Jobs and CronJobs

Kubernetes Deployments are designed for long-running services: web servers, APIs, background workers that process indefinitely. But many workloads are finite. Generate a nightly report. Process a batch of 10,000 images. Send weekly email digests. Clean up expired sessions. Run a database backup.

These tasks need to start, do their work, and stop. Running them as a Deployment is wrong in two ways. First, the Deployment controller restarts the pod after it exits, which means your completed backup starts running again for no reason. Second, the Deployment keeps the pod alive permanently, consuming CPU and memory while doing nothing between runs.

Jobs are Kubernetes' answer to finite tasks. A Job creates one or more pods, waits for them to complete successfully, and marks itself as done. The pod is not restarted on success. A CronJob is a Job on a schedule, like cron on Linux but with Kubernetes' reliability guarantees: retry on failure, concurrency control, deadline enforcement, and history tracking.

Stripe runs thousands of CronJobs across their Kubernetes fleet for tasks like billing cycle processing, compliance report generation, and data pipeline orchestration. Each CronJob has explicit timeouts, retry limits, and monitoring, because a missed nightly billing run means real revenue impact.

Job: Runs a task to completion.

apiVersion: batch/v1
kind: Job
metadata:
  name: db-backup
spec:
  completions: 1        # How many successful pod runs needed
  parallelism: 1        # How many pods run simultaneously
  backoffLimit: 3       # Retry attempts before marking Failed
  activeDeadlineSeconds: 3600  # Kill the job after 1 hour
  ttlSecondsAfterFinished: 86400  # Clean up 24h after completion
  template:
    spec:
      restartPolicy: Never
      containers:
      - name: backup
        image: postgres:16
        command: ["pg_dump", "-h", "db", "-o", "/backup/dump.sql"]

CronJob: Runs a Job on a schedule.

apiVersion: batch/v1
kind: CronJob
metadata:
  name: nightly-backup
spec:
  schedule: "0 2 * * *"           # Every day at 2 AM UTC
  concurrencyPolicy: Forbid       # Skip if previous run still going
  startingDeadlineSeconds: 600    # Give up if 10 min late
  successfulJobsHistoryLimit: 3   # Keep last 3 successful Jobs
  failedJobsHistoryLimit: 1       # Keep last 1 failed Job
  jobTemplate:
    spec:
      # ... same as Job spec above