scheduler: fix monitor terminating pod (#2331)

Signed-off-by: saintube <[email protected]>
Co-authored-by: shenxin <[email protected]>

pkg/scheduler/frameworkext/scheduler_monitor.go

@@ -91,6 +91,15 @@ func (m *SchedulerMonitor) RecordNextPod(podInfo *framework.QueuedPodInfo) {
 		return
 	}
 	pod := podInfo.Pod
+
+	// clean up from the cache when the pod is terminating
+	if pod.DeletionTimestamp != nil {
+		m.lock.Lock()
+		delete(m.schedulingPods, pod.UID)
+		m.lock.Unlock()
+		return
+	}
+
 	now := time.Now()
 	scheduleState := podScheduleState{
 		namespace:       pod.Namespace,

pkg/scheduler/frameworkext/scheduler_monitor_test.go

@@ -125,4 +125,17 @@ func TestSchedulerMonitor_StartAndCompleteMonitoring(t *testing.T) {
 	if _, exists := monitor.schedulingPods[pod.UID]; exists {
 		t.Errorf("Pod should be removed from schedulingPods after Complete")
 	}
+
+	terminatingPod := podInfo.DeepCopy()
+	deleted := metav1.Now()
+	terminatingPod.Pod.DeletionTimestamp = &deleted
+	queuePodInfo = &framework.QueuedPodInfo{
+		Timestamp: time.Now(),
+		PodInfo:   terminatingPod,
+	}
+	monitor.RecordNextPod(queuePodInfo)
+	_, ok = monitor.schedulingPods[pod.UID]
+	if ok {
+		t.Fatal("Pod should be removed in schedulingPods when terminating")
+	}
 }