Workloads

The controller is the unit of truth

Do not operate pods as individuals. Operate the controller: it owns the template, the rollout, the scaling posture, and the steady-state you will converge back to.

When a workload fails, you must identify whether the failure is template-level (config, image, probes, resources) or environment-level (capacity, policy, dependencies).

• Deployments roll forward and back; ReplicaSets are the historical record.
• DaemonSets enforce presence per node and amplify blast radius when misconfigured.
• StatefulSets carry identity and storage—treat them as operational contracts.

A rollout-first diagnostic sequence

If you can’t describe a rollout, you can’t safely change it.

kubectl rollout status deploy/<name> -n <ns>
kubectl describe deploy/<name> -n <ns>
kubectl get rs -n <ns> -o wide
kubectl describe pod <pod> -n <ns>