Obviously, the first thing you need to do is make sure you always have access to the current account as keyed by the subdomain. This means a before filter on any controller that runs under the subdomain that loads in your account. The idea is that once you load the account, you never have to pull it from the database again:

class AccountSubdomainController < ApplicationController
  before_filter :current_account

protected
  def current_account
    @account ||= params[:current_account] ||= get_account_by_subdomain
    params[:user][:current_account] = @account if params[:user]
    @account
  end

  def get_account_by_subdomain
    Account.where(:subdomain => request.subdomain.downcase).first
  end
end

If you’re not using an account-specific subdomain, just modify this to pull out the account from the URL or something.

Once we have our account, we need to inject the current account into the params to keep it over the whole request. We also need them in the users subhash if authentication is currently being run. That means you need to define current_account in your authentication keys, either in the devise.rb initializer or on your model’s devise invocation:

config.authentication_keys = [:email, :current_account]

This tells devise to use not only the email request parameter but also the current_account parameter to look up users. Now we just need to override the lookup method for authentication:

  def self.find_for_database_authentication(conditions)
    acct = conditions[:current_account] || Account.where("users.email" => conditions[:email]).first
    acct && acct.users.where(:email => conditions[:email]).first
  end

That’s sufficient for the initial login, but if you stop there your app will authenticate correctly, store the user in the session, but never be able to pull it back out. We need to a way to get to the params to pull that current_account out and provide it to the user model for a proper lookup. This is where shit gets a little hacky, because we’re going to create an initializer that overrides how Warden does something:

class Warden::SessionSerializer
  def deserialize(keys)
    klass_name, *args = keys

    # add current account into mix so we don't have to pull it from the db again!
    args << params[:current_account] if params[:current_account]

    begin
      klass = ActiveSupport::Inflector.constantize(klass_name)
      if klass.respond_to? :serialize_from_session
        klass.serialize_from_session(*args)
      else
        Rails.logger.warn "[Devise] Stored serialized class #{klass_name} seems not to be Devise enabled anymore. Did you do that on purpose?"
        nil
      end
    rescue NameError => e
      if e.message =~ /uninitialized constant/
        Rails.logger.debug "[Devise] Trying to deserialize invalid class #{klass_name}"
        nil
      else
        raise
      end
    end
  end
end

That’s the link between the current_account in the request and our user model. Now we just need a way to use it, and instead of using self.find like that old gist, we’d be better off implementing this on our model:

def self.serialize_from_session(*args)
  key, salt, account = args
  single_key = key.is_a?(Array) ? key.first : key
  account.users.find single_key
end

You should be all set now. Note that this hack has not been well tested, but I thought it was high time somebody shared a different approach. Please advise if you have criticisms or a better way.

But what if you want to reorder the embedded documents? Should be simple to sort an array. Our ODM - Mongoid in this case - would never represent the embedded collection as an array and not let us work with it as an array, right?

class Container
	include Mongoid::Document
	embeds_many :items
end

class Item
	include Mongoid::Document
	embedded_in :container
	
	field :title, :as => String
end

c = Container.create
c.items.create :title => "first"
c.items.create :title => "second"

>> c.items
=> [#<Item _id: 4dd2d971322bcdab7c000003, title: "first", _id: BSON::ObjectId('4dd2d971322bcdab7c000003'), _type: nil>, #<Item _id: 4dd2d981322bcdab7c000004, title: "second", _id: BSON::ObjectId('4dd2d981322bcdab7c000004'), _type: nil>]
>> c.items.reverse!
=> [#<Item _id: 4dd2d981322bcdab7c000004, title: "second", _id: BSON::ObjectId('4dd2d981322bcdab7c000004'), _type: nil>, #<Item _id: 4dd2d971322bcdab7c000003, title: "first", _id: BSON::ObjectId('4dd2d971322bcdab7c000003'), _type: nil>]
>> c.save
=> true
>> c.reload.items
=> [#<Item _id: 4dd2d971322bcdab7c000003, title: "first", _id: BSON::ObjectId('4dd2d971322bcdab7c000003'), _type: nil>, #<Item _id: 4dd2d981322bcdab7c000004, title: "second", _id: BSON::ObjectId('4dd2d981322bcdab7c000004'), _type: nil>]

OK, so not that easy, but maybe this means we just need to set the new array explicitly.

>> c.items = c.items.reverse
=> []

Yikes. So we can treat it as an array as much as we want - as long as we don’t need to persist it. Keep in mind this is a MongoDB trait, not a failing of the ODM per se (though one would expect the ODM to help us out here!).

