Closed audrey-mux closed 3 months ago
I think this is probably another example of this bug: https://github.com/argoproj/argo-cd/issues/18212
I think the comment by @todaywasawesome here was prescient: a generator failure might mean that it's time to stop the world.
I'm going to revert #17062 until the author has time to make it safer.
Ah, yep that’s likely it. It’s happening even with cluster generators
Hey @crenshaw-dev
Was wondering if there's an ETA on #18781 getting merged and a new release cut?
@audrey-mux I'll cherry-pick the change to 2.12 and 2.12 and plan to cut a release today or tomorrow.
Checklist:
argocd version
.Describe the bug
We had a sudden deletion of a handful of applications created by appsets. The applicationset controller looks like it lost its connection to the kube-api service for less than a second. This caused errors in the application generation. The connection issue resolved quickly, but within a few seconds of the event the affected applications were deleted by the applicationset controller.
They were recreated a few seconds later, but the damage was done.
Since set the application controller policy to
create-update
and addingTo all applicationset manifests. Will that be enough to prevent deletion if this sort of error were to happen again?
To Reproduce
Break the applicationset controllers access to the local kube-api service.
Expected behavior
Expected at a minimum a retry, not application deletion.
Screenshots
Version
Logs
The API connection errors
and the deletion (app names partially redacted)