Skip to content

keypup-io/cloudtasker

Repository files navigation

Build Status Gem Version

Cloudtasker

Background jobs for Ruby using Google Cloud Tasks.

Cloudtasker provides an easy to manage interface to Google Cloud Tasks for background job processing. Workers can be defined programmatically using the Cloudtasker DSL and enqueued for processing using a simple to use API.

Cloudtasker is particularly suited for serverless applications only responding to HTTP requests and where running a dedicated job processing server is not an option (e.g. deploy via Cloud Run). All jobs enqueued in Cloud Tasks via Cloudtasker eventually get processed by your application via HTTP requests.

Cloudtasker also provides optional modules for running cron jobs, batch jobs, unique jobs and storable jobs.

A local processing server is also available for development. This local server processes jobs in lieu of Cloud Tasks and allows you to work offline.

Summary

  1. Installation
  2. Get started with Rails
  3. Get started with Rails & ActiveJob
  4. Configuring Cloudtasker
    1. Cloud Tasks authentication & permissions
    2. Cloudtasker initializer
  5. Enqueuing jobs
  6. Managing worker queues
    1. Creating queues
    2. Assigning queues to workers
  7. Extensions
  8. Working locally
    1. Option 1: Cloudtasker local server
    2. Option 2: Using ngrok
  9. Logging
    1. Configuring a logger
    2. Logging context
    3. Truncating log arguments
    4. Searching logs: Job ID vs Task ID
  10. Error Handling
    1. HTTP Error codes
    2. Worker callbacks
    3. Global callbacks
    4. Max retries
    5. Conditional reenqueues using retry errors
    6. Dispatch deadline
  11. Testing
    1. Test helper setup
    2. In-memory queues
    3. Unit tests
  12. Best practices building workers

Installation

Add this line to your application's Gemfile:

gem 'cloudtasker'

And then execute:

$ bundle

Or install it yourself with:

$ gem install cloudtasker

Get started with Rails

Cloudtasker is pre-integrated with Rails. Follow the steps below to get started.

Install redis on your machine (this is required by the Cloudtasker local processing server)

# E.g. using brew
brew install redis

Add the following initializer

# config/initializers/cloudtasker.rb

Cloudtasker.configure do |config|
  #
  # Adapt the server port to be the one used by your Rails web process
  #
  config.processor_host = 'http://localhost:3000'

  #
  # If you do not have any Rails secret_key_base defined, uncomment the following
  # This secret is used to authenticate jobs sent to the processing endpoint
  # of your application.
  #
  # config.secret = 'some-long-token'
end

Define your first worker:

# app/workers/dummy_worker.rb

class DummyWorker
  include Cloudtasker::Worker

  def perform(some_arg)
    logger.info("Job run with #{some_arg}. This is working!")
  end
end

Launch Rails and the local Cloudtasker processing server (or add cloudtasker to your foreman config as a worker process)

# In one terminal
> rails s -p 3000

# In another terminal
> cloudtasker

Open a Rails console and enqueue some jobs

  # Process job as soon as possible
  DummyWorker.perform_async('foo')

  # Process job in 60 seconds
  DummyWorker.perform_in(60, 'foo')

Your Rails logs should display the following:

Started POST "/cloudtasker/run" for ::1 at 2019-11-22 09:20:09 +0100

Processing by Cloudtasker::WorkerController#run as */*
  Parameters: {"worker"=>"DummyWorker", "job_id"=>"d76040a1-367e-4e3b-854e-e05a74d5f773", "job_args"=>["foo"], "job_meta"=>{}}

I, [2019-11-22T09:20:09.319336 #49257]  INFO -- [Cloudtasker][d76040a1-367e-4e3b-854e-e05a74d5f773] Starting job...: {:worker=>"DummyWorker", :job_id=>"d76040a1-367e-4e3b-854e-e05a74d5f773", :job_meta=>{}}
I, [2019-11-22T09:20:09.319938 #49257]  INFO -- [Cloudtasker][d76040a1-367e-4e3b-854e-e05a74d5f773] Job run with foo. This is working!: {:worker=>"DummyWorker", :job_id=>"d76040a1-367e-4e3b-854e-e05a74d5f773", :job_meta=>{}}
I, [2019-11-22T09:20:09.320966 #49257]  INFO -- [Cloudtasker][d76040a1-367e-4e3b-854e-e05a74d5f773] Job done: {:worker=>"DummyWorker", :job_id=>"d76040a1-367e-4e3b-854e-e05a74d5f773", :job_meta=>{}}

That's it! Your job was picked up by the Cloudtasker local server and sent for processing to your Rails web process.

Now jump to the next section to configure your app to use Google Cloud Tasks as a backend.

Get started with Rails & ActiveJob

Note: ActiveJob is supported since 0.11.0
Note: Cloudtasker extensions (cron, batch, unique jobs and storable) are not available when using cloudtasker via ActiveJob.

Cloudtasker is pre-integrated with ActiveJob. Follow the steps below to get started.

