Workload uses 50 threads each of which repeatedly creates a new document, then grows an array in the document and does some other updates; document grows to 100-200 kB. The test script lets this workload run for a random time from 0-60 seconds then abruptly kills the mongod processes, repeating this cycle of running the workload then killing the mongod until a mongod fails to recover during startup. Recovery will frequently fail with a BSON "bad type" error, usually within a few iterations.
This is a recurrence in 4.4.4-rc0 of a problem that was thought to have been fixed by WT-7117 and is using the same test that led to that ticket.
- is caused by
-
WT-7117 RTS to skip modifies that are more recent than on-disk base update while restoring an update
- Closed
- related to
-
SERVER-54301 Add correctness tests of recovery using history store
- Closed
-
WT-13349 Improve testing for history store crash recovery
- Open