-
Type:
Bug
-
Resolution: Fixed
-
Priority:
Major - P3
-
Affects Version/s: None
-
Component/s: Schema Management
-
None
-
Storage Engines - Foundations
-
601.753
-
SE Foundations - 2026-10-13
-
3
Problem:
With schema epochs and preserve_prepared=true, eviction can remove prepared state while a table awaits publication. A later checkpoint succeeds but is missing that state.
AssertionError: 0 not greater than 0 : Stat 2418: expected increase after checkpoint
At the point of time this error was found, statistic 2418 is: rec_time_window_prepared.
Context:
This was found while running the Python suite with schema publishing enabled as part of WT-18031. The failing test is:
test_prepare46.test_prepare46.test_prepared_cell_preserved_after_eviction_at_unstable_prepare_ts
This is similar to WT-18704. This test itself is already the reproducer (when running the schema epoch hook).
Definition of Done:
- Investigate and apply the bug fix for this scenario
- Add regression tests covering this particular scenario
- Remove the skip in test_prepare46
- is depended on by
-
WT-18803 Remove the unreachable "awaiting publication" handling from reconciliation
-
- Closed
-
- is related to
-
WT-18704 Schema epoch hook: prepared insert rollback causes “leaked prepared update” assertion
-
- Closed
-
-
WT-18031 Extend disagg Python hook to publish schema ops and advance schema epoch
-
- Closed
-
- related to
-
WT-18706 Schema epoch hook: publication racing with debug mode eviction loses committed data
-
- Closed
-
-
WT-18787 test_prepare21.py assertion fires 7 != 10
-
- Open
-
-
WT-18746 failed: unit-test-hook-disagg-leader-publish-bucket01 on ubuntu2004-nonstandalone [wiredtiger @ ca050f0e]
-
- Closed
-