When a $lookup is using the index strategy, the index fetching code is doing the seek-deduplicate-fetch loop in the leaf node of the plan, but above this section there is a loop over all the possible search keys that belong to the same outer document. This loop performs a second deduplicate check to remove identical foreign documents matching different parts of the same local document. The result is correct, but we end up re-fetching documents that are later discarded.

Option 1: delay fetching, but this would imply delaying also the filtering and any manipulation done in the subpipeline representing the foreign stream

Option 2: switch to a more complex index scan representation that receives the entire set of search keys