We have a test failure wherein an error occurs to the effect that a WT transaction is already open when the MDB layer calls WT::begin_transaction – it has already been called once and calling it again is not supported.

It isn't obvious where the issue is, that the MDB layer is double calling or not. The RecoveryUnit is supposed to track state for storage engine access – active, inactive, WUOW, etc. --, so it might be helpful to log some of that information before proceeding to call invariantWTOK on the WT::begin_transaction() calls. Assuming the RecoveryUnit doesn't have a bug, that is. Hopefully RU state will be enlightening regardless.