Allow decrease reclaimable pods to 0 for suspended job (#1277)

kubernetes-sigs · Oct 28, 2023 · 1684bf8 · 1684bf8
1 parent d37b3ce
commit 1684bf8
Show file tree

Hide file tree

Showing 2 changed files with 36 additions and 1 deletion.
diff --git a/pkg/webhooks/workload_webhook.go b/pkg/webhooks/workload_webhook.go
@@ -388,14 +388,17 @@ func validateReclaimablePodsUpdate(newObj, oldObj *kueue.Workload, basePath *fie
 	for i := range newObj.Status.ReclaimablePods {
 		newCount := &newObj.Status.ReclaimablePods[i]
 		newNames.Insert(newCount.Name)
+		if !workload.HasQuotaReservation(newObj) && newCount.Count == 0 {
+			continue
+		}
 		oldCount, found := knowPodSets[newCount.Name]
 		if found && newCount.Count < oldCount.Count {
 			ret = append(ret, field.Invalid(basePath.Key(newCount.Name).Child("count"), newCount.Count, fmt.Sprintf("cannot be less then %d", oldCount.Count)))
 		}
 	}
 
 	for name := range knowPodSets {
-		if !newNames.Has(name) {
+		if workload.HasQuotaReservation(newObj) && !newNames.Has(name) {
 			ret = append(ret, field.Required(basePath.Key(name), "cannot be removed"))
 		}
 	}

diff --git a/pkg/webhooks/workload_webhook_test.go b/pkg/webhooks/workload_webhook_test.go
@@ -546,6 +546,38 @@ func TestValidateWorkloadUpdate(t *testing.T) {
 				field.Required(field.NewPath("status", "reclaimablePods").Key("ps2"), ""),
 			},
 		},
+		"reclaimable pod count can go to 0 if the job is suspended": {
+			before: testingutil.MakeWorkload(testWorkloadName, testWorkloadNamespace).
+				PodSets(
+					*testingutil.MakePodSet("ps1", 3).Obj(),
+					*testingutil.MakePodSet("ps2", 3).Obj(),
+				).
+				ReserveQuota(
+					testingutil.MakeAdmission("cluster-queue").
+						PodSets(kueue.PodSetAssignment{Name: "ps1"}, kueue.PodSetAssignment{Name: "ps2"}).
+						Obj(),
+				).
+				ReclaimablePods(
+					kueue.ReclaimablePod{Name: "ps1", Count: 2},
+					kueue.ReclaimablePod{Name: "ps2", Count: 1},
+				).
+				Obj(),
+			after: testingutil.MakeWorkload(testWorkloadName, testWorkloadNamespace).
+				PodSets(
+					*testingutil.MakePodSet("ps1", 3).Obj(),
+					*testingutil.MakePodSet("ps2", 3).Obj(),
+				).
+				AdmissionChecks(kueue.AdmissionCheckState{
+					PodSetUpdates: []kueue.PodSetUpdate{{Name: "ps1"}, {Name: "ps2"}},
+					State:         kueue.CheckStateReady,
+				}).
+				ReclaimablePods(
+					kueue.ReclaimablePod{Name: "ps1", Count: 0},
+					kueue.ReclaimablePod{Name: "ps2", Count: 1},
+				).
+				Obj(),
+			wantErr: nil,
+		},
 		"priorityClassSource should not be updated": {
 			before: testingutil.MakeWorkload(testWorkloadName, testWorkloadNamespace).Queue("q").
 				PriorityClass("test-class").PriorityClassSource(constants.PodPriorityClassSource).