aggregation framework - Three Way Many-to-Many in MongoDb -

i have seen plenty of examples on internet , answers on over various ways establish many-to-many relation between 2 entities. example of users , roles used. problem quite straight forward , simple understand me has 2 months experience mongodb , half of time has been spent fighting against rdbms schema patterns aren't necessary in mongodb.

however, little lost trying establish 3 way many-to-many relation seems straight forward in rdbms world i'm unable think through how done in mongodb.

here's example.

let's have 3 entities: project, user, projectrole.

each user can have multiple projects, each project can have multiple users.

each user can have multiple projectroles , each projectrole can have multiple users.

another condition projectroles can assigned users in relation project. is, users can't have projectrole if they're not part of project.

similarly, projectroles related projects via users.

so there ways have thought creating these relationships of these (or maybe all) ways direct result of rdbms way of doing things.

1.

create join collection these 3 entities such each document looks this:

{     "user_id" : objectid("someuserid"),     "project_id" : objectid("someprojectid"),     "projectrole_id" : objectid("someprojectroleid") } 

2.

copy on user (embed) object project document projectrole object/information embedded, this:

{     "name" : "projectname",     "users" : [{ "username" : "firstusername", "projectrole" : "somerole" },                { "username" : "secondusername", "projectrole": "someotherrole"}] } 

3.

copy on (embed) project object in user document associated projectrole user in project, this:

{     "name" : "username",     "projects" : [{ "projectname" : "someproject", "projectrole" : "somerole"},                   { "projectname" : "someotherproject", "projectrole" : "someotherrole"}] } 

option 1 has best chance of me being able keep data clean , keep integrity intact. going pain read since goes on 3 collections (and possibly more project gets more complex).

option 2 , 3 seem, me @ least, mongodb style embedded information. prefer 2 on 3 because projects sort of higher in hierarchy followed users. in other words, theoretically each project have lot more users each user , number of projects have been assigned.

problem option 2 data integrity. can dealt making sure on each update, example, of user server goes , searches through project collection , updates instances of user object (matching on id). potentially extremely slow operation (perhaps indexing on userid make faster) priority read-speed rather quick updates. assumption user data remain unchanged on period of time (on other hand project data updated quite frequently).

past point i'm not entirely sure how proceed. @ least on right track or solutions based on fundamental misunderstanding of either mongodb or actual problem itself?