resque-retry

A Resque plugin. Requires Resque ~> 1.25 or Resque ~> 2.0 & resque-scheduler ~> 4.0.

This gem provides retry, delay and exponential backoff support for resque jobs.

• Redis backed retry count/limit.
• Retry on all or specific exceptions.
• Exponential backoff (varying the delay between retrys).
• Multiple failure backend with retry suppression & resque-web tab.
• Small & Extendable - plenty of places to override retry logic/settings.

Install & Quick Start

To install:

$ gem install resque-retry

If you're using Bundler to manage your dependencies, you should add gem 'resque-retry' to your Gemfile.

Add this to your Rakefile:

require 'resque/tasks'
require 'resque/scheduler/tasks'

The delay between retry attempts is provided by resque-scheduler. You'll want to run the scheduler process, otherwise delayed retry attempts will never perform:

$ rake resque:scheduler

Use the plugin:

require 'resque-retry'

class ExampleRetryJob
  extend Resque::Plugins::Retry
  @queue = :example_queue

  @retry_limit = 3
  @retry_delay = 60

  def self.perform(*args)
    # your magic/heavy lifting goes here.
  end
end

Then start up a resque worker as normal:

$ QUEUE=* rake resque:work

Now if you ExampleRetryJob fails, it will be retried 3 times, with a 60 second delay between attempts.

For more explanation and examples, please see the remaining documentation.

Failure Backend & Resque Web Additions

Lets say you're using the Redis failure backend of resque (the default). Every time a job fails, the failure queue is populated with the job and exception details.

Normally this is useful, but if your jobs retry... it can cause a bit of a mess.

For example: given a job that retried 4 times before completing successful. You'll have a lot of failures for the same job and you wont be sure if it actually completed successfully just by just using the resque-web interface.

Failure Backend

MultipleWithRetrySuppression is a multiple failure backend, with retry suppression.

Here's an example, using the Redis failure backend:

require 'resque-retry'
require 'resque/failure/redis'

# require your jobs & application code.

Resque::Failure::MultipleWithRetrySuppression.classes = [Resque::Failure::Redis]
Resque::Failure.backend = Resque::Failure::MultipleWithRetrySuppression

If a job fails, but can and will retry, the failure details wont be logged in the Redis failed queue (visible via resque-web).

If the job fails, but can't or won't retry, the failure will be logged in the Redis failed queue, like a normal failure (without retry) would.

Resque Web Additions

If you're using the MultipleWithRetrySuppression failure backend, you should also checkout the resque-web additions!

The new Retry tab displays delayed jobs with retry information; the number of attempts and the exception details from the last failure.

Configuring and running the Resque-Web Interface

Using a Rack configuration:

One alternative is to use a rack configuration file. To use this, make sure you include this in your config.ru or similar file:

require 'resque-retry'
require 'resque-retry/server'

# Make sure to require your workers & application code below this line:
# require '[path]/[to]/[jobs]/your_worker'

# Run the server
run Resque::Server.new

As an example, you could run this server with the following command:

rackup -p 9292 config.ru

When using bundler, you can also run the server like this:

bundle exec rackup -p 9292 config.ru

Using the 'resque-web' command with a configuration file:

Another alternative is to use resque's built-in 'resque-web' command with the additional resque-retry tabs. In order to do this, you must first create a configuration file. For the sake of this example we'll create the configuration file in a 'config' directory, and name it 'resque_web_config.rb'. In practice you could rename this configuration file to anything you like and place in your project in a directory of your choosing. The contents of the configuration file would look like this:

# [app_dir]/config/resque_web_config.rb
require 'resque-retry'
require 'resque-retry/server'

# Make sure to require your workers & application code below this line:
# require '[path]/[to]/[jobs]/your_worker'

Once you have the configuration file ready, you can pass the configuration file to the resque-web command as a parameter, like so:

resque-web [app_dir]/config/resque_web_config.rb

Retry Options & Logic

Please take a look at the yardoc/code for more details on methods you may wish to override.

Customisation is pretty easy, the below examples should give you some ideas =), adapt for your own usage and feel free to pick and mix!

