#Quickstart

This page walks through a minimal JobSet demo to confirm that the operator is working and to introduce the most common fields of the JobSet custom resource.

#Prerequisites

• The JobSet Operator is installed and the controller pod in jobset-system is Running. See Install JobSet.
• kubectl access to the target cluster.
• A namespace to run the demo in. The examples below use the default namespace; replace it as needed.

#JobSet CR Overview

The most commonly used fields in a JobSet CR are:

#Run a Simple Demo

The example below defines a JobSet with two groups:

• leader: one Job with one pod, simulating the driver process.
• workers: one Job with four parallel pods, simulating the workers.

The JobSet is marked successful as soon as both leader and workers complete.

Save the manifest as jobset-quickstart.yaml:

apiVersion: jobset.x-k8s.io/v1alpha2
kind: JobSet
metadata:
  name: jobset-quickstart
  namespace: default
spec:
  successPolicy:
    operator: All
  replicatedJobs:
  - name: leader
    replicas: 1
    template:
      spec:
        parallelism: 1
        completions: 1
        backoffLimit: 0
        template:
          spec:
            restartPolicy: Never
            containers:
            - name: leader
              image: busybox:1.36
              command:
              - sh
              - -c
              - |
                echo "leader $(hostname) starting"
                sleep 30
                echo "leader done"
  - name: workers
    replicas: 1
    template:
      spec:
        parallelism: 4
        completions: 4
        backoffLimit: 0
        template:
          spec:
            restartPolicy: Never
            containers:
            - name: worker
              image: busybox:1.36
              command:
              - sh
              - -c
              - |
                echo "worker $(hostname) starting"
                sleep 30
                echo "worker done"

Apply it:

kubectl apply -f jobset-quickstart.yaml

#Inspect the Resources

Check that the JobSet has been accepted and child Jobs have been created:

kubectl get jobset jobset-quickstart -n default
kubectl get jobs -n default -l jobset.sigs.k8s.io/jobset-name=jobset-quickstart
kubectl get pods -n default -l jobset.sigs.k8s.io/jobset-name=jobset-quickstart

Expected results:

• One Job named jobset-quickstart-leader-0 with one pod.
• One Job named jobset-quickstart-workers-0 with four pods.
• A headless Service named jobset-quickstart for stable DNS hostnames of the pods.

Tail logs from a pod, for example:

kubectl logs -n default -l jobset.sigs.k8s.io/replicatedjob-name=leader -f

#Verify Completion

After roughly 30 seconds, all pods complete. Confirm the JobSet status:

kubectl get jobset jobset-quickstart -n default -o jsonpath='{.status.conditions}'

A successful JobSet reports a condition with type: Completed and status: "True". Child Jobs report COMPLETIONS reaching their configured value (1/1 for leader and 4/4 for workers).

#Clean Up

Remove the demo resources:

kubectl delete -f jobset-quickstart.yaml

Deleting the JobSet cascades to its child Jobs, pods, and the headless Service that was created for the workload.

#Next Steps

For more advanced patterns, refer to the upstream examples and documentation:

• JobSet concepts: https://jobset.sigs.k8s.io/docs/concepts/
• Upstream examples: https://github.com/kubernetes-sigs/jobset/tree/main/examples
• Success and failure policies: configurable success criteria (Any/All) and per-error-type failure handling.
• Startup sequencing: spec.startupPolicy.startupPolicyOrder: InOrder enforces a leader-before-workers startup order.
• Exclusive placement: set alpha.jobset.sigs.k8s.io/exclusive-topology on the JobSet to colocate each child Job's pods inside a single topology domain.
• Kueue integration: pair JobSet with Alauda Build of Kueue to manage queues and resource quotas for batch workloads.