Kubernetes

Kubernetes is a control loop: you declare the state you want, and controllers work continuously to make the cluster match it. Almost everything that feels confusing at first — why a pod came back after you deleted it, why a rollout stalled — follows from that one idea.

The gap between a working cluster and a well-run one is mostly resource limits, probes and rollout strategy. Those three decide whether a bad deploy degrades one pod or takes down the service.