Sidekiq::FairTenant

Throttle “greedy” clients’ jobs to ensure more or less fair distribution of resources between clients.

This tiny Sidekiq middleware will re-route client's jobs after certain threshold to throttled queues (defined by you), where they will be processed with reduced priority.

“Weighted queues” feature of Sidekiq allows to de-prioritize jobs in throttled queues, so they will not block jobs from other clients, at the same time preserving overall throughput.

1. Install the gem and add to the application's Gemfile by executing:

   bundle add sidekiq-fair_tenant

2. Add fair_tenant_queues section to sidekiq_options in your job class:

    class SomeJob
      sidekiq_options \
        queue: 'default',
   +    fair_tenant_queues: [
   +     { queue: 'throttled_2x', threshold: 100, per: 1.hour },
   +     { queue: 'throttled_4x', threshold:  10, per: 1.minute },
   +    ]
    end

3. Add tenant detection login into your job class:

    class SomeJob
   +  def self.fair_tenant(*_perform_arguments)
   +    # Return any string that will be used as tenant name
   +    "tenant_1"
   +  end
    end

4. Add throttled queues with reduced weights to your Sidekiq configuration:

    # config/sidekiq.yml
    :queues:
      - [default, 4]
   +  - [throttled_2x, 2]
   +  - [throttled_4x, 1]

   See Sidekiq Advanced options for Queues to learn more about queue weights.

In your job class, add fair_tenant_queues section to sidekiq_options as array of hashes with following keys:

• queue - throttled queue name to re-route jobs into.
• threshold - maximum number of jobs allowed to be enqueued within per seconds.
• per - sliding time window in seconds to count jobs (you can use ActiveSupport Durations in Rails).

You can specify multiple rules and they all will be checked. Last matching rule will be used, so order rules from least to most restrictive.

Example:

sidekiq_options \
  queue: 'default',
  fair_tenant_queues: [
    # First rule is less restrictive, reacting to a large number of jobs enqueued in a long time window
    { queue: 'throttled_2x', threshold: 1_000, per: 1.day },
    # Next rule is more restrictive, reacting to spikes of jobs in a short time window
    { queue: 'throttled_4x', threshold:    10, per: 1.minute },
  ]

1. Explicitly during job enqueuing:

   SomeJob.set(fair_tenant: 'tenant_1').perform_async

2. Dynamically using fair_tenant class-level method in your job class (receives same arguments as perform)

   class SomeJob
     def self.fair_tenant(*_perform_arguments)
       # Return any string that will be used as tenant name
       "tenant_1"
     end
   end

3. Set fair_tenant job option in a custom middleware earlier in the stack.

4. Or let this gem automatically pick tenant name from apartment-sidekiq if you're using apartment gem.

Configuration is handled by anyway_config gem. With it you can load settings from environment variables (which names are constructed from config key upcased and prefixed with SIDEKIQ_FAIR_TENANT_), YAML files, and other sources. See anyway_config docs for details.

If number of jobs enqueued by a single client exceeds some threshold per a sliding time window, their jobs would be re-routed to another queue, with lower priority.

This gem tracks single client's jobs in a Redis sorted set with job id as a key and enqueuing timestamp as a score. When a new job is enqueued, it is added to the set, and then the set is trimmed to contain only jobs enqueued within the last max_throttling_window seconds.

On every enqueue attempt, the set is checked for number of jobs enqueued within the last per seconds of every rule. If the number of jobs in this time window exceeds threshold, the job is enqueued to a throttled queue, otherwise it is enqueued to the default queue. If multiple rules match, last one is used.

You are expected to configure Sidekiq to process throttled queues with lower priority using queue weights.

• If fast queues are empty then slow queues are processed at full speed (no artificial delays)
• If fast queues are full, slow queues are still processed, but slower (configurable), so application doesn’t “stall” for throttled users
• Minimal changes to the application code are required.

• As Sidekiq does not support mixing ordered and weighted queue modes (as stated in Sidekiq Wiki on queue configuration), you can’t make the same worker process execute some super important queue always first, ignoring other queues. Run separate worker to solve this.
• You have to keep track of all your queues and their weights.

After checking out the repo, run bin/setup to install dependencies. Then, run rake spec to run the tests. You can also run bin/console for an interactive prompt that will allow you to experiment.

To install this gem onto your local machine, run bundle exec rake install. To release a new version, update the version number in version.rb, and then run bundle exec rake release, which will create a git tag for the version, push git commits and the created tag, and push the .gem file to rubygems.org.

Bug reports and pull requests are welcome on GitHub at https://github.com/Envek/sidekiq-fair_tenant.

The gem is available as open source under the terms of the MIT License.