Install redis on your machine (this is required by the Cloudtasker local processing server)

# E.g. using brew
brew install redis

Add the following initializer

# config/initializers/cloudtasker.rb

Cloudtasker.configure do |config|
  #
  # Adapt the server port to be the one used by your Rails web process
  #
  config.processor_host = 'http://localhost:3000'

  #
  # If you do not have any Rails secret_key_base defined, uncomment the following
  # This secret is used to authenticate jobs sent to the processing endpoint
  # of your application.
  #
  # config.secret = 'some-long-token'
end

Configure ActiveJob to use Cloudtasker. You can also configure ActiveJob per environment via the config/environments/:env.rb files

# config/application.rb

require_relative 'boot'
require 'rails/all'

Bundler.require(*Rails.groups)

module Dummy
  class Application < Rails::Application
    # Initialize configuration defaults for originally generated Rails version.
    config.load_defaults 6.0

    # Settings in config/environments/* take precedence over those specified here.
    # Application configuration can go into files in config/initializers
    # -- all .rb files in that directory are automatically loaded after loading
    # the framework and any gems in your application.

    # Use cloudtasker as the ActiveJob backend:
    config.active_job.queue_adapter = :cloudtasker
  end
end

Define your first job:

# app/jobs/example_job.rb

class ExampleJob < ApplicationJob
  queue_as :default

  def perform(some_arg)
    logger.info("Job run with #{some_arg}. This is working!")
  end
end

Launch Rails and the local Cloudtasker processing server (or add cloudtasker to your foreman config as a worker process)

# In one terminal
> rails s -p 3000

# In another terminal
> cloudtasker

Open a Rails console and enqueue some jobs

  # Process job as soon as possible
  ExampleJob.perform_later('foo')

  # Process job in 60 seconds
  ExampleJob.set(wait: 60).perform_later('foo')

Configuring Cloudtasker

Cloud Tasks authentication & permissions

The Google Cloud library authenticates via the Google Cloud SDK by default. If you do not have it setup then we recommend you install it.

Other options are available such as using a service account. You can see all authentication options in the Google Cloud Authentication guide.

In order to function properly Cloudtasker requires the authenticated account to have the following IAM permissions:

  • cloudtasks.tasks.get
  • cloudtasks.tasks.create
  • cloudtasks.tasks.delete

To get started quickly you can add the roles/cloudtasks.admin role to your account via the IAM Console. This is not required if your account is a project admin account.

The GCP project ID and region values are not loaded automatically by the Google Cloud library, and must be explicitly defined in the initializer when using Google Cloud Tasks.

Cloudtasker initializer

The gem can be configured through an initializer. See below all the available configuration options.

# config/initializers/cloudtasker.rb

Cloudtasker.configure do |config|
  #
  # If you do not have any Rails secret_key_base defined, uncomment the following.
  # This secret is used to authenticate jobs sent to the processing endpoint
  # of your application.
  #
  # Default with Rails: Rails.application.credentials.secret_key_base
  #
  # config.secret = 'some-long-token'

  #
  # Specify the details of your Google Cloud Task location. 
  # 
  # This is required when the mode of operation is set to :production
  # 
  config.gcp_location_id = 'us-central1' # defaults to 'us-east1'
  config.gcp_project_id = 'my-gcp-project'

  #
  # Specify the namespace for your Cloud Task queues.
  #
  # Specifying a namespace is optional but strongly recommended to keep
  # queues organised, especially in a micro-service environment.
  #
  # The gem assumes that a least a default queue named 'my-app-default'
  # exists in Cloud Tasks. You can create this default queue using the
  # gcloud SDK or via the `rake cloudtasker:setup_queue` task if you use Rails.
  #
  # Workers can be scheduled on different queues. The name of the queue
  # in Cloud Tasks is always assumed to be prefixed with the prefix below.
  #
  # E.g.
  # Setting `cloudtasker_options queue: 'critical'` on a worker means that
  # the worker will be pushed to 'my-app-critical' in Cloud Tasks.
  #
  # Specific queues can be created in Cloud Tasks using the gcloud SDK or
  # via the `rake cloudtasker:setup_queue name=<queue_name>` task.
  #
  config.gcp_queue_prefix = 'my-app'

  #
  # Specify the publicly accessible host for your application
  #
  # > E.g. in development, using the cloudtasker local server
  # config.processor_host = 'http://localhost:3000'
  #
  # > E.g. in development, using `config.mode = :production` and ngrok
  # config.processor_host = 'https://111111.ngrok.io'
  #
  config.processor_host = 'https://app.mydomain.com'

  #
  # Specify the mode of operation:
  # - :development => jobs will be pushed to Redis and picked up by the Cloudtasker local server
  # - :production => jobs will be pushed to Google Cloud Tasks. Requires a publicly accessible domain.
  #
  # Defaults to :development unless CLOUDTASKER_ENV or RAILS_ENV or RACK_ENV is set to something else.
  #
  # config.mode = Rails.env.production? || Rails.env.my_other_env? ? :production : :development

  #
  # Specify the logger to use
  #
  # Default with Rails: Rails.logger
  # Default without Rails: Logger.new(STDOUT)
  #
  # config.logger = MyLogger.new(STDOUT)

