I am trying to figure out the best way to store (and query) relationships between documents in a MongoDB database.

I have three collections in my database: say, one called 'Bag', another called 'Item', and a third called 'Relationship'. Documents of type 'Relationship' allow me to describe how a given 'Item' is related to a given 'Bag'. Each 'Relationship' document also contain DBRefs to both a 'Bag' and an 'Item' document. Several 'Relationship' documents can exists if there are many ways the same 'Item' is linked to the same 'Bag'.

Basically, I am storing a multi-edge directed graph with nodes being 'Bag' or 'Item' documents, and edges being 'Relationship' documents.

All of this works beautifully to model my data, but so far I have trouble to get fast answers to the query 'give me all Items that are related to this Bag' when I want to filter both the Items and the Relationships. E.g., "give me all Items with key/value { a: 1 } that are linked to my Bag through a relationship with key/value { type: 'part-of' }"

What would be your opinion? I know I could try other technologies (e.g., RDF or graph databases), but for various technical reasons I really want to keep using MongoDB. Also, storing the list of Items a Bag is connected to in the Bag object itself looks unpractical, as I also need to store informations about this connection. Plus, there are about hundreds of thousands of Items that can be in one given Bag.

I am considering using server-side functions, and maybe mapreduce, but I am not sure if this tool (which I never used) could be of help here.

My current strategy is

(1) to ask, client side, for the list of Relationship documents that match a filter (e.g., { type: 'part-of' }), then

(2) to extract the $id of the Item objects these Relationship points to, then

(3) query all Item objects that match a filter (e.g., { a: 1 }) which I supplemented with an { $id: { $in: [long list of Item identifiers I got from querying the Relationships] }}.

This is slow, and I even have to split the list of identifiers by chunks of few thousands to avoid the 4Mb limit.

Best,

Aurelien Mazurie