For the 'local' database, the collection catalog (_mdb_catalog) and the view catalog (local.system.views) tables are recovered at different timestamps after a crash:

• For _mdb_catalog, we disable WiredTiger's table logging, so after a crash it gets rolled back to the last checkpoint (rollback to stable timestamp).
• For local.system.views, since the local DB is unreplicated, we enable WiredTiger's journal (table logging/WAL) so it gets rolled back to the last journal flush.

This can lead to inconsistencies, for example:

• At the stable timestamp "local.a" is a collection.
• Since then, "local.a" has been dropped and re-created as a view, the insert to "local.system.views" has been journaled by WT.
• The node crashes.
• _mdb_catalog gets recovered to the stable timestamp ("local.a" is a collection).
• local.system.views gets recovered from WiredTiger's journal ("local.a" is a view).

A reproducer is attached.

"local.system.views" should get the same recovery guarantees as "_mdb_catalog".