Both drop database and drop collection on sharded cluster currently only invalidate the information on the DatabaseShardingState and the CollectionShardingRuntime after the database/collection have been dropped. The rename collection operation also cleans up the CollectionShardingRuntime.

This is enough to ensure that subsequent requests will be causal consistent with the DDL operation.

In order to free-up some memory, we should also proactively evict the cached entries for the dropped db/coll or renamed coll in the catalog cache.