Here are a list of the options provided (click to jump):

• Retry Defaults
• Custom Retry
• Sleep After Requeuing
• Exponential Backoff
• Retry Specific Exceptions
• Fail Fast For Specific Exceptions
• Custom Retry Criteria Check Callbacks
• Retry Arguments
• Job Retry Identifier/Key
• Expire Retry Counters From Redis
• Try Again and Give Up Callbacks
• Ignored Exceptions
• Debug Plugin Logging

Retry Defaults

Retry the job once on failure, with zero delay.

require 'resque-retry'

class DeliverWebHook
  extend Resque::Plugins::Retry
  @queue = :web_hooks

  def self.perform(url, hook_id, hmac_key)
    heavy_lifting
  end
end

When a job runs, the number of retry attempts is checked and incremented in Redis. If your job fails, the number of retry attempts is used to determine if we can requeue the job for another go.

Custom Retry

class DeliverWebHook
  extend Resque::Plugins::Retry
  @queue = :web_hooks

  @retry_limit = 10
  @retry_delay = 120

  def self.perform(url, hook_id, hmac_key)
    heavy_lifting
  end
end

The above modification will allow your job to retry up to 10 times, with a delay of 120 seconds, or 2 minutes between retry attempts.

You can override the retry_delay method to set the delay value dynamically. For example:

class ExampleJob
  extend Resque::Plugins::Retry
  @queue = :testing

  def self.retry_delay(exception)
    if exception == SocketError
      10
    else
      1
    end
  end

  def self.perform(*args)
    heavy_lifting
  end
end

Or, if you'd like the delay to be dependent on job arguments:

class ExampleJob
  extend Resque::Plugins::Retry
  @queue = :testing

  def self.retry_delay(exception, *args)
    # the delay is dependent on the arguments passed to the job
    # in this case, "3" is passed as the arg and that is used as the delay
    # make sure this method returns a integer
    args.first.to_i
  end

  def self.perform(*args)
    heavy_lifting
  end
end

Resque.enqueue(ExampleJob, '3')

Sleep After Requeuing

Sometimes it is useful to delay the worker that failed a job attempt, but still requeue the job for immediate processing by other workers. This can be done with @sleep_after_requeue:

class DeliverWebHook
  extend Resque::Plugins::Retry
  @queue = :web_hooks

  @sleep_after_requeue = 5

  def self.perform(url, hook_id, hmac_key)
    heavy_lifting
  end
end

This retries the job once and causes the worker that failed to sleep for 5 seconds after requeuing the job. If there are multiple workers in the system this allows the job to be retried immediately while the original worker heals itself. For example failed jobs may cause other (non-worker) OS processes to die. A system monitor such as monit or god can fix the server while the job is being retried on a different worker.

@sleep_after_requeue is independent of @retry_delay. If you set both, they both take effect.

You can override the sleep_after_requeue method to set the sleep value dynamically.

Exponential Backoff

Use this if you wish to vary the delay between retry attempts:

class DeliverSMS
  extend Resque::Plugins::ExponentialBackoff
  @queue = :mt_messages

  def self.perform(mt_id, mobile_number, message)
    heavy_lifting
  end
end

Default Settings

key: m = minutes, h = hours

                    0s, 1m, 10m,   1h,    3h,    6h
@backoff_strategy = [0, 60, 600, 3600, 10800, 21600]
@retry_delay_multiplicand_min = 1.0
@retry_delay_multiplicand_max = 1.0

The first delay will be 0 seconds, the 2nd will be 60 seconds, etc... Again, tweak to your own needs.

The number of retries is equal to the size of the backoff_strategy array, unless you set retry_limit yourself.

The delay values will be multiplied by a random Float value between retry_delay_multiplicand_min and retry_delay_multiplicand_max (both have a default of 1.0). The product (delay_multiplicand) is recalculated on every attempt. This feature can be useful if you have a lot of jobs fail at the same time (e.g. rate-limiting/throttling or connectivity issues) and you don't want them all retried on the same schedule.

Retry Specific Exceptions

The default will allow a retry for any type of exception. You may change it so only specific exceptions are retried using retry_exceptions:

class DeliverSMS
  extend Resque::Plugins::Retry
  @queue = :mt_messages

  @retry_exceptions = [NetworkError]

  def self.perform(mt_id, mobile_number, message)
    heavy_lifting
  end
end

The above modification will only retry if a NetworkError (or subclass) exception is thrown.

