Details
-
Bug
-
Status: Resolved
-
Major
-
Resolution: Fixed
-
2.0.0
-
None
Description
We see occasional instances where we kill the workers, wait for them to die, force delete the worker directory successfully, and the heartbeat directory is then re-created.
It looks like when new local states get created, these will create the directory. There is probably a race condition between the worker calling LocalState VersionStore mkdirs() and being killed and the supervisor directory cleanup.
Containers already guarantee this directory exists, so a fix could be to add an option to allow the LocalStates to create the dir or not.
Attachments
Issue Links
- links to