Merge pull request kubernetes#130649 from natasha41575/pod-conditions-scheduler

[FG:PodObservedGenerationTracking] scheduler sets observedGeneration on pod conditions

Author: k8s-ci-robot

pkg/scheduler/framework/preemption/preemption.go

@@ -171,10 +171,11 @@ func NewEvaluator(pluginName string, fh framework.Handle, i Interface, enableAsy
 			logger.V(2).Info("Preemptor pod rejected a waiting pod", "preemptor", klog.KObj(preemptor), "waitingPod", klog.KObj(victim), "node", c.Name())
 		} else {
 			condition := &v1.PodCondition{
-				Type:    v1.DisruptionTarget,
-				Status:  v1.ConditionTrue,
-				Reason:  v1.PodReasonPreemptionByScheduler,
-				Message: fmt.Sprintf("%s: preempting to accommodate a higher priority pod", preemptor.Spec.SchedulerName),
+				Type:               v1.DisruptionTarget,
+				ObservedGeneration: apipod.GetPodObservedGenerationIfEnabledOnCondition(&victim.Status, victim.Generation, v1.DisruptionTarget),
+				Status:             v1.ConditionTrue,
+				Reason:             v1.PodReasonPreemptionByScheduler,
+				Message:            fmt.Sprintf("%s: preempting to accommodate a higher priority pod", preemptor.Spec.SchedulerName),
 			}
 			newStatus := victim.Status.DeepCopy()
 			updated := apipod.UpdatePodCondition(newStatus, condition)

pkg/scheduler/schedule_one.go

@@ -1097,10 +1097,11 @@ func (sched *Scheduler) handleSchedulingFailure(ctx context.Context, fwk framewo
 	msg := truncateMessage(errMsg)
 	fwk.EventRecorder().Eventf(pod, nil, v1.EventTypeWarning, "FailedScheduling", "Scheduling", msg)
 	if err := updatePod(ctx, sched.client, pod, &v1.PodCondition{
-		Type:    v1.PodScheduled,
-		Status:  v1.ConditionFalse,
-		Reason:  reason,
-		Message: errMsg,
+		Type:               v1.PodScheduled,
+		ObservedGeneration: podutil.GetPodObservedGenerationIfEnabledOnCondition(&pod.Status, pod.Generation, v1.PodScheduled),
+		Status:             v1.ConditionFalse,
+		Reason:             reason,
+		Message:            errMsg,
 	}, nominatingInfo); err != nil {
 		logger.Error(err, "Error updating pod", "pod", klog.KObj(pod))
 	}

test/e2e/node/pods.go

@@ -529,7 +529,7 @@ var _ = SIGDescribe("Pods Extended (pod generation)", feature.PodObservedGenerat
 			}
 		})
 
-		ginkgo.It("custom-set generation on new pods should be overwritten to 1", func(ctx context.Context) {
+		ginkgo.It("custom-set generation on new pods and graceful delete", func(ctx context.Context) {
 			ginkgo.By("creating the pod")
 			name := "pod-generation-" + string(uuid.NewUUID())
 			value := strconv.Itoa(time.Now().Nanosecond())
@@ -598,6 +598,10 @@ var _ = SIGDescribe("Pods Extended (pod generation)", feature.PodObservedGenerat
 			gomega.Expect(pod.Generation).To(gomega.BeEquivalentTo(expectedPodGeneration))
 		})
 
+		// This is the same test as https://github.com/kubernetes/kubernetes/blob/aa08c90fca8d30038d3f05c0e8f127b540b40289/test/e2e/node/pod_admission.go#L35,
+		// except that this verifies the pod generation and observedGeneration, which is
+		// currently behind a feature gate. When we GA observedGeneration functionality,
+		// we can fold these tests together into one.
 		ginkgo.It("pod rejected by kubelet should have updated generation and observedGeneration", func(ctx context.Context) {
 			node, err := e2enode.GetRandomReadySchedulableNode(ctx, f.ClientSet)
 			framework.ExpectNoError(err, "Failed to get a ready schedulable node")
@@ -631,6 +635,18 @@ var _ = SIGDescribe("Pods Extended (pod generation)", feature.PodObservedGenerat
 				return podClient.Delete(ctx, pod.Name, metav1.DeleteOptions{})
 			})
 
+			// Wait for the scheduler to update the pod status
+			err = e2epod.WaitForPodNameUnschedulableInNamespace(ctx, f.ClientSet, pod.Name, pod.Namespace)
+			framework.ExpectNoError(err)
+
+			// Fetch the pod to verify that the scheduler has set the PodScheduled condition
+			// with observedGeneration.
+			pod, err = podClient.Get(ctx, pod.Name, metav1.GetOptions{})
+			framework.ExpectNoError(err)
+			gomega.Expect(len(pod.Status.Conditions)).To(gomega.BeEquivalentTo(1))
+			gomega.Expect(pod.Status.Conditions[0].Type).To(gomega.BeEquivalentTo(v1.PodScheduled))
+			gomega.Expect(pod.Status.Conditions[0].ObservedGeneration).To(gomega.BeEquivalentTo(1))
+
 			// Force assign the Pod to a node in order to get rejection status.
 			binding := &v1.Binding{
 				ObjectMeta: metav1.ObjectMeta{
@@ -650,7 +666,7 @@ var _ = SIGDescribe("Pods Extended (pod generation)", feature.PodObservedGenerat
 			framework.ExpectNoError(err)
 
 			// Fetch the rejected Pod and verify the generation and observedGeneration.
-			gotPod, err := f.ClientSet.CoreV1().Pods(pod.Namespace).Get(ctx, pod.Name, metav1.GetOptions{})
+			gotPod, err := podClient.Get(ctx, pod.Name, metav1.GetOptions{})
 			framework.ExpectNoError(err)
 			gomega.Expect(gotPod.Generation).To(gomega.BeEquivalentTo(1))
 			gomega.Expect(gotPod.Status.ObservedGeneration).To(gomega.BeEquivalentTo(1))