Details
-
Bug
-
Status: Open
-
Major
-
Resolution: Unresolved
-
None
-
None
Description
I believe a race condition exists that will cause a disk store backup to not backup one or more krf files for the oplogs it backs up. You can still restore from this backup but the recovery may go much slower.
Backup should be changed to wait for the async krf creation to complete instead of skipping over that krf file. I think this can be done in Oplog.java at the spot it currently calls "hasKrf". I think that if it instead first called "finishKrf" that it would wait for an in-progress krf creation to finish.