Details
-
Bug
-
Resolution: Unresolved
-
Major
-
6.5.2, 6.6.5, 7.0.4, 7.1.1
-
None
-
Triaged
-
0
-
No
Description
We purge aborts in the same way as tombstones, as they must be kept for the duration of the metadata purge interval to ensure stream continuity (to clean up any prepares that have not been committed). This means that purging an abort moves the purge seqno forwards. Both prepares and aborts are filtered for any non-replication clients. If a cluster were in a bad state for the duration of the metadata purge interval and the only operations that occurred were prepares and their aborts, then it is possible for a client to be disconnected and reconnected behind the purge seqno, even if they have seen every item that they ought to.