   "node1",
+			unschedulablePod.UID: "node5",
+		},
+		nominatedPods: map[string][]*v1.Pod{
+			"node1": {&medPriorityPod},
+			"node5": {&unschedulablePod},
+		},
+	}
+	if !reflect.DeepEqual(q.nominatedPods, expectedNominatedPods) {
+		t.Errorf("Unexpected nominated map after deleting pods. Expected: %v, got: %v", expectedNominatedPods, q.nominatedPods)
+	}
+}
+
 func TestUnschedulablePodsMap(t *testing.T) {
 	var pods = []*v1.Pod{
 		{

pkg/scheduler/scheduler.go

@@ -318,11 +318,19 @@ func (sched *Scheduler) preempt(preemptor *v1.Pod, scheduleErr error) (string, e
 	var nodeName = ""
 	if node != nil {
 		nodeName = node.Name
+		// Update the scheduling queue with the nominated pod information. Without
+		// this, there would be a race condition between the next scheduling cycle
+		// and the time the scheduler receives a Pod Update for the nominated pod.
+		sched.config.SchedulingQueue.UpdateNominatedPodForNode(preemptor, nodeName)
+
+		// Make a call to update nominated node name of the pod on the API server.
 		err = sched.config.PodPreemptor.SetNominatedNodeName(preemptor, nodeName)
 		if err != nil {
 			klog.Errorf("Error in preemption process. Cannot update pod %v/%v annotations: %v", preemptor.Namespace, preemptor.Name, err)
+			sched.config.SchedulingQueue.DeleteNominatedPodIfExists(preemptor)
 			return "", err
 		}
+
 		for _, victim := range victims {
 			if err := sched.config.PodPreemptor.DeletePod(victim); err != nil {
 				klog.Errorf("Error preempting pod %v/%v: %v", victim.Namespace, victim.Name, err)

test/integration/scheduler/BUILD

@@ -26,6 +26,7 @@ go_test(
         "//cmd/kube-scheduler/app:go_default_library",
         "//cmd/kube-scheduler/app/config:go_default_library",
         "//pkg/api/legacyscheme:go_default_library",
+        "//pkg/api/v1/pod:go_default_library",
         "//pkg/controller/nodelifecycle:go_default_library",
         "//pkg/controller/volume/persistentvolume:go_default_library",
         "//pkg/controller/volume/persistentvolume/options:go_default_library",

test/integration/scheduler/util.go

@@ -584,6 +584,17 @@ func podScheduled(c clientset.Interface, podNamespace, podName string) wait.Cond
 
 // podUnschedulable returns a condition function that returns true if the given pod
 // gets unschedulable status.
+func podSchedulableCondition(c clientset.Interface, podNamespace, podName string) (*v1.PodCondition, error) {
+	pod, err := c.CoreV1().Pods(podNamespace).Get(podName, metav1.GetOptions{})
+	if err != nil {
+		return nil, err
+	}
+	_, cond := podutil.GetPodCondition(&pod.Status, v1.PodScheduled)
+	return cond, nil
+}
+
+// podUnschedulable returns a condition function that returns true if the given pod
+// gets unschedulable status.
 func podUnschedulable(c clientset.Interface, podNamespace, podName string) wait.ConditionFunc {
 	return func() (bool, error) {
 		pod, err := c.CoreV1().Pods(podNamespace).Get(podName, metav1.GetOptions{})
@@ -710,7 +721,7 @@ func cleanupPods(cs clientset.Interface, t *testing.T, pods []*v1.Pod) {
 		}
 	}
 	for _, p := range pods {
-		if err := wait.Poll(time.Second, wait.ForeverTestTimeout,
+		if err := wait.Poll(time.Millisecond, wait.ForeverTestTimeout,
 			podDeleted(cs, p.Namespace, p.Name)); err != nil {
 			t.Errorf("error while waiting for pod  %v/%v to get deleted: %v", p.Namespace, p.Name, err)
 		}