mongo db - map reduce and lookup

Question

Is it possible to perform both a map reduce with a lookup in the same query pipeline efficiently?

Let's say I've two collections:

• items: { _id, group_id, createdAt }
• purchases: { _id, item_id }

I want to get the top n item groups, based on the number of purchases on the most recent x items per group.

If I had the number of purchases available in the item documents, then I could aggregate and sort, but this is not the case.

I can get the most recent x items per group as so:

let x = 3;
let map = function () {
  emit(this.group_id, { items: [this] });
};
let reduce = function (key, values) {
  return { items: getLastXItems(x, values.map(v => v.items[0])) };
};
let scope = { x };

db.items.mapReduce(map, reduce, { out: { inline: 1 }, scope }, function(err, res) {
  if (err) {
    ...
  } else {
    // res is an array of { group_id, items } where items is the last x items of the group
  }
});

But I'm missing purchase count so I can't use it to sort groups, and output the top n groups (which btw I'm not even sure I can do)

I'm using this on a web server, and running the query with scope variable depending on the user context, so I don't want to output the result to another collection and have to do everything inline.

=== edit 1 === add data example:

Sample data could be:

// items
{ _id: '1, group_id: 'a', createdAt: 0 }
{ _id: '2, group_id: 'a', createdAt: 2 }
{ _id: '3, group_id: 'a', createdAt: 4 }
{ _id: '4, group_id: 'b', createdAt: 1 }
{ _id: '5, group_id: 'b', createdAt: 3 }
{ _id: '6, group_id: 'b', createdAt: 5 }
{ _id: '7, group_id: 'b', createdAt: 7 }
{ _id: '8, group_id: 'c', createdAt: 5 }
{ _id: '9, group_id: 'd', createdAt: 5 }

// purchases
{ _id: '1', item_id: '1' }
{ _id: '2', item_id: '1' }
{ _id: '3', item_id: '3' }
{ _id: '4', item_id: '5' }
{ _id: '5', item_id: '5' }
{ _id: '6', item_id: '6' }
{ _id: '7', item_id: '7' }
{ _id: '8', item_id: '7' }
{ _id: '9', item_id: '7' }
{ _id: '10', item_id: '3' }
{ _id: '11', item_id: '9' }

and sample result with n = 3 and x = 2 would be:

[
  group_id: 'a', numberOfPurchasesOnLastXItems: 4,
  group_id: 'b', numberOfPurchasesOnLastXItems: 3,
  group_id: 'c', numberOfPurchasesOnLastXItems: 1,
]

Answer 1

I think this can be solved with the aggregation pipeline, but I've no idea on how bad this is, especially performance wise.

Concerns I have are:

• will the aggregation pipeline be able to benefits from indexes, on lookup and sort?
• can the lookup + projection that's only used to count matching items be simplified

Anyway, I think one solution I could be:

x = 2;
n = 3;

items.aggregate([
  {
    $lookup: {
      from: 'purchases',
      localField: '_id',
      foreignField: 'item_id',
      as: 'purchases',
    },
  },
  /*
  after the join, the data is like {
    _id: <itemId>,
    group_id: <itemGroupId>,
    createdAt: <itemCreationDate>,
    purchases: <arrayOfPurchases>,
  }
  */

  {
    $project: {
      group_id: 1,
      createdAt: 1,
      pruchasesCount: { $size: '$purchases' },
    }
  }
  /*
  after the projection, the data is like {
    _id: <itemId>,
    group_id: <itemGroupId>,
    createdAt: <itemCreationDate>,
    purchasesCount: <numberOfPurchases>,
  }
  */

  {
    $sort: { createdAt: 1 }
  },

  {
    $group: {
      _id: '$group_id',
      items: {
        $push: '$purchasesCount',
      }
    }
  }
  /*
  after the group, the data is like {
    _id: <groupId>,
    items: <array of number of purchases per item, sorted per item creation date>,
  }
  */

  {
    $project: {
      numberOfPurchasesOnMostRecentItems: { $sum: { $slice: ['$purchasesCount', x] } },
    }
  }
  /*
  after the projection, the data is like {
    _id: <groupId>,
    numberOfPurchasesOnMostRecentItems: <number of purchases on the last x items>,
  }
  */

  {
    $sort: { numberOfPurchasesOnMostRecentItems: 1 }
  },

  { $limit : n }
]);