In logical initial sync, during collection copy if the source collection is dropped the destination's CollectionBulkLoader aborts index builds and drops all indexes, including the _id index. However, it doesn't seem to drop the actual collection - the record store - that contains the documents. The test initial_sync_drop_collection.js demonstrates this.

Unfortunately when we move to oplog application, we still attempt to perform document updates / deletes to the collection if we see such oplog entries. It's not clear what the effect of not having an _id index is - do we end up resorting to a collection scan to perform each update / delete?

See comments for more info.