faye-websocket

This is a general-purpose WebSocket implementation extracted from the Faye project. It provides classes for easily building WebSocket servers and clients in Ruby. It does not provide a server itself, but rather makes it easy to handle WebSocket connections within an existing Rack application. It does not provide any abstraction other than the standard WebSocket API.

It also provides an abstraction for handling EventSource connections, which are one-way connections that allow the server to push data to the client. They are based on streaming HTTP responses and can be easier to access via proxies than WebSockets.

The following web servers are supported. Other servers that implement the rack.hijack API should also work.

• Goliath
• Phusion Passenger >= 4.0 with nginx >= 1.4
• Puma >= 2.0
• Rainbows
• Thin

$ gem install faye-websocket

You can handle WebSockets on the server side by listening for requests using the Faye::WebSocket.websocket? method, and creating a new socket for the request. This socket object exposes the usual WebSocket methods for receiving and sending messages. For example this is how you'd implement an echo server:

# app.rb
require 'faye/websocket'

App = lambda do |env|
  if Faye::WebSocket.websocket?(env)
    ws = Faye::WebSocket.new(env)

    ws.on :message do |event|
      ws.send(event.data)
    end

    ws.on :close do |event|
      p [:close, event.code, event.reason]
      ws = nil
    end

    # Return async Rack response
    ws.rack_response

  else
    # Normal HTTP request
    [200, { 'Content-Type' => 'text/plain' }, ['Hello']]
  end
end

Note that under certain circumstances (notably a draft-76 client connecting through an HTTP proxy), the WebSocket handshake will not be complete after you call Faye::WebSocket.new because the server will not have received the entire handshake from the client yet. In this case, calls to ws.send will buffer the message in memory until the handshake is complete, at which point any buffered messages will be sent to the client.

If you need to detect when the WebSocket handshake is complete, you can use the onopen event.

If the connection's protocol version supports it, you can call ws.ping() to send a ping message and wait for the client's response. This method takes a message string, and an optional callback that fires when a matching pong message is received. It returns true if and only if a ping message was sent. If the client does not support ping/pong, this method sends no data and returns false.

ws.ping 'Mic check, one, two' do
  # fires when pong is received
end

The client supports both the plain-text ws protocol and the encrypted wss protocol, and has exactly the same interface as a socket you would use in a web browser. On the wire it identifies itself as hybi-13.

require 'faye/websocket'
require 'eventmachine'

EM.run {
  ws = Faye::WebSocket::Client.new('ws://www.example.com/')

  ws.on :open do |event|
    p [:open]
    ws.send('Hello, world!')
  end

  ws.on :message do |event|
    p [:message, event.data]
  end

  ws.on :close do |event|
    p [:close, event.code, event.reason]
    ws = nil
  end
}

The WebSocket client also lets you inspect the status and headers of the handshake response via its status and headers methods.

To connect via a proxy, set the proxy option to the HTTP origin of the proxy, including any authorization information and custom headers you require:

ws = Faye::WebSocket::Client.new('ws://www.example.com/', [], {
  :proxy => {
    :origin  => 'http://username:password@proxy.example.com',
    :headers => { 'User-Agent' => 'ruby' }
  }
})

The WebSocket protocol allows peers to select and identify the application protocol to use over the connection. On the client side, you can set which protocols the client accepts by passing a list of protocol names when you construct the socket:

ws = Faye::WebSocket::Client.new('ws://www.example.com/', ['irc', 'amqp'])

On the server side, you can likewise pass in the list of protocols the server supports after the other constructor arguments:

ws = Faye::WebSocket.new(env, ['irc', 'amqp'])

If the client and server agree on a protocol, both the client- and server-side socket objects expose the selected protocol through the ws.protocol property.

faye-websocket is based on the websocket-extensions framework that allows extensions to be negotiated via the Sec-WebSocket-Extensions header. To add extensions to a connection, pass an array of extensions to the :extensions option. For example, to add permessage-deflate:

require 'permessage_deflate'

ws = Faye::WebSocket.new(env, [], :extensions => [PermessageDeflate])

