Health and resources
Liveness probes restart a wedged container, readiness probes remove it from load balancing, and startup probes protect slow boots. Requests drive scheduling while limits cap usage — a memory limit breach means OOMKill, and aggressive CPU limits cause silent throttling.
Scaling and scheduling
The Horizontal Pod Autoscaler adds replicas on CPU or custom metrics; the Vertical Pod Autoscaler tunes requests; the Cluster Autoscaler adds nodes. PodDisruptionBudgets, affinity rules and topology spread constraints keep replicas apart during upgrades and failures.
Security and delivery
Use RBAC with per-namespace service accounts, network policies for east–west traffic, and Pod security standards to block privileged containers. Deploy with Helm or Kustomize, adopt GitOps (Argo CD, Flux) so the cluster reconciles from git, and roll out canary or blue-green with automated rollback.