-
Type:
Task
-
Resolution: Unresolved
-
Priority:
Major - P3
-
None
-
Affects Version/s: None
-
Component/s: Cursors, Transactions
-
None
-
Storage Engines - Foundations
-
128.6
-
None
-
8
Business outcome
Allow the affected untimestamped snapshot reads to survive stepdown with their original snapshot. Today, their WT_ROLLBACK can become a customer-visible read failure, breaching the transparent-stepdown contract.
Why reads roll back
Follower checkpoints mask local transaction IDs, so the original leader snapshot cannot exclude later commits. Retrying the bind does not restore that view, and replacing the snapshot can change results already returned.
Direction to investigate
Evaluate Alex Gorrod's proposal to keep eligible pre-stepdown snapshots on this node's final live stable view, retaining local transaction IDs. This has not been prototyped.
- Find and open the correct view, including tables first touched after stepdown and cursor reopens.
- Retain read-only access through stepdown, sweep and eviction, limited to eligible snapshots.
- Keep follower checkpoint pickup from changing that view; assess effects on follower progress.
- Check whether history-store lookups need the same treatment.
- Assess cleanup through WT_GEN_DISAGG_ROLE after the last reader finishes. Continue refusing snapshots spanning multiple role changes.
Next step
Record viability and required tests. If the investigation is quick and the fix contained, implement and test here. Otherwise, create a linked implementation ticket.
For more detail, see the design document.