In Kubernetes, restarting a deployment is a common way to refresh application pods, apply updated configurations, or recover from a stuck pod without changing the deployment definition itself. Here's everything you need to know about restarting deployments in Kubernetes.
Why Restart a Deployment?
There are several scenarios where restarting a deployment is useful:
- When we want to apply ConfigMaps or Secrets. If we change ConfigMaps or Secrets, pods don't automatically pick up these changes. This is a really good reason to restart a pod.
- Restart a pod that got stuck or on an unstable state: If pods are failing, stuck in a crash loop, or misbehaving, restarting can bring them back to a healthy state.
- Force application refresh: In cases where you need to reload app code, even if the image tag hasn’t changed.
How to restart:
Chances are that you just don't have Pods running by themself and that you have a Replicaset and a deployment making sure that your pods are running. To restart all the pods that a deployment is managing using kubectl, you need to type this command:
kubectl rollout restart deployment <deployment-name>Example:
kubectl rollout restart deployment my-appThis command triggers a rolling restart—terminating old pods and gradually spinning up new ones while maintaining availability. What Happens Behind the Scenes? The rollout restart command updates the deployment’s spec.template.metadata.annotations field with a new timestamp. This forces Kubernetes to treat the deployment as updated, triggering the rolling update strategy defined in your deployment.
Other Ways to Trigger a Restart
While kubectl rollout restart is the cleanest approach, here are some alternatives:
1. Patch the Deployment with a Dummy Change
kubectl patch deployment <deployment-name> -p \
"{\"spec\":{\"template\":{\"metadata\":{\"annotations\":{\"kubectl.kubernetes.io/restartedAt\":\"$(date +%s)\"}}}}}"This manually simulates what rollout restart does under the hood.
2. Update an Environment Variable
Changing an env var forces a new pod template hash, triggering a rollout:
env:
- name: FORCE_RESTART
value: "timestamp-$(date +%s)"3. Delete Pods (Not Recommended for Rollouts have downtime)
kubectl delete pod -l app=<app-label>This will force the deployment to recreate the pods, but it's not a rolling restart and may cause downtime.