Use this in a Rails 3.1 app. Right now the only supported pubsub messaging system is Faye faye.jcoglan.com/.
This assumes you already have a Backbone.js + Rails app.
Run a Faye server. It's pretty straightforward, check out `example_faye/run.sh` in this repo.
Tell your app where the faye server is. This may differ per Rails.env.
For now, let's say we add `config/initializers/backbone_sync_rails_faye.rb` with:
“`ruby BackboneSync::Rails::Faye.root_address = 'localhost:9292' “`
(Until I rewrite it, one of 'em uses coffeescript, so you'll need that.)
Open a connection to Faye from your clients, somewhere on your page (in the layout?):
Observe model changes in Rails, and broadcast them. We provide the guts of
an observer for you, so add a file like `app/models/user_observer.rb`:
“`ruby class UserObserver < ActiveRecord::Observer
and enable it in `config/application.rb` like any good observer:
“`ruby module MyApp
class Application < Rails::Application # snip... # Activate observers that should always be running. config.active_record.observers = :user_observer # snip... end
Instantiate a new `BackboneSync.RailsFayeSynchronizer` for *each instance*
of a Backbone collection you instantiate. You could do this in the collection's constructor, or do it by hand:
Check it out! Open two browsers, do some stuff in one, and see your changes cascade to the other. It also assumes that your Backbone views will need to observe events on the collection like `change`, `add`, and `remove`.
Installing on Rails < 3.1
If you're on a version of Rails < 3.1, you'll probably have to copy some files into your app by hand, like the `vendor/assets` files. You'll probably have to require the `lib/backbone_sync-rails/faye.rb` file yourself, too.
In short, we augment the `Backbone.Collection.prototype._add` function so that adding multiple models to the same collection with the same `id` attribute (or your `idAttribute`-specified attribute of choice) will pass silently.
In a distributed messaging system, messages should be idempotent: this means that, for any message, an actor should be able to execute that message several times with no ill effect.
Why? Consider the following situation.
There are two clients, Alice and Bob.
Alice creates a new model in Backbone.
The server receives her request and persists it. It also distributes a
“create” message to all subscribed clients. 4. Alice's new model is added to her local collection in the normal due course of `Backbone.Model.prototype.save`.
Bob receives the create message and creates a model in his local collection.
All is well until this point. Now, Alice receives the create message (she
is subscribed just as Bob is) and creates a duplicate model into her collection.
There is actually a race condition in that Alice's HTTP request to create (and therefore her normal `save()`-based addition to the collection)_ may complete before or after the pubsub notification informs her collection to add a new member.
One approach to solving this would be for each update message to be tagged with its originating client, and for each client to filter out those messages. This would prove difficult, particularly since, in this implementation, the `ActiveModel::Observer` subclass is decoupled from the originating client.
I'm more than happy to hear about better approaches from people with more experience in distributed messaging systems.
Copyright © 2011 Jason Morrison. See MIT-LICENSE for details.