Skip to content

Commit

Permalink
add e2e test
Browse files Browse the repository at this point in the history
# Conflicts:
#	pkg/webhooks/jobset_webhook.go
  • Loading branch information
mimowo committed Jul 26, 2024
1 parent 001e6d6 commit 38e5520
Show file tree
Hide file tree
Showing 3 changed files with 109 additions and 9 deletions.
7 changes: 6 additions & 1 deletion pkg/controllers/jobset_controller.go
Original file line number Diff line number Diff line change
Expand Up @@ -205,7 +205,7 @@ func (r *JobSetReconciler) reconcile(ctx context.Context, js *jobset.JobSet, upd
// Handle suspending a jobset or resuming a suspended jobset.
jobsetSuspended := jobSetSuspended(js)
if jobsetSuspended {
if err := r.suspendJobs(ctx, js, ownedJobs.active, updateStatusOpts); err != nil {
if err := r.deleteJobs(ctx, ownedJobs.active); err != nil {
log.Error(err, "suspending jobset")
return ctrl.Result{}, err
}
Expand Down Expand Up @@ -495,6 +495,11 @@ func (r *JobSetReconciler) reconcileReplicatedJobs(ctx context.Context, js *jobs

status := findReplicatedJobStatus(replicatedJobStatus, replicatedJob.Name)

// Don't create Jobs if the JobSet is suspended
if jobSetSuspended(js) {
continue
}

// For startup policy, if the replicatedJob is started we can skip this loop.
// Jobs have been created.
if !jobSetSuspended(js) && inOrderStartupPolicy(startupPolicy) && allReplicasStarted(replicatedJob.Replicas, status) {
Expand Down
27 changes: 19 additions & 8 deletions pkg/webhooks/jobset_webhook.go
Original file line number Diff line number Diff line change
Expand Up @@ -313,14 +313,25 @@ func (j *jobSetWebhook) ValidateUpdate(ctx context.Context, old, newObj runtime.
// This is needed for integration with Kueue/DWS.
if ptr.Deref(oldJS.Spec.Suspend, false) || ptr.Deref(js.Spec.Suspend, false) {
for index := range js.Spec.ReplicatedJobs {
// Pod values which must be mutable for Kueue are defined here: https://github.com/kubernetes-sigs/kueue/blob/a50d395c36a2cb3965be5232162cf1fded1bdb08/apis/kueue/v1beta1/workload_types.go#L256-L260
mungedSpec.ReplicatedJobs[index].Template.Spec.Template.Annotations = oldJS.Spec.ReplicatedJobs[index].Template.Spec.Template.Annotations
mungedSpec.ReplicatedJobs[index].Template.Spec.Template.Labels = oldJS.Spec.ReplicatedJobs[index].Template.Spec.Template.Labels
mungedSpec.ReplicatedJobs[index].Template.Spec.Template.Spec.NodeSelector = oldJS.Spec.ReplicatedJobs[index].Template.Spec.Template.Spec.NodeSelector
mungedSpec.ReplicatedJobs[index].Template.Spec.Template.Spec.Tolerations = oldJS.Spec.ReplicatedJobs[index].Template.Spec.Template.Spec.Tolerations

// Pod Scheduling Gates can be updated for batch/v1 Job: https://github.com/kubernetes/kubernetes/blob/ceb58a4dbc671b9d0a2de6d73a1616bc0c299863/pkg/apis/batch/validation/validation.go#L662
mungedSpec.ReplicatedJobs[index].Template.Spec.Template.Spec.SchedulingGates = oldJS.Spec.ReplicatedJobs[index].Template.Spec.Template.Spec.SchedulingGates
munge := true
// Don't allow to unsuspend if there are still active Jobs with
// different PodTemplate. We want to wait for the Jobs to be deleted
// first ast the JobSet remains suspended.
if !ptr.Deref(js.Spec.Suspend, false) {
if len(js.Status.ReplicatedJobsStatus) < index || js.Status.ReplicatedJobsStatus[index].Active > 0 {
munge = false
}
}
if munge {
// Pod values which must be mutable for Kueue are defined here: https://github.com/kubernetes-sigs/kueue/blob/a50d395c36a2cb3965be5232162cf1fded1bdb08/apis/kueue/v1beta1/workload_types.go#L256-L260
mungedSpec.ReplicatedJobs[index].Template.Spec.Template.Annotations = oldJS.Spec.ReplicatedJobs[index].Template.Spec.Template.Annotations
mungedSpec.ReplicatedJobs[index].Template.Spec.Template.Labels = oldJS.Spec.ReplicatedJobs[index].Template.Spec.Template.Labels
mungedSpec.ReplicatedJobs[index].Template.Spec.Template.Spec.NodeSelector = oldJS.Spec.ReplicatedJobs[index].Template.Spec.Template.Spec.NodeSelector
mungedSpec.ReplicatedJobs[index].Template.Spec.Template.Spec.Tolerations = oldJS.Spec.ReplicatedJobs[index].Template.Spec.Template.Spec.Tolerations

