Merge pull request #130650 from natasha41575/pod-conditions-controller

[FG:PodObservedGenerationTracking] controller sets observedGeneration on pod conditions

pkg/api/v1/pod/util.go

@@ -428,3 +428,21 @@ func GetPodObservedGenerationIfEnabled(pod *v1.Pod) int64 {
 	}
 	return 0
 }
+
+// We will emit condition.observedGeneration if the feature is enabled OR if condition.observedGeneration is already set.
+// This protects against an infinite loop of kubelet trying to clear the value after the FG is turned off, and
+// the API server preserving existing values when an incoming update tries to clear it.
+func GetPodObservedGenerationIfEnabledOnCondition(pod *v1.Pod, conditionType v1.PodConditionType) int64 {
+	if pod == nil {
+		return 0
+	}
+	if utilfeature.DefaultFeatureGate.Enabled(features.PodObservedGenerationTracking) {
+		return pod.Generation
+	}
+	for _, condition := range pod.Status.Conditions {
+		if condition.Type == conditionType && condition.ObservedGeneration != 0 {
+			return pod.Generation
+		}
+	}
+	return 0
+}

pkg/controller/disruption/disruption.go

@@ -787,6 +787,7 @@ func (dc *DisruptionController) syncStalePodDisruption(ctx context.Context, key
 	newPod := pod.DeepCopy()
 	updated := apipod.UpdatePodCondition(&newPod.Status, &v1.PodCondition{
 		Type:               v1.DisruptionTarget,
+		ObservedGeneration: apipod.GetPodObservedGenerationIfEnabledOnCondition(newPod, v1.DisruptionTarget),
 		Status:             v1.ConditionFalse,
 	})
 	if !updated {

pkg/controller/podgc/gc_controller.go

@@ -247,6 +247,7 @@ func (gcc *PodGCController) gcOrphaned(ctx context.Context, pods []*v1.Pod, node
 		logger.V(2).Info("Found orphaned Pod assigned to the Node, deleting", "pod", klog.KObj(pod), "node", klog.KRef("", pod.Spec.NodeName))
 		condition := &v1.PodCondition{
 			Type:               v1.DisruptionTarget,
+			ObservedGeneration: apipod.GetPodObservedGenerationIfEnabledOnCondition(pod, v1.DisruptionTarget),
 			Status:             v1.ConditionTrue,
 			Reason:             "DeletionByPodGC",
 			Message:            "PodGC: node no longer exists",

pkg/controller/tainteviction/taint_eviction.go

@@ -134,6 +134,7 @@ func addConditionAndDeletePod(ctx context.Context, c clientset.Interface, name,
 	newStatus := pod.Status.DeepCopy()
 	updated := apipod.UpdatePodCondition(newStatus, &v1.PodCondition{
 		Type:               v1.DisruptionTarget,
+		ObservedGeneration: apipod.GetPodObservedGenerationIfEnabledOnCondition(pod, v1.DisruptionTarget),
 		Status:             v1.ConditionTrue,
 		Reason:             "DeletionByTaintManager",
 		Message:            "Taint manager: deleting due to NoExecute taint",