worker-killer

Kill any workers by memory and/or request counts or take custom reaction. Inspired by unicorn-worker-killer.

worker-killer gem provides automatic restart of Web-server and/or background job processor based on 1) max number of requests, and 2) process memory size (RSS). This will greatly improves site’s stability by avoiding unexpected memory exhaustion at the application nodes.

Features:

• generic middleware implementation
• Phusion Passenger support(through passenger-config detach-process <PID>)
• Puma support through native plugin
• DelayedJob support
• custom reaction hook

Install

No external process like god is required. Just install one gem: worker-killer.

  gem 'worker-killer'

Usage

Rack-based Web-server

Add these lines to your config.ru or application.rb. (These lines should be added above the require ::File.expand_path('../config/environment', __FILE__) line.

  # self-process killer
  require 'worker_killer/middleware'

  killer = WorkerKiller::Killer::Passenger.new

  # Max requests per worker
  middleware.insert_before(
    Rack::Runtime,
    WorkerKiller::Middleware::RequestsLimiter, killer: killer, min: 3072, max: 4096
  )

  # Max memory size (RSS) per worker
  middleware.insert_before(
    Rack::Runtime,
    WorkerKiller::Middleware::OOMLimiter, killer: killer, min: nil, max: 0.5, check_cycle: 16
  )

DelayedJob background processor

Add these lines to your initializers/delayed_job.rb or application.rb.

  # self-process killer
  require 'worker_killer/delayed_job_plugin'

  Delayed::Worker.plugins.tap do |plugins|
    killer = WorkerKiller::Killer::DelayedJob.new

    plugins << WorkerKiller::DelayedJobPlugin::JobsLimiter.new(
      killer: killer, min: 200, max: 300
    )

    plugins << WorkerKiller::DelayedJobPlugin::OOMLimiter.new(
      killer: killer, min: 500 * (1024**2), max: 600 * (1024**2)
    )
  end

Puma Web-server

Add these lines to your puma.rb AND application.rb.

  # puma.rb

  require 'worker_killer/puma/plugin/worker_killer'
  plugin('worker_killer')

  # application.rb
  if defined?(::Puma::Client)
    killer = ::WorkerKiller::PumaPlugin.instance.killer

    config.middleware.insert_before(
      Rack::Runtime,
      WorkerKiller::Middleware::OOMLimiter, killer: killer, min: nil, max: 0.5, verbose: false, check_cycle: 16
    )

    config.middleware.insert_before(
      Rack::Runtime,
      WorkerKiller::Middleware::RequestsLimiter, killer: killer, min: 3000, max: 4000, verbose: false
    )
  end

This gem provides two modules: WorkerKiller::CountLimiter and WorkerKiller::MemoryLimiter, some Rack integration, DelayedJob plugin and Puma plugin.

WorkerKiller::Middleware::RequestsLimiter and WorkerKiller::DelayedJobPlugin::JobsLimiter

This module automatically restarts/kills the workers, based on the number of requests/jobs which worker processed.

min and max specify the min and max of maximum requests per worker. The actual limit is decided by rand() between min and max per worker, to prevent all workers to be dead at the same time. Once the number exceeds the limit, that worker is automatically restarted.

If verbose is set to true, then after every request, your log will show the requests left before restart. This logging is done at the info level.

WorkerKiller::Middleware::OOMLimiter and WorkerKiller::DelayedJobPlugin::OOMLimiter

This module automatically restarts/kills the workers, based on its memory size.

min and max specify the min and max of maximum memory in bytes per worker. The actual limit is decided by rand() between min and max per worker, to prevent all workers to be dead at the same time. Once the memory size exceeds memory_size, that worker is automatically restarted.

The memory size check is done in every check_cycle requests.

If verbose is set to true, then every memory size check will be shown in your logs. This logging is done at the info level.

WorkerKiller::PumaPlugin

Works in Puma Cluster Mode. Work correctly with phased-restart and fork_workers, but has maximum effectivity (memory usage) in simple Cluster Mode with preload_app!.

Example production config:

  port ENV.fetch('PUMA_PORT', 3000).to_i

  environment ENV.fetch('RAILS_ENV', 'development')

  pidfile ENV.fetch('PUMA_PIDFILE', 'tmp/pids/server.pid')

  threads ENV.fetch('PUMA_THREADS', 4).to_i

  workers ENV.fetch('PUMA_WORKERS', 3).to_i

  require 'worker_killer/puma/plugin/worker_killer'
  plugin('worker_killer')

  preload_app!

Special Thanks

• @hotchpotch for the original idea
• @kzk for the unicorn-worker-killer