In a build failure on MongoDB master, we observed the following behavior:
- There is a validation failure on the local.replset.initialSyncId collection. That collection is unusual in a few different ways:
- It is an immediately durable (journaled) collection.
- It only ever has a single document in it.
- Writes to the collection are not done at a timestamp.
- The validation failure tends to happen shortly after startup, and results in either the value in the collection or index not being found by the validation.
- The document and its index are inserted as part of a single WiredTiger transaction, so they should either both be visible or neither - that's a guarantee that WiredTiger gives, which is not being honored here.
- is depended on by
-
SERVER-66430 Revert SERVER-66428 when obsolete
- Closed
- is duplicated by
-
WT-9462 test_bug010.test_checkpoint_dirty assertion 3 != 4
- Closed
- is related to
-
SERVER-66428 Have testing crash the server on local.replset.initialSyncId validation errors
- Closed
-
WT-9417 Update assertions in reconciliation code to be enabled in release builds where appropriate
- Closed
-
WT-9418 Introduce the ref track build option
- Closed
- related to
-
SERVER-66917 InsertAndSeekKeyString unit test should use makeKeyStringForSeek
- Closed
-
SERVER-66920 Crash in debug builds due to WiredTigerIndexCursorBase::seekWTCursor assertion not following search_near comparison semantics
- Closed
-
WT-9731 Add test case encouraging race conditions on collection creation
- Open
-
WT-9414 Add information about ref state transitions to developer docs
- Open
-
WT-9251 Perform log flush before the metadata checkpoint
- Closed