I would like the ability in the aggregation framework to group and select the top K elements in each group.

Here is an example:

Let's say I have documents with the following structure:

{ "_id":..., "Group": 1, "Value": 4 } { "_id":..., "Group": 2, "Value": 8 } { "_id":..., "Group": 1, "Value": 10 }

Given X, Y and K, I would like to be able to use the aggregation framework to do the following:

• Select all documents with the "Group" property between X and Y
• Group by the "Group" property
• In each group: pick only the K documents with the largest "Value" property

Because K will be much smaller than the size of the collection itself, I would like to be able to do that in a way that is memory efficient.

There is a new feature in 2.4 where we can efficiently sort then limit with the aggregation framework. This is basically the same thing, working on groups.

I have no preference on which operator should enable this scenario, as long as it is possible to achieve what I just described.