  #
  # Specify how many retries are allowed on jobs. This number of retries excludes any
  # connectivity error due to the application being down or unreachable.
  #
  # Default: 25
  #
  # config.max_retries = 10

  #
  # Specify the redis connection hash.
  #
  # This is ONLY required in development for the Cloudtasker local server and in
  # all environments if you use any cloudtasker extension (unique jobs, cron jobs,
  # batch jobs or storable jobs)
  #
  # See https://github.com/redis/redis-rb for examples of configuration hashes.
  #
  # Default: redis-rb connects to redis://127.0.0.1:6379/0
  #
  # config.redis = { url: 'redis://localhost:6379/5' }

  #
  # Set to true to store job arguments in Redis instead of sending arguments as part
  # of the job payload to Google Cloud Tasks.
  #
  # This is useful if you expect to process jobs with payloads exceeding 100KB, which
  # is the limit enforced by Google Cloud Tasks.
  #
  # You can set this configuration parameter to a KB value if you want to store jobs
  # args in redis only if the JSONified arguments payload exceeds that threshold.
  #
  # Supported since: v0.10.0
  #
  # Default: false
  #
  # Store all job payloads in Redis:
  # config.store_payloads_in_redis = true
  #
  # Store all job payloads in Redis exceeding 50 KB:
  # config.store_payloads_in_redis = 50

  #
  # Specify the dispatch deadline for jobs in Cloud Tasks, in seconds.
  # Jobs taking longer will be retried by Cloud Tasks, even if they eventually
  # complete on the server side.
  #
  # Note that this option is applied when jobs are enqueued job. Changing this value
  # will not impact already enqueued jobs.
  #
  # This option can also be configured on a per worker basis via
  # the cloudtasker_options directive.
  #
  # Supported since: v0.12.0
  #
  # Default: 600 seconds (10 minutes)
  # Min: 15 seconds
  # Max: 1800 seconds (30 minutes)
  #
  # config.dispatch_deadline = 600

  #
  # Specify a proc to be invoked every time a job fails due to a runtime
  # error.
  #
  # This hook is not invoked for DeadWorkerError. See on_dead instead.
  #
  # This is useful when you need to apply general exception handling, such
  # as reporting errors to a third-party service like Rollbar or Bugsnag.
  #
  # Note: the worker argument might be nil, such as when InvalidWorkerError is raised.
  #
  # Supported since: v0.12.0
  # 
  # Default: no operation
  #
  # config.on_error = ->(error, worker) { Rollbar.error(error) }

  #
  # Specify a proc to be invoked every time a job dies due to too many
  # retries.
  #
  # This is useful when you need to apply general exception handling, such
  # logging specific messages/context when a job dies.
  #
  # Supported since: v0.12.0
  # 
  # Default: no operation
  #
  # config.on_dead = ->(error, worker) { Rollbar.error(error) }

  #
  # Specify the Open ID Connect (OIDC) details to connect to a protected GCP service, such
  # as a private Cloud Run application.
  #
  # The configuration supports the following details:
  # - service_account_email: This is the "act as" user. It can be found under the security details
  #   of the Cloud Run service.
  # - audience: The audience is usually the publicly accessible host for the Cloud Run service
  #   (which is the same value configured as the processor_host). If no audiences are provided
  #   it will be set to the processor_host.
  #
  # Note: If the OIDC token is used for a Cloud Run service make sure to include the
  # `iam.serviceAccounts.actAs` permission on the service account.
  #
  # See https://cloud.google.com/tasks/docs/creating-http-target-tasks#sa for more information on
  # setting up service accounts for use with Cloud Tasks.
  #
  # Supported since: v0.14.0 (upcoming)
  #
  # Default: nil 
  #
  # config.oidc = { service_account_email: '[email protected]' }
  # config.oidc = { service_account_email: '[email protected]', audience: 'https://api.example.net' }

  #
  # Enable/disable the verification of SSL certificates on the local processing server when
  # sending tasks to the processor.
  #
  # Set to false to disable SSL verification (OpenSSL::SSL::VERIFY_NONE).
  #
  # Default: true
  #
  # config.local_server_ssl_verify = true
end

If the default queue <gcp_queue_prefix>-default does not exist in Cloud Tasks you should create it using the gcloud sdk.

Alternatively with Rails you can simply run the following rake task if you have queue admin permissions (cloudtasks.queues.get and cloudtasks.queues.create).

bundle exec rake cloudtasker:setup_queue

Enqueuing jobs

Cloudtasker provides multiple ways of enqueuing jobs.

