There is a race condition in the current order, which we first mark the prepared txn as aborted, copy the previous update to the update chain, and then fix the history store record.
We may race with other sessions adding more updates and checkpoint moving them to the history store.
The solution is to change the order to first append the previous update to the update chain, changed the status of the prepared update, and then fix the history store record.
- depends on
-
WT-6245 Handle aborting prepared onpage tombstones
-
- Closed
-
- is depended on by
-
WT-6167 format failure, cursor out-of-order return on a cursor prev operation
-
- Closed
-
-
-
- Closed
-
- is duplicated by
-
-
- Closed
-