Add integration test for disable preemption

test/integration/scheduler/preemption_test.go

@@ -41,8 +41,8 @@ import (
 
 var lowPriority, mediumPriority, highPriority = int32(100), int32(200), int32(300)
 
-func waitForNominatedNodeName(cs clientset.Interface, pod *v1.Pod) error {
-	if err := wait.Poll(100*time.Millisecond, wait.ForeverTestTimeout, func() (bool, error) {
+func waitForNominatedNodeNameWithTimeout(cs clientset.Interface, pod *v1.Pod, timeout time.Duration) error {
+	if err := wait.Poll(100*time.Millisecond, timeout, func() (bool, error) {
 		pod, err := cs.CoreV1().Pods(pod.Namespace).Get(pod.Name, metav1.GetOptions{})
 		if err != nil {
 			return false, err
@@ -57,6 +57,10 @@ func waitForNominatedNodeName(cs clientset.Interface, pod *v1.Pod) error {
 	return nil
 }
 
+func waitForNominatedNodeName(cs clientset.Interface, pod *v1.Pod) error {
+	return waitForNominatedNodeNameWithTimeout(cs, pod, wait.ForeverTestTimeout)
+}
+
 // TestPreemption tests a few preemption scenarios.
 func TestPreemption(t *testing.T) {
 	// Enable PodPriority feature gate.
@@ -285,6 +289,88 @@ func TestPreemption(t *testing.T) {
 	}
 }
 
+// TestDisablePreemption tests disable pod preemption of scheduler works as expected.
+func TestDisablePreemption(t *testing.T) {
+	// Enable PodPriority feature gate.
+	utilfeature.DefaultFeatureGate.Set(fmt.Sprintf("%s=true", features.PodPriority))
+	// Initialize scheduler, and disable preemption.
+	context := initTestDisablePreemption(t, "disable-preemption")
+	defer cleanupTest(t, context)
+	cs := context.clientSet
+
+	tests := []struct {
+		description  string
+		existingPods []*v1.Pod
+		pod          *v1.Pod
+	}{
+		{
+			description: "pod preemption will not happen",
+			existingPods: []*v1.Pod{
+				initPausePod(context.clientSet, &pausePodConfig{
+					Name:      "victim-pod",
+					Namespace: context.ns.Name,
+					Priority:  &lowPriority,
+					Resources: &v1.ResourceRequirements{Requests: v1.ResourceList{
+						v1.ResourceCPU:    *resource.NewMilliQuantity(400, resource.DecimalSI),
+						v1.ResourceMemory: *resource.NewQuantity(200, resource.BinarySI)},
+					},
+				}),
+			},
+			pod: initPausePod(cs, &pausePodConfig{
+				Name:      "preemptor-pod",
+				Namespace: context.ns.Name,
+				Priority:  &highPriority,
+				Resources: &v1.ResourceRequirements{Requests: v1.ResourceList{
+					v1.ResourceCPU:    *resource.NewMilliQuantity(300, resource.DecimalSI),
+					v1.ResourceMemory: *resource.NewQuantity(200, resource.BinarySI)},
+				},
+			}),
+		},
+	}
+
+	// Create a node with some resources and a label.
+	nodeRes := &v1.ResourceList{
+		v1.ResourcePods:   *resource.NewQuantity(32, resource.DecimalSI),
+		v1.ResourceCPU:    *resource.NewMilliQuantity(500, resource.DecimalSI),
+		v1.ResourceMemory: *resource.NewQuantity(500, resource.BinarySI),
+	}
+	_, err := createNode(context.clientSet, "node1", nodeRes)
+	if err != nil {
+		t.Fatalf("Error creating nodes: %v", err)
+	}
+
+	for _, test := range tests {
+		pods := make([]*v1.Pod, len(test.existingPods))
+		// Create and run existingPods.
+		for i, p := range test.existingPods {
+			pods[i], err = runPausePod(cs, p)
+			if err != nil {
+				t.Fatalf("Test [%v]: Error running pause pod: %v", test.description, err)
+			}
+		}
+		// Create the "pod".
+		preemptor, err := createPausePod(cs, test.pod)
+		if err != nil {
+			t.Errorf("Error while creating high priority pod: %v", err)
+		}
+		// Ensure preemptor should keep unschedulable.
+		if err := waitForPodUnschedulable(cs, preemptor); err != nil {
+			t.Errorf("Test [%v]: Preemptor %v should not become scheduled",
+				test.description, preemptor.Name)
+		}
+
+		// Ensure preemptor should not be nominated.
+		if err := waitForNominatedNodeNameWithTimeout(cs, preemptor, 5*time.Second); err == nil {
+			t.Errorf("Test [%v]: Preemptor %v should not be nominated",
+				test.description, preemptor.Name)
+		}
+
+		// Cleanup
+		pods = append(pods, preemptor)
+		cleanupPods(cs, t, pods)
+	}
+}
+
 func mkPriorityPodWithGrace(tc *TestContext, name string, priority int32, grace int64) *v1.Pod {
 	defaultPodRes := &v1.ResourceRequirements{Requests: v1.ResourceList{
 		v1.ResourceCPU:    *resource.NewMilliQuantity(100, resource.DecimalSI),

