When performing some DDL operations like dropDatabase, dropCollection, etc. we try to send the drop to all shards and clear their local cached collection metadata. This serves as a "barrier" between previously and later created collections on the same namespace.

However, this is not done for the config server, which can later transition into an embedded config server that can also serve as a data shard. The shard will then have stale cached collection metadata.

There could be two approaches to solve this:

1. Make DDL coordinators send the command to the config server as well to make sure cached metadata is cleared there
2. Whenever a dedicated config server transitions to a data shard, refresh the local catalog to clear the stale metadata