Support Elastic JobSets

[kannon92]

What would you like to be added:

With Elastic Indexed jobs, it is possible to change completions/parallelism to down/up scale your jobs.

It would be nice to have something similar for JobSet.

Why is this needed:

Elastic jobs are an important usecase for autoscaling and other cases.

Implementation:

At a quick glance of the API this may be possible as replicas of a replicated job are not immutable so I think someone could patch the replicas of a ReplicatedJob to downscale or upscale.

But then I wonder what should we do with the existing replicated job?

And should we support ElasticIndexedJob with JobSet (so someone could patch the JobTemplate in a single replicated job?

[ahg-g]

Yes, I think we should think about allowing to autoscale the number of replicas in a replicatedJob! For example, the number of tpuslices (or more generally accelerator islands) supporting a large scale training job could scale down in case of failures.

But then I wonder what should we do with the existing replicated job?

The jobs are indexed, so a scale down means removing the higher order ones.

And should we support ElasticIndexedJob with JobSet (so someone could patch the JobTemplate in a single replicated job?

It is possible.

If the child jobs themselves should be elastic, then the operator could change the individual jobs directly. But I guess we could also allow changing that in bulk for all job replicas, but I need to hear a use case first.

[kannon92]

So @ahg-g it sounds like this is supported as you are correct. Both replicas in ReplicatedJob are mutable and JobTemplate is mutable..

Maybe we should consider a task for this to at least document that this is possible?

I think Kueue or other use cases would be interested in this but not sure what we need in this repo.

[ahg-g]

We need to have tests for that to verify the behavior though.

[kannon92]

Well good thing I tried it haha.

I think this code is blocking us from doing this.

https://github.com/kubernetes-sigs/jobset/blob/main/api/jobset/v1alpha2/jobset_webhook.go#L172

I tried a simple example and did a kubectl edit jobset and tried changing the replica of a ReplicatedJob.

Got:

