We had logics to handle cluster topology changes during a database creation. We might need to do something similar from the primary shard now that the createDatabase command runs through the new DDL coordinator.