sidekiq-scheduler

sidekiq-scheduler is an extension to Sidekiq that pushes jobs in a scheduled way, mimicking cron utility.

Note: If you are looking for version 2.2.*, go to 2.2-stable branch.

Installation

gem install sidekiq-scheduler

Usage

Hello World

# hello-scheduler.rb

require 'sidekiq-scheduler'

class HelloWorld
  include Sidekiq::Worker

  def perform
    puts 'Hello world'
  end
end

# config/sidekiq.yml

:schedule:
  hello_world:
    cron: '0 * * * * *'   # Runs once per minute
    class: HelloWorld

Run sidekiq:

sidekiq -r ./hello-scheduler.rb

You'll see the following output:

2016-12-10T11:53:08.561Z 6452 TID-ovouhwvm4 INFO: Loading Schedule
2016-12-10T11:53:08.561Z 6452 TID-ovouhwvm4 INFO: Scheduling HelloWorld {"cron"=>"0 * * * * *", "class"=>"HelloWorld"}
2016-12-10T11:53:08.562Z 6452 TID-ovouhwvm4 INFO: Schedules Loaded

2016-12-10T11:54:00.212Z 6452 TID-ovoulivew HelloWorld JID-b35f36a562733fcc5e58444d INFO: start
Hello world
2016-12-10T11:54:00.213Z 6452 TID-ovoulivew HelloWorld JID-b35f36a562733fcc5e58444d INFO: done: 0.001 sec

2016-12-10T11:55:00.287Z 6452 TID-ovoulist0 HelloWorld JID-b7e2b244c258f3cd153c2494 INFO: start
Hello world
2016-12-10T11:55:00.287Z 6452 TID-ovoulist0 HelloWorld JID-b7e2b244c258f3cd153c2494 INFO: done: 0.001 sec

Configuration options

Configuration options are placed inside sidekiq.yml config file.

Available options are:

:dynamic: <if true the schedule can be modified in runtime [false by default]>
:dynamic_every: <if dynamic is true, the schedule is reloaded every interval [5s by default]>
:enabled: <enables scheduler if true [true by default]>
:scheduler:
  :listened_queues_only: <push jobs whose queue is being listened by sidekiq [false by default]>

Schedule configuration

The schedule is configured through the :schedule config entry in the sidekiq config file:

:schedule:
  CancelAbandonedOrders:
    cron: '0 */5 * * * *'   # Runs when second = 0, every 5 minutes

  queue_documents_for_indexing:
    cron: '0 0 * * * *'   # Runs every hour

    # By default the job name will be taken as worker class name.
    # If you want to have a different job name and class name, provide the 'class' option
    class: QueueDocuments

    queue: slow
    args: ['*.pdf']
    description: "This job queues pdf content for indexing in solr"

    # Enable the `metadata` argument which will pass a Hash containing the schedule metadata
    # as the last argument of the `perform` method. `false` by default.
    include_metadata: true

    # Enable / disable a job. All jobs are enabled by default.
    enabled: true

Schedule metadata

You can configure Sidekiq-scheduler to pass an argument with metadata about the scheduling process to the worker's perform method.

In the configuration file add the following on each worker class entry:

  SampleWorker:
    include_metadata: true

On your perform method, expect an additional argument:

  def perform(args, ..., metadata)
    # Do something with the metadata
  end

The metadata hash contains the following keys:

  metadata.keys =>
    [
      :scheduled_at # The epoch when the job was scheduled to run
    ]

Schedule types

Supported types are cron, every, interval, at, in.

Cron, every, and interval types push jobs into sidekiq in a recurrent manner.

cron follows the same pattern as cron utility, with seconds resolution.

:schedule:
  HelloWorld:
    cron: '0 * * * * *' # Runs when second = 0

every triggers following a given frequency:

    every: '45m'    # Runs every 45 minutes

The value is parsed by Fugit::Duration.parse. It understands quite a number of formats, including human-readable ones:

    every: 45 minutes
    every: 2 hours and 30 minutes
    every: 1.5 hours

interval is similar to every, the difference between them is that interval type schedules the next execution after the interval has elapsed counting from its last job enqueue.

Note that every and interval count from when the Sidekiq process (re)starts. So every: '48h' will never run if the Sidekiq process is restarted daily, for example. You can do every: ['48h', first_in: '0s'] to make the job run immediately after a restart, and then have the worker check when it was last run.

At, and in types push jobs only once. at schedules in a point in time:

    at: '3001/01/01'

You can specify any string that DateTime.parse and Chronic understand. To enable Chronic strings, you must add it as a dependency.

in triggers after a time duration has elapsed:

    in: 1h # pushes a sidekiq job in 1 hour, after start-up

You can provide options to every or cron via an Array:

    every: ['30s', first_in: '120s']

See https://github.com/jmettraux/rufus-scheduler for more information.

Load the schedule from a different file

You can place the schedule configuration in a separate file from config/sidekiq.yml

# sidekiq_scheduler.yml

clear_leaderboards_contributors:
  cron: '0 30 6 * * 1'
  class: ClearLeaderboards
  queue: low
  args: contributors
  description: 'This job resets the weekly leaderboard for contributions'

Please notice that the schedule root key is not present in the separate file.

To load the schedule:

require 'sidekiq'
require 'sidekiq-scheduler'

Sidekiq.configure_server do |config|
  config.on(:startup) do
    Sidekiq.schedule = YAML.load_file(File.expand_path('../../sidekiq_scheduler.yml', __FILE__))
    SidekiqScheduler::Scheduler.instance.reload_schedule!
  end
end

The above code can be placed in an initializer (in config/initializers) that runs every time the app starts up.

