Mastering Kubernetes Jobs: Serial & Parallel Execution with Real‑World Examples

Job Controller Overview

The Job controller runs Pods that perform a one‑time task; when the container exits successfully the Pod reaches a Completed state, while failures trigger a restart based on the configured restart policy. If a Pod is terminated unexpectedly due to node failure, it is rescheduled.

Job Execution Modes

Jobs can run either serially (single work queue) or in parallel (multiple work queues). In the serial mode only one Pod is active at a time, while the parallel mode can launch several Pods simultaneously, each handling a portion of the total work.

Creating a Job Object

The spec of a Job only requires a template field; a selector is generated automatically, making it similar to a Deployment except for this omission.

1. Define the Job manifest – the example below uses the busybox image and sleeps for 120 seconds before exiting.

cat busybox-job.yaml</code>
<code>apiVersion: batch/v1</code>
<code>kind: Job</code>
<code>metadata:</code>
<code>  name: busybox-job</code>
<code>spec:</code>
<code>  template:</code>
<code>    spec:</code>
<code>      containers:</code>
<code>      - name: busybox</code>
<code>        image: busybox:latest</code>
<code>        command: ["/bin/sh", "-c", "sleep 120s"]</code>
<code>      restartPolicy: Never

2. Apply the Job kubectl apply -f busybox-job.yaml 3. Check Job and Pod status

kubectl get job -o wide</code>
<code>kubectl get pods -l job-name=busybox-job -o wide

After 120 seconds the Pod reaches the Completed state.

The default spec.restartPolicy for Pods is Always , which is unsuitable for Jobs; use Never or OnFailure instead.

Serial Job Example

Set parallelism: 1 and completions: 5 to run five tasks one after another.

cat busybox-serial-job.yaml</code>
<code>apiVersion: batch/v1</code>
<code>kind: Job</code>
<code>metadata:</code>
<code>  name: busybox-job</code>
<code>spec:</code>
<code>  parallelism: 1</code>
<code>  completions: 5</code>
<code>  template:</code>
<code>    spec:</code>
<code>      containers:</code>
<code>      - name: busybox</code>
<code>        image: busybox:latest</code>
<code>        command: ["/bin/sh", "-c", "sleep 20s"]</code>
<code>      restartPolicy: OnFailure

kubectl apply -f busybox-serial-job.yaml

Monitor the Job with kubectl get job -w or watch the Pods.

Parallel Job Example

Set parallelism and completions to the same value (e.g., 5) to launch five Pods concurrently.

cat busybox-parallel-job.yaml</code>
<code>apiVersion: batch/v1</code>
<code>kind: Job</code>
<code>metadata:</code>
<code>  name: busybox-job</code>
<code>spec:</code>
<code>  parallelism: 5</code>
<code>  completions: 5</code>
<code>  template:</code>
<code>    spec:</code>
<code>      containers:</code>
<code>      - name: busybox</code>
<code>        image: busybox:latest</code>
<code>        command: ["/bin/sh", "-c", "sleep 20s"]</code>
<code>      restartPolicy: OnFailure

kubectl apply -f busybox-parallel-job.yaml

All Pods start at the same time; their creation timestamps are identical.

Deleting a Job

Completed Pods free resources automatically, but you can delete them manually if needed. To prevent endless restarts, configure backoffLimit (max retry attempts) and activeDeadlineSeconds (overall timeout).

cat busybox-failure-job.yaml</code>
<code>apiVersion: batch/v1</code>
<code>kind: Job</code>
<code>metadata:</code>
<code>  name: busybox-job</code>
<code>spec:</code>
<code>  backoffLimit: 5</code>
<code>  activeDeadlineSeconds: 100</code>
<code>  parallelism: 1</code>
<code>  completions: 5</code>
<code>  template:</code>
<code>    spec:</code>
<code>      containers:</code>
<code>      - name: busybox</code>
<code>        image: busybox:latest</code>
<code>        command: ["/bin/sh", "-c", "sleep 30s"]</code>
<code>      restartPolicy: OnFailure

Apply the manifest and monitor; the Job will fail after five retries or after 100 seconds, whichever occurs first.

☆ END ☆