feat: evict pod only called api successed for cpu eviction

Signed-off-by: j4ckstraw <j4ckstraw@foxmail.com>

pkg/koordlet/qosmanager/plugins/cpuevict/cpu_evict.go

@@ -58,6 +58,7 @@ type cpuEvictor struct {
 	metricCache           metriccache.MetricCache
 	evictor               *framework.Evictor
 	lastEvictTime         time.Time
+	onlyEvictByAPI        bool
 }
 
 func New(opt *framework.Options) framework.QOSStrategy {
@@ -68,6 +69,7 @@ func New(opt *framework.Options) framework.QOSStrategy {
 		statesInformer:        opt.StatesInformer,
 		metricCache:           opt.MetricCache,
 		lastEvictTime:         time.Now(),
+		onlyEvictByAPI:        opt.Config.OnlyEvictByAPI,
 	}
 }
 
@@ -292,24 +294,31 @@ func (c *cpuEvictor) killAndEvictBEPodsRelease(node *corev1.Node, bePodInfos []*
 		node.Name, cpuNeedMilliRelease)
 
 	cpuMilliReleased := int64(0)
-	var killedPods []*corev1.Pod
+	hasKillPods := false
 	for _, bePod := range bePodInfos {
 		if cpuMilliReleased >= cpuNeedMilliRelease {
 			break
 		}
 
-		podKillMsg := fmt.Sprintf("%s, kill pod: %s", message, util.GetPodKey(bePod.pod))
-		helpers.KillContainers(bePod.pod, podKillMsg)
-
-		killedPods = append(killedPods, bePod.pod)
-		cpuMilliReleased = cpuMilliReleased + bePod.milliRequest
+		if c.onlyEvictByAPI {
+			if c.evictor.EvictPodIfNotEvicted(bePod.pod, node, resourceexecutor.EvictPodByBECPUSatisfaction, message) {
+				cpuMilliReleased = cpuMilliReleased + bePod.milliRequest
+				klog.V(5).Infof("cpuEvict pick pod %s to evict", util.GetPodKey(bePod.pod))
+				hasKillPods = true
+			} else {
+				klog.V(5).Infof("cpuEvict pick pod %s to evict, failed", util.GetPodKey(bePod.pod))
+			}
+		} else {
+			podKillMsg := fmt.Sprintf("%s, kill pod: %s", message, util.GetPodKey(bePod.pod))
+			helpers.KillContainers(bePod.pod, podKillMsg)
 
-		klog.V(5).Infof("cpuEvict pick pod %s/%s to evict", util.GetPodKey(bePod.pod))
+			cpuMilliReleased = cpuMilliReleased + bePod.milliRequest
+			klog.V(5).Infof("cpuEvict pick pod %s to evict", util.GetPodKey(bePod.pod))
+			hasKillPods = true
+		}
 	}
 
-	c.evictor.EvictPodsIfNotEvicted(killedPods, node, resourceexecutor.EvictPodByBECPUSatisfaction, message)
-
-	if len(killedPods) > 0 {
+	if hasKillPods {
 		c.lastEvictTime = time.Now()
 	}
 	klog.V(5).Infof("killAndEvictBEPodsRelease finished! cpuNeedMilliRelease(%d) cpuMilliReleased(%d)",

pkg/koordlet/qosmanager/plugins/cpuevict/cpu_evict_test.go

@@ -574,6 +574,65 @@ func Test_killAndEvictBEPodsRelease(t *testing.T) {
 
 	cpuEvictor.killAndEvictBEPodsRelease(node, podEvictInfosSorted, 18*1000)
 
+	getNotEvictObject, err := client.Tracker().Get(testutil.PodsResource, podEvictInfosSorted[2].pod.Namespace, podEvictInfosSorted[2].pod.Name)
+	assert.Nil(t, err, "get pod fail", err)
+	assert.IsType(t, &corev1.Pod{}, getNotEvictObject, "no need evict", podEvictInfosSorted[2].pod.Name)
+	assert.True(t, cpuEvictor.lastEvictTime.After(time.Now().Add(-5*time.Second)), "checkLastTime")
+}
+
+func Test_killAndEvictBEPodsReleaseWithEvictionAPIOnly(t *testing.T) {
+	podEvictInfosSorted := []*podEvictCPUInfo{
+		{
+			pod:          mockBEPodForCPUEvict("pod_be_3_priority10", 16*1000, 10),
+			milliRequest: 16 * 1000,
+		},
+		{
+			pod:          mockBEPodForCPUEvict("pod_be_2_priority100", 16*1000, 100),
+			milliRequest: 16 * 1000,
+		},
+		{
+			pod:          mockBEPodForCPUEvict("pod_be_1_priority100", 16*1000, 100),
+			milliRequest: 16 * 1000,
+		},
+	}
+
+	ctl := gomock.NewController(t)
+	defer ctl.Finish()
+
+	fakeRecorder := &testutil.FakeRecorder{}
+	client := clientsetfake.NewSimpleClientset()
+
+	stop := make(chan struct{})
+	evictor := framework.NewEvictor(client, fakeRecorder, policyv1beta1.SchemeGroupVersion.Version)
+	evictor.Start(stop)
+	defer func() { stop <- struct{}{} }()
+
+	node := testutil.MockTestNode("100", "500G")
+	runtime.DockerHandler = handler.NewFakeRuntimeHandler()
+	// create pod
+	var containers []*critesting.FakeContainer
+	for _, podInfo := range podEvictInfosSorted {
+		pod := podInfo.pod
+		_, _ = client.CoreV1().Pods(pod.Namespace).Create(context.TODO(), pod, metav1.CreateOptions{})
+		for _, containerStatus := range pod.Status.ContainerStatuses {
+			_, containerId, _ := util.ParseContainerId(containerStatus.ContainerID)
+			fakeContainer := &critesting.FakeContainer{
+				SandboxID:       string(pod.UID),
+				ContainerStatus: runtimeapi.ContainerStatus{Id: containerId},
+			}
+			containers = append(containers, fakeContainer)
+		}
+	}
+	runtime.DockerHandler.(*handler.FakeRuntimeHandler).SetFakeContainers(containers)
+
+	cpuEvictor := &cpuEvictor{
+		evictor:        evictor,
+		onlyEvictByAPI: true,
+		lastEvictTime:  time.Now().Add(-5 * time.Minute),
+	}
+
+	cpuEvictor.killAndEvictBEPodsRelease(node, podEvictInfosSorted, 18*1000)
+
 	getEvictObject, err := client.Tracker().Get(testutil.PodsResource, podEvictInfosSorted[0].pod.Namespace, podEvictInfosSorted[0].pod.Name)
 	assert.NotNil(t, getEvictObject, "evictPod Fail, err: %v", err)
 	assert.IsType(t, &policyv1beta1.Eviction{}, getEvictObject, "evictPod: %s Fail", podEvictInfosSorted[0].pod.Name)