Details
-
Bug
-
Resolution: Fixed
-
Major
-
6.5.0
-
Untriaged
-
No
-
KV-Engine Mad-Hatter Beta
Description
Allowing prepares and commits in the same checkpoint (using the diskdockey instead of the ordinary dockey as an identifier) would make things easier in regards to disk backfills where we don't have checkpoint markers. Would allow us to simplify things on the replica side as we currently handle an exception if we attempt to put a prepare and a commit in the same checkpoint when doing the initial backfill. This would have to be expanded to cover both the initial disk backfill and any subsequent resuming disk backfill which requires some extra state to track (or maybe re-use of backfillPhase). If we use diskdockeys instead then we wouldn't have this problem and could revert the error checking to ignore backfill entirely.