HTTP Request Rate Limiter for Rack Applications

This is a Rack middleware that provides logic for rate-limiting incoming HTTP requests to Rack applications. You can use Rack::Throttle with any Ruby web framework based on Rack, including with Ruby on Rails 3.0 and with Sinatra.

http://github.com/bensomers/improved-rack-throttle

Features

Throttles a Rack application by enforcing a minimum time interval between subsequent HTTP requests from a particular client, as well as by defining a maximum number of allowed HTTP requests per a given time period (hourly or daily).
Scopes throttling rules by request path, http method, or user_agent for applications that need a variety of rules
Compatible with any Rack application and any Rack-based framework.
Stores rate-limiting counters in any key/value store implementation that responds to #[]/#[]= (like Ruby's hashes) or to #get/#set (like memcached or Redis). This makes it easy to use global counters across multiple web servers.
Compatible with the gdbm binding included in Ruby's standard library.
Compatible with the memcached, memcache-client, memcache and redis gems.
Compatible with Heroku's memcached add-on

Examples

Adding throttling to a Rails 3.x application

# config/application.rb
require 'rack/throttle'

class Application < Rails::Application
  config.middleware.use Rack::Throttle::Interval
end

Adding throttling to a Sinatra application

#!/usr/bin/env ruby -rubygems
require 'sinatra'
require 'rack/throttle'

use Rack::Throttle::Interval

get('/hello') { "Hello, world!\n" }

Adding throttling to a Rackup application

#!/usr/bin/env rackup
require 'rack/throttle'

use Rack::Throttle::Interval

run lambda { |env| [200, {'Content-Type' => 'text/plain'}, "Hello, world!\n"] }

Enforcing a minimum 3-second interval between requests

use Rack::Throttle::Interval, :min => 3.0

Allowing a maximum of 100 requests per hour, for the foo basic auth

user

use Rack::Throttle::Hourly,   :max => 100, :rules => {:basic_auth =>

/foo/}

Allowing a maximum of 100 requests per hour

use Rack::Throttle::Hourly,   :max => 100

Allowing a maximum of 1,000 requests per day

use Rack::Throttle::Daily,    :max => 1000

Allowing 1 request per second, with bursts of up to 5 requests

use Rack::Throttle::SlidingWindow, :average => 1, :burst => 5

Combining various throttling constraints into one overall policy

use Rack::Throttle::Daily,    :max => 1000  # requests
use Rack::Throttle::Hourly,   :max => 100   # requests
use Rack::Throttle::Interval, :min => 3.0   # seconds

Storing the rate-limiting counters in a GDBM database

require 'gdbm'

use Rack::Throttle::Interval, :cache => GDBM.new('tmp/throttle.db')

Storing the rate-limiting counters on a Memcached server

require 'memcached'

use Rack::Throttle::Interval, :cache => Memcached.new, :key_prefix => :throttle

Storing the rate-limiting counters on a Redis server

require 'redis'

use Rack::Throttle::Interval, :cache => Redis.new, :key_prefix => :throttle

Scoping the rate-limit to a specific path and method

use Rack::Throttle::Interval, :rules => {:url => /api/, :method => :post}

Throttling Strategies

Rack::Throttle supports four built-in throttling strategies:

Rack::Throttle::Interval: Throttles the application by enforcing a minimum interval (by default, 1 second) between subsequent HTTP requests.
Rack::Throttle::Hourly: Throttles the application by defining a maximum number of allowed HTTP requests per hour (by default, 3,600 requests per 60 minutes, which works out to an average of 1 request per second).
Rack::Throttle::Daily: Throttles the application by defining a maximum number of allowed HTTP requests per day (by default, 86,400 requests per 24 hours, which works out to an average of 1 request per second).
Rack::Throttle::SlidingWindow: Throttles the application by defining an average request rate, and a burst allowance that clients can hit (as long as they stay within the average). By default, this is 1 request per second, with bursts of up to 5 requests at a time. Users who exceed the average and who have used up their burst will have all of their requests denied until they comply with the policy.

You can fully customize the implementation details of any of these strategies by simply subclassing one of the aforementioned default implementations. And, of course, should your application-specific requirements be significantly more complex than what we've provided for, you can also define entirely new kinds of throttling strategies by subclassing the Rack::Throttle::Limiter base class directly.

Scoping Rules

Rack::Throttle ships with a Rack::Throttle::Matcher base class, and three implementations. Rack::Throttle::UrlMatcher and Rack::Throttle::UserAgentMatcher allow you to pass in regular expressions for request path and user agent, while Rack::Throttle::MethodMatcher will filter by request method when passed a Symbol :get, :post, :put, or :delete. These rules are additive, so you can throttle just POST requests to your '/login' page, for example.

HTTP Client Identification

The rate-limiting counters stored and maintained by Rack::Throttle are keyed to unique HTTP clients.

By default, HTTP clients are uniquely identified by their IP address as returned by Rack::Request#ip. If you wish to instead use a more granular, application-specific identifier such as a session key or a user account name, you can subclass a throttling strategy implementation and override the #client_identifier method.

HTTP Response Codes and Headers

403 Forbidden (Rate Limit Exceeded)

When a client exceeds their rate limit, Rack::Throttle by default returns a "403 Forbidden" response with an associated "Rate Limit Exceeded" message in the response body.

An HTTP 403 response means that the server understood the request, but is refusing to respond to it and an accompanying message will explain why. This indicates an error on the client's part in exceeding the rate limits outlined in the acceptable use policy for the site, service, or API.

503 Service Unavailable (Rate Limit Exceeded)

However, there exists a widespread practice of instead returning a "503 Service Unavailable" response when a client exceeds the set rate limits. This is technically dubious because it indicates an error on the server's part, which is certainly not the case with rate limiting - it was the client that committed the oops, not the server.

An HTTP 503 response would be correct in situations where the server was genuinely overloaded and couldn't handle more requests, but for rate limiting an HTTP 403 response is more appropriate. Nonetheless, if you think otherwise, Rack::Throttle does allow you to override the returned HTTP status code by passing in a :code => 503 option when constructing a Rack::Throttle::Limiter instance.

Documentation

http://rubydoc.info/gems/improved-rack-throttle

{Rack::Throttle}
- {Rack::Throttle::Interval}
- {Rack::Throttle::Daily}
- {Rack::Throttle::Hourly}
- {Rack::Throttle::SlidingWindow}
- {Rack::Throttle::Matcher}
- {Rack::Throttle::MethodMatcher}
- {Rack::Throttle::UrlMatcher}
- {Rack::Throttle::UserAgentMatcher}
- {Rack::Throttle::BasicAuthMatcher}

Dependencies

Rack (>= 1.0.0)

Installation

The recommended installation method is via RubyGems. To install the latest official release of the gem, do:

% [sudo] gem install improved-rack-throttle

Download

To get a local working copy of the development repository, do:

% git clone git://github.com/bensomers/improved-rack-throttle.git

Alternatively, you can download the latest development version as a tarball as follows:

% wget http://github.com/bensomers/improved-rack-throttle/tarball/master

Authors

License

Rack::Throttle is free and unencumbered public domain software. For more information, see http://unlicense.org/ or the accompanying UNLICENSE file.

Support

Recent work on improved-rack-throttle has been funded by Rafter, Inc.