I'm not sure if this is a bug or just a change in behavior from 4.0 to 3.6. If it is just a change, it should be documented.

Previously if you made a bulk op with 1000 operations it would execute it as one operation, but now it splits it into two batches of 999 and 1.

This is additional evidence that we absolutely need https://jira.mongodb.org/browse/NODE-2768 because we have no way of anticipating how the driver splits operations into batches so we cannot track that information on our side.