A successful apply means the API accepted the change. It takes a separate check to learn whether the new Pods are serving traffic. Put that check in the deployment workflow so a failed rollout produces a clear stop.

sh
kubectl -n staging rollout history deployment/api
kubectl -n staging rollout status deployment/api --timeout=3m
kubectl -n staging get pods -l app=api -o wide

Choose the smoke test before deploying

For a catalog service, read one known item through the same hostname and load balancer that clients use. Check the response shape and a field that should exist. Capture the version and request ID where possible so you can tie failures to the new replica.

When the rollout stalls, read events for image pulls, scheduling, and probe failures. If Kubernetes reports a progress deadline, the controller surfaces the problem for an operator or deployment tool to handle. Decide whether to fix forward or restore a known good revision.

Know what a rollback restores

sh
# Replace 12 with the tested revision from rollout history.
kubectl -n staging rollout undo deployment/api --to-revision=12
kubectl -n staging rollout status deployment/api --timeout=3m

The rollback restores the Deployment’s Pod template. A database migration, changed external secret, or API contract can outlive it. During a schema change, make sure the old and new application versions can both run against the transitional schema before sending any production traffic to the new version.

References