MongoDB MapReduce Stats Engine

I'm trying to develop a generic reporting engine for a MongoDB system which will aggregate information from a set of documents. I won't know the structure of the documents in advance of the query running.

e.g. One set of documents might look like this:

{"drink": "Tea","accompaniments": ["Biscuits","Crisps"]}
{"drink": "Tea","accompaniments": ["Biscuits"]}
{"drink": "Coffee","accompaniments": ["Biscuits","Crisps"]}
{"drink": "Coffee","accompaniments": ["Nuts","Biscuits"]}
{"drink": "Beer","accompaniments": ["Nuts","Crisps"]}
{"drink": "Beer","accompaniments": ["Nuts"]}
{"drink": "Beer","accompaniments": ["Crisps"]}

Here, I need results such as:

{drink:{Tea:2,Coffee:2,Beer:3},accompaniments:{biscuits:4,Crisps:4,Nuts:3}}

Another document set might look like this:

{"favcolor": "Red","shapes": ["Square","Circle"]}
{"favcolor": "Red","shapes": ["Square"]}
{"favcolor": "Green","shapes": ["Circle"]}

There may be any number of different document formats.

So, I can't 'hard-code' the map-reduce functions with named keys. Also, notice that I need to grab 开发者_开发百科info from nested arrays and/or documents - I am in control of how the documents are built, so, to use one example, I could make the accompaniments array a proper set of nested documents if need be (in this example they're built from multiple-choice check-boxes).

What would be the most efficient map-reduce function to grab all the stats from each set of documents? I'm really hoping that there will be 'one' map-reduce that will be able to grab the data from each set. Thanks very much.

---

I'm not sure about my concept, it's just an idea.

Your two example datasets have common structure. Both contain documents with to properties: one has string value, second is an array. So your map-reduce should iterate over document properties and check the type of particular property. Then agregate according to type and property name.