Since merging branches with MDL-48302 we have had problems with many behat test failures (only 27/28, not master, oddly); this appears to allow some dirty data in MUC, possibly access or groups related.

This is a problem for us only because of some sort of caching customisation (I'm not sure which exactly) but the new ordering seems like it will become problematic eventually for core too.

Basically I think in behat_util::reset_all_data, the DB reset needs to be first, before others like accesslib_clear_all_caches and friends.