Both the server- and client-side classes allow an options hash to be passed in at initialization time, for example:

ws = Faye::WebSocket.new(env, protocols, options)
ws = Faye::WebSocket::Client.new(url, protocols, options)

protocols as an array of subprotocols as described above, or nil. options is an optional hash containing any of these keys:

• :extensions - an array of websocket-extensions compatible extensions, as described above
• :headers - a hash containing key-value pairs representing HTTP headers to be sent during the handshake process
• :max_length - the maximum allowed size of incoming message frames, in bytes. The default value is 2^26 - 1, or 1 byte short of 64 MiB.
• :ping - an integer that sets how often the WebSocket should send ping frames, measured in seconds
• :tls - a hash containing key-value pairs for specifying TLS parameters. These are passed along to EventMachine and you can find more details here

Starting with version 0.11.0, Faye::WebSocket::Client will verify the server certificate for wss connections. This is not the default behaviour for EventMachine's TLS interface, and so our defaults for the :tls option are a little different.

First, :verify_peer is enabled by default. Our implementation checks that the chain of certificates sent by the server is trusted by your root certificates, and that the final certificate's hostname matches the hostname in the request URL.

By default, we use your system's root certificate store by invoking OpenSSL::X509::Store#set_default_paths. If you want to use a different set of root certificates, you can pass them via the :root_cert_file option, which takes a path or an array of paths to the certificates you want to use.

ws = Faye::WebSocket::Client.new('wss://example.com/', [], :tls => {
  :root_cert_file => ['path/to/certificate.pem']
})

If you want to switch off certificate verification altogether, then set :verify_peer to false.

ws = Faye::WebSocket::Client.new('wss://example.com/', [], :tls => {
  :verify_peer => false
})

Both the server- and client-side WebSocket objects support the following API:

• on(:open) { |event| } fires when the socket connection is established. Event has no attributes.
• on(:message) { |event| } fires when the socket receives a message. Event has one attribute, data, which is either a String (for text frames) or an Array of unsigned integers, i.e. integers in the range 0..255 (for binary frames).
• on(:error) { |event| } fires when there is a protocol error due to bad data sent by the other peer. This event is purely informational, you do not need to implement error recovery.
• on(:close) { |event| } fires when either the client or the server closes the connection. Event has two optional attributes, code and reason, that expose the status code and message sent by the peer that closed the connection.
• send(message) accepts either a String or an Array of byte-sized integers and sends a text or binary message over the connection to the other peer; binary data must be encoded as an Array.
• ping(message, &callback) sends a ping frame with an optional message and fires the callback when a matching pong is received.
• close(code, reason) closes the connection, sending the given status code and reason text, both of which are optional.
• version is a string containing the version of the WebSocket protocol the connection is using.
• protocol is a string (which may be empty) identifying the subprotocol the socket is using.

EventSource connections provide a very similar interface, although because they only allow the server to send data to the client, there is no onmessage API. EventSource allows the server to push text messages to the client, where each message has an optional event-type and ID.

# app.rb
require 'faye/websocket'

App = lambda do |env|
  if Faye::EventSource.eventsource?(env)
    es = Faye::EventSource.new(env)
    p [:open, es.url, es.last_event_id]

    # Periodically send messages
    loop = EM.add_periodic_timer(1) { es.send('Hello') }

    es.on :close do |event|
      EM.cancel_timer(loop)
      es = nil
    end

    # Return async Rack response
    es.rack_response

  else
    # Normal HTTP request
    [200, { 'Content-Type' => 'text/plain' }, ['Hello']]
  end
end

The send method takes two optional parameters, :event and :id. The default event-type is 'message' with no ID. For example, to send a notification event with ID 99:

es.send('Breaking News!', :event => 'notification', :id => '99')

The EventSource object exposes the following properties:

• url is a string containing the URL the client used to create the EventSource.
• last_event_id is a string containing the last event ID received by the client. You can use this when the client reconnects after a dropped connection to determine which messages need resending.

