Example of using a simple future library for parallel HTTP requests

0_app.rb

class TicketsController < ApplicationController
  def show
    tickets = params[:tickets].split(",")

    ticket_data = tickets.map do |ticket|
      parallel { Faraday.get("http://tickets.local/#{ticket}") }
    end

    render json: { tickets: ticket_data.map(&:result) }
  end
end

1_parallel.rb

require "future"

module ActionController
  module ParallelIO
    def parallel(&block)
      Future.new(&block)
    end
  end
end

ActiveSupport.on_load(:action_controller) do
  include ActionController::ParallelIO
end

futures.rb

require "thread"

class Future
  attr_reader :exception, :cancelled

  def initialize(&block)
    @thread = Thread.new(&block)
    @thread.abort_on_exception = false
    @exception = nil
    @cancelled = false
  end

  def running?
    @thread.alive?
  end

  def cancelled?
    @cancelled
  end

  def done?
    [email protected]?
  end

  def cancel
    @cancelled = true
    @thread.kill
  end

  def result(limit=nil)
    value(limit)
  rescue Exception => e
    @exception = e
    nil
  end

  def exception(limit=nil)
    result(limit)
    @exception
  end

private
  def value(limit)
    if @thread.join(limit)
      @thread.value
    else
      nil
    end
  end
end

q = Queue.new

require "uri"
require "net/http"
futures = {}

["http://www.google.com", "http://www.heroku.com", "http://www.github.com"].each do |url|
  futures[url] = Future.new { Net::HTTP.get(URI(url)) }
end

futures["http://www.github.com"].cancel

puts "heroku.com:"
p futures["http://www.heroku.com"].cancelled?
puts futures["http://www.heroku.com"].result
puts futures["http://www.heroku.com"].exception
puts

puts "google.com:"
p futures["http://www.google.com"].cancelled?
puts futures["http://www.google.com"].result
puts futures["http://www.google.com"].exception

puts "github.com:"
p futures["http://www.github.com"].cancelled?
puts futures["http://www.github.com"].result
puts futures["http://www.github.com"].exception

oh, it is so cool!

Concurrent? Yes. Parallel? Not really. Not on MRI at least. Or am I confused about GIL?

In case I am not confused, then possibly fibers would make a better job than threads?

@wycats: Thanks for the gist. May I suggest you provide a quick explanation of the upsides and benefits of such a technique for less technical users? i.e. explain what that code achieves and how exactly one would leverage such a feature in one's code?

@artemave JRuby and Rubinius both support parallel multithreading because they have fine-grained locking models instead of a GIL. So this is extremely relevant to those platforms.

Even on MRI/YARV, you can do parallel blocking I/O, and C extensions can explicitly release the GIL for computationally intensive activities as well.