Question

tricky mongodb querying problem: I have a collection ‘Accounts’ with a bunch of documents that (simplified and with actual data swapped for non-real-world values) look like this:

{"_id": "<Mongo Binary Data>",
 "Roles": {
     "D7753879C7020F8ECF947122FA211413": {
       "_id": "<Mongo Binary Data>",
       "OrgName": "ACME",
       "Rolename": "Coyote Liaison",
    },

     "CFA7722E6799170706E4C5FFF3F01E63": {
       "_id": "<Mongo Binary Data>",
       "OrgName": "ACME",      
       "Rolename": "MembershipAdmin",
    },

     "C7020F8ECF947122FAGIGHFVFF3F7753": {
       "_id": "<Mongo Binary Data>",
       "OrgName": "Initech",       
       "Rolename": "MembershipAdmin",
    }   
  } 
}

The keys in the Roles array are a combination of a role id and an org id which are then hashed, which makes it very quick to query (once an Account object has been loaded up from MongoDB and into C#) for if an Account has a given role for a given org, ie, is a user a MembershipAdmin for Initech.

Now I want to query for users who have a role for ANY org, which in pseudo-SQL could be expressed as ‘SELECT all accounts which have at least one role object where Rolename = THISROLENAME’. Ie, get all users which are MembershipAdmins.

I have tried this:

{
  Roles.Rolename: "MembershipAdmin"
}

and this

{
  Roles: {"Rolename": "MembershipAdmin"}
}

and this

{
  Roles: {"$elemMatch": {"Rolename": "MembershipAdmin"}}
}

…to no avail and have seen several answers, etc, saying that the only way out of this is to push the associative array key down into the subobjects, which I don’t want to do as it makes the primary function of this data structure (checking if an account has a given role for a given organisation) very quick indeed. The use case I describe above doesn’t have to be mega-fast as it is part of an admin user’s responsibility so I’m happy to make them wait a few moments – so queries with excessive recursion, etc, are OK in this case.

Does anyone have any ideas how to make this work without refactoring the data structure? At my wit’s end with this.

Many thanks,

G

[EDIT: the structure above is not queryable, see the accepted answer for an erudite but quick explanation as to why not and what you should properly do to fix it. If you’re OK with a hacky workaround, though, you can store a copy of the data in a BsonArray alongside it and query against that using $elemMatch]

Answer 1

There is no way to do with you want with that schema, nor is the schema very practical to begin with. You should change your schema to this :

{"_id": "<Mongo Binary Data>",
 "Roles": [
    {
       "hash": "D7753879C7020F8ECF947122FA211413",
       "_id": "<Mongo Binary Data>",
       "OrgName": "ACME",
       "Rolename": "Coyote Liaison",
    },

    {
       "hash": "CFA7722E6799170706E4C5FFF3F01E63",
       "_id": "<Mongo Binary Data>",
       "OrgName": "ACME",      
       "Rolename": "MembershipAdmin",
    },

    {
       "hash": "C7020F8ECF947122FAGIGHFVFF3F7753",
       "_id": "<Mongo Binary Data>",
       "OrgName": "Initech",       
       "Rolename": "MembershipAdmin",
    }   
  ] 
}

The reason why using nonstable values as document field names is so unpractical is exactly because it complicates most querying. The only upside is that some queries might be a little faster but given the fact that it also causes issues for indexing it’s almost always a bad idea.

I would strongly suggest to change your schema to the above rather than look for solutions that allow your current schema.

EDIT: As mentioned in the discussion below it is technically possible to create the needed query using the $where operator. If you cannot do it any other way it means you have a schema smell and a potential scaling and performance bottleneck that will be diabolically hard to fix when your software goes live. Don’t use $where. Ever.