# Worker will be processed as soon as possible
MyWorker.perform_async(arg1, arg2)

# Worker will be processed in 5 minutes
MyWorker.perform_in(5 * 60, arg1, arg2)
# or with Rails
MyWorker.perform_in(5.minutes, arg1, arg2)

# Worker will be processed on a specific date
MyWorker.perform_at(Time.parse('2025-01-01 00:50:00Z'), arg1, arg2)
# also with Rails
MyWorker.perform_at(3.days.from_now, arg1, arg2)

# With all options, including which queue to run the worker on.
MyWorker.schedule(args: [arg1, arg2], time_at: Time.parse('2025-01-01 00:50:00Z'), queue: 'critical')
# or
MyWorker.schedule(args: [arg1, arg2], time_in: 5 * 60, queue: 'critical')

Cloudtasker also provides a helper for re-enqueuing jobs. Re-enqueued jobs keep the same job id. Some middlewares may rely on this to track the fact that that a job didn't actually complete (e.g. Cloustasker batch). This is optional and you can always fallback to using exception management (raise an error) to retry/re-enqueue jobs.

E.g.

# app/workers/fetch_resource_worker.rb

class FetchResourceWorker
  include Cloudtasker::Worker

  def perform(id)
    # ...do some logic...
    if some_condition
      # Stop and re-enqueue the job to be run again in 10 seconds.
      # Also see the section on Cloudtasker::RetryWorkerError for a different
      # approach on reenqueuing.
      return reenqueue(10)
    else
      # ...keep going...
    end
  end
end

Managing worker queues

Cloudtasker allows you to manage several queues and distribute workers across them based on job priority. By default jobs are pushed to the default queue, which is <gcp_queue_prefix>-default in Cloud Tasks.

Creating queues

More queues can be created using the gcloud sdk or the cloudtasker:setup_queue rake task.

E.g. Create a critical queue with a concurrency of 5 via the gcloud SDK

gcloud tasks queues create <gcp_queue_prefix>-critical --max-concurrent-dispatches=5

E.g. Create a real-time queue with a concurrency of 15 via the rake task (Rails only)

rake cloudtasker:setup_queue name=real-time concurrency=15

When running the Cloudtasker local processing server, you can specify the concurrency for each queue using:

cloudtasker -q critical,5 -q important,4 -q default,3

Assigning queues to workers

Queues can be assigned to workers via the cloudtasker_options directive on the worker class:

# app/workers/critical_worker.rb

class CriticalWorker
  include Cloudtasker::Worker

  cloudtasker_options queue: :critical

  def perform(some_arg)
    logger.info("This is a critical job run with arg=#{some_arg}.")
  end
end

Queues can also be assigned at runtime when scheduling a job:

CriticalWorker.schedule(args: [1], queue: :important)

Extensions

Note: Extensions are not available when using cloudtasker via ActiveJob.

Cloudtasker comes with three optional features:

  • Cron Jobs [docs]: Run jobs at fixed intervals.
  • Batch Jobs [docs]: Run jobs in jobs and track completion of the overall batch.
  • Unique Jobs [docs]: Ensure uniqueness of jobs based on job arguments.
  • Storable Jobs [docs]: Park jobs until they are ready to be enqueued.

Working locally

Cloudtasker pushes jobs to Google Cloud Tasks, which in turn sends jobs for processing to your application via HTTP POST requests to the /cloudtasker/run endpoint of the publicly accessible domain of your application.

When working locally on your application it is usually not possible to have a public domain. So what are the options?

Option 1: Cloudtasker local server

The Cloudtasker local server is a ruby daemon that looks for jobs pushed to Redis and sends them to your application via HTTP POST requests. The server mimics the way Google Cloud Tasks works, but locally!

You can configure your application to use the Cloudtasker local server using the following initializer:

# config/initializers/cloudtasker.rb

Cloudtasker.configure do |config|
  # ... other options

  # Push jobs to redis and let the Cloudtasker local server collect them
  # This is the default mode unless CLOUDTASKER_ENV or RAILS_ENV or RACK_ENV is set
  # to a non-development environment
  config.mode = :development
end

The Cloudtasker server can then be started using:

bundle exec cloudtasker

You can as well define a Procfile to manage the cloudtasker process via foreman. Then use foreman start to launch both your Rails server and the Cloudtasker local server.

# Procfile
web: bundle exec rails s
worker: bundle exec cloudtasker

Note that the local development server runs with 5 concurrent threads by default. You can tune the number of threads per queue by running cloudtasker the following options:

bundle exec cloudtasker -q critical,5 -q important,4 -q default,3

Option 2: Using ngrok

Want to test your application end to end with Google Cloud Task? Then ngrok is the way to go.

First start your ngrok tunnel:

ngrok http 3000

Take note of your ngrok domain and configure Cloudtasker to use Google Cloud Task in development via ngrok.

# config/initializers/cloudtasker.rb

Cloudtasker.configure do |config|
  # Specify your Google Cloud Task queue configuration
  config.gcp_location_id = 'us-central1'
  config.gcp_project_id = 'my-gcp-project'
  config.gcp_queue_prefix = 'my-app'

  # Use your ngrok domain as the processor host
  config.processor_host = 'https://your-tunnel-id.ngrok.io'

  # Force Cloudtasker to use Google Cloud Tasks in development
  config.mode = :production
end

Finally start Rails to accept jobs from Google Cloud Tasks

bundle exec rails s

Logging

There are several options available to configure logging and logging context.

Configuring a logger

Cloudtasker uses Rails.logger if Rails is available and falls back on a plain ruby logger Logger.new(STDOUT) if not.

