Resharding's management of the oldest timestamp pin scans over all of the documents in the donor collection (one document per concurrent resharding operation), but only writes to a single document. In a future world where there can be multiple resharding operations happening, this can result in write-skew that can manifest as "leaking" history until a restart, rollback or new resharding operation.
This ticket should also uncomment the locking invariant.
- is depended on by
-
SERVER-55680 Generate, pin, and replicate minFetchTimestamp on donor shards in resharding atomically
- Closed
- related to
-
SERVER-57195 Convert resharding document deletion to use PersistentTaskStore instead of creating a WriteUnitOfWork
- Closed