根据价值折叠子文档的投影

Question

I have a data structure with a subdoc containing an array of department IDs, which are objects containing both IDs and department names. People may be in a large number of departments and have a separate ID for each, so that their record might look like this:

{ 
    "_id" : "xxxxx", 
    "dept_ids" : [
        {
            "dept_id" : "dd7867535", 
            "dept_name" : "d1"
        }, 
        {
            "dept_id" : "dl97087079", 
            "dept_name" : "d2"
        }
    ]
}

I'm putting together an aggregate query with a known subset of departments and I'd like to use the value of "dept_name" to roll up the values, so that the resulting document looks like this:

{ 
    "_id" : "xxxxx", 
    "d1" : "dd7867535",
    "d2" : "dl97087079"
}

I'm having trouble finding a way to do this within the aggregation framework. Any suggestions would be welcome.

Thanks, Alex

Answer 1

If you actually have a MongoDB suporting $replaceRoot and $arrayToObject then you can use this:

db.collection.aggregate([
  { "$replaceRoot": {
    "newRoot": {
      "$mergeObjects": [
        { "_id": "$_id" },
        { "$arrayToObject": {
          "$map": {
            "input": "$dept_ids",
            "in": { "k": "$$this.dept_name", "v": "$$this.dept_id" }
          }
        }}
      ]
    }
  }}
])

But you don't even need that, and simply transform the returned documents instead:

db.collection.find().map(d =>
  Object.assign(
    { _id: d._id },
    d.dept_ids.reduce((acc,dep) => Object.assign(acc,{ [dep.dept_name]: dep.dept_id }), {})
  )
);

In modern ECMASCRIPT environments, not the Mongo shell but something like NodeJS you can clean up the syntax a bit. ie with the NodeJS driver:

const mapper = ({ _id, dept_ids }) => ({
  _id, ...dept_ids.reduce((acc, { dept_name: k, dept_id: v }) => ({ ...acc, [k]: v }),{})
});

let results = await db.collection('departments').find().map(mapper).toArray();

Both produce the same result:

{ "_id" : "xxxxx", "d1" : "dd7867535", "d2" : "dl97087079" }

So you really don't need to process things through the aggregation pipeline that can be done quite simply from the received data on the client. You're not "reducing" any data, which is actually the point of the aggregation framework. So tranformations like this really "should" be done in post processing the cursor results instead.