Grouping records in nested documents - mongodb

Question

I have a document like this:
{
"_id" : ObjectId("533e6ab0ef2188940b00002c"),
"uin" : "1396599472869",
"vm" : {
"0" : {
"draw" : "01s",
"count" : "2",
"type" : "",
"data" : {
"title" : "K1"
},
"child" : [
"1407484608965"
]
},
"1407484608965" : {
"data" : {
"title" : "K2",
"draw" : "1407473540857",
"count" : "1",
"type" : "Block"
},
"child" : [
"1407484647012"
]
},
"1407484647012" : {
"data" : {
"title" : "K3",
"draw" : "03.8878.98",
"count" : "1",
"type" : "SB"
},
"child" : [
"1407484762473"
]
},
"1407484762473" : {
"data" : {
"type" : "SB",
"title" : "D1",
"draw" : "7984",
"count" : "1"
},
"child" : []
}
}
}
How to group all records with condition (type="Block")?
I've tried:
db.ITR.aggregate({$match:{"uin":"1396599472869"}},{$project:{"vm":1}},{$group:{_id:null,r1:{$push:"$vm"}}},{$unwind:"$r1"},{$group:{_id:null,r2:{$push:"$r1"}}},{$unwind:"$r2"})
But the result is still in the form of an object and not an array. With "MapReduce" I did not get.

Answer 1

Your problem here is basically with the way you currently have your document structured. The usage of "keys" under "vm" here that actually identify data points does not play well with the standard query forms and the aggregation framework in general.
It also is generally not a very good pattern, as in order to access any part under "vm" you need to specify the "exact path" to the data. So looking for type "Block" requires this:
db.collection.find({
"$or": [
{ "vm.0.type": "Block" },
{ "vm.1407484608965.type": "Block" }
{ ... }
]
})
And so on. You cannot "wildcard" field names like this so the exact path is required.
A better approach to modelling is to use an array instead, and move that inner key inside the documents:
{
"_id" : ObjectId("533e6ab0ef2188940b00002c"),
"uin" : "1396599472869",
"vm" : [
{
"key": 0,
"draw" : "01s",
"count" : "2",
"type" : "",
"data" : {
"title" : "K1"
},
"child" : [
"1407484608965"
]
},
{
"key": "1407484608965",
"title" : "K2",
"draw" : "1407473540857",
"count" : "1",
"type" : "Block",
"child" : [
"1407484647012"
]
},
{
"key": "1407484647012",
"title" : "K3",
"draw" : "03.8878.98",
"count" : "1",
"type" : "SB",
"child" : [
"1407484762473"
]
}
]
}
This allows you to query for documents that contain the matching property by a common path, which greatly simplifies things:
db.collection.find({ "vm.type": "Block" })
Or if you want to "filter" the array contents so that only those "sub-documents" that match are returned you can do this:
db.collection.aggregate([
{ "$match": { "vm.type": "Block" } },
{ "$unwind": "$vm" },
{ "$match": { "vm.type": "Block" } },
{ "$group": {
"_id": "$_id",
"uin": { "$first": "$uin" },
"vm": { "$push": "$vm" }
}}
])
Or even possibly this with MongoDB 2.6 or greater:
db.collection.aggregate([
{ "$match": { "vm.type": "Block" } },
{ "$project": {
"uin": 1,
"vm": {
"$setDifference": [
{ "$map": {
"input": "$vm",
"as": "el",
"in": {"$cond": [
{ "$eq": [ "$$el.type", "Block" ] },
"$$el",
false
]}
}},
[false]
]
}
}}
])
Or any other operation, which is simplified to traverse now the data is structured that way. But as your data presently stands your only option to "traverse keys" is to use JavaScript operations, which is much slower than being able to query in a proper way:
db.collection.find(function() {
return Object.keys(this.vm).some(function(x) {
return this.vm[x].type == "Block"
})
})
Or with similar object processing using mapReduce but essentially with no other way to access the fields with fixed paths that vary all the time.
Perhaps this was a design entered into to avoid having "nested arrays" which is where the "child" element would be placed. Of course this poses a problem with updates. But really if any element should not be an array it is probably the "inner" element such as "child", which could have some kind of structure that does not use an array.
So the key is to look at restructuring, as this will likely suit the patterns that you want without causing performance problems that JavaScript traversal will introduce.