Open TomBillietKlarrio opened 1 month ago
Hm, i need to investigate on that in a test cluster.
If restarting nodes triggers this, then i should be able to reproduce it.
But we should also be able to reproduce it then by killing the provider pod right? (You can also start multiple replicas for HA, maybe that changes it) - However, that should not matter because the real state should be stored in ETCD and not in memory.
I did try quite a lot of different things to reproduce it, but have not been able to find a root cause what exactly triggers it unfortunately. Just killing the keycloak-crossplane (or crossplane) pod does not trigger it. We're not indeed trying to run 2 instances for the keycloak-crossplane provider to see if that helps
Hi,
We have quite some openid clients we create trough crossplane. However, from time to time, they seem to loose the
external-name
annotation in the metadata section, causing crossplane to give errors on those objects as it will try to create new ones in keycloak and can't do that. We're unsure why this happens, we seems to trigger it sometimes when we restart some k8s nodes during a software release, but haven't been able to pinpoint the exact root. I've extracted some logs that are related to a client at the moment it got corrupted. It looks like it looses the object from the cache, and tries to rebuild it? But then why would it remove the external-name annotation? Any help is appreciated.