When you initialize an EventSource with Faye::EventSource.new, you can pass configuration options after the env parameter. Available options are:

• :headers is a hash containing custom headers to be set on the EventSource response.
• :retry is a number that tells the client how long (in seconds) it should wait after a dropped connection before attempting to reconnect.
• :ping is a number that tells the server how often (in seconds) to send 'ping' packets to the client to keep the connection open, to defeat timeouts set by proxies. The client will ignore these messages.

For example, this creates a connection that allows access from any origin, pings every 15 seconds and is retryable every 10 seconds if the connection is broken:

es = Faye::EventSource.new(es,
  :headers => { 'Access-Control-Allow-Origin' => '*' },
  :ping    => 15,
  :retry   => 10
)

You can send a ping message at any time by calling es.ping. Unlike WebSocket the client does not send a response to this; it is merely to send some data over the wire to keep the connection alive.

The following describes how to run a WebSocket application using all our supported web servers.

If you use Thin to serve your application you need to include this line after loading faye/websocket:

Faye::WebSocket.load_adapter('thin')

Thin can be started via the command line if you've set up a config.ru file for your application:

$ thin start -R config.ru -p 9292

Or, you can use rackup. In development mode, this adds middlewares that don't work with async apps, so you must start it in production mode:

$ rackup config.ru -s thin -E production -p 9292

It can also be started using the Rack::Handler interface common to many Ruby servers. You can configure Thin further in a block passed to run:

require 'eventmachine'
require 'rack'
require 'thin'
require './app'

Faye::WebSocket.load_adapter('thin')

thin = Rack::Handler.get('thin')

thin.run(App, :Port => 9292) do |server|
  # You can set options on the server here, for example to set up SSL:
  server.ssl_options = {
    :private_key_file => 'path/to/ssl.key',
    :cert_chain_file  => 'path/to/ssl.crt'
  }
  server.ssl = true
end

faye-websocket requires either Passenger for Nginx or Passenger Standalone. Apache doesn't work well with WebSockets at this time. You do not need any special configuration to make faye-websocket work, it should work out of the box on Passenger provided you use at least Passenger 4.0.

However, you do need to insert the following code in config.ru for optimal WebSocket performance in Passenger. This is documented in the Passenger manual.

if defined?(PhusionPassenger)
  PhusionPassenger.advertised_concurrency_level = 0
end

Run your app on Passenger for Nginx by creating a virtual host entry which points to your app's "public" directory:

server {
  listen 9292;
  server_name yourdomain.local;
  root /path-to-your-app/public;
  passenger_enabled on;
}

Or run your app on Passenger Standalone:

$ passenger start -p 9292

More information can be found on the Passenger website.

Puma has a command line interface for starting your application:

$ puma config.ru -p 9292

Or, you can use rackup. In development mode, this adds middlewares that don't work with async apps, so you must start it in production mode:

$ rackup config.ru -s puma -E production -p 9292

If you're using version 4.4 or lower of Rainbows, you need to run it with the EventMachine backend and enable the adapter. Put this in your rainbows.conf file:

Rainbows! { use :EventMachine }

And make sure you load the adapter in your application:

Faye::WebSocket.load_adapter('rainbows')

Version 4.5 of Rainbows does not need this adapter.

You can run your config.ru file from the command line. Again, Rack::Lint will complain unless you put the application in production mode.

$ rainbows config.ru -c path/to/rainbows.conf -E production -p 9292

If you use Goliath to server your application you need to include this line after loading faye/websocket:

Faye::WebSocket.load_adapter('goliath')

Goliath can be made to run arbitrary Rack apps by delegating to them from a Goliath::API instance. A simple server looks like this:

require 'goliath'
require './app'
Faye::WebSocket.load_adapter('goliath')

class EchoServer < Goliath::API
  def response(env)
    App.call(env)
  end
end

Faye::WebSocket can also be used inline within a Goliath app:

require 'goliath'
require 'faye/websocket'
Faye::WebSocket.load_adapter('goliath')

class EchoServer < Goliath::API
  def response(env)
    ws = Faye::WebSocket.new(env)

    ws.on :message do |event|
      ws.send(event.data)
    end

    ws.rack_response
  end
end