flagger icon indicating copy to clipboard operation
flagger copied to clipboard

Raise the priority of rollback check

Open imuxin opened this issue 3 years ago • 2 comments

Describe the bug

Some errors might happen in canary initialization, like primary or canary deploy not running because of image pull failure. Then we want to rollback to cancel canary initialization. But it does not work, because the rollback check is after the deployment ready check task. ref code: https://github.com/fluxcd/flagger/blob/main/pkg/controller/scheduler.go#L331

To Reproduce

Expected behavior

To raise the priority of rollback check before deployment ready check. Or a better strategy?

imuxin avatar Sep 16 '22 10:09 imuxin

If the initial health check fails, then rollback is triggered automatically after spec.progressDeadlineSeconds has passed.

stefanprodan avatar Sep 19 '22 10:09 stefanprodan

Well, passed spec.progressDeadlineSeconds can stop to retry initial health check, and I believe it works. In addition, I think it's not a precondition to trigger rollback. FYI, rollback is a remedy action, which can happen in any time of the canary flow tasks.

imuxin avatar Sep 20 '22 06:09 imuxin