Implement the PodTermination failure recovery action

Author: kshalot

internal/mocks/controller/jobframework/interface.go

@@ -44,4 +44,9 @@ const (
 	// ManagedByKueueLabelKey label that signalize that an object is managed by Kueue
 	ManagedByKueueLabelKey   = "kueue.x-k8s.io/managed"
 	ManagedByKueueLabelValue = "true"
+
+	// PodSetLabel is a label set on the Job's PodTemplate to indicate the name
+	// of the PodSet of the admitted Workload corresponding to the PodTemplate.
+	// The label is set when starting the Job, and removed on stopping the Job.
+	PodSetLabel = "kueue.x-k8s.io/podset"
 )

pkg/constants/constants.go

@@ -43,9 +43,4 @@ const (
 
 	// MaxExecTimeSecondsLabel is the label key in the job that holds the maximum execution time.
 	MaxExecTimeSecondsLabel = `kueue.x-k8s.io/max-exec-time-seconds`
-
-	// PodSetLabel is a label set on the Job's PodTemplate to indicate the name
-	// of the PodSet of the admitted Workload corresponding to the PodTemplate.
-	// The label is set when starting the Job, and removed on stopping the Job.
-	PodSetLabel = "kueue.x-k8s.io/podset"
 )

pkg/controller/constants/constants.go

@@ -25,6 +25,7 @@ import (
 	qcache "sigs.k8s.io/kueue/pkg/cache/queue"
 	schdcache "sigs.k8s.io/kueue/pkg/cache/scheduler"
 	"sigs.k8s.io/kueue/pkg/constants"
+	"sigs.k8s.io/kueue/pkg/controller/failurerecovery"
 	"sigs.k8s.io/kueue/pkg/features"
 )
 
@@ -63,6 +64,13 @@ func SetupControllers(mgr ctrl.Manager, qManager *qcache.Manager, cc *schdcache.
 		watchers = append(watchers, cohortRec)
 	}
 
+	if features.Enabled(features.ZombiePodTermination) {
+		zpRec := failurerecovery.NewTerminatingPodReconciler(mgr.GetClient())
+		if err := zpRec.SetupWithManager(mgr); err != nil {
+			return "ZombiePodTermination", err
+		}
+	}
+
 	cqRec := NewClusterQueueReconciler(
 		mgr.GetClient(),
 		qManager,