Dynamic schedule

The schedule can be modified after startup. To add / update a schedule, you have to:

Sidekiq.set_schedule('heartbeat', { 'every' => ['1m'], 'class' => 'HeartbeatWorker' })

If the schedule did not exist it will be created, if it existed it will be updated.

When :dynamic flag is set to true, schedule changes are loaded every 5 seconds. Use the :dynamic_every flag for a different interval.

# config/sidekiq.yml
:dynamic: true

If :dynamic flag is set to false, you'll have to reload the schedule manually in sidekiq side:

SidekiqScheduler::Scheduler.instance.reload_schedule!

Invoke Sidekiq.get_schedule to obtain the current schedule:

Sidekiq.get_schedule
#  => { 'every' => '1m', 'class' => 'HardWorker' }

Time zones

Note that if you use the cron syntax and are not running a Rails app, this will be interpreted in the server time zone.

In a Rails app, rufus-scheduler (>= 3.3.3) will use the config.time_zone specified in Rails.

You can explicitly specify the time zone that rufus-scheduler will use:

    cron: '0 30 6 * * 1 Europe/Stockholm'

Also note that config.time_zone in Rails allows for a shorthand (e.g. "Stockholm") that rufus-scheduler does not accept. If you write code to set the scheduler time zone from the config.time_zone value, make sure it's the right format, e.g. with:

ActiveSupport::TimeZone.find_tzinfo(Rails.configuration.time_zone).name

Notes about connection pooling

If you're configuring your own Redis connection pool, you need to make sure the size is adequate to be inclusive of both Sidekiq's own connection pool and Rufus Scheduler's.

That's a minimum of concurrency + 5 (per the Sidekiq wiki) + Rufus::Scheduler::MAX_WORK_THREADS (28 as of this writing; per the Rufus README), for a total of 58 with the default concurrency of 25.

You can also override the thread pool size in Rufus Scheduler by setting e.g.:

SidekiqScheduler::Scheduler.instance.rufus_scheduler_options = { max_work_threads: 5 }

Notes about running on Multiple Hosts

Under normal conditions, cron and at jobs are pushed once regardless of the number of sidekiq-scheduler running instances, assumming that time deltas between hosts is less than 24 hours.

Non-normal conditions that could push a specific job multiple times are:

high cpu load + a high number of jobs scheduled at the same time, like 100 jobs
network / redis latency + 28 (see MAX_WORK_THREADS https://github.com/jmettraux/rufus-scheduler/blob/master/lib/rufus/scheduler.rb#L41) or more jobs scheduled within the same network latency window

every, interval and in jobs will be pushed once per host.

Notes on when sidekiq worker is down

For a cron/at (and all other) job to be successfully enqueued, you need at least one sidekiq worker with scheduler to be up at that moment. Handling this is up to you and depends on your application.

Possible solutions include:

Simply ignoring this fact, if you only run frequent periodic jobs, that can tolerate some increased interval
Abstaining from deploys/restarts during time when critical jobs are usually scheduled
Making your infrequent jobs idempotent (so that they can be enqueued multiple times but still produce result as if was run once) and scheduling them multiple times to reduce likelihood of not being run
Zero downtime deploy for sidekiq workers: keep at least one worker up during whole deploy and only restart/shut it down after when new one has started
Running scheduler inside your unicorn/rails processes (if you already have zero downtime deploy set up for these)

Each option has it's own pros and cons.

Sidekiq Web Integration

sidekiq-scheduler provides an extension to the Sidekiq web interface that adds a Recurring Jobs page.

# config.ru

require 'sidekiq/web'
require 'sidekiq-scheduler/web'

run Sidekiq::Web

ActiveJob integration

When using sidekiq-scheduler with ActiveJob your jobs can just extend ApplicationJob as usual, without the require and include boilerplate. Under the hood Rails will load up the scheduler and include the worker module for you.

class HelloWorld < ApplicationJob
  def perform
    puts 'Hello world'
  end
end

The Spring preloader and Testing your initializer via Rails console

If you're pulling in your schedule from a YML file via an initializer as shown, be aware that the Spring application preloader included with Rails will interefere with testing via the Rails console.

Spring will not reload initializers unless the initializer is changed. Therefore, if you're making a change to your YML schedule file and reloading Rails console to see the change, Spring will make it seem like your modified schedule is not being reloaded.

To see your updated schedule, be sure to reload Spring by stopping it prior to booting the Rails console.

Run spring stop to stop Spring.

For more information, see this issue and Spring's README.

Manage tasks from Unicorn/Rails server

If you want start sidekiq-scheduler only from Unicorn/Rails, but not from sidekiq you can have something like this in an initializer:

# config/initializers/sidekiq_scheduler.rb
require 'sidekiq'
require 'sidekiq-scheduler'

puts "Sidekiq.server? is #{Sidekiq.server?.inspect}"
puts "defined?(Rails::Server) is #{defined?(Rails::Server).inspect}"
puts "defined?(Unicorn) is #{defined?(Unicorn).inspect}"

if Rails.env == 'production' && (defined?(Rails::Server) || defined?(Unicorn))
  Sidekiq.configure_server do |config|

    config.on(:startup) do
      Sidekiq.schedule = YAML.load_file(File.expand_path('../../scheduler.yml', __FILE__))
      SidekiqScheduler::Scheduler.instance.reload_schedule!
    end
  end
else
  SidekiqScheduler::Scheduler.instance.enabled = false
  puts "SidekiqScheduler::Scheduler.instance.enabled is #{SidekiqScheduler::Scheduler.instance.enabled.inspect}"
end

License

MIT License

channainfo/sidekiq-scheduler