Shard servers disallow mongodb transactions if writeConcernMajorityJournalDefault=false. Currently we check that for each transaction or retryable write here but the ordering of the assert means we always call into the ReplicationCoordinator and lock its mutex even though retryable writes can never fail this check. If we reorder the check to check for !autocommit first, then we should be able to reduce contention on the replication coordinator mutex from retryable writes.