Don't Panic — Kubernetes Operations

Kubernetes operations is the work of keeping the cluster itself alive while applications continue their busy little lives on top of it. This is not the same job as deploying an application. One job concerns Deployments and Services; the other concerns upgrades, nodes, certificates, capacity, backups, and the machinery that remembers where everything is. The machinery has not volunteered to remember its own instructions, naturally.

The useful picture is cattle around a precious core. A node is replaceable. You can stop scheduling onto it, move eligible Pods away, update or replace the machine, and put it back. The precious core is etcd, the cluster's memory of every API object. Nodes are expected to come and go. etcd is expected to be protected, backed up, and restored in rehearsal before it is needed in anger.

The first surprise is that an upgrade is not a button with a reassuring progress bar. Kubernetes components follow a version skew policy: the API server leads, kubelets may lag but never lead, and minor versions are crossed one at a time. That is why a quiet maintenance calendar matters. Release notes, deprecated API cleanup, and add-on compatibility are not preamble; they are the part that stops a routine upgrade becoming an archaeological dig.

The operational verb to remember is drain. First cordon the node so new Pods do not arrive. Then drain it so eligible Pods are evicted while PodDisruptionBudgets enforce the availability limits declared for the workload. A blocked drain is not the cluster being difficult for sport. It is the cluster pointing at an unmet condition: perhaps one replica, insufficient capacity, or a disruption budget that cannot be honored. The correct response is inspection, not force.

Managed control planes make the precious-core duties smaller, which is welcome. They do not make the remaining work vanish into a tasteful mist. Node strategy, upgrade cadence, add-ons, workload disruption readiness, and application-state backup still belong to somebody. Automation can reduce repeated effort, but it also becomes another component with an upgrade path and failure modes.

Read the Intro for the full map of the control plane, nodes, etcd, certificates, and capacity. Use Slides when the relationships need to fit on one page. Keep the Cheatsheet nearby for skew rules, drain commands, and backup layers. Then use the practice reference and exercise to make a maintenance change in a disposable cluster, where the only endangered asset is an afternoon.