You may also want to specify different retry delays for different exception types. You may optionally set @retry_exceptions to a hash where the keys are your specific exception classes to retry on, and the values are your retry delays in seconds or an array of retry delays to be used similar to exponential backoff.

class DeliverSMS
  extend Resque::Plugins::Retry
  @queue = :mt_messages

  @retry_exceptions = { NetworkError => 30, SystemCallError => [120, 240] }

  def self.perform(mt_id, mobile_number, message)
    heavy_lifting
  end
end

In the above example, Resque would retry any DeliverSMS jobs which throw a NetworkError or SystemCallError. If the job throws a NetworkError it will be retried 30 seconds later, if it throws SystemCallError it will first retry 120 seconds later then subsequent retry attempts 240 seconds later.

Fail Fast For Specific Exceptions

The default will allow a retry for any type of exception. You may change it so specific exceptions fail immediately by using fatal_exceptions:

class DeliverSMS
  extend Resque::Plugins::Retry
  @queue = :mt_divisions

  @fatal_exceptions = [NetworkError]

  def self.perform(mt_id, mobile_number, message)
    heavy_lifting
  end
end

In the above example, Resque would retry any DeliverSMS jobs that throw any type of error other than NetworkError. If the job throws a NetworkError it will be marked as "failed" immediately.

You should use either @fatal_exceptions or @retry_exceptions. If you specify @fatal_exceptions the @retry_exceptions are ignored.

Custom Retry Criteria Check Callbacks

You may define custom retry criteria callbacks:

class TurkWorker
  extend Resque::Plugins::Retry
  @queue = :turk_job_processor

  @retry_exceptions = [NetworkError]

  retry_criteria_check do |exception, *args|
    if exception.message =~ /SpecialErrorMessageToRetry/
      return true
    end

    false
  end

  def self.perform(job_id)
    heavy_lifting
  end
end

Similar to the previous example, this job will retry if either a NetworkError (or subclass) exception is thrown or any of the callbacks return true.

You'll want to return false by default in the retry_criteria_check callback since the result of this callback is OR'd with the result of your retry_exceptions or fatal_exceptions configuration. In other words, if you returned true your retry_exceptions configuration would never be used.

If you want to AND the result of fatal_exceptions or retry_exceptions with custom retry criteria, you'll need to implement your own logic in a retry_criteria_check to check for fatal_exceptions or retry_exceptions.

You can also register a retry criteria check with a Symbol if the method is already defined on the job class:

class AlwaysRetryJob
  extend Resque::Plugins::Retry

  retry_criteria_check :yes

  def self.yes(ex, *args)
    true
  end
end

Use @retry_exceptions = [] and @fatal_exceptions = [] to only use your custom retry criteria checks to determine if the job should retry.

NB: Your callback must be able to accept the exception and job arguments as passed parameters, or else it cannot be called. e.g., in the example above, defining def self.yes; true; end would not work.

Retry Arguments

You may override retry_args, which is passed the current job arguments, to modify the arguments for the next retry attempt.

class DeliverViaSMSC
  extend Resque::Plugins::Retry
  @queue = :mt_smsc_messages

  # retry using the emergency SMSC.
  def self.retry_args(smsc_id, mt_message)
    [999, mt_message]
  end

  def self.perform(smsc_id, mt_message)
    heavy_lifting
  end
end

Alternatively, if you require finer control of the args based on the exception thrown, you may override retry_args_for_exception, which is passed the exception and the current job arguments, to modify the arguments for the next retry attempt.

class DeliverViaSMSC
  extend Resque::Plugins::Retry
  @queue = :mt_smsc_messages

  # retry using the emergency SMSC.
  def self.retry_args_for_exception(exception, smsc_id, mt_message)
    [999, mt_message + exception.message]
  end

  def self.perform(smsc_id, mt_message)
    heavy_lifting
  end
end

Custom Retry Queues

By default, when a job is retried, it is added to the @queue specified in the worker. However, you may want to push the job into another (lower or higher priority) queue when the job fails. You can do this by dynamically specifying the retry queue. For example:

class ExampleJob
  extend Resque::Plugins::Retry
  @queue = :testing
  @retry_delay = 1

  def self.work(*args)
    user_id, user_mode, record_id = *args
    
    Resque.enqueue_to(
      target_queue_for_args(user_id, user_mode, record_id),
      self,
      *args
     )
  end

  def self.retry_queue(exception, *args)
    target_queue_for_args(*args)
  end

  def self.perform(*args)
    heavy_lifting
  end

  def self.target_queue_for_args(*args)
    user_id, user_mode, record_id = *args

    if user_mode
      'high
    else
      'low'
    end
  end
end

Job Retry Identifier/Key

The retry attempt is incremented and stored in a Redis key. The key is built using the retry_identifier. If you have a lot of arguments or really long ones, you should consider overriding retry_identifier to define a more precise or loose custom retry identifier.

