-
Type:
Bug
-
Resolution: Fixed
-
Priority:
Major - P3
-
Affects Version/s: None
-
Component/s: Checkpoints, Test Model
-
Storage Engines - Persistence
-
6,932.447
-
SE Persistence backlog
-
None
-
76
Currently, test/model assumes that any checkpoint_crash_point will result in an incomplete non-recoverable checkpoint. However, this assumption does not hold when logging is enabled. From WT perspective, a crash that occurs after the checkpoint txn is committed (but before the metadata/turtle file is updated) will still result in a valid, recoverable checkpoint.
To avoid test/model failures under this mismatch, I've temporarily prevented crash points from occurring after commit (specifically CKPT_CRASH_BEFORE_METADATA_SYNC) when logging is enabled. This is just a workaround and should be removed as part of this ticket.
Definition of done:
- test/model supports and correctly verifies crash scenarios that may result in recoverable checkpoints
- Remove the workaround code before the log flush that is preventing us from reaching the post commit crash point when logging is enabled.
- Ensure correct, deterministic behaviour under these crash scenarios with logging enabled and disabled
- Add appropriate tests.
- depends on
-
WT-18392 Checkpoint crash points after the checkpoint transaction commit have a timing-dependent outcome
-
- Closed
-
- duplicates
-
WT-18299 failed: model-test-long on ubuntu2004-asan [wiredtiger @ d6a87e24]
-
- Closed
-
- is related to
-
WT-18363 model_test ignores the -G conn_logging, timing_stress and checkpoint_crash generator overrides
-
- Closed
-
- related to
-
WT-16832 List all the planned changes to __checkpoint_db_internal
-
- Closed
-
-
WT-14394 Run test/model with logging enabled
-
- Closed
-
-
WT-18299 failed: model-test-long on ubuntu2004-asan [wiredtiger @ d6a87e24]
-
- Closed
-
-
WT-18511
test/model should reconcile its connection logging belief with WiredTiger at open time
-
- Closed
-