error: jobsets.jobset.x-k8s.io "simple-no-ttl" could not be patched: admission webhook "vjobset.kb.io" denied the request: spec.replicatedJobs: Invalid value: []v1alpha2.ReplicatedJob{v1alpha2.ReplicatedJob{Name:"leader", Template:v1.JobTemplateSpec{ObjectMeta:v1.ObjectMeta{Name:"", GenerateName:"", Namespace:"", SelfLink:"", UID:"", ResourceVersion:"", Generation:0, CreationTimestamp:time.Date(1, time.January, 1, 0, 0, 0, 0, time.UTC), DeletionTimestamp:<nil>, DeletionGracePeriodSeconds:(*int64)(nil), Labels:map[string]string(nil), Annotations:map[string]string(nil), OwnerReferences:[]v1.OwnerReference(nil), Finalizers:[]string(nil), ManagedFields:[]v1.ManagedFieldsEntry(nil)}, Spec:v1.JobSpec{Parallelism:(*int32)(0xc0001fe320), Completions:(*int32)(0xc0001fe324), ActiveDeadlineSeconds:(*int64)(nil), PodFailurePolicy:(*v1.PodFailurePolicy)(nil), BackoffLimit:(*int32)(0xc0001fe328), BackoffLimitPerIndex:(*int32)(nil), MaxFailedIndexes:(*int32)(nil), Selector:(*v1.LabelSelector)(nil), ManualSelector:(*bool)(nil), Template:v1.PodTemplateSpec{ObjectMeta:v1.ObjectMeta{Name:"", GenerateName:"", Namespace:"", SelfLink:"", UID:"", ResourceVersion:"", Generation:0, CreationTimestamp:time.Date(1, time.January, 1, 0, 0, 0, 0, time.UTC), DeletionTimestamp:<nil>, DeletionGracePeriodSeconds:(*int64)(nil), Labels:map[string]string(nil), Annotations:map[string]string(nil), OwnerReferences:[]v1.OwnerReference(nil), Finalizers:[]string(nil), ManagedFields:[]v1.ManagedFieldsEntry(nil)}, Spec:v1.PodSpec{Volumes:[]v1.Volume(nil), InitContainers:[]v1.Container(nil), Containers:[]v1.Container{v1.Container{Name:"leader", Image:"bash:latest", Command:[]string{"bash", "-xc", "sleep 10000\n"}, Args:[]string(nil), WorkingDir:"", Ports:[]v1.ContainerPort(nil), EnvFrom:[]v1.EnvFromSource(nil), Env:[]v1.EnvVar(nil), Resources:v1.ResourceRequirements{Limits:v1.ResourceList(nil), Requests:v1.ResourceList(nil), Claims:[]v1.ResourceClaim(nil)}, ResizePolicy:[]v1.ContainerResizePolicy(nil), RestartPolicy:(*v1.ContainerRestartPolicy)(nil), VolumeMounts:[]v1.VolumeMount(nil), VolumeDevices:[]v1.VolumeDevice(nil), LivenessProbe:(*v1.Probe)(nil), ReadinessProbe:(*v1.Probe)(nil), StartupProbe:(*v1.Probe)(nil), Lifecycle:(*v1.Lifecycle)(nil), TerminationMessagePath:"", TerminationMessagePolicy:"", ImagePullPolicy:"", SecurityContext:(*v1.SecurityContext)(nil), Stdin:false, StdinOnce:false, TTY:false}}, EphemeralContainers:[]v1.EphemeralContainer(nil), RestartPolicy:"OnFailure", TerminationGracePeriodSeconds:(*int64)(nil), ActiveDeadlineSeconds:(*int64)(nil), DNSPolicy:"", NodeSelector:map[string]string(nil), ServiceAccountName:"", DeprecatedServiceAccount:"", AutomountServiceAccountToken:(*bool)(nil), NodeName:"", HostNetwork:false, HostPID:false, HostIPC:false, ShareProcessNamespace:(*bool)(nil), SecurityContext:(*v1.PodSecurityContext)(nil), ImagePullSecrets:[]v1.LocalObjectReference(nil), Hostname:"", Subdomain:"", Affinity:(*v1.Affinity)(nil), SchedulerName:"", Tolerations:[]v1.Toleration(nil), HostAliases:[]v1.HostAlias(nil), PriorityClassName:"", Priority:(*int32)(nil), DNSConfig:(*v1.PodDNSConfig)(nil), ReadinessGates:[]v1.PodReadinessGate(nil), RuntimeClassName:(*string)(nil), EnableServiceLinks:(*bool)(nil), PreemptionPolicy:(*v1.PreemptionPolicy)(nil), Overhead:v1.ResourceList(nil), TopologySpreadConstraints:[]v1.TopologySpreadConstraint(nil), SetHostnameAsFQDN:(*bool)(nil), OS:(*v1.PodOS)(nil), HostUsers:(*bool)(nil), SchedulingGates:[]v1.PodSchedulingGate(nil), ResourceClaims:[]v1.PodResourceClaim(nil)}}, TTLSecondsAfterFinished:(*int32)(nil), CompletionMode:(*v1.CompletionMode)(0xc00099fa60), Suspend:(*bool)(nil), PodReplacementPolicy:(*v1.PodReplacementPolicy)(nil)}}, Replicas:4}, v1alpha2.ReplicatedJob{Name:"workers", Template:v1.JobTemplateSpec{ObjectMeta:v1.ObjectMeta{Name:"", GenerateName:"", Namespace:"", SelfLink:"", UID:"", ResourceVersion:"", Generation:0, CreationTimestamp:time.Date(1, time.January, 1, 0, 0, 0, 