For The Web

Getting Started


net/http is pretty much not good. Additionally, DNS behavior in ruby changes quite frequently.

I primarily want two things in both client and server operations:

  • A consistent API with good documentation, readable code, and high quality tests.
  • Modern web features: websockets, spdy, etc.

Desired features:

  • Awesome documentation
  • A HTTP client that acts as a full user agent, not just a single connections. (With connection reuse)
  • HTTP and SPDY support.
  • WebSockets support.
  • SSL/TLS support.
  • Browser Agent features like cookies and caching
  • An API that lets me do what I need.
  • Server and Client modes.
  • Support for both normal operation and EventMachine would be nice.

For reference:

Agent API

Reference: FTW::Agent

Common case

agent =

request = agent.get("")
response = request.execute

# Simpler
response = agent.get!("").read


  • This is not implemented yet

SPDY should automatically be attempted. The caller should be unaware.

I do not plan on exposing any direct means for invoking SPDY.


# 'http(s)' or 'ws(s)' urls are valid here. They will mean the same thing.
websocket = agent.websocket!("http://somehost/endpoint")

websocket.publish("Hello world")
websocket.each do |message|
  puts :received => message

Web Server API

I have implemented a rack server, Rack::Handler::FTW. It does not comply fully with the Rack spec. See 'Rack Compliance Issues' below.

Under the FTW rack handler, there is an environment variable added, "ftw.connection". This will be a FTW::Connection you can use for CONNECT, Upgrades, etc.

There's also a websockets wrapper, FTW::WebSockets::Rack, that will help you specifically with websocket requests and such.

Rack Compliance issues

Due to some awkward and bad requirements - specifically those around the specified behavior of 'rack.input' - I can't support the rack specification fully.

The 'rack.input' must be an IO-like object supporting #rewind which rewinds to the beginning of the request.

For high-data connections (like uploads, HTTP CONNECT, and HTTP Upgrade), it's not practical to hold the entire history of time in a buffer. We'll run out of memory, you crazy fools!

Details here:

Other Projects

Here are some related projects that I have no affiliation with:

Given some of the above (especially the server-side stuff), I'm likely try and integrate with those projects. For example, writing a Faye handler that uses the FTW server, if the FTW web server even stays around.