Ship a new image version with zero downtime, watch the rollout, and roll back instantly if it fails.
Change the image: kubectl set image deploy/web web=nginx:1.28 (or re-apply the manifest)
Without a readinessProbe Kubernetes sends traffic to new pods as soon as they start, so a broken version can take the service down before you notice.