MongoDB：连接多个数组

Question

I have 3 arrays of ObjectIds I want to concatenate into a single array, and then sort by creation date. $setUnion does precisely what I want, but I'd like to try without using it.

Schema of object I want to sort:

var chirpSchema = new mongoose.Schema({
    interactions: { 
      _liked      : ["55035390d3e910505be02ce2"] // [{ type: $oid, ref: "interaction" }]
    , _shared     : ["507f191e810c19729de860ea", "507f191e810c19729de860ea"] //  [{ type: $oid, ref: "interaction" }]
    , _viewed     : ["507f1f77bcf86cd799439011"] //  [{ type: $oid, ref: "interaction" }]
  }
});

Desired result : Concatenate _liked, _shared, and _viewed into a single array, and then sort them by creation date using aggregate pipeline. See below

["507f1f77bcf86cd799439011", "507f191e810c19729de860ea", "507f191e810c19729de860ea", "55035390d3e910505be02ce2"]

I know I'm suppose to use $push , $each , $group , and $unwind in some combination or other, but I'm having trouble piecing together the documenation to make this happen.

Update: Query

model_user.aggregate([
      { $match    : { '_id' : { $in : following } } }
    , { $project  : { 'interactions' : 1 } }
    , { $project  : {
          "combined": { $setUnion : [ 
            "$interactions._liked"
          , "$interactions._shared"
          , "$interactions._viewed"
        ]}
      }}
])
.exec(function (err, data) {
  if (err) return next(err);
  next(data); // Combined is returning null
})

Answer 1

If all the Object _id values are "unique" then $setUnion is your best option. It is of course not "ordered" in any way as it works with a "set", and that does not guarantee order. But you can always unwind and $sort.

[
    { "$project": {
        "combined": { "$setUnion": [ 
            { "$ifNull": [ "$interactions._liked", [] ] },
            { "$ifNull": [ "$interactions._shared", [] ] },
            { "$ifNull", [ "$interactions._viewed", [] ] }
        ]}
    }},
    { "$unwind": "$combined" },
    { "$sort": { "combined": 1 } },
    { "$group": {
        "_id": "$_id",
        "combined": { "$push": "$combined" }
    }}
]

Of course again since this is a "set" of distinct values you can do the old way instead with $addToSet , after processing $unwind on each array:

[
    { "$unwind": "$interactions._liked" },
    { "$unwind": "$interactions._shared" },
    { "$unwind": "$interactions._viewed" },
    { "$project": {
        "interactions": 1,
        "type": { "$const": [ "liked", "shared", "viewed" ] }
    }}
    { "$unwind": "$type" },
    { "$group": {
        "_id": "$_id",
        "combined": {
            "$addToSet": {
                "$cond": [
                   { "$eq": [ "$type", "liked" ] },
                   "$interactions._liked",
                   { "$cond": [
                       { "$eq": [ "$type", "shared" ] },
                       "$interactions._shared",
                       "$interactions._viewed"
                   ]}
                ]
            }
        }
    }},
    { "$unwind": "$combined" },
    { "$sort": { "combined": 1 } },
    { "$group": {
        "_id": "$_id",
        "combined": { "$push": "$combined" }
    }}
]

But still the same thing applies to ordering.

Future releases even have the ability to concatenate arrays without reducing to a "set":

[
    { "$project": {
        "combined": { "$concatArrays": [ 
            "$interactions._liked",
            "$interactions._shared",
            "$interactions._viewed"
        ]}
    }},
    { "$unwind": "$combined" },
    { "$sort": { "combined": 1 } },
    { "$group": {
        "_id": "$_id",
        "combined": { "$push": "$combined" }
    }}
]

But still there is no way to re-order the results without procesing $unwind and $sort .

You might therefore consider that unless you need this grouped across multiple documents, that the basic "contenate and sort" operation is best handled in client code. MongoDB has no way to do this "in place" on the array at present, so per document in client code is your best bet.

But if you do need to do this grouping over multiple documents, then the sort of approaches as shown here are for you.

Also note that "creation" here means creation of the ObjectId value itself and not other properties from your referenced objects. If you need those, then you perform a populate on the id values after the aggregation or query instead, and of course sort in client code.