add support for Network.Subdomain

Problem: A user that wants to create jobs on the same network, or an existing
network, currently cannot, as the default with DNS enabled is to create
different networks.
Solution: Allow the user to specify a subdomain for the Network.

Signed-off-by: vsoch <vsoch@users.noreply.github.com>

api/v1alpha1/jobset_types.go

@@ -130,6 +130,11 @@ type Network struct {
 	// <jobSet.name>-<spec.replicatedJob.name>-<job-index>-<pod-index>.<jobSet.name>-<spec.replicatedJob.name>
 	// +optional
 	EnableDNSHostnames *bool `json:"enableDNSHostnames,omitempty"`
+
+	// Subdomain is an explicit choice for a network subdomain name
+	// When set, the replicated job is added to this network. This would allow sharing common networks
+	// +optional
+	Subdomain string `json:"subdomain,omitempty"`
 }
 
 // Operator defines the target of a SuccessPolicy or FailurePolicy.

config/components/crd/bases/jobset.x-k8s.io_jobsets.yaml

@@ -79,6 +79,11 @@ spec:
                             fully qualified pod hostname, which is in the format:
                             <jobSet.name>-<spec.replicatedJob.name>-<job-index>-<pod-index>.<jobSet.name>-<spec.replicatedJob.name>'
                           type: boolean
+                        subdomain:
+                          description: Subdomain is an explicit choice for a network
+                            subdomain name When set, the replicated job is added to
+                            this network. This would allow sharing common networks
+                          type: string
                       type: object
                     replicas:
                       default: 1

pkg/controllers/jobset_controller.go

@@ -380,8 +380,8 @@ func (r *JobSetReconciler) createJobs(ctx context.Context, js *jobset.JobSet, ow
 func (r *JobSetReconciler) createHeadlessSvcIfNotExist(ctx context.Context, js *jobset.JobSet, rjob *jobset.ReplicatedJob) error {
 	log := ctrl.LoggerFrom(ctx)
 
-	// Check if service already exists. Service name is <jobSetName>-<replicatedJobName>.
-	// If the service does not exist, create it.
+	// Check if service already exists. Default service name is <jobSetName>-<replicatedJobName> unless otherwise specified.
+	// If a subdomain is provided already, it should also be in the same namespace.
 	var headlessSvc corev1.Service
 	subdomain := GenSubdomain(js, rjob)
 	if err := r.Get(ctx, types.NamespacedName{Name: subdomain, Namespace: js.Namespace}, &headlessSvc); err != nil {
@@ -390,6 +390,7 @@ func (r *JobSetReconciler) createHeadlessSvcIfNotExist(ctx context.Context, js *
 				Name:      subdomain,
 				Namespace: js.Namespace,
 			},
+			// TODO how to handle replicated job selectors here when more than one job share a service?
 			Spec: corev1.ServiceSpec{
 				ClusterIP: "None",
 				Selector: map[string]string{
@@ -573,7 +574,8 @@ func constructJob(js *jobset.JobSet, rjob *jobset.ReplicatedJob, jobIdx int) (*b
 	labelAndAnnotateObject(&job.Spec.Template, js, rjob, jobIdx)
 
 	// If enableDNSHostnames is set, update job spec to set subdomain as
-	// job name (a headless service with same name as job will be created later).
+	// job name (a headless service with a chosen name or the
+	// same name as job will be created or found later).
 	if dnsHostnamesEnabled(rjob) {
 		job.Spec.Template.Spec.Subdomain = GenSubdomain(js, rjob)
 	}
@@ -681,6 +683,11 @@ func genJobName(js *jobset.JobSet, rjob *jobset.ReplicatedJob, jobIndex int) str
 }
 
 func GenSubdomain(js *jobset.JobSet, rjob *jobset.ReplicatedJob) string {
+
+	// If we have selected an explicit network name, use it
+	if rjob.Network.Subdomain != "" {
+		return rjob.Network.Subdomain
+	}
 	return fmt.Sprintf("%s-%s", js.Name, rjob.Name)
 }
 

pkg/util/testing/wrappers.go

@@ -115,6 +115,12 @@ func (r *ReplicatedJobWrapper) EnableDNSHostnames(val bool) *ReplicatedJobWrappe
 	return r
 }
 
+// NetworkSubdomain sets the value of ReplicatedJob.Network.Subdomain
+func (r *ReplicatedJobWrapper) NetworkSubdomain(val string) *ReplicatedJobWrapper {
+	r.ReplicatedJob.Network.Subdomain = val
+	return r
+}
+
 // Replicas sets the value of the ReplicatedJob.Replicas.
 func (r *ReplicatedJobWrapper) Replicas(val int) *ReplicatedJobWrapper {
 	r.ReplicatedJob.Replicas = val

test/integration/controller/jobset_controller_test.go

@@ -390,6 +390,18 @@ var _ = ginkgo.Describe("JobSet controller", func() {
 				},
 			},
 		}),
+		ginkgo.Entry("jobset with DNS hostnames enabled and shared network name should create 1 headless service and succeed when all jobs succeed", &testCase{
+			makeJobSet: sharedNetworkJobSet,
+			updates: []*update{
+				{
+					checkJobSetState: checkSharedService,
+				},
+				{
+					jobUpdateFn:          completeAllJobs,
+					checkJobSetCondition: testutil.JobSetCompleted,
+				},
+			},
+		}),
 		ginkgo.Entry("succeeds from first run", &testCase{
 			makeJobSet: testJobSet,
 			updates: []*update{
@@ -813,6 +825,17 @@ func jobActive(job *batchv1.Job) bool {
 	return active
 }
 
+// Check one headless service across all replicated jobs
+func checkSharedService(js *jobset.JobSet) {
+	gomega.Eventually(func() (int, error) {
+		var svcList corev1.ServiceList
+		if err := k8sClient.List(ctx, &svcList, client.InNamespace(js.Namespace)); err != nil {
+			return -1, err
+		}
+		return len(svcList.Items), nil
+	}).Should(gomega.Equal(1))
+}
+
 // 2 replicated jobs:
 // - one with 1 replica
 // - one with 3 replicas and DNS hostnames enabled
@@ -829,3 +852,22 @@ func testJobSet(ns *corev1.Namespace) *testing.JobSetWrapper {
 			Replicas(3).
 			Obj())
 }
+
+// 2 replicated jobs:
+// - same as above, but with both having DNS enabled and a shared network
+func sharedNetworkJobSet(ns *corev1.Namespace) *testing.JobSetWrapper {
+	return testing.MakeJobSet("shared-network-js", ns.Name).
+		SuccessPolicy(&jobset.SuccessPolicy{Operator: jobset.OperatorAll, TargetReplicatedJobs: []string{}}).
+		ReplicatedJob(testing.MakeReplicatedJob("replicated-job-a").
+			Job(testing.MakeJobTemplate("test-job-A", ns.Name).PodSpec(testing.TestPodSpec).Obj()).
+			EnableDNSHostnames(true).
+			NetworkSubdomain("shared-network").
+			Replicas(1).
+			Obj()).
+		ReplicatedJob(testing.MakeReplicatedJob("replicated-job-b").
+			Job(testing.MakeJobTemplate("test-job-B", ns.Name).PodSpec(testing.TestPodSpec).CompletionMode(batchv1.IndexedCompletion).Obj()).
+			NetworkSubdomain("shared-network").
+			EnableDNSHostnames(true).
+			Replicas(3).
+			Obj())
+}