Configure Automated Rollbacks in the Kubernetes Provider

There are clear benefits to having your stored manifests (in Git, GCS, etc…) match as closely as possible to what’s running in your cluster: you have auditability, versioned changes, and a high confidence in what code you have running. However, in some cases, pushing changes from someplace like GitHub through your build system all the way to production can simply take too long, and an escape hatch is needed when something has gone seriously wrong. Automated rollback is that escape hatch.

When using a Deployment object, you can see the history of rollouts in the “Clusters” tab:

Spinnaker exposes “Undo Rollout” functionality in two places, in the Clusters tab , and as a pipeline stage .

Ad-hoc rollbacks

In cases where you see something is immediately wrong and isolated to a resource in the “Clusters” tab, you can select “Undo Rollout” from the “Actions” dropdown:

And select the healthy revision to make active again:

Notice that the old configuration (version 3) will be rolled forward into a new version (version 5):

Automated rollbacks

You can also configure automated rollbacks inside of Spinnaker pipelines. These stages are best configured to run when other stages or branches fail, indicating that something has gone wrong in your rollout.

One parameter to watch out for is Revisions Back, which counts how many revisions the current active revision should be rolled back by. If you have the following state:

nginx-deployment-2d8178b77 (Revision 5) # active
nginx-deployment-7bdd110f7 (Revision 4)
nginx-deployment-0b13cc8c1 (Revision 1)

And roll back by “1” revision, (Revision 4) will be active again. Roll back by “2” revisions and (Revision 1) will be active again.

Keep in mind that Kubernetes will implicitly rollforward the old configuration, creating (Revision 6) in both cases.

Parameterized rollbacks

It’s worth mentioning that you can parameterize the target resource to roll back. It can point to something specified using pipeline parameters, and upstream deploy stage, or another stage’s outputs. See more details in the pipeline expressions guide .

Pitfalls

If the artifacts deployed in your manifest (Docker image, ConfigMap, Secret, …) are not versioned, rolling back your manifest will likely not roll back your code or config changes. See more details here .