The default identifier is just your job arguments joined with a dash '-'.

By default the key uses this format: 'resque-retry:<job class name>:<retry_identifier>'.

Or you can define the entire key by overriding redis_retry_key.

class DeliverSMS
  extend Resque::Plugins::Retry
  @queue = :mt_messages

  def self.retry_identifier(mt_id, mobile_number, message)
    "#{mobile_number}:#{mt_id}"
  end

  def self.perform(mt_id, mobile_number, message)
    heavy_lifting
  end
end

Expire Retry Counters From Redis

Allow the Redis to expire stale retry counters from the database by setting @expire_retry_key_after:

class DeliverSMS
  extend Resque::Plugins::Retry
  @queue = :mt_messages
  @expire_retry_key_after = 3600 # expire key after `retry_delay` plus 1 hour

  def self.perform(mt_id, mobile_number, message)
    heavy_lifting
  end
end

This saves you from having to run a "house cleaning" or "errand" job.

The expiary timeout is "pushed forward" or "touched" after each failure to ensure it's not expired too soon.

Try Again and Give Up Callbacks

Resque's on_failure callbacks are always called, regardless of whether the job is going to be retried or not. If you want to run a callback only when the job is being retried, you can add a try_again_callback:

class LoggedJob
  extend Resque::Plugins::Retry

  try_again_callback do |exception, *args|
    logger.info("Received #{exception}, retrying job #{self.name} with #{args}")
  end
end

Similarly, if you want to run a callback only when the job has failed, and is not retrying, you can add a give_up_callback:

class LoggedJob
  extend Resque::Plugins::Retry

  give_up_callback do |exception, *args|
    logger.error("Received #{exception}, job #{self.name} failed with #{args}")
  end
end

You can register a callback with a Symbol if the method is already defined on the job class:

class LoggedJob
  extend Resque::Plugins::Retry

  give_up_callback :log_give_up

  def self.log_give_up(exception, *args)
    logger.error("Received #{exception}, job #{self.name} failed with #{args}")
  end
end

You can register multiple callbacks, and they will be called in the order that they were registered. You can also set callbacks by setting @try_again_callbacks or @give_up_callbacks to an array where each element is a Proc or Symbol.

class CallbackJob
  extend Resque::Plugins::Retry

  @try_again_callbacks = [
    :call_me_first,
    :call_me_second,
    lambda { |*args| call_me_third(*args) }
  ]

  def self.call_me_first(ex, *args); end
  def self.call_me_second(ex, *args); end
  def self.call_me_third(ex, *args); end
end

Warning: Make sure your callbacks do not throw any exceptions. If they do, subsequent callbacks will not be triggered, and the job will not be retried (if it was trying again). The retry counter also will not be reset.

Ignored Exceptions

If there is an exception for which you want to retry, but you don't want it to increment your retry counter, you can add it to @ignore_exceptions.

One use case: Restarting your workers triggers a Resque::TermException. You may want your workers to retry the job that they were working on, but without incrementing the retry counter.

class RestartResilientJob
  extend Resque::Plugins::Retry

  @retry_exceptions = [Resque::TermException]
  @ignore_exceptions = [Resque::TermException]
end

Reminder: @ignore_exceptions should be a subset of @retry_exceptions.

Debug Plugin Logging

The inner-workings of the plugin are output to the Resque Logger when Resque.logger.level is set to Logger::DEBUG.

Add VVERBOSE=true as an environment variable to easily set the log level to debug.

Testing

To run a specific test and inspect logging output

bundle exec rake TEST=the_test_file.rb VVERBOSE=true

There are many example jobs implementing various use-cases for this gem in test_jobs.rb

Contributing/Pull Requests

• Yes please!
• Fork the project.
• Make your feature addition or bug fix.
• Add tests for it.
• In a seperate commit, update the HISTORY.md file please.
• Send us a pull request. Bonus points for topic branches.
• If you edit the gemspec/version etc, please do so in another commit.