It is also possible to configure your own logger. For example you can setup Cloudtasker with semantic_logger by doing the following in your initializer:

# config/initializers/cloudtasker.rb

Cloudtasker.configure do |config|
  config.logger = SemanticLogger[Cloudtasker]
end

Logging context

Cloudtasker provides worker contextual information to the worker logger method inside your worker methods.

For example:

# app/workers/dummy_worker.rb

class DummyWorker
  include Cloudtasker::Worker

  def perform(some_arg)
    logger.info("Job run with #{some_arg}. This is working!")
  end
end

Will generate the following log with context {:worker=> ..., :job_id=> ..., :job_meta=> ...}

[Cloudtasker][d76040a1-367e-4e3b-854e-e05a74d5f773] Job run with foo. This is working!: {:worker=>"DummyWorker", :job_id=>"d76040a1-367e-4e3b-854e-e05a74d5f773", :job_meta=>{}, :task_id => "4e755d3f-6de0-426c-b4ac-51edd445c045"}

The way contextual information is displayed depends on the logger itself. For example with semantic_logger contextual information might not appear in the log message but show up as payload data on the log entry itself (e.g. using the fluentd adapter).

Contextual information can be customised globally and locally using a log context_processor. By default the Cloudtasker::WorkerLogger is configured the following way:

Cloudtasker::WorkerLogger.log_context_processor = ->(worker) { worker.to_h.slice(:worker, :job_id, :job_meta, :job_queue, :task_id) }

You can decide to add a global identifier for your worker logs using the following:

# config/initializers/cloudtasker.rb

Cloudtasker::WorkerLogger.log_context_processor = lambda { |worker|
  worker.to_h.slice(:worker, :job_id, :job_meta, :job_queue, :task_id).merge(app: 'my-app')
}

You could also decide to log all available context - including arguments passed to perform - for specific workers only:

# app/workers/full_context_worker.rb

class FullContextWorker
  include Cloudtasker::Worker

  cloudtasker_options log_context_processor: ->(worker) { worker.to_h }

  def perform(some_arg)
    logger.info("This log entry will have full context!")
  end
end

See the Cloudtasker::Worker class for more information on attributes available to be logged in your log_context_processor proc.

Truncating log arguments

Supported since: v0.14.0 (upcoming)

By default Cloudtasker does not log job arguments as arguments can contain sensitive data and generate voluminous logs, which may lead to noticeable costs with your log provider (e.g. GCP Logging). Also some providers (e.g. GCP Logging) will automatically truncate log entries that are too big and reduce their searchability.

Job arguments can be logged for all workers by configuring the following log context processor in your Cloudtasker initializer:

Cloudtasker::WorkerLogger.log_context_processor = ->(worker) { worker.to_h }

In order to reduce the size of logged job arguments, the following truncate utility is provided by Cloudtasker:

# string_limit: The maximum size for strings. Default is 64. Set to -1 to disable.
# array_limit: The maximum length for arrays. Default is 10. Set to -1 to disable.
# max_depth: The maximum recursive depth. Default is 3. Set to -1 to disable.
Cloudtasker::WorkerLogger.truncate(payload, string_limit: 64, array_limit: 10, max_depth: 3)

You may use it the following way:

Cloudtasker::WorkerLogger.log_context_processor = lambda do |worker|
  payload = worker.to_h

  # Using default options
  payload[:job_args] = Cloudtasker::WorkerLogger.truncate(payload[:job_args])

  # Using custom options
  # payload[:job_args] = Cloudtasker::WorkerLogger.truncate(payload[:job_args], string_limit: 32, array_limit: 5, max_depth: 2)

  # Return the payload to log
  payload
end

To further reduce logging cost, you may also log a reasonably complete version of job arguments at start then log a watered down version for the remaining log entries:

Cloudtasker::WorkerLogger.log_context_processor = lambda do |worker|
  payload = worker.to_h

  # Adjust the log payload based on the lifecycle of the job
  payload[:job_args] = if worker.perform_started_at
                         # The job start has already been logged. Log the job primitive arguments without depth.
                         # Arrays and hashes will be masked.
                         Cloudtasker::WorkerLogger.truncate(payload[:job_args], max_depth: 0)
                       else
                         # This is the job start. Log a more complete version of the job args.
                         Cloudtasker::WorkerLogger.truncate(payload[:job_args])
                       end

  # Return the payload to log
  payload
end

Searching logs: Job ID vs Task ID

Note: task_id field is available in logs starting with 0.10.0

Job instances are assigned two different different IDs for tracking and logging purpose: job_id and task_id. These IDs are found in each log entry to facilitate search.

Field Definition
job_id This ID is generated by Cloudtasker. It identifies the job along its entire lifecyle. It is persistent across retries and reschedules.
task_id This ID is generated by Google Cloud Tasks. It identifies a job instance on the Google Cloud Task side. It is persistent across retries but NOT across reschedules.

