Return only matching subdocs with Mongo aggregation - mongodb

Question

I have a schema with subdocs.
// Schema
var company = {
_id: ObjectId,
publish: Boolean,
divisions: {
employees: [ObjectId]
}
};
I need to find all the subdocs (divisions) that match my query. It appears that I have to use 2 matches - one to filter out initial docs and a second one to filter out the matching subdocs from the resulting $unwind operation. Is there a more efficient way?
// Query
this.aggregate({
$match: {
'publish': 1,
'divisions.employees': new ObjectId(userid)
}
}, {
$unwind: '$divisions'
}, {
$match: {
'divisions.employees': new ObjectId(userid)
}
}
I found this ticket but I am unsure this does what I need.

Answer 1

Doing both matches is the right thing here. You could eliminate the first match stage and just unwind, but having an initial $match allows you to narrow down the pipeline to exactly those documents that will produce at least one output document (i.e. those documents for which publish : true and some employees ObjectId matches the given ObjectId). You will be able to use indexes, like an index on { publish : 1, divisions.employees : 1 }, to perform the first match stage quickly.
However, you should ask yourself why you are using the aggregation pipeline here and if it is appropriate. Will you commonly be querying for a given employee that's part of a company with publish : 1? Is this one of the main queries for your use case? If it's infrequent or not critical then the aggregation is a good way to do it. Otherwise, you should reconsider the schema. You could make this query easy with a schema like
{
"_id" : ObjectId,
"publish" : Boolean,
"company" : (unique identifier, possibly a String or ObjectId)
}
that models employees as documents and denormalizes company information into the employee document. Then your query is as easy as
db.employees.find({ "_id" : ObjectId(userid), "publish" : true })
which will be much quicker than the aggregation (not that the aggregation will be slow; this is just relatively quicker). I'm not telling you to do it this way - use your own knowledge of your use case to make the right call.