Action Cable – Integrated websockets for Rails

Build Status

Action Cable seamlessly integrates websockets with the rest of your Rails application. It allows for real-time features to be written in Ruby in the same style and form as the rest of your Rails application, while still being performant and scalable. It's a full-stack offering that provides both a client-side JavaScript framework and a server-side Ruby framework. You have access to your full domain model written with ActiveRecord or your ORM of choice.

Terminology

A single Action Cable server can handle multiple connection instances. It has one connection instance per websocket connection. A single user may have multiple websockets open to your application if they use multiple browser tabs or devices. The client of a websocket connection is called the consumer.

Each consumer can in turn subscribe to multiple cable channels. Each channel encapsulates a logical unit of work, similar to what a controller does in a regular MVC setup. For example, you could have a ChatChannel and a AppearancesChannel, and a consumer could be subscribed to either or to both of these channels. At the very least, a consumer should be subscribed to one channel.

When the consumer is subscribed to a channel, they act as a subscriber. The connection between the subscriber and the channel is, surprise-surprise, called a subscription. A consumer can act as a subscriber to a given channel any number of times. For example, a consumer could subscribe to multiple chat rooms at the same time. (And remember that a physical user may have multiple consumers, one per tab/device open to your connection).

Each channel can then again be streaming zero or more broadcastings. A broadcasting is a pubsub link where anything transmitted by the broadcaster is sent directly to the channel subscribers who are streaming that named broadcasting.

As you can see, this is a fairly deep architectural stack. There's a lot of new terminology to identify the new pieces, and on top of that, you're dealing with both client and server side reflections of each unit.

A full-stack example

The first thing you must do is define your ApplicationCable::Connection class in Ruby. This is the place where you authorize the incoming connection, and proceed to establish it if all is well. Here's the simplest example starting with the server-side connection class:

# app/channels/application_cable/connection.rb
module ApplicationCable
  class Connection < ActionCable::Connection::Base
    identified_by :current_user

    def connect
      self.current_user = find_verified_user
    end

    protected
      def find_verified_user
        if current_user = User.find(cookies.signed[:user_id])
          current_user
        else
          reject_unauthorized_connection
        end
      end
  end
end

Then you should define your ApplicationCable::Channel class in Ruby. This is the place where you put shared logic between your channels.

# app/channels/application_cable/channel.rb
module ApplicationCable
  class Channel < ActionCable::Channel::Base
  end
end

This relies on the fact that you will already have handled authentication of the user, and that a successful authentication sets a signed cookie with the user_id. This cookie is then automatically sent to the connection instance when a new connection is attempted, and you use that to set the current_user. By identifying the connection by this same current_user, you're also ensuring that you can later retrieve all open connections by a given user (and potentially disconnect them all if the user is deleted or deauthorized).

The client-side needs to setup a consumer instance of this connection. That's done like so:

# app/assets/javascripts/application_cable.coffee
#= require cable

@App = {}
App.cable = Cable.createConsumer "ws://cable.example.com"

The ws://cable.example.com address must point to your set of Action Cable servers, and it must share a cookie namespace with the rest of the application (which may live under http://example.com). This ensures that the signed cookie will be correctly sent.

That's all you need to establish the connection! But of course, this isn't very useful in itself. This just gives you the plumbing. To make stuff happen, you need content. That content is defined by declaring channels on the server and allowing the consumer to subscribe to them.

Channel example 1: User appearances

Here's a simple example of a channel that tracks whether a user is online or not and what page they're on. (This is useful for creating presence features like showing a green dot next to a user name if they're online).

First you declare the server-side channel:

# app/channels/appearance_channel.rb
class AppearanceChannel < ApplicationCable::Channel
  def subscribed
    current_user.appear
  end

  def unsubscribed
    current_user.disappear
  end

  def appear(data)
    current_user.appear on: data['appearing_on']
  end

  def away
    current_user.away
  end
end

The #subscribed callback is invoked when, as we'll show below, a client-side subscription is initiated. In this case, we take that opportunity to say "the current user has indeed appeared". That appear/disappear API could be backed by Redis or a database or whatever else. Here's what the client-side of that looks like:

# app/assets/javascripts/cable/subscriptions/appearance.coffee
App.appearance = App.cable.subscriptions.create "AppearanceChannel",
  connected: ->
    # Called once the subscription has been successfully completed

  appear: ->
    @perform 'appear', appearing_on: @appearingOn()

  away: ->
    @perform 'away'

  appearingOn: ->
    $('main').data 'appearing-on'

$(document).on 'page:change', ->
  App.appearance.appear()

$(document).on 'click', '[data-behavior~=appear_away]', ->
  App.appearance.away()
  false

Simply calling App.cable.subscriptions.create will setup the subscription, which will call AppearanceChannel#subscribed, which in turn is linked to original App.consumer -> ApplicationCable::Connection instances.

We then link App.appearance#appear to AppearanceChannel#appear(data). This is possible because the server-side channel instance will automatically expose the public methods declared on the class (minus the callbacks), so that these can be reached as remote procedure calls via App.appearance#perform.

Finally, we expose App.appearance to the machinations of the application itself by hooking the #appear call into the Turbolinks page:change callback and allowing the user to click a data-behavior link that triggers the #away call.

