The concept you are talking about can be called forward paging. A good reason for that is unlike using .skip() and .limit() modifiers this cannot be used to go back to a previous page or indeed skip to a specific page. At least not with a great deal of effort to store seen or discovered pages, so if that type of links to page paging is what you want, then you are best off sticking with the .skip() and .limit() approach, despite the performance drawbacks.

If it is a viable option to you to only move forward, then here is the basic concept:

db.junk.find().limit(3)



{ _id : ObjectId(54c03f0c2f63310180151877), a : 1, b : 1 }

{ _id : ObjectId(54c03f0c2f63310180151878), a : 4, b : 4 }

{ _id : ObjectId(54c03f0c2f63310180151879), a : 10, b : 10 }

Of course that's your first page with a limit of 3 items. Consider that now with code iterating the cursor:

var lastSeen = null;

var cursor = db.junk.find().limit(3);



while (cursor.hasNext()) {

   var doc = cursor.next();

   printjson(doc);

   if (!cursor.hasNext())

     lastSeen = doc._id;

}

So that iterates the cursor and does something, and when it is true that the last item in the cursor is reached you store the lastSeen value to the present _id:

ObjectId(54c03f0c2f63310180151879)

In your subsequent iterations you just feed that _id value which you keep ( in session or whatever ) to the query:

var cursor = db.junk.find({ _id: { $gt: lastSeen } }).limit(3);



while (cursor.hasNext()) {

   var doc = cursor.next();

   printjson(doc);

   if (!cursor.hasNext())

     lastSeen = doc._id;

}



{ _id : ObjectId(54c03f0c2f6331018015187a), a : 1, b : 1 }

{ _id : ObjectId(54c03f0c2f6331018015187b), a : 6, b : 6 }

{ _id : ObjectId(54c03f0c2f6331018015187c), a : 7, b : 7 }

And the process repeats over and over until no more results can be obtained.

That's the basic process for a natural order such as _id. For something else it gets a bit more complex. Consider the following:

{ _id: 4, rank: 3 }

{ _id: 8, rank: 3 }

{ _id: 1, rank: 3 }    

{ _id: 3, rank: 2 }

To split that into two pages sorted by rank then what you essentially need to know is what you have already seen and exclude those results. So looking at a first page:

var lastSeen = null;

var seenIds = [];

var cursor = db.junk.find().sort({ rank: -1 }).limit(2);



while (cursor.hasNext()) {

   var doc = cursor.next();

   printjson(doc);

   if ( lastSeen != null && doc.rank != lastSeen )

       seenIds = [];

   seenIds.push(doc._id);

   if (!cursor.hasNext() || lastSeen == null)

     lastSeen = doc.rank;

}



{ _id: 4, rank: 3 }

{ _id: 8, rank: 3 }

On the next iteration you want to be less or equal to the lastSeen rank score, but also excluding those already seen documents. You do this with the $nin operator:

var cursor = db.junk.find(

    { _id: { $nin: seenIds }, rank: $lte: lastSeen }

).sort({ rank: -1 }).limit(2);



while (cursor.hasNext()) {

   var doc = cursor.next();

   printjson(doc);

   if ( lastSeen != null && doc.rank != lastSeen )

       seenIds = [];

   seenIds.push(doc._id);

   if (!cursor.hasNext() || lastSeen == null)

     lastSeen = doc.rank;

}



{ _id: 1, rank: 3 }    

{ _id: 3, rank: 2 }

How many seenIds you actually hold on to depends on how granular your results are where that value is likely to change. In this case you can check if the current rank score is not equal to the lastSeen value and discard the present seenIds content so it does not grow to much.

That's the basic concepts of forward paging for you to practice and learn.