test/integration/scheduler/util.go

@@ -63,7 +63,7 @@ type TestContext struct {
 	scheduler              *scheduler.Scheduler
 }
 
-// createConfiguratorWithPodInformer create a configurator for scheduler with given informer factory, custom name and  pod informer.
+// createConfiguratorWithPodInformer creates a configurator for scheduler.
 func createConfiguratorWithPodInformer(
 	schedulerName string,
 	clientSet clientset.Interface,
@@ -116,7 +116,14 @@ func initTestMaster(t *testing.T, nsPrefix string, admission admission.Interface
 	}
 
 	// 2. Create kubeclient
-	context.clientSet = clientset.NewForConfigOrDie(&restclient.Config{QPS: -1, Host: s.URL, ContentConfig: restclient.ContentConfig{GroupVersion: testapi.Groups[v1.GroupName].GroupVersion()}})
+	context.clientSet = clientset.NewForConfigOrDie(
+		&restclient.Config{
+			QPS: -1, Host: s.URL,
+			ContentConfig: restclient.ContentConfig{
+				GroupVersion: testapi.Groups[v1.GroupName].GroupVersion(),
+			},
+		},
+	)
 	return &context
 }
 
@@ -129,6 +136,21 @@ func initTestScheduler(
 	setPodInformer bool,
 	policy *schedulerapi.Policy,
 ) *TestContext {
+	// Pod preemption is enabled by default scheduler configuration, but preemption only happens when PodPriority
+	// feature gate is enabled at the same time.
+	return initTestSchedulerWithOptions(t, context, controllerCh, setPodInformer, policy, false)
+}
+
+// initTestSchedulerWithOptions initializes a test environment and creates a scheduler with default
+// configuration and other options.
+func initTestSchedulerWithOptions(
+	t *testing.T,
+	context *TestContext,
+	controllerCh chan struct{},
+	setPodInformer bool,
+	policy *schedulerapi.Policy,
+	disablePreemption bool,
+) *TestContext {
 	// Enable EnableEquivalenceClassCache for all integration tests.
 	defer utilfeaturetesting.SetFeatureGateDuringTest(
 		t,
@@ -167,19 +189,29 @@ func initTestScheduler(
 		context.schedulerConfig.StopEverything = controllerCh
 	}
 