The Google Cloud Task UI (GCP console) lists all the tasks pending/retrying and their associated task id (also called "Task name"). From there you can:

  1. Use a task ID to lookup the logs of a specific job instance in Stackdriver Logging (or any other logging solution).
  2. From (1) you can retrieve the job_id attribute of the job.
  3. From (2) you can use the job_id to lookup the job logs along its entire lifecycle.

Error Handling

Jobs failures will return an HTTP error to Cloud Task and trigger a retry at a later time. The number of Cloud Task retries depends on the configuration of your queue in Cloud Tasks.

HTTP Error codes

Jobs failing will automatically return the following HTTP error code to Cloud Tasks, based on the actual reason:

Code Description
204 The job was processed successfully
205 The job is dead and has been removed from the queue
404 The job has specified an incorrect worker class.
422 An error happened during the execution of the worker (perform method)

Worker callbacks

Workers can implement the on_error(error) and on_dead(error) callbacks to do things when a job fails during its execution:

E.g.

# app/workers/handle_error_worker.rb

class HandleErrorWorker
  include Cloudtasker::Worker

  def perform
    raise(ArgumentError)
  end

  # The runtime error is passed as an argument.
  def on_error(error)
    logger.error("The following error happened: #{error}")
  end

  # The job has been retried too many times and will be removed
  # from the queue.
  def on_dead(error)
    logger.error("The job died with the following error: #{error}")
  end
end

Global callbacks

Supported since: 0.12.0

If you need to apply general exception handling logic to your workers you can specify on_error and on_dead hooks in the Cloudtasker configuration.

This is useful if you need to report errors to third-party services such as Rollbar or Bugsnag.

# config/initializers/cloudtasker.rb

Cloudtasker.configure do |config|
  #
  # Report runtime and dead worker errors to Rollbar
  #
  config.on_error = -> (error, _worker) { Rollbar.error(error) }
  config.on_dead = -> (error, _worker) { Rollbar.error(error) }
end

Max retries

By default jobs are retried 25 times - using an exponential backoff - before being declared dead. This number of retries can be customized locally on workers and/or globally via the Cloudtasker initializer.

Note that the number of retries set on your Cloud Task queue should be many times higher than the number of retries configured in Cloudtasker because Cloud Task also includes failures to connect to your application. Ideally set the number of retries to unlimited in Cloud Tasks.

Note: Versions prior to v0.14.0 (upcoming) use the X-CloudTasks-TaskRetryCount header for retries instead of the X-CloudTasks-TaskExecutionCount header to detect the number of retries, because there a previous bug on the GCP side which made the X-CloudTasks-TaskExecutionCount stay at zero instead of increasing on successive executions. Versions prior to v0.14.0 (upcoming) count any failure as failure, including failures due to the backend being unavailable (HTTP 503). Versions v0.14.0 (upcoming) and later only count application failure (HTTP 4xx) as failure for retry purpose.

E.g. Set max number of retries globally via the cloudtasker initializer.

# config/initializers/cloudtasker.rb

Cloudtasker.configure do |config|
  #
  # Specify how many retries are allowed on jobs. This number of retries excludes any
  # connectivity error that would be due to the application being down or unreachable.
  #
  # Default: 25
  #
  config.max_retries = 10
end

E.g. Set max number of retries to 3 on a given worker

# app/workers/some_error_worker.rb

class SomeErrorWorker
  include Cloudtasker::Worker

  # This will override the global setting
  cloudtasker_options max_retries: 3

  def perform
    raise(ArgumentError)
  end
end

E.g. Evaluate the number of max retries at runtime (Supported since: v0.10.1)

# app/workers/some_error_worker.rb

class SomeErrorWorker
  include Cloudtasker::Worker

  # Return the number of max retries based on
  # worker arguments.
  #
  # If this method returns nil then max_retries
  # will delegate to the class `max_retries` setting or Cloudtasker
  # `max_retries` configuration otion.
  def max_retries(arg1, arg2)
    arg1 == 'foo' ? 13 : nil
  end

  def perform(arg1, arg2)
    raise(ArgumentError)
  end
end

Conditional reenqueues using retry errors

Supported since: v0.14.0 (upcoming)

If your worker is waiting for some precondition to occur and you want to re-enqueue it until the condition has been met, you can raise a Cloudtasker::RetryWorkerError. This special error will fail your job without logging an error while still increasing the number of retries.

This is a safer approach than using the reenqueue helper, which can lead to forever running jobs if not used properly.

# app/workers/my_worker.rb

class MyWorker
  include Cloudtasker::Worker

  def perform(project_id)
    # Abort if project does not exist
    return unless (project = Project.find_by(id: project_id))

    # Trigger a retry if the project is still in "discovering" status
    # This error will NOT log an error. It only triggers a retry.
    raise Cloudtasker::RetryWorkerError if project.status == 'discovering'

    # The previous approach was to use `reenqueue`. This works but since it
    # does not increase the number of retries, you may end up with forever running
    # jobs
    # return reenqueue(10) if project.status == 'discovering'

    # Do stuff when project is not longer discovering
    do_some_stuff
  end

  # You can then specify what should be done if we've been waiting for too long
  def on_dead(error)
    logger.error("Looks like the project is forever discovering. Time to give up.")

