A Failed Rollout Can Keep Retrying

What the deadline reports

A Deployment can stall while bringing up its newest ReplicaSet. Image pull errors, failed readiness probes, insufficient quota, and application misconfiguration are among the possible causes. The .spec.progressDeadlineSeconds field tells the Deployment controller how long to wait without progress before reporting the stall. When the deadline is exceeded, the Deployment gets a Progressing condition with status False and reason ProgressDeadlineExceeded. Kubernetes documents the condition and its causes.

The condition describes rollout progress. It does not, by itself, tell you whether every old Pod has stopped serving. The documentation shows a Deployment with Available: True and Progressing: False at the same time. Check both conditions before judging the service’s current state. See the failed Deployment example.

What happens afterward

Kubernetes reports the missed deadline without automatically rolling back the Deployment. The rollout can still finish if the problem is resolved. In the documentation’s quota example, satisfying the quota allows the controller to complete the rollout. The Progressing condition then changes to True with reason NewReplicaSetAvailable. A higher-level orchestrator may use the failure condition to trigger a rollback, so its behavior also matters. See the controller behavior and recovery example.

This distinction matters for automation. kubectl rollout status returns an error when the progress deadline is exceeded. That error is a useful signal to investigate. It is not evidence that Kubernetes has frozen the Deployment or restored the previous revision. Kubernetes describes the command’s result.

What to do

1. Run kubectl describe deployment NAME and read the conditions and events. Check the new Pods and ReplicaSet for the cause, such as an image pull error or a quota failure.
2. Fix the cause, then check kubectl rollout status deployment/NAME and the Deployment conditions again. A rollout that completes reports NewReplicaSetAvailable.
3. If the new revision needs to be withdrawn, inspect the rollout history and choose a known stable revision. Kubernetes supports kubectl rollout undo deployment/NAME for that action. The Deployment guide shows these checks and the rollback command.