Redisse

Redisse is a Redis-backed Ruby library for creating Server-Sent Events, publishing them from your application, and serving them to your clients.

• Homepage: github.com/tigerlily/redisse
• Documentation: tigerlily.github.io/redisse

Features

• Pub/Sub split into channels for privacy & access rights handling.

• SSE history via the Last-Event-Id header and the lastEventId query parameter, with a limit of 100 events per channel.

• Long-polling via the polling query parameter. Allows to send several events at once for long-polling clients by waiting one second before closing the connection.

• Lightweight: only one Redis connection for history and one for all subscriptions, no matter the number of connected clients.

• missedevents event fired when the full requested history could not be found, to allow the client to handle the case where events were missed.

• lastEventId event fired if no event was sent, to make sure the client knows an event id if the connection fails before it receives a regular event.

• Event types from SSE are left untouched for your application code, but keep in mind that a client will receive events of all types from their channels. To handle access rights, use channels instead.

Rationale

Redisse’s design comes from these requirements:

• The client wants to listen to several channels but use only one connection. (e.g. a single EventSource object is created in the browser but you want events coming from different Redis channels.)

• A server handles the concurrent connections so that the application servers don't need to (e.g. Unicorn workers).

• The application is written in Ruby, so there needs to be a Ruby API to publish events.

• The application is written on top of Rack, so the code that lists the Redis Pub/Sub channels to subscribe to needs to be able to use Rack middlewares and should receive a Rack environment. (e.g. you can use Warden as a middleware and simply use env['warden'].user to decide which channels the user can access.)

Redirect endpoint

The simplest way that last point can be fulfilled is by actually loading and running your code in the Redisse server. Unfortunately since it’s EventMachine-based, if your method takes a while to return the channels, all the other connected clients will be blocked too. You'll also have some duplication between your Rack config and Redisse server config.

Another way if you use nginx is instead to use a endpoint in your main application that will use the header X-Accel-Redirect to redirect to the Redisse server, which is now free from your blocking code. The channels will be sent instead via the redirect URL. See the section on nginx for more info.

Installation

Add this line to your application's Gemfile:

gem 'redisse', '~> 0.4.0'

Usage

Configure Redisse (e.g. in config/initializers/redisse.rb):

require 'redisse'

Redisse.channels do |env|
  %w[ global ]
end

Use the endpoint in your main application (in config.ru or your router):

# config.ru Rack
map "/events" do
  run Redisse.redirect_endpoint
end

# config/routes.rb Rails
get "/events" => Redisse.redirect_endpoint

Run the server:

$ bundle exec redisse --stdout --verbose

Get ready to receive events (with HTTPie or cURL):

$ http localhost:8080 Accept:text/event-stream --stream
$ curl localhost:8080 -H 'Accept: text/event-stream'

Send a Server-Sent Event:

Redisse.publish('global', success: "It's working!")

Check out the stats from you server

$ http localhost:8080 Accept:application/json
$ curl localhost:8080 -H 'Accept: application/json'

{
  "connected": 2,
  "served":    3,
  "events":    42,
  "missing":   0,
}

See what the stats mean.

Testing

In the traditional Rack app specs or tests, use Redisse.test_mode!:

describe "SSE" do
  before do
    Redisse.test_mode!
  end

  it "should send a Server-Sent Event" do
    post '/publish', channel: 'global', message: 'Hello'
    expect(Redisse.published.size).to be == 1
  end
end

See the example app specs.

Behind nginx

When running behind nginx as a reverse proxy, you should disable buffering (proxy_buffering off) and close the connection to the server when the client disconnects (proxy_ignore_client_abort on) to preserve resources (otherwise connections to Redis will be kept alive longer than necessary).

You should take advantage of the redirect endpoint instead of directing the SSE requests to the SSE server. Let your Rack application determine the channels, but have the request served by the SSE server with a redirect (X-Accel-Redirect) to an internal location.

In this case, and if you have a large number of long-named channels, the internal redirect URL will be long and you might need to increase proxy_buffer_size from its default in your Rack application location configuration. For example, 8k will allow you about 200 channels with UUIDs as names, which is quite a lot.

You can check the nginx conf of the example for all the details.

Contributing

1. Fork it
2. Create your feature branch (git checkout -b my-new-feature)
3. Commit your changes (git commit -am 'Add some feature')
4. Push to the branch (git push origin my-new-feature)
5. Create new Pull Request