Open
Description
There are functions depending on what they saw in _bucket, but which don't make a follow-up WAL write to be rolled back in case the _bucket write fails. For instance vshard.storage.bucket_ref*(). Need to investigate if it can lead to any issues if such a function sees uncommitted _bucket record, which is then rolled back. May be related to #173 in case _bucket changes will be able to invalidate the local refs.