Details
-
Task
-
Resolution: Unresolved
-
Major
-
None
-
None
-
5
Description
The root cause of K8S-1329 is that some resources, when generated, depend on things that may have been deleted.
- External services rely on the service list for port filtering
- Recovery relies on the name, volume mounts and pod template etc.
This stuff disappears we have to make compromises that are prone to go wrong (things that should be filled in aren't), or just panic and get stuck unrecoverable (I should be able to trigger this condition by renaming a class while all pods are down – esoteric but customers will be customers!)
The simple – aka correct – fix is to persist this data so it's available when the class gets deleted. This makes the handling of these conditions far more simple and less buggy in the long term.
Attachments
Issue Links
- relates to
-
K8S-1329 Removal of server class using NodePorts fails
- Closed