    # This is of course an imaginary method
    send_slack_notification_to_internal_support_team(worker: self.class, args: job_args)
  end
end

Dispatch deadline

Supported since: 0.12.0

By default Cloud Tasks will automatically timeout your jobs after 10 minutes, independently of your server HTTP timeout configuration.

You can modify the dispatch deadline for jobs at a global level or on a per job basis.

E.g. Set the default dispatch deadline to 20 minutes.

# config/initializers/cloudtasker.rb

Cloudtasker.configure do |config|
  #
  # Specify the dispatch deadline for jobs in Cloud Tasks, in seconds. 
  # Jobs taking longer will be retried by Cloud Tasks, even if they eventually
  # complete on the server side.
  #
  # Note that this option is applied when jobs are enqueued job. Changing this value
  # will not impact already enqueued jobs.
  #
  # Default: 600 (10 minutes)
  #
  config.dispatch_deadline = 20 * 60 # 20 minutes
end

E.g. Set a dispatch deadline of 5 minutes on a specific worker

# app/workers/some_error_worker.rb

class SomeFasterWorker
  include Cloudtasker::Worker

  # This will override the global setting
  cloudtasker_options dispatch_deadline: 5 * 60

  def perform
    # ... do things ...
  end
end

Testing

Cloudtasker provides several options to test your workers.

Test helper setup

Require cloudtasker/testing in your rails_helper.rb (Rspec Rails) or spec_helper.rb (Rspec) or test unit helper file then enable one of the three modes:

require 'cloudtasker/testing'

# Mode 1 (default): Push jobs to Google Cloud Tasks (env != development) or Redis (env == development)
Cloudtasker::Testing.enable!

# Mode 2: Push jobs to an in-memory queue. Jobs will not be processed until you call
# Cloudtasker::Worker.drain_all (process all jobs) or MyWorker.drain (process jobs for specific worker)
Cloudtasker::Testing.fake!

# Mode 3: Push jobs to an in-memory queue. Jobs will be processed immediately.
Cloudtasker::Testing.inline!

You can query the current testing mode with:

Cloudtasker::Testing.enabled?
Cloudtasker::Testing.fake?
Cloudtasker::Testing.inline?

Each testing mode accepts a block argument to temporarily switch to it:

# Enable fake mode for all tests
Cloudtasker::Testing.fake!

# Enable inline! mode temporarily for a given test
Cloudtasker.inline! do
   MyWorker.perform_async(1,2)
end

Note that extension middlewares - e.g. unique job, batch job etc. - run in test mode. You can disable middlewares in your tests by adding the following to your test helper:

# Remove all middlewares
Cloudtasker.configure do |c|
  c.client_middleware.clear
  c.server_middleware.clear
end

# Remove all unique job middlewares
Cloudtasker.configure do |c|
  c.client_middleware.remove(Cloudtasker::UniqueJob::Middleware::Client)
  c.server_middleware.remove(Cloudtasker::UniqueJob::Middleware::Server)
end

In-memory queues

The fake! or inline! modes use in-memory queues, which can be queried and controlled using the following methods:

# Perform all jobs in queue
Cloudtasker::Worker.drain_all

# Remove all jobs in queue
Cloudtasker::Worker.clear_all

# Perform all jobs in queue for a specific worker type
MyWorker.drain

# Return the list of jobs in queue for a specific worker type
MyWorker.jobs

Unit tests

Below are examples of rspec tests. It is assumed that Cloudtasker::Testing.fake! has been set in the test helper.

Example 1: Testing that a job is scheduled

describe 'worker scheduling'
  subject(:enqueue_job) { MyWorker.perform_async(1,2) }

  it { expect { enqueue_job }.to change(MyWorker.jobs, :size).by(1) }
end

Example 2: Testing job execution logic

describe 'worker calls api'
  subject { Cloudtasker::Testing.inline! { MyApiWorker.perform_async(1,2) } }

  before { expect(MyApi).to receive(:fetch).and_return([]) }
  it { is_expected.to be_truthy }
end

Best practices building workers

Below are recommendations and notes about creating workers.

Use primitive arguments

Pushing a job via MyWorker.perform_async(arg1, arg2) will serialize all arguments as JSON. Cloudtasker does not do any magic marshalling and therefore passing user-defined class instance as arguments is likely to make your jobs fail because of JSON serialization/deserialization.

When defining your worker perform method, use primitive arguments (integers, strings, hashes).

Don't do that:

# app/workers/user_email_worker.rb

class UserEmailWorker
  include Cloudtasker::Worker

  def perform(user)
    user.reload.send_email
  end
end

Do that:

# app/workers/user_email_worker.rb

class UserEmailWorker
  include Cloudtasker::Worker

  def perform(user_id)
    User.find_by(id: user_id)&.send_email
  end
end

Assume hash arguments are stringified

Because of JSON serialization/deserialization hashes passed to perform_* methods will eventually be passed as stringified hashes to the worker perform method.

