Add global-job-replicas label/annotation

api/jobset/v1alpha2/jobset_types.go

@@ -22,6 +22,10 @@ import (
 const (
 	JobSetNameKey         string = "jobset.sigs.k8s.io/jobset-name"
 	ReplicatedJobReplicas string = "jobset.sigs.k8s.io/replicatedjob-replicas"
+	// GlobalReplicasKey is a label/annotation set to the total number of replicatedJob replicas.
+	// For each JobSet, this value will be equal to the sum of `replicas`, where `replicas`
+	// is equal to jobset.spec.replicatedJobs[*].replicas.
+	GlobalReplicasKey string = "jobset.sigs.k8s.io/global-replicas"
 	// ReplicatedJobNameKey is used to index into a Jobs labels and retrieve the name of the parent ReplicatedJob
 	ReplicatedJobNameKey string = "jobset.sigs.k8s.io/replicatedjob-name"
 	// JobIndexKey is a label/annotation set to the index of the Job replica within its parent replicatedJob.

pkg/controllers/jobset_controller.go

@@ -728,6 +728,7 @@ func labelAndAnnotateObject(obj metav1.Object, js *jobset.JobSet, rjob *jobset.R
 	labels[jobset.ReplicatedJobNameKey] = rjob.Name
 	labels[constants.RestartsKey] = strconv.Itoa(int(js.Status.Restarts))
 	labels[jobset.ReplicatedJobReplicas] = strconv.Itoa(int(rjob.Replicas))
+	labels[jobset.GlobalReplicasKey] = globalReplicas(js)
 	labels[jobset.JobIndexKey] = strconv.Itoa(jobIdx)
 	labels[jobset.JobKey] = jobHashKey(js.Namespace, jobName)
 	labels[jobset.JobGlobalIndexKey] = globalJobIndex(js, rjob.Name, jobIdx)
@@ -738,6 +739,7 @@ func labelAndAnnotateObject(obj metav1.Object, js *jobset.JobSet, rjob *jobset.R
 	annotations[jobset.ReplicatedJobNameKey] = rjob.Name
 	annotations[constants.RestartsKey] = strconv.Itoa(int(js.Status.Restarts))
 	annotations[jobset.ReplicatedJobReplicas] = strconv.Itoa(int(rjob.Replicas))
+	annotations[jobset.GlobalReplicasKey] = globalReplicas(js)
 	annotations[jobset.JobIndexKey] = strconv.Itoa(jobIdx)
 	annotations[jobset.JobKey] = jobHashKey(js.Namespace, jobName)
 	annotations[jobset.JobGlobalIndexKey] = globalJobIndex(js, rjob.Name, jobIdx)
@@ -1063,3 +1065,12 @@ func globalJobIndex(js *jobset.JobSet, replicatedJobName string, jobIdx int) str
 	}
 	return ""
 }
+
+// globalReplicas calculates the total number of replicas across all replicated jobs in a JobSet.
+func globalReplicas(js *jobset.JobSet) string {
+	currGlobalReplicas := 0
+	for _, rjob := range js.Spec.ReplicatedJobs {
+		currGlobalReplicas += int(rjob.Replicas)
+	}
+	return strconv.Itoa(currGlobalReplicas)
+}

pkg/controllers/jobset_controller_test.go

@@ -692,6 +692,7 @@ func TestConstructJobsFromTemplate(t *testing.T) {
 			// full JobSet spec to calculate a unique ID for each Job.
 			for _, expectedJob := range tc.want {
 				addJobGlobalIndex(t, tc.js, expectedJob)
+				addGlobalReplicas(t, tc.js, expectedJob)
 			}
 
 			// Now get the actual output of constructJobsFromTemplate, and diff the results.
@@ -728,6 +729,18 @@ func addJobGlobalIndex(t *testing.T, js *jobset.JobSet, job *batchv1.Job) {
 	job.Spec.Template.Annotations[jobset.JobGlobalIndexKey] = globalJobIndex(js, rjobName, jobIdx)
 }
 
+func addGlobalReplicas(t *testing.T, js *jobset.JobSet, job *batchv1.Job) {
+	t.Helper()
+
+	// Job label/annotation
+	job.Labels[jobset.GlobalReplicasKey] = globalReplicas(js)
+	job.Annotations[jobset.GlobalReplicasKey] = globalReplicas(js)
+
+	// Job template spec label/annotation
+	job.Spec.Template.Labels[jobset.GlobalReplicasKey] = globalReplicas(js)
+	job.Spec.Template.Annotations[jobset.GlobalReplicasKey] = globalReplicas(js)
+}
+
 func TestUpdateConditions(t *testing.T) {
 	var (
 		jobSetName        = "test-jobset"
@@ -1471,3 +1484,57 @@ func TestGlobalJobIndex(t *testing.T) {
 		})
 	}
 }
+
+func TestGlobalReplicas(t *testing.T) {
+	tests := []struct {
+		name                   string
+		jobSet                 *jobset.JobSet
+		expectedGlobalReplicas string
+	}{
+		{
+			name: "empty jobset",
+			jobSet: &jobset.JobSet{
+				Spec: jobset.JobSetSpec{},
+			},
+			expectedGlobalReplicas: "0",
+		},
+		{
+			name: "single replicated job",
+			jobSet: &jobset.JobSet{
+				Spec: jobset.JobSetSpec{
+					ReplicatedJobs: []jobset.ReplicatedJob{
+						{
+							Replicas: 3,
+						},
+					},
+				},
+			},
+			expectedGlobalReplicas: "3",
+		},
+		{
+			name: "multiple replicated jobs",
+			jobSet: &jobset.JobSet{
+				Spec: jobset.JobSetSpec{
+					ReplicatedJobs: []jobset.ReplicatedJob{
+						{
+							Replicas: 2,
+						},
+						{
+							Replicas: 5,
+						},
+					},
+				},
+			},
+			expectedGlobalReplicas: "7",
+		},
+	}
+
+	for _, tc := range tests {
+		t.Run(tc.name, func(t *testing.T) {
+			actualGlobalReplicas := globalReplicas(tc.jobSet)
+			if diff := cmp.Diff(tc.expectedGlobalReplicas, actualGlobalReplicas); diff != "" {
+				t.Errorf("unexpected global replicas (-want/+got): %s", diff)
+			}
+		})
+	}
+}