Back to Kubernetes Course

Why Jobs Exist

Deployments and StatefulSets are built around the assumption that a Pod should run forever — if it exits, it's restarted. That's the wrong behavior for a one-off task: a database migration, a batch report, or a data-processing script that's supposed to finish and exit successfully. A Job is the controller designed for run-to-completion workloads.

---

Basic Job

apiVersion: batch/v1
kind: Job
metadata:
  name: db-migration
spec:
  template:
    spec:
      containers:
        - name: migrate
          image: myapp-migrations:1.0
          command: ["./migrate.sh"]
      restartPolicy: Never
  backoffLimit: 4

• restartPolicy inside a Job's Pod template must be Never or OnFailure — never Always (the Deployment default), since a Job Pod is expected to terminate.
• backoffLimit caps how many times Kubernetes retries a failed Pod before marking the Job itself as failed (with exponential backoff between retries).

kubectl apply -f job.yaml
kubectl get jobs
kubectl logs job/db-migration
kubectl describe job db-migration

Once the Job's Pod exits with status 0, the Job is marked Complete and the Pod is not restarted — this is the key behavioral difference from every other controller covered so far.

---

Parallel Jobs

Jobs can run multiple Pods in parallel to process a batch of work faster:

spec:
  completions: 10    # Total successful Pod completions needed
  parallelism: 3     # Max Pods running at once
  template:
    spec:
      containers:
        - name: worker
          image: batch-worker:1.0
      restartPolicy: OnFailure

• completions: 10 with parallelism: 3 means Kubernetes runs up to 3 Pods at a time until 10 have succeeded in total.
• Omitting completions entirely turns this into a work queue pattern, where the Job is considered done once any one Pod exits successfully (each Pod is expected to pull its own unit of work from an external queue).

---

TTL Cleanup

Completed Jobs (and their Pods) stick around by default, which clutters kubectl get jobs output over time. ttlSecondsAfterFinished automatically deletes a Job (and its Pods) some time after it finishes:

spec:
  ttlSecondsAfterFinished: 3600   # Clean up 1 hour after completion

---

CronJobs

A CronJob creates Jobs on a repeating schedule — the Kubernetes equivalent of a Unix cron entry, used for nightly backups, scheduled report generation, periodic cache cleanup, and similar recurring tasks.

apiVersion: batch/v1
kind: CronJob
metadata:
  name: nightly-backup
spec:
  schedule: "0 2 * * *"          # 2:00 AM every day, in the cluster's configured time zone
  concurrencyPolicy: Forbid       # Don't start a new run if the previous one is still going
  successfulJobsHistoryLimit: 3
  failedJobsHistoryLimit: 1
  jobTemplate:
    spec:
      template:
        spec:
          containers:
            - name: backup
              image: db-backup-tool:1.0
              command: ["./backup.sh"]
          restartPolicy: OnFailure

• schedule uses standard cron syntax (minute hour day-of-month month day-of-week).
• concurrencyPolicy controls what happens if a scheduled run is still executing when the next one is due: Allow (default, runs concurrently), Forbid (skips the new run), or Replace (cancels the running one and starts fresh).
• successfulJobsHistoryLimit / failedJobsHistoryLimit cap how many old Job records are kept around for inspection (kubectl logs on a past run), rather than growing unbounded.

kubectl get cronjobs
kubectl get jobs --watch                     # See Jobs appear as the schedule fires
kubectl create job manual-run --from=cronjob/nightly-backup   # Trigger a run immediately, for testing

---

Jobs, CronJobs, and Pod Health Checks

Unlike long-running Deployments, Job Pods are not expected to pass liveness/readiness probes to be considered "up" — success is defined entirely by the container's exit code. Exit code 0 means success; anything else is treated as a failure subject to backoffLimit retries. Keep this distinction in mind: a script that runs forever and never exits will never let a Job complete, no matter how "healthy" it looks.