ProxyMachine

By Tom Preston-Werner (tom@mojombo.com)

ProxyMachine is a simple content aware (layer 7) TCP routing proxy built on EventMachine that lets you configure the routing logic in Ruby.

If you need to proxy connections to different backend servers depending on the contents of the transmission, then ProxyMachine will make your life easy!

The idea here is simple. For each client connection, start receiving data chunks and placing them into a buffer. Each time a new chunk arrives, send the buffer to a user specified block. The block's job is to parse the buffer to determine where the connection should be proxied. If the buffer contains enough data to make a determination, the block returns the address and port of the correct backend server. If not, it can choose to do nothing and wait for more data to arrive, close the connection, or close the connection after sending custom data. Once the block returns an address, a connection to the backend is made, the buffer is replayed to the backend, and the client and backend connections are hooked up to form a transparent proxy. This bidirectional proxy continues to exist until either the client or backend close the connection.

ProxyMachine was developed for GitHub's federated architecture and is successfully used in production to proxy millions of requests every day. The performance and memory profile have both proven to be excellent.

$ gem install proxymachine -s http://gemcutter.org

Usage:
  proxymachine -c <config file> [-h <host>] [-p <port>]

Options:
  -c, --config CONFIG              Configuration file
  -h, --host HOST                  Hostname to bind. Default 0.0.0.0
  -p, --port PORT                  Port to listen on. Default 5432

QUIT - Graceful shutdown. Stop accepting connections immediately and
       wait as long as necessary for all connections to close.

TERM - Fast shutdown. Stop accepting connections immediately and wait
       up to 10 seconds for connections to close before forcing
       termination.

INT  - Same as TERM

class GitRouter
  # Look at the routing table and return the correct address for +name+
  # Returns "<host>:<port>" e.g. "ae8f31c.example.com:9418"
  def self.lookup(name)
    ...
  end
end

# Perform content-aware routing based on the stream data. Here, the
# header information from the Git protocol is parsed to find the 
# username and a lookup routine is run on the name to find the correct
# backend server. If no match can be made yet, do nothing with the
# connection.
proxy do |data|
  if data =~ %r{^....git-upload-pack /([\w\.\-]+)/[\w\.\-]+\000host=\w+\000}
    name = $1
    { :remote => GitRouter.lookup(name) }
  else
    { :noop => true }
  end
end

proxy do |data|
  return  if data.size < 9
  v, c, port, o1, o2, o3, o4, user = data.unpack("CCnC4a*")
  return { :close => "\0\x5b\0\0\0\0\0\0" }  if v != 4 or c != 1
  return  if ! idx = user.index("\0")
  { :remote => "#{[o1,o2,o3,o4]*'.'}:#{port}",
    :reply => "\0\x5a\0\0\0\0\0\0",
    :data => data[idx+9..-1] }
end

{ :remote => String } - String is the host:port of the backend server that will be proxied.
{ :remote => String, :data => String } - Same as above, but send the given data instead.
{ :remote => String, :data => String, :reply => String} - Same as above, but reply with given data back to the client { :noop => true } - Do nothing.
{ :close => true } - Close the connection.
{ :close => String } - Close the connection after sending the String.

It's possible to register a custom callback for handling connection errors. The callback is passed the remote when a connection is either rejected or a connection timeout occurs:

proxy do |data|
  if data =~ /your thing/
    { :remote => 'localhost:1234', :connect_timeout => 1.0 }
  else
    { :noop => true }
  end
end

proxy_connect_error do |remote|
  puts "error connecting to #{remote}"
end

You must provide a :connect_timeout value in the proxy return value to enable connection timeouts. The :connect_timeout value is a float representing the number of seconds to wait before a connection is established. Hard connection rejections always trigger the callback, even when no :connect_timeout is provided.

Inactivity timeouts work like connect timeouts but are triggered after the configured amount of time elapses without receiving the first byte of data from an already connected server:

proxy do |data|
  { :remote => 'localhost:1234', :inactivity_timeout => 10.0 }
end

proxy_inactivity_error do |remote|
  puts "#{remote} did not send any data for 10 seconds"
end

If no :inactivity_timeout is provided, the proxy_inactivity_error callback is never triggered.

If you'd like to hack on ProxyMachine, start by forking my repo on GitHub:

http://github.com/mojombo/proxymachine

To get all of the dependencies, install the gem first. The best way to get your changes merged back into core is as follows:

1. Clone down your fork
2. Create a topic branch to contain your change
3. Hack away
4. Add tests and make sure everything still passes by running rake
5. If you are adding new functionality, document it in the README.md
6. Do not change the version number, I will do that on my end
7. If necessary, rebase your commits into logical chunks, without errors
8. Push the branch up to GitHub
9. Send me (mojombo) a pull request for your branch

Copyright (c) 2009 Tom Preston-Werner. See LICENSE for details.