-
Type:
Bug
-
Resolution: Fixed
-
Priority:
Major - P3
-
Affects Version/s: 9.1.0-rc0
-
Component/s: None
-
None
-
Catalog and Routing
-
Fully Compatible
-
ALL
-
CAR Team 2026-11-09
-
0
-
None
-
None
-
None
-
None
-
None
-
None
-
None
Issue
During a DSC step down, DDLs can bubble up a WriteConflict error to the user (instead of the critical section code internally retrying it).
Analysis
Acquiring, promoting or releasing the critical section writes via DBDirectClient with locks held.
In a DSC cluster, during stepdown we may hit a WriteConflict there due to WT-17895 ("step down straddler").
Because we hold locks, the query code called by the DBDirectClient can't yield so it can't retry the WriteConflict.
It's up to the caller to do it, but it currently doesn't and thus the WriteConflict is incorrectly bubbled up to the user.
Proposed fix
Add a writeConflictRetry loop around the CS acquire/promote/release code, including the DBDirectClient and the locks.
- is related to
-
WT-17895 Core Async Stepdown Implementation
-
- Closed
-