We used to purge database routing cache in the end for both successful and failed database creation to make sure stale metadata is updated. This might be due to database creation not being an atomic operation without using a DDL coordinator. We shouldn't need to do this anymore.