Workloads & Scheduling

A Deployment manages a ReplicaSet, and the ReplicaSet manages the Pods. The Deployment controller creates a new ReplicaSet on each template change and uses ReplicaSets to roll out and roll back.

spec.replicas (desired Pod count), spec.selector.matchLabels (which Pods the Deployment owns), and spec.template (the Pod template). The selector must match the labels in the template, or the API rejects the Deployment.

The selector determines which Pods the controller owns and reconciles. If the selector and template labels disagree, the Deployment cannot adopt the Pods it creates; the API server validates this and rejects the manifest.

kubectl create deployment web --image=nginx. Add --replicas=N to set the replica count, or generate a manifest with --dry-run=client -o yaml.

kubectl scale deployment <name> --replicas=N. The controller adds or removes Pods to reconcile to the new desired count.

kubectl create is imperative and fails if the object already exists. kubectl apply -f is declarative: it creates the object if missing and otherwise reconciles existing objects to the manifest, tracking the last-applied configuration.