Channel example 2: Receiving new web notifications

The appearance example was all about exposing server functionality to client-side invocation over the websocket connection. But the great thing about websockets is that it's a two-way street. So now let's show an example where the server invokes action on the client.

This is a web notification channel that allows you to trigger client-side web notifications when you broadcast to the right streams:

# app/channels/web_notifications_channel.rb
class WebNotificationsChannel < ApplicationCable::Channel
   def subscribed
     stream_from "web_notifications_#{current_user.id}"
   end
 end
# Somewhere in your app this is called, perhaps from a NewCommentJob
ActionCable.server.broadcast \
  "web_notifications_#{current_user.id}", { title: 'New things!', body: 'All the news that is fit to print' }

# Client-side which assumes you've already requested the right to send web notifications
App.cable.subscriptions.create "WebNotificationsChannel",
  received: (data) ->
    new Notification data['title'], body: data['body']

The ActionCable.server.broadcast call places a message in the Redis' pubsub queue under a separate broadcasting name for each user. For a user with an ID of 1, the broadcasting name would be web_notifications_1. The channel has been instructed to stream everything that arrives at web_notifications_1 directly to the client by invoking the #received(data) callback. The data is the hash sent as the second parameter to the server-side broadcast call, JSON encoded for the trip across the wire, and unpacked for the data argument arriving to #received.

More complete examples

See the rails/actioncable-examples repository for a full example of how to setup Action Cable in a Rails app and adding channels.

Configuration

The only must-configure part of Action Cable is the Redis connection. By default, ActionCable::Server::Base will look for a configuration file in Rails.root.join('config/redis/cable.yml'). The file must follow the following format:

production: &production
  :url: redis://10.10.3.153:6381
  :host: 10.10.3.153
  :port: 6381
  :timeout: 1
development: &development
  :url: redis://localhost:6379
  :host: localhost
  :port: 6379
  :timeout: 1
  :inline: true
test: *development

This format allows you to specify one configuration per Rails environment. You can also change the location of the Redis config file in a Rails initializer with something like:

ActionCable.server.config.redis_path = Rails.root('somewhere/else/cable.yml')

The other common option to configure is the log tags applied to the per-connection logger. Here's close to what we're using in Basecamp:

ActionCable.server.config.log_tags = [
  -> request { request.env['bc.account_id'] || "no-account" },
  :action_cable,
  -> request { request.uuid }
]

For a full list of all configuration options, see the ActionCable::Server::Configuration class.

Also note that your server must provide at least the same number of database connections as you have workers. The default worker pool is set to 100, so that means you have to make at least that available. You can change that in config/database.yml through the pool attribute.

Starting the cable server

As mentioned, the cable server(s) is separated from your normal application server. It's still a rack application, but it is its own rack application. The recommended basic setup is as follows:

# cable/config.ru
require ::File.expand_path('../../config/environment',  __FILE__)
Rails.application.eager_load!

require 'action_cable/process/logging'

run ActionCable.server

Then you start the server using a binstub in bin/cable ala:

#!/bin/bash
bundle exec puma -p 28080  cable/config.ru

The above will start a cable server on port 28080. Remember to point your client-side setup against that using something like: App.cable.createConsumer('ws://basecamp.dev:28080').

Beware that currently the cable server will not auto-reload any changes in the framework. As we've discussed, long-running cable connections mean long-running objects. We don't yet have a way of reloading the classes of those objects in a safe manner. So when you change your channels, or the model your channels use, you must restart the cable server.

Note: We'll get all this abstracted properly when the framework is integrated into Rails.

Dependencies

Action Cable is currently tied to Redis through its use of the pubsub feature to route messages back and forth over the websocket cable connection. This dependency may well be alleviated in the future, but for the moment that's what it is. So be sure to have Redis installed and running.

The Ruby side of things is built on top of faye-websocket and celluloid.

Deployment

Action Cable is powered by a combination of EventMachine and threads. The framework plumbing needed for connection handling is handled in the EventMachine loop, but the actual channel, user-specified, work is handled in a normal Ruby thread. This means you can use all your regular Rails models with no problem, as long as you haven't committed any thread-safety sins.

But this also means that Action Cable needs to run in its own server process. So you'll have one set of server processes for your normal web work, and another set of server processes for the Action Cable. The former can be single-threaded, like Unicorn, but the latter must be multi-threaded, like Puma.

Alpha disclaimer

Action Cable is currently considered alpha software. The API is almost guaranteed to change between now and its first production release as part of Rails 5.0. Real applications using the framework are all well underway, but as of July 8th, 2015, there are no deployments in the wild yet.

So this current release, which resides in rails/actioncable, is primarily intended for the adventurous kind, who do not mind reading the full source code of the framework. And it serves as an invitation for all those crafty folks to contribute to and test what we have so far, in advance of that general production release.

Action Cable will move from rails/actioncable to rails/rails and become a full-fledged default framework alongside Action Pack, Active Record, and the like once we cross the bridge from alpha to beta software (which will happen once the API and missing pieces have solidified).

Finally, note that testing is a unfinished/unstarted area of this framework. The framework has been developed in-app up until this point. We need to find a good way to allow the user to test their connection and channel logic.

License

Action Cable is released under the MIT license:

Support

Bug reports can be filed for the alpha development project here: