enova / sidekiq-rate-limiter

Redis-backed, per-worker rate limits for job processing.
http://rubygems.org/gems/sidekiq-rate-limiter
MIT License
193 stars 43 forks source link

sidekiq-rate-limiter

Gem Version Test Status Coverage Status

Redis-backed, per-worker rate limits for job processing.

Compatibility

sidekiq-rate-limiter is actively tested against MRI versions 2.7 and 3.1.

sidekiq-rate-limiter works by using a custom fetch class, the class responsible for pulling work from the queue stored in Redis. Consequently, you'll want to be careful about using other gems that use the same strategy, sidekiq-priority being one example.

I've attempted to support the same options as used by sidekiq-throttler. So, if your worker already looks like this example I lifted from the sidekiq-throttler wiki:

class MyWorker
  include Sidekiq::Worker

  sidekiq_options throttle: { threshold: 50, period: 1.hour }

  def perform(user_id)
    # Do some heavy API interactions.
  end
end

Then you wouldn't need to change anything.

Installation

Add this line to your application's Gemfile:

gem 'sidekiq-rate-limiter'

And then execute:

$ bundle

Or install it yourself as:

$ gem install sidekiq-rate-limiter

Configuration

See server.rb for an example of how to configure sidekiq-rate-limiter. Alternatively, you can add the following to your initializer or what-have-you:

require 'sidekiq-rate-limiter/server'

Or, if you prefer, amend your Gemfile like so:

gem 'sidekiq-rate-limiter', require: 'sidekiq-rate-limiter/server'

By default, the limiter uses the name sidekiq-rate-limiter. You can define the constant Sidekiq::RateLimiter::DEFAULT_LIMIT_NAME prior to requiring to change this. Alternatively, you can include a name parameter in the configuration hash included in sidekiq_options

For example, the following:

  class Job
    include Sidekiq::Worker

    sidekiq_options queue: 'some_silly_queue',
                    rate: {
                      name:   'my_super_awesome_rate_limit',
                      limit:  50,
                      period: 3600, ## An hour
                    }

    def perform(*args)
      ## do stuff
      ## ...

The configuration above would result in any jobs beyond the first 50 in a one-hour period being delayed. The server will continue to fetch items from Redis, & will place any items that are beyond the threshold at the back of their queue.

Dynamic Configuration

The simplest way to set the rate-limiting options (:name, :limit, and :period) is to assign them each a static value (as above). In some cases, you may wish to calculate values for these options for each specific job. You can do this by supplying a Proc for any or all of these options.

The Proc may receive as its arguments the same values that will be passed to perform when the job is finally performed.

class Job
  include Sidekiq::Worker

  sidekiq_options queue: "my_queue",
                  rate: {
                    name:   ->(user_id, rate_limit) { user_id },
                    limit:  ->(user_id, rate_limit) { rate_limit },
                    period: ->{ Date.today.monday? ? 2.hours : 4.hours }, # can ignore arguments
                  }

  def perform(user_id, rate_limit)
    ## do something

Caveat: Normally, Sidekiq stores the sidekiq_options with the job on your Redis server at the time the job is enqueued, and it is these stored values that are used for rate-limiting. This means that if you deploy a new version of your code with different sidekiq_options, the already-queued jobs will continue to behave according to the options that were in place when they were created. When you supply a Proc for one or more of your configuration options, your rate-limiting options can no longer be stored in Redis, but must instead be calculated when the job is fetched by your Sidekiq server for potential execution. If your application code changes while a job is in the queue, it may run with different sidekiq_options than existed when it was first enqueued.

Motivation

Sidekiq::Throttler is great for smaller quantities of jobs but falls down a bit for larger queues (see issue #8). In addition, jobs that are limited multiple times are counted as 'processed' each time, so the stats balloon quickly.

TODO

Contributing

  1. Fork
  2. Commit
  3. Pull Request

License

MIT. See LICENSE for details.