mongodb multiple aggregations in single operation - mongodb

Question

I have an item collection with following documents.
{ "item" : "i1", "category" : "c1", "brand" : "b1" }
{ "item" : "i2", "category" : "c2", "brand" : "b1" }
{ "item" : "i3", "category" : "c1", "brand" : "b2" }
{ "item" : "i4", "category" : "c2", "brand" : "b1" }
{ "item" : "i5", "category" : "c1", "brand" : "b2" }
I want to separate aggregation results --> count by category, count by brand. Please note, it is not count by (category,brand)
I am able to do this using map-reduce using following code.
map = function(){
emit({type:"category",category:this.category},1);
emit({type:"brand",brand:this.brand},1);
}
reduce = function(key, values){
return Array.sum(values)
}
db.item.mapReduce(map,reduce,{out:{inline:1}})
And the result is
{
"results" : [
{
"_id" : {
"type" : "brand",
"brand" : "b1"
},
"value" : 3
},
{
"_id" : {
"type" : "brand",
"brand" : "b2"
},
"value" : 2
},
{
"_id" : {
"type" : "category",
"category" : "c1"
},
"value" : 3
},
{
"_id" : {
"type" : "category",
"category" : "c2"
},
"value" : 2
}
],
"timeMillis" : 21,
"counts" : {
"input" : 5,
"emit" : 10,
"reduce" : 4,
"output" : 4
},
"ok" : 1,
}
I can get same results by firing two different aggregation commands as below.
db.item.aggregate({$group:{_id:"$category",count:{$sum:1}}})
db.item.aggregate({$group:{_id:"$brand",count:{$sum:1}}})
Is there anyway I can do the same using aggregation framework by single aggregation command.
I have simplified my case here, but in actual I need this grouping from fields in array of subdocuments. Assume the above is structure after I do unwind.
It is a real-time query (someone waiting for response), though on smaller dataset, so execution time is important.
I am using MongoDB 2.4.

Answer 1

Starting in Mongo 3.4, the $facet aggregation stage greatly simplifies this type of use case by processing multiple aggregation pipelines within a single stage on the same set of input documents:
// { "item" : "i1", "category" : "c1", "brand" : "b1" }
// { "item" : "i2", "category" : "c2", "brand" : "b1" }
// { "item" : "i3", "category" : "c1", "brand" : "b2" }
// { "item" : "i4", "category" : "c2", "brand" : "b1" }
// { "item" : "i5", "category" : "c1", "brand" : "b2" }
db.collection.aggregate(
{ $facet: {
categories: [{ $group: { _id: "$category", count: { "$sum": 1 } } }],
brands: [{ $group: { _id: "$brand", count: { "$sum": 1 } } }]
}}
)
// {
// "categories" : [
// { "_id" : "c1", "count" : 3 },
// { "_id" : "c2", "count" : 2 }
// ],
// "brands" : [
// { "_id" : "b1", "count" : 3 },
// { "_id" : "b2", "count" : 2 }
// ]
// }

Answer 2

Over a large data set I would say that your current mapReduce approach would be the best one, because the aggregation technique for this would not work well with large data. But possibly over a reasonably small size it might just be what you need:
db.items.aggregate([
{ "$group": {
"_id": null,
"categories": { "$push": "$category" },
"brands": { "$push": "$brand" }
}},
{ "$project": {
"_id": {
"categories": "$categories",
"brands": "$brands"
},
"categories": 1
}},
{ "$unwind": "$categories" },
{ "$group": {
"_id": {
"brands": "$_id.brands",
"category": "$categories"
},
"count": { "$sum": 1 }
}},
{ "$group": {
"_id": "$_id.brands",
"categories": { "$push": {
"category": "$_id.category",
"count": "$count"
}},
}},
{ "$project": {
"_id": "$categories",
"brands": "$_id"
}},
{ "$unwind": "$brands" },
{ "$group": {
"_id": {
"categories": "$_id",
"brand": "$brands"
},
"count": { "$sum": 1 }
}},
{ "$group": {
"_id": null,
"categories": { "$first": "$_id.categories" },
"brands": { "$push": {
"brand": "$_id.brand",
"count": "$count"
}}
}}
])
Not really the same as the mapReduce output, you could throw in some more stages to change the output format, but this should be usable:
{
"_id" : null,
"categories" : [
{
"category" : "c2",
"count" : 2
},
{
"category" : "c1",
"count" : 3
}
],
"brands" : [
{
"brand" : "b2",
"count" : 2
},
{
"brand" : "b1",
"count" : 3
}
]
}
As you can see, this involves a fair bit of shuffling between arrays in order to group each set of either "category" or "brand" within the same pipeline process. Again I will say, this will not do well for large data, but for something like "items in an order" it would probably do nicely.
Of course as you say, you have simplified somewhat, so the first grouping key on null is either going to be something else or either narrowed down to do that null case by an earlier $match stage, which is probably what you want to do.