fix: flake integration test

sanposhiho · sanposhiho · commit 6d11c3646d7d · 2025-07-28T18:59:34.000+09:00
diff --git a/pkg/scheduler/framework/preemption/preemption.go b/pkg/scheduler/framework/preemption/preemption.go
@@ -496,10 +496,20 @@ func (ev *Evaluator) prepareCandidateAsync(c Candidate, pod *v1.Pod, pluginName
 	}
 
 	errCh := parallelize.NewErrorChannel()
+	// Whether all victim pods are already deleted before making API call.
+	allPodsAlreadyDeleted := true
 	preemptPod := func(index int) {
 		victim := victimPods[index]
-		if err := ev.PreemptPod(ctx, c, pod, victim, pluginName); err != nil {
+		err := ev.PreemptPod(ctx, c, pod, victim, pluginName)
+		switch {
+		case apierrors.IsNotFound(err):
+			logger.V(2).Info("Victim Pod is already deleted", "preemptor", klog.KObj(pod), "node", c.Name(), "err", err)
+		case err != nil:
 			errCh.SendErrorWithCancel(err, cancel)
+		default:
+			ev.mu.Lock()
+			allPodsAlreadyDeleted = false
+			ev.mu.Unlock()
 		}
 	}
 
@@ -511,11 +521,11 @@ func (ev *Evaluator) prepareCandidateAsync(c Candidate, pod *v1.Pod, pluginName
 		startTime := time.Now()
 		result := metrics.GoroutineResultSuccess
 
-		// Whether all victim pods are already deleted before making API call.
-		allPodsAlreadyDeleted := true
 		defer metrics.PreemptionGoroutinesDuration.WithLabelValues(result).Observe(metrics.SinceInSeconds(startTime))
 		defer metrics.PreemptionGoroutinesExecutionTotal.WithLabelValues(result).Inc()
 		defer func() {
+			ev.mu.Lock()
+			defer ev.mu.Unlock()
 			// When API call isn't successful, the Pod may get stuck in the unschedulable pod pool in the worst case.
 			// So, we should move the Pod to the activeQ.
 			if result == metrics.GoroutineResultError ||
@@ -547,15 +557,9 @@ func (ev *Evaluator) prepareCandidateAsync(c Candidate, pod *v1.Pod, pluginName
 			// and the pod could end up stucking at the unschedulable pod pool
 			// by all the pod removal events being ignored.
 			ev.Handler.Parallelizer().Until(ctx, len(victimPods)-1, preemptPod, ev.PluginName)
-			err := errCh.ReceiveError()
-			switch {
-			case apierrors.IsNotFound(err):
-				logger.V(2).Info("Victim Pod is already deleted", "preemptor", klog.KObj(pod), "node", c.Name(), "err", err)
-			case err != nil:
+			if err := errCh.ReceiveError(); err != nil {
 				utilruntime.HandleErrorWithContext(ctx, err, "Error occurred during async preemption")
 				result = metrics.GoroutineResultError
-			default:
-				allPodsAlreadyDeleted = false
 			}
 		}
 
@@ -569,6 +573,7 @@ func (ev *Evaluator) prepareCandidateAsync(c Candidate, pod *v1.Pod, pluginName
 			logger.V(2).Info("Victim Pod is already deleted", "preemptor", klog.KObj(pod), "node", c.Name(), "err", err)
 		case err != nil:
 			utilruntime.HandleErrorWithContext(ctx, err, "Error occurred during async preemption")
+			cancel()
 			result = metrics.GoroutineResultError
 		default:
 			allPodsAlreadyDeleted = false
diff --git a/test/integration/scheduler/preemption/preemption_test.go b/test/integration/scheduler/preemption/preemption_test.go
@@ -1569,17 +1569,18 @@ func TestNominatedNodeCleanUp(t *testing.T) {
 					st.MakePod().Name("low-4").Priority(lowPriority).Req(map[v1.ResourceName]string{v1.ResourceCPU: "1"}).Obj(),
 				},
 				{
-					st.MakePod().Name("medium").Priority(mediumPriority).Req(map[v1.ResourceName]string{v1.ResourceCPU: "4"}).Obj(),
+					st.MakePod().Name("medium").Priority(mediumPriority).Req(map[v1.ResourceName]string{v1.ResourceCPU: "3"}).Obj(),
 				},
 				{
-					st.MakePod().Name("high").Priority(highPriority).Req(map[v1.ResourceName]string{v1.ResourceCPU: "3"}).Obj(),
+					st.MakePod().Name("high").Priority(highPriority).Req(map[v1.ResourceName]string{v1.ResourceCPU: "4"}).Obj(),
 				},
 			},
 			postChecks: []func(ctx context.Context, cs clientset.Interface, pod *v1.Pod) error{
 				testutils.WaitForPodToSchedule,
 				testutils.WaitForNominatedNodeName,
 				testutils.WaitForNominatedNodeName,
 			},
+			podNamesToDelete: []string{"low-1", "low-2", "low-3", "low-4"},
 		},
 		{
 			name:         "mid-priority pod preempts low-priority pod, followed by a high-priority pod without additional preemption",