# Enqueuing a job with:
MyWorker.perform_async({ foo: 'bar', 'baz' => { key: 'value' } })

# will be processed as
MyWorker.new.perform({ 'foo' => 'bar', 'baz' => { 'key' => 'value' } })

Be careful with default arguments

Default arguments passed to the perform method are not actually considered as job arguments. Default arguments will therefore be ignored in contextual logging and by extensions relying on arguments such as the unique job extension.

Consider the following worker:

# app/workers/user_email_worker.rb

class UserEmailWorker
  include Cloudtasker::Worker

  cloudtasker_options lock: :until_executed

  def perform(user_id, time_at = Time.now.iso8601)
    User.find_by(id: user_id)&.send_email(Time.parse(time_at))
  end
end

If you enqueue this worker by omitting the second argument MyWorker.perform_async(123) then:

  • The time_at argument will not be included in contextual logging
  • The time_at argument will be ignored by the unique-job extension, meaning that job uniqueness will be only based on the user_id argument.

Handling big job payloads

Google Cloud Tasks enforces a limit of 100 KB for job payloads. Taking into accounts Cloudtasker authentication headers and meta information this leave ~85 KB of free space for JSONified job arguments.

Any excessive job payload (> 100 KB) will raise a Cloudtasker::MaxTaskSizeExceededError, both in production and development mode.

Option 1: Use Cloudtasker optional support for payload storage in Redis

Supported since: 0.10.0

Cloudtasker provides optional support for storing argument payloads in Redis instead of sending them to Google Cloud Tasks.

To enable it simply put the following in your Cloudtasker initializer:

# config/initializers/cloudtasker.rb

Cloudtasker.configure do |config|
  # Enable Redis support. Specify your redis connection
  config.redis = { url: 'redis://localhost:6379/5' }

  # Store all job payloads in Redis:
  config.store_payloads_in_redis = true

  # OR: store all job payloads in Redis exceeding 50 KB:
  # config.store_payloads_in_redis = 50
end

Option 2: Do it yourself solution

If you feel that a job payload is going to get big, prefer to store the payload using a datastore (e.g. Redis) and pass a reference to the job to retrieve the payload inside your job perform method.

E.g. Define a job like this

# app/workers/big_payload_worker.rb

class BigPayloadWorker
  include Cloudtasker::Worker

  def perform(payload_id)
    data = Rails.cache.fetch(payload_id)
    # ...do some processing...
  end
end

Then enqueue your job like this:

# Fetch and store the payload
data = ApiClient.fetch_thousands_of_records
payload_id = SecureRandom.uuid
Rails.cache.write(payload_id, data)

# Enqueue the processing job
BigPayloadWorker.perform_async(payload_id)

Sizing the concurrency of your queues

When defining the max concurrency of your queues (max_concurrent_dispatches in Cloud Tasks) you must keep in mind the maximum number of threads that your application provides. Otherwise your application threads may eventually get exhausted and your users will experience outages if all your web threads are busy running jobs.

With server based applications

Let's consider an application deployed in production with 3 instances, each having RAILS_MAX_THREADS set to 20. This gives us a total of 60 threads available.

Now let's say that we distribute jobs across two queues: default and critical. We can set the concurrency of each queue depending on the profile of the application:

E.g. 1: The application serves requests from web users and runs backgrounds jobs in a balanced way

concurrency for default queue: 20
concurrency for critical queue: 10

Total threads consumed by jobs at most: 30
Total threads always available to web users at worst: 30

E.g. 2: The application is a micro-service API heavily focused on running jobs (e.g. data processing)

concurrency for default queue: 35
concurrency for critical queue: 15

Total threads consumed by jobs at most: 50
Total threads always available to API clients at worst: 10

Also always ensure that your total number of threads does not exceed the available number of database connections (if you use any).

With serverless applications

In a serverless context your application will be scaled up/down based on traffic. When we say 'traffic' this includes requests from Cloud Tasks to run jobs.

Because your application is auto-scaled - and assuming you haven't set a maximum - your job processing capacity if theoretically unlimited. The main limiting factor in a serverless context becomes external constraints such as the number of database connections available.

To size the concurrency of your queues you should therefore take the most limiting factor - which is often the database connection pool size of relational databases - and use the calculations of the previous section with this limiting factor as the capping parameter instead of threads.

Development

After checking out the repo, run bin/setup to install dependencies.

For tests, run rake to run the tests. Note that Rails is not in context by default, which means Rails-specific test will not run. For tests including Rails-specific tests, run bundle exec appraisal rails_7.0 rake For all context-specific tests (incl. Rails), run the appraisal tests using bundle exec appraisal rake.

You can 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 tags, and push the .gem file to rubygems.org.

Contributing

Bug reports and pull requests are welcome on GitHub at https://github.com/keypup-io/cloudtasker. This project is intended to be a safe, welcoming space for collaboration, and contributors are expected to adhere to the Contributor Covenant code of conduct.

License

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

Code of Conduct

Everyone interacting in the Cloudtasker project’s codebases, issue trackers, chat rooms and mailing lists is expected to follow the code of conduct.

Author

Provided with ❤️ by keypup.io