Open
Description
If a candidate pod is not allowed (e.g., another one was chosen, in a different target cluster), by design, it fails the PreBind candidate scheduler plugin by timeout. Unfortunately, this generates errors in the scheduler's logs, which are misleading, e.g.
E0215 19:45:37.807953 1 framework.go:967] "Failed running PreBind plugin" err="timed out waiting for the condition" plugin="candidate" pod="default/follow-7564df7669-2v5n6-2f456"
E0215 19:45:37.808232 1 factory.go:225] "Error scheduling pod; retrying" err="running PreBind plugin \"candidate\": timed out waiting for the condition" pod="default/follow-7564df7669-2v5n6-2f456"
I0215 19:45:37.808304 1 factory.go:231] "Pod doesn't exist in informer cache" pod="default/follow-7564df7669-2v5n6-2f456" err="pod \"follow-7564df7669-2v5n6-2f456\" not found"
E0215 19:45:37.815203 1 scheduler.go:322] "Error updating pod" err="pods \"follow-7564df7669-2v5n6-2f456\" not found" pod="default/follow-7564df7669-2v5n6-2f456"
Note: regarding the two last lines, by timeout, the candidate pod actually doesn't exist anymore, because it's been deleted by the proxy scheduler.
We should find a way to silently abort PreBind.
Metadata
Metadata
Assignees
Labels
No labels