Rollout controller

A rollout controller moves traffic from the old version to the new one in steps, checks the new one against metrics at each step, and rolls back when a check fails. It is part of the deployment platform, found in its configuration rather than assumed.

Finding it

• The deployment manifests: a Rollout resource (Argo Rollouts) or a Canary resource (Flagger) beside or in place of a plain Deployment.
• The analysis: an AnalysisTemplate in Argo Rollouts, or the metric checks in Flagger, naming the queries and their limits.
• Access through kubectl, and the controller's own plugin or dashboard (kubectl argo rollouts).

Using it

• Follow a rollout to its end: kubectl rollout status deployment/<name>, or kubectl argo rollouts get rollout <name> --watch.
• Compare the new version with the old under the same traffic, not only with a fixed limit.
• Set a step with no metric data to stop the rollout, rather than let it through.
• Check the rollback itself (kubectl rollout undo, kubectl argo rollouts abort): the old version is healthy again afterwards.

Examples: kubectl rollout (Kubernetes), Argo Rollouts, Flagger; Kayenta for the canary analysis alone (Spinnaker).