Restart Policies

RestartPolicy determines what happens to a Pod when its containers exit or are terminated.

Available Restart Policies

Always (Default)

Container is always restarted if it exits, regardless of exit code.

spec:
  restartPolicy: Always
  containers:
  - name: app
    image: myapp:1.0

Use Case: Deployments, Services, web servers that should always be running.

OnFailure

Container is restarted only if it exits with a non-zero exit code.

spec:
  restartPolicy: OnFailure
  containers:
  - name: job
    image: batch-job:1.0

Use Case: Jobs that should retry on failure but not restart on successful completion.

Never

Container is never restarted, even if it fails.

spec:
  restartPolicy: Never
  containers:
  - name: one-time-task
    image: init-task:1.0

Use Case: Jobs, one-time scripts, initialization tasks.

Restart Behavior Details

• Restart delay increases exponentially - 100ms, 200ms, 400ms, 800ms, 1.6s, 3.2s, etc.
• Max restart delay is capped at 5 minutes
• Restarts count toward resource quotas
• CrashLoopBackOff state occurs when container keeps crashing

Pod-Level vs Container-Level

RestartPolicy is specified at Pod level and applies to all containers in the Pod.

apiVersion: v1
kind: Pod
metadata:
  name: multi-container-pod
spec:
  restartPolicy: OnFailure
  containers:
  - name: container1
    image: image1:1.0
  - name: container2
    image: image2:1.0

Both containers follow the same restart policy.

Important Notes

• If a container exit code is 0, it's considered successful (no restart with OnFailure)
• Deployments override RestartPolicy for individual container restarts
• StatefulSets and DaemonSets typically use Always
• CronJobs and Jobs typically use Never or OnFailure

Checking Restart Count

kubectl get pods
kubectl describe pod <pod-name>  # Shows 'Restarts' column