0, time.UTC), DeletionTimestamp:<nil>, DeletionGracePeriodSeconds:(*int64)(nil), Labels:map[string]string(nil), Annotations:map[string]string(nil), OwnerReferences:[]v1.OwnerReference(nil), Finalizers:[]string(nil), ManagedFields:[]v1.ManagedFieldsEntry(nil)}, Spec:v1.JobSpec{Parallelism:(*int32)(0xc0001fe32c), Completions:(*int32)(0xc0001fe330), ActiveDeadlineSeconds:(*int64)(nil), PodFailurePolicy:(*v1.PodFailurePolicy)(nil), BackoffLimit:(*int32)(0xc0001fe334), BackoffLimitPerIndex:(*int32)(nil), MaxFailedIndexes:(*int32)(nil), Selector:(*v1.LabelSelector)(nil), ManualSelector:(*bool)(nil), Template:v1.PodTemplateSpec{ObjectMeta:v1.ObjectMeta{Name:"", GenerateName:"", Namespace:"", SelfLink:"", UID:"", ResourceVersion:"", Generation:0, CreationTimestamp:time.Date(1, time.January, 1, 0, 0, 0, 0, time.UTC), DeletionTimestamp:<nil>, DeletionGracePeriodSeconds:(*int64)(nil), Labels:map[string]string(nil), Annotations:map[string]string(nil), OwnerReferences:[]v1.OwnerReference(nil), Finalizers:[]string(nil), ManagedFields:[]v1.ManagedFieldsEntry(nil)}, Spec:v1.PodSpec{Volumes:[]v1.Volume(nil), InitContainers:[]v1.Container(nil), Containers:[]v1.Container{v1.Container{Name:"worker", Image:"bash:latest", Command:[]string{"bash", "-xc", "sleep 100000\n"}, Args:[]string(nil), WorkingDir:"", Ports:[]v1.ContainerPort(nil), EnvFrom:[]v1.EnvFromSource(nil), Env:[]v1.EnvVar(nil), Resources:v1.ResourceRequirements{Limits:v1.ResourceList(nil), Requests:v1.ResourceList(nil), Claims:[]v1.ResourceClaim(nil)}, ResizePolicy:[]v1.ContainerResizePolicy(nil), RestartPolicy:(*v1.ContainerRestartPolicy)(nil), VolumeMounts:[]v1.VolumeMount(nil), VolumeDevices:[]v1.VolumeDevice(nil), LivenessProbe:(*v1.Probe)(nil), ReadinessProbe:(*v1.Probe)(nil), StartupProbe:(*v1.Probe)(nil), Lifecycle:(*v1.Lifecycle)(nil), TerminationMessagePath:"", TerminationMessagePolicy:"", ImagePullPolicy:"", SecurityContext:(*v1.SecurityContext)(nil), Stdin:false, StdinOnce:false, TTY:false}}, EphemeralContainers:[]v1.EphemeralContainer(nil), RestartPolicy:"OnFailure", TerminationGracePeriodSeconds:(*int64)(nil), ActiveDeadlineSeconds:(*int64)(nil), DNSPolicy:"", NodeSelector:map[string]string(nil), ServiceAccountName:"", DeprecatedServiceAccount:"", AutomountServiceAccountToken:(*bool)(nil), NodeName:"", HostNetwork:false, HostPID:false, HostIPC:false, ShareProcessNamespace:(*bool)(nil), SecurityContext:(*v1.PodSecurityContext)(nil), ImagePullSecrets:[]v1.LocalObjectReference(nil), Hostname:"", Subdomain:"", Affinity:(*v1.Affinity)(nil), SchedulerName:"", Tolerations:[]v1.Toleration(nil), HostAliases:[]v1.HostAlias(nil), PriorityClassName:"", Priority:(*int32)(nil), DNSConfig:(*v1.PodDNSConfig)(nil), ReadinessGates:[]v1.PodReadinessGate(nil), RuntimeClassName:(*string)(nil), EnableServiceLinks:(*bool)(nil), PreemptionPolicy:(*v1.PreemptionPolicy)(nil), Overhead:v1.ResourceList(nil), TopologySpreadConstraints:[]v1.TopologySpreadConstraint(nil), SetHostnameAsFQDN:(*bool)(nil), OS:(*v1.PodOS)(nil), HostUsers:(*bool)(nil), SchedulingGates:[]v1.PodSchedulingGate(nil), ResourceClaims:[]v1.PodResourceClaim(nil)}}, TTLSecondsAfterFinished:(*int32)(nil), CompletionMode:(*v1.CompletionMode)(0xc00099fa70), Suspend:(*bool)(nil), PodReplacementPolicy:(*v1.PodReplacementPolicy)(nil)}}, Replicas:1}}: field is immutable

[kannon92]

/retitle Support Elastic JobSets

[kannon92]

I opened up #465 for discussion. We were treating the entire replicated job as immutable. It isn't clear to me what validation logic we want to have for a replicated job. We could go with just changing replicas (name and JobTemplate are immutable).

[k8s-triage-robot]

The Kubernetes project currently lacks enough contributors to adequately respond to all issues.

This bot triages un-triaged issues according to the following rules:

• After 90d of inactivity, lifecycle/stale is applied
• After 30d of inactivity since lifecycle/stale was applied, lifecycle/rotten is applied
• After 30d of inactivity since lifecycle/rotten was applied, the issue is closed

You can:

• Mark this issue as fresh with /remove-lifecycle stale
• Close this issue with /close
• Offer to help out with Issue Triage

Please send feedback to sig-contributor-experience at kubernetes/community.

/lifecycle stale

[ahg-g]

/remove-lifecycle stale

[kannon92]

@andreyvelich @tenzen-y Could you comment on what kind of behavior you'd expect from JobSet for Elastic PyTorch?