Closed scrayos closed 1 year ago
Thanks for the bug! Looks like something has changed in recent versions of Kubernetes. I'll take a look -- I wanted to look at this anyway for another reason.
Just noting I'm seeing this consistently within 1.27.x clusters. The test namespaces aren't deleting.
I've also filed this as a bug with Kubernetes: https://github.com/kubernetes/kubernetes/issues/119662 - since the new feature (Aggregated Discovery) seems to break backward compatibility.
I have faced with that issue after upgrading agones release from 1.34 to 1.41. The cause was misconfiguration - I accidentally turn off caBundle for extensions, so it became a reason of that behavior.
agones:
extensions:
allocationApiService:
disableCaBundle: true
So, if you need to disable it, you have to manage you own certs for that.
What happened: Any namespace deletion is stuck in
Terminating
even though Agones is still installed and theallocation.agones.dev/v1
API is available. The namespaces are stuck in this state indefinitely and can only be deleted by removing theNamespaceDeletionDiscoveryFailure
finalizer. This is because the finalizer fails:The apiserver log reports
DiscoveryManager: Failed to download discovery for agones-system/agones-controller-service:443: 404 404 page not found
.I've looked in the Kubernetes implementation and this error is thrown here. Any error returned by this method leads to the error in the
NamespaceDeletionDiscoveryFailure
, reporting stale GroupVersion discovery.And while looking into the
webhooks
port of theagones-controller-service
, indeed404
is returned for/apis
.What you expected to happen: I'd expect namespaces to delete normally, even with Agones installed.
How to reproduce it (as minimally and precisely as possible):
Anything else we need to know?:
Helm-Values:
Environment:
kubectl version
): 1.26.5