How to implement has_many :through relationships with Mongoid and mongodb?

Using this modified example from the Rails guides, how does one model a relational "has_many :through" association using mongoid?

The challenge is that mongoid does not support has_many :through as ActiveRecord does.

# doctor checking out patient
class Physician < ActiveRecord::Base
  has_many :appointments
  has_many :patients, :through => :appointments
  has_many :meeting_notes, :through => :appointments
end

# notes taken during the appointment
class MeetingNote < ActiveRecord::Base
  has_many :appointments
  has_many :patients, :through => :appointments
  has_many :physicians, :through => :appointments
end

# the patient
class Patient < ActiveRecord::Base
  has_many :appointments
  has_many :physicians, :through => :appointments
  has_many :meeting_notes, :through => :appointments
end

# the appointment
class Appointment < ActiveRecord::Base
  belongs_to :physician
  belongs_to :patient
  belongs_to :meeting_note
  # has timestamp attribute
end

Mongoid doesn't have has_many :through or an equivalent feature. It would not be so useful with MongoDB because it does not support join queries so even if you could reference a related collection via another it would still require multiple queries.

https://github.com/mongoid/mongoid/issues/544

Normally if you have a many-many relationship in a RDBMS you would model that differently in MongoDB using a field containing an array of 'foreign' keys on either side. For example:

class Physician
  include Mongoid::Document
  has_and_belongs_to_many :patients
end

class Patient
  include Mongoid::Document
  has_and_belongs_to_many :physicians
end

In other words you would eliminate the join table and it would have a similar effect to has_many :through in terms of access to the 'other side'. But in your case thats probably not appropriate because your join table is an Appointment class which carries some extra information, not just the association.

How you model this depends to some extent on the queries that you need to run but it seems as though you will need to add the Appointment model and define associations to Patient and Physician something like this:

class Physician
  include Mongoid::Document
  has_many :appointments
end

class Appointment
  include Mongoid::Document
  belongs_to :physician
  belongs_to :patient
end

class Patient
  include Mongoid::Document
  has_many :appointments
end

With relationships in MongoDB you always have to make a choice between embedded or associated documents. In your model I would guess that MeetingNotes are a good candidate for an embedded relationship.

class Appointment
  include Mongoid::Document
  embeds_many :meeting_notes
end

class MeetingNote
  include Mongoid::Document
  embedded_in :appointment
end

This means that you can retrieve the notes together with an appointment all together, whereas you would need multiple queries if this was an association. You just have to bear in mind the 16MB size limit for a single document which might come into play if you have a very large number of meeting notes.


Just to expand on this, here's the models extended with methods that act very similar to the has_many :through from ActiveRecord by returning a query proxy instead of an array of records:

class Physician
  include Mongoid::Document
  has_many :appointments

  def patients
    Patient.in(id: appointments.pluck(:patient_id))
  end
end

class Appointment
  include Mongoid::Document
  belongs_to :physician
  belongs_to :patient
end

class Patient
  include Mongoid::Document
  has_many :appointments

  def physicians
    Physician.in(id: appointments.pluck(:physician_id))
  end
end

Steven Soroka solution is really great! I don't have the reputation to comment an answer(That's why I'm adding a new answer :P) but I think using map for a relationship is expensive(specially if your has_many relationship have hunders|thousands of records) because it gets the data from database, build each record, generates the original array and then iterates over the original array to build a new one with the values from the given block.

Using pluck is faster and maybe the fastest option.

class Physician
  include Mongoid::Document
  has_many :appointments

  def patients
    Patient.in(id: appointments.pluck(:patient_id))
  end
end

class Appointment
  include Mongoid::Document
  belongs_to :physician
  belongs_to :patient 
end

class Patient
  include Mongoid::Document
  has_many :appointments 

  def physicians
    Physician.in(id: appointments.pluck(:physician_id))
  end
end

Here some stats with Benchmark.measure:

> Benchmark.measure { physician.appointments.map(&:patient_id) }
 => #<Benchmark::Tms:0xb671654 @label="", @real=0.114643818, @cstime=0.0, @cutime=0.0, @stime=0.010000000000000009, @utime=0.06999999999999984, @total=0.07999999999999985> 

> Benchmark.measure { physician.appointments.pluck(:patient_id) }
 => #<Benchmark::Tms:0xb6f4054 @label="", @real=0.033517774, @cstime=0.0, @cutime=0.0, @stime=0.0, @utime=0.0, @total=0.0> 

I am using just 250 appointments. Don't forget to add indexes to :patient_id and :physician_id in Appointment document!

I hope it helps, Thanks for reading!