Kubernetes can roll out a new version without dropping a single request — but it will also happily cause a brief outage if you let it. The difference comes down to a handful of settings that tell the cluster what "healthy" means.
Health checks are the foundation
Readiness probes tell Kubernetes when a pod is actually ready to serve traffic, and liveness probes tell it when a pod is stuck and should be restarted. Without accurate probes, the cluster routes traffic to pods that are not ready — the classic cause of "downtime" during a deploy.
Rollout strategies
Once health is honest, you choose how new versions reach users.
- Rolling update — replace pods gradually while keeping enough healthy ones serving.
- Blue-green — stand up the new version alongside the old, then switch traffic at once.
- Canary — send a small slice of traffic to the new version, watch the metrics, then ramp up.
Make rollback automatic
The final piece is not getting the deploy right — it is getting recovery right. Tie your rollout to real signals (error rate, latency) so a bad version is rolled back automatically before most users ever notice.