// Pod Scheduling Gates can be updated for batch/v1 Job: https://github.com/kubernetes/kubernetes/blob/ceb58a4dbc671b9d0a2de6d73a1616bc0c299863/pkg/apis/batch/validation/validation.go#L662
mungedSpec.ReplicatedJobs[index].Template.Spec.Template.Spec.SchedulingGates = oldJS.Spec.ReplicatedJobs[index].Template.Spec.Template.Spec.SchedulingGates
}
}
}

Expand Down
84 changes: 84 additions & 0 deletions test/e2e/e2e_test.go
Original file line number Diff line number Diff line change
Expand Up @@ -24,7 +24,9 @@ import (
corev1 "k8s.io/api/core/v1"
metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
"k8s.io/apimachinery/pkg/types"
"k8s.io/utils/ptr"

"sigs.k8s.io/controller-runtime/pkg/client"
jobset "sigs.k8s.io/jobset/api/jobset/v1alpha2"
"sigs.k8s.io/jobset/pkg/util/testing"
"sigs.k8s.io/jobset/test/util"
Expand Down Expand Up @@ -131,6 +133,66 @@ var _ = ginkgo.Describe("JobSet", func() {
})
})

ginkgo.When("job is unsuspended and suspend", func() {
for rep := 0; rep < 100; rep++ {
ginkgo.FIt(fmt.Sprintf("should allow to update PodTemplate on unsuspend and restore the PodTemplate on suspend-%v", rep), func() {
ctx := context.Background()
js := shortSleepTestJobSet(ns).Obj()

ginkgo.By("Create a suspended JobSet", func() {
js.Spec.Suspend = ptr.To(true)
js.Spec.TTLSecondsAfterFinished = ptr.To[int32](5)
gomega.Expect(k8sClient.Create(ctx, js)).Should(gomega.Succeed())
})

ginkgo.By("Unsuspend the JobSet setting nodeSelectors that prevent pods from being scheduled", func() {
gomega.Eventually(func() error {
gomega.Expect(k8sClient.Get(ctx, client.ObjectKeyFromObject(js), js)).Should(gomega.Succeed())
js.Spec.Suspend = ptr.To(false)
podTemplate := &js.Spec.ReplicatedJobs[0].Template.Spec.Template
if podTemplate.Spec.NodeSelector == nil {
podTemplate.Spec.NodeSelector = make(map[string]string)
}
podTemplate.Spec.NodeSelector["kubernetes.io/hostname"] = "non-existing-node"
if podTemplate.Labels == nil {
podTemplate.Labels = make(map[string]string)
}
podTemplate.Labels["custom-label-key"] = "custom-label-value"
if podTemplate.Annotations == nil {
podTemplate.Annotations = make(map[string]string)
}
podTemplate.Annotations["custom-annotation-key"] = "custom-annotation-value"
return k8sClient.Update(ctx, js)
}, timeout, interval).Should(gomega.Succeed())
})

ginkgo.By("Suspend the JobSet restoring the PodTemplate properties", func() {
gomega.Eventually(func() error {
gomega.Expect(k8sClient.Get(ctx, client.ObjectKeyFromObject(js), js)).Should(gomega.Succeed())
js.Spec.Suspend = ptr.To(true)
podTemplate := &js.Spec.ReplicatedJobs[0].Template.Spec.Template
delete(podTemplate.Spec.NodeSelector, "kubernetes.io/hostname")
delete(podTemplate.Labels, "custom-label-key")
delete(podTemplate.Annotations, "custom-annotation-key")
return k8sClient.Update(ctx, js)
}, timeout, interval).Should(gomega.Succeed())
})

ginkgo.By("Unsuspending the JobSet again with PodTemplate allowing completion", func() {
gomega.Eventually(func() error {
gomega.Expect(k8sClient.Get(ctx, client.ObjectKeyFromObject(js), js)).Should(gomega.Succeed())
js.Spec.Suspend = ptr.To(false)
return k8sClient.Update(ctx, js)
}, timeout, interval).Should(gomega.Succeed())
})

ginkgo.By("Await for the JobSet to complete successfully", func() {
util.JobSetCompleted(ctx, k8sClient, js, timeout)
})
})
}
})

}) // end of Describe

// getPingCommand returns ping command for 4 hostnames
Expand Down Expand Up @@ -246,3 +308,25 @@ func sleepTestJobSet(ns *corev1.Namespace) *testing.JobSetWrapper {
Replicas(int32(replicas)).
Obj())
}

func shortSleepTestJobSet(ns *corev1.Namespace) *testing.JobSetWrapper {
jsName := "js"
rjobName := "rjob"
replicas := 3
return testing.MakeJobSet(jsName, ns.Name).
ReplicatedJob(testing.MakeReplicatedJob(rjobName).
Job(testing.MakeJobTemplate("job", ns.Name).
PodSpec(corev1.PodSpec{
RestartPolicy: "Never",
Containers: []corev1.Container{
{
Name: "short-sleep-test-container",
Image: "bash:latest",
Command: []string{"bash", "-c"},
Args: []string{"sleep 1"},
},
},
}).Obj()).
Replicas(int32(replicas)).
Obj())
}

0 comments on commit 38e5520

Please sign in to comment.