-
Type:
Bug
-
Resolution: Unresolved
-
Priority:
Major - P3
-
None
-
Affects Version/s: None
-
Component/s: Truncate
-
Storage Engines - Foundations
-
2,934.109
-
SE Foundations - 2026-09-29
-
8
This is a spin-off of WT-17272, which described three correctness gaps in the follower's in-memory truncate-list write-conflict detection. That ticket was closed with items (2) and (3) resolved. Item (1) was explicitly dropped from the PR scope and is still TODO.
Gap (1) — search_near ingest truncate misses uncommitted ingest updates
The follower truncate path positions via search_near before calling _wt_range_truncate. Because search_near is a read operation, it does not trigger _wt_txn_update_check. As a result, write conflicts against uncommitted ingest updates that fall within the truncate range are silently missed — conflicting operations can commit when one should roll back.
Proposed fix location: __wt_insert_truncate_entry — the conflict check needs to be invoked there so that truncate-range positioning via search_near is covered.
- blocks
-
WT-18641 Enable fast truncate in test/format disaggregated switch mode runs
-
- Backlog
-
-
WT-18642 Expand write conflict test coverage for layered fast truncate
-
- Backlog
-
- is related to
-
WT-17272 Fix follower fast-truncate write-conflict detection gaps
-
- Closed
-
-
WT-16812 (Follower mode) validate fast truncate write‑conflict detection semantics
-
- Closed
-
- related to
-
WT-18734 Document the fast-truncate implementation across ASC, disaggregated leader and follower
-
- Needs Scheduling
-
-
WT-17775 test/format (mode=switch+fast_truncate) WiredTiger assertion failed: __rec_append_orig_value
-
- Open
-