So what do we do? Well, I worked around it by explicitly rebuilding the array of embedded documents:

>> reordered_items = c.items.reverse
=> [#<Item _id: 4dd2e0b7322bcdae0d000002, title: "second", _id: BSON::ObjectId('4dd2e0b7322bcdae0d000002'), _type: nil>, #<Item _id: 4dd2e0b6322bcdae0d000001, title: "first", _id: BSON::ObjectId('4dd2e0b6322bcdae0d000001'), _type: nil>]
>> c.items.clear
=> []
>> reordered_items.each { |i| c.items.create i.attributes }
=> [#<Item _id: 4dd2e0b7322bcdae0d000002, title: "second", _id: BSON::ObjectId('4dd2e0b7322bcdae0d000002'), _type: nil>, #<Item _id: 4dd2e0b6322bcdae0d000001, title: "first", _id: BSON::ObjectId('4dd2e0b6322bcdae0d000001'), _type: nil>]
>> c.reload.items
=> [#<Item _id: 4dd2e0b7322bcdae0d000002, title: "second", _id: BSON::ObjectId('4dd2e0b7322bcdae0d000002'), _type: nil>, #<Item _id: 4dd2e0b6322bcdae0d000001, title: "first", _id: BSON::ObjectId('4dd2e0b6322bcdae0d000001'), _type: nil>]

This works for me, but it’s not very elegant. Any problems with this approach? Feel free to let me know.

The tests are there but nominal; I’d love to see them fleshed out. I also had some ideas for making it cooler; for example, I keep an array of an object’s descendants in the object, allowing me to hierarchically order objects. This opens up some novel means to simplify how I implement the parent and children accessor. Imagine this:

field :ancestry, type => Array # contains ids of all ancestors including self, already exists

# but instead of a parent_id accessor

def parent_id
  ancestry.at(-2) # the parent can be fetched from the ancestry list
end

This also allows all descendants of a given object to be easily fetched - if the id shows up in the ancestry, return it! It’s this kind of out-of-the-box thinking that has really endeared MongoDB to me. I hope you can benefit from this and help me improve it. Or help with greedy. I’d love to get said help at CVREG’s upcoming Why Day hackfest.

Since this project is something I’m doing in my off time, I decided to experiment with MongoDB using the Mongoid framework. I had played with MongoMapper before, but always felt like I was using an ActiveRecord clone that didn’t take advantage of the full capabilities of a document database and was forcing and ActiveRecord-style approach on me. With Mongoid you get has_many, has_one, and belongs_to relationships that map to MongoDB concepts like embedded documents. Mongoid is fully compatible with the ActiveModel interface for Rails3, and things like associations and nested attributes work out of the box.

I also had been hearing great things about CarrierWave from co-workers. It employs the concept of an “uploader” outside of the MVC ecosystem. The uploader handles resizing, storage, and all other details. In your model, you simply “mount” the uploader and you’re golden. Of course, for this project the killer feature is the GridFS storage option, which is something I wanted to play with.

GridFS is a feature of MongoDB that allows storage of large files inside the database. It chunks the file into pieces and assigns a database index so the file can be associated with other documents. While CarrierWave takes care of getting the file into GridFS, serving the file back out is a bit trickier. CarrierWave gives you all you need to configure the url at which your file will be served, but you have to roll your own mechanism for serving it. But I’ll walk you through it.

I’ll assume you have the right version of Rails 3 installed - it’s changing too frequently to document here. Once you’ve got that, create your app with rails appname --skip-activerecord and navigate up in that piece.

I’ve discovered through trial and error that your Bundler config for all the gems mentioned here should track their respective master git branches, since a lot of the Rails 3 compatibility issues are still being worked out. I also track Rails edge because, at this point, why wouldn’t you? So, get your Gemfile looking like this:

source :rubygems
gem 'bson_ext'

gem 'rails', :git => 'https://github.com/rails/rails.git', :branch => 'master'
gem 'mongoid', :git => 'git://github.com/durran/mongoid.git'
gem 'carrierwave', :git => "git://github.com/jnicklas/carrierwave.git"
gem 'mini_magick', :git => 'git://github.com/probablycorey/mini_magick.git'

I included MiniMagick, but you can choose your own image processing library - just remember to change lines referencing MiniMagick in future code examples.

Now let’s install your gem bundle. I recommend calling bundle install vendor to get all your gems vendored in your Rails app, which makes tracking gems much more straightforward.

Next step is to run rails generate mongoid:config which generates a config/mongoid.yml file. You’ll need to fill in the details, but I recommend something a bit like this just to get started:

defaults: &defaults
  host: localhost

development:
  <<: *defaults
  database: appname_development

test:
  <<: *defaults
  database: appname_test

The generator will also place require 'mongoid/railtie' at the top of your config/application.rb file. I recommend setting up your generators with Mongoid by adding this line in the config block:

config.generators do |g|
  g.orm :mongoid
  g.template_engine :erb # this could be :haml or whatever
  g.test_framework :test_unit, :fixture => false # this could be :rpsec or whatever
end

Those generator settings will allow the resource generator to give you exactly the models, views, and controllers you want by invoking rails generate scaffold thing. Now let’s go into app/models/thing.rb and add the following:

require 'carrierwave/orm/mongoid'

class Thing
  include Mongoid::Document
  mount_uploader :image, ImageUploader
end

This is the equivalent of Paperclip’s has_attached_file or Attachment_fu’s has_attachment - except that you don’t do all the configuration for upload processing there. Instead you do it in the uploader class, which you can generate by invoking rails generate uploader image. This will create an uploaders/image.rb file, which is a slight quirk because Rails doesn’t know how to find this file when looking up the ImageUploader class. We’re going to change the name of the file to uploaders/image_uploader.rb so it conforms to ruby’s conventions for class definition files.