+	// set DisablePreemption option
+	context.schedulerConfig.DisablePreemption = disablePreemption
+
 	// set setPodInformer if provided.
 	if setPodInformer {
 		go podInformer.Informer().Run(context.schedulerConfig.StopEverything)
 	}
 
 	eventBroadcaster := record.NewBroadcaster()
-	context.schedulerConfig.Recorder = eventBroadcaster.NewRecorder(legacyscheme.Scheme, v1.EventSource{Component: v1.DefaultSchedulerName})
-	eventBroadcaster.StartRecordingToSink(&clientv1core.EventSinkImpl{Interface: context.clientSet.CoreV1().Events("")})
+	context.schedulerConfig.Recorder = eventBroadcaster.NewRecorder(
+		legacyscheme.Scheme,
+		v1.EventSource{Component: v1.DefaultSchedulerName},
+	)
+	eventBroadcaster.StartRecordingToSink(&clientv1core.EventSinkImpl{
+		Interface: context.clientSet.CoreV1().Events(""),
+	})
 
 	context.informerFactory.Start(context.schedulerConfig.StopEverything)
 	context.informerFactory.WaitForCacheSync(context.schedulerConfig.StopEverything)
 
-	context.scheduler, err = scheduler.NewFromConfigurator(&scheduler.FakeConfigurator{Config: context.schedulerConfig}, nil...)
+	context.scheduler, err = scheduler.NewFromConfigurator(&scheduler.FakeConfigurator{
+		Config: context.schedulerConfig},
+		nil...)
 	if err != nil {
 		t.Fatalf("Couldn't create scheduler: %v", err)
 	}
@@ -193,6 +225,13 @@ func initTest(t *testing.T, nsPrefix string) *TestContext {
 	return initTestScheduler(t, initTestMaster(t, nsPrefix, nil), nil, true, nil)
 }
 
+// initTestDisablePreemption initializes a test environment and creates master and scheduler with default
+// configuration but with pod preemption disabled.
+func initTestDisablePreemption(t *testing.T, nsPrefix string) *TestContext {
+	return initTestSchedulerWithOptions(
+		t, initTestMaster(t, nsPrefix, nil), nil, true, nil, true)
+}
+
 // cleanupTest deletes the scheduler and the test namespace. It should be called
 // at the end of a test.
 func cleanupTest(t *testing.T, context *TestContext) {
@@ -206,7 +245,8 @@ func cleanupTest(t *testing.T, context *TestContext) {
 
 // waitForReflection waits till the passFunc confirms that the object it expects
 // to see is in the store. Used to observe reflected events.
-func waitForReflection(t *testing.T, nodeLister corelisters.NodeLister, key string, passFunc func(n interface{}) bool) error {
+func waitForReflection(t *testing.T, nodeLister corelisters.NodeLister, key string,
+	passFunc func(n interface{}) bool) error {
 	nodes := []*v1.Node{}
 	err := wait.Poll(time.Millisecond*100, wait.ForeverTestTimeout, func() (bool, error) {
 		n, err := nodeLister.Get(key)
@@ -345,7 +385,8 @@ func createPausePod(cs clientset.Interface, p *v1.Pod) (*v1.Pod, error) {
 // createPausePodWithResource creates a pod with "Pause" image and the given
 // resources and returns its pointer and error status. The resource list can be
 // nil.
-func createPausePodWithResource(cs clientset.Interface, podName string, nsName string, res *v1.ResourceList) (*v1.Pod, error) {
+func createPausePodWithResource(cs clientset.Interface, podName string,
+	nsName string, res *v1.ResourceList) (*v1.Pod, error) {
 	var conf pausePodConfig
 	if res == nil {
 		conf = pausePodConfig{
@@ -439,7 +480,8 @@ func podUnschedulable(c clientset.Interface, podNamespace, podName string) wait.
 			return false, nil
 		}
 		_, cond := podutil.GetPodCondition(&pod.Status, v1.PodScheduled)
-		return cond != nil && cond.Status == v1.ConditionFalse && cond.Reason == v1.PodReasonUnschedulable, nil
+		return cond != nil && cond.Status == v1.ConditionFalse &&
+			cond.Reason == v1.PodReasonUnschedulable, nil
 	}
 }
 
@@ -481,7 +523,8 @@ func cleanupPods(cs clientset.Interface, t *testing.T, pods []*v1.Pod) {
 		}
 	}
 	for _, p := range pods {
-		if err := wait.Poll(time.Second, wait.ForeverTestTimeout, podDeleted(cs, p.Namespace, p.Name)); err != nil {
+		if err := wait.Poll(time.Second, 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)
 		}
 	}