KubeCHANGE DELIBERATELY
Workload operations
Select a workload in its resource table before opening an action. Commands capture
the selected context, namespace, name and UID. They require an interactive
selection rather than a --command startup view.
| Command | Resources | Behavior |
|---|---|---|
:scale or :scale 3 |
Deployments, ReplicaSets, StatefulSets | Review replicas, then separately Confirm a guarded /scale patch. |
:restart |
Deployments, StatefulSets, DaemonSets | Review a pod-template timestamp change, then Confirm. |
:rollback or :rollback 1 |
Deployments, StatefulSets, DaemonSets | Restore the template from one explicit retained controller revision. |
:rollout |
Deployments, StatefulSets, DaemonSets | Read actual controller progress, including in read-only mode. |
Cancel has default focus. Input Enter prepares a review without writing. Review identifies the resource/version/effect; separate Confirm consumes one prepared change. Changed input invalidates the review. Counts range from 0 to 2147483647; zero stops replicas. StatefulSet scale-down review discloses PVC deletion when its actual retention policy requests it.
Scale requires workload get and /scale get/patch access, without parent patch
permission. Namespace HPA list access checks ownership; an HPA targeting this
workload blocks manual scale. Absent autoscaling v2 falls back to v1; denied or
invalid checks block preparation. An HPA can appear after the check: this is not
a cross-resource atomic transaction. ReplicaSet owners can reconcile manual replicas.
Restart/rollback require parent get/patch. Paused Deployments refuse these actions.
OnDelete StatefulSets/DaemonSets need deliberate pod deletion; automatic restart/
rollback is refused. Rolling partitions are respected, without resetting strategy.
Rollback reads Deployment ReplicaSets or StatefulSet/DaemonSet ControllerRevisions owned by the captured UID. One explicit positive retained revision is required; absent, ambiguous or unsupported history is refused. Its UID/version is checked again before writing. Only the template is restored; replicas and strategy stay current. Failed StatefulSet recovery may require deliberate removal of a failed pod after rollback; this action does not force-delete pods.
The shared mutation guards provide one-use confirmation, atomic
UID/version tests and no automatic replay. Conflicts need a new review. :writes
retains confirmed/uncertain outcomes without manifest values. Leaving a confirmed
write view does not undo or repeat the request. Context change/exit drains it.
Monitoring reads observed generation, controller counters, Deployment deadline
failure, paused state and StatefulSet revisions/partitions. API confirmation does
not imply rollout completion. Cancel/Escape stops monitoring without claiming an
undo. After five minutes it times out; reopen progress with :rollout.
Sources: Deployment controllers, StatefulSet controllers, DaemonSet controllers. See acceptance evidence for measured scope and limits.