I’m going to use MiniMagick to process thumbnails, so here’s my fully configured uploader file:

require 'carrierwave/processing/mini_magick'

class ImageUploader < CarrierWave::Uploader::Base
  include CarrierWave::MiniMagick
  
  version :thumb do
    process :resize_to_fill => [80,80]
  end
end

There’s some CarrierWave settings we should set up globally, such as all the MongoDB and GridFS stuff. An initializer is the best place for that junk, so stick this in a new file called config/initializers/carrierwave.rb:

CarrierWave.configure do |config|
  config.grid_fs_database = Mongoid.database.name
  config.grid_fs_host = Mongoid.config.master.connection.host
  config.storage = :grid_fs
  config.grid_fs_access_url = "/images"
end

Note the config.grid_fs_access_url = "/images" line, which helps CarrierWave figure out what url to serve this under. The actual url it generates will look like /images/uploads/version_filename.jpg, which is fine for now - you can configure this to your liking later.

Now we just need to update the form to allowing file uploads. Make your “thing” form partial look like this, noting the file field and mulitpart form lines in particular:

<%= form_for(@thing, :html => { :multipart => true }) do |f| %>
  <% if @thing.errors.any? %>
    <div id="error_explanation">
      <h2><%= pluralize(@thing.errors.count, "error") %> prohibited this thing from being saved:</h2>

      <ul>
      <% @thing.errors.full_messages.each do |msg| %>
        <li><%= msg %></li>
      <% end %>
      </ul>
    </div>
  <% end %>
  
  <%= f.label :image %>
  <%= f.file_field :image %>

  <div class="actions">
    <%= f.submit %>
  </div>
<% end %>

And let’s amend the show view to display the image. Notice we pass the version into the url method to access a particular version.

<p id="notice"><%= notice %></p>
<%= image_tag @thing.image.url(:thumb) %>

<%= link_to 'Edit', edit_thing_path(@thing) %> |
<%= link_to 'Back', things_path %>

We can start up the server by invoking rails server. Navigating to https://localhost:3000/things/new should give us a form where we can select an image to upload. The image should get uploaded and the “thing” saved without a hitch, but when it redirects you to view the “thing” there will be a broken image waiting for you. This is because Rails has no freakin’ clue how to access the file via GridFS. So we need to tell it how.

In order to serve the image as quickly as possible, we need a way to access GridFS without involving the entire Rails slow-ass stack. Enter Rails Metal, which allows you to process requests directly from Rack. While Rails 2 required you to place metal processing in its own directory under app, Rails 3 bakes Rack support directly into the inheritance hierarchy of ActionController, allowing you to do something like this:

require 'mongo'

class GridfsController < ActionController::Metal
  def serve
    gridfs_path = env["PATH_INFO"].gsub("/images/", "")
    begin
      gridfs_file = Mongo::GridFileSystem.new(Mongoid.database).open(gridfs_path, 'r')
      self.response_body = gridfs_file.read
      self.content_type = gridfs_file.content_type
    rescue
      self.status = :file_not_found
      self.content_type = 'text/plain'
      self.response_body = ''
    end
  end
end

Save this file as app/controllers/gridfs_controller. Notice that we’re pulling details about the request path directly out of the request. By default, CarrierWave stores files in GridFS under “uploads/filename”. Therefore, we need to turn the request path (/images/uploads/filename) into a GridFS file path by simply removing the “/images/” (note both slashes). All of these settings are fully configurable in CarrierWave, but that’s beyond the scope of this article - just don’t forget to modify this controller if you change the url or GridFS storage path.

Now the last part is to set up the route for the image. Open up config/routes.rb and add a line for our GridfsController:

Example::Application.routes.draw do |map|
  match "/images/uploads/*path" => "gridfs#serve"
  resources :things
end

This maps a url like /images/uploads/thumb_image.jpg to the GridfsController’s serve action.

You should now see an thumbnail image on the show view for the “thing”. Congratulations - you’re cooking with GridFS now! There’s a ton of other cool stuff in Rails 3, CarrierWave, and Mongoid, but this should give you the basics for how to handle uploads with GridFS. Have fun, and let me know if I fucked anything up!