Know Your Sidekiq Testing Rights


Sidekiq is a background job processor for Ruby that’s relied on by thousands of Rails apps. It’s regularly used to process slow or resource intensive work in the background. The results and operations of background jobs are often critical to a business and its users. Like sending email, for example.

It’s essential that new signups get a confirmation email, accounts are charged for usage, and incoming data is indexed for searching. All of these activities are ideal candidates for being implemented as background jobs. As with any critical part of your application infrastructure, proper testing is paramount. By no coincidence, Sidekiq has fantastic support for various means of testing.

The performance and reliability of your application’s test suite hinges on the granularity with which you test each component. Features that exercise the entire stack are tested at a higher level than the individual methods that comprise a particular class. That same granular approach applies directly to testing background jobs and the Sidekiq workers that process them.

You need to know which testing modes Sidekiq is packaged with and the best practices for testing both jobs and workers. This post will walk you through evaluating each testing paradigm and show examples of how to work effectively with each.

Before proceeding, let’s define a couple of commonly used Sidekiq terms: job and worker. A job is an operation to be processed in the background. Sidekiq manages a queue of jobs within Redis as simple JSON data structures. A worker is a Ruby class, with a little Sidekiq sugar mixed in, that’s responsible for executing a job. When Sidekiq is ready to process a job, a corresponding worker is initialized and passed the job’s arguments.

Try Codeship – The simplest Continuous Delivery service out there.

Know Your Modes in Sidekiq

There are three distinct testing “modes” that Sidekiq ships with. Here’s the breakdown, straight from the GitHub wiki:

  • Fake. No job processing is performed. Jobs are only enqueued.
  • Inline. Jobs are enqueued and then processed synchronously.
  • Disabled. No special handling for testing. Jobs are enqueued in Redis and processed normally.

Before diving into the particulars for each paradigm, let’s review how to properly configure Sidekiq within your test suite. Changing the testing mode is done on the Sidekiq::Testing module, which is a constant and therefore global. This can cause unexpected behavior for randomized or parallelized tests, so it’s advisable to change the mode within the context of a block when possible.

require 'minitest/autorun' 
require 'sidekiq/testing'

class IndexWorkerTest < Minitest::Test 
  def setup 

  def test_that_mode_can_change_within_a_block 
    Sidekiq::Testing.inline do 

      # This would fail in `fake` mode, no indexing would happen
      assert model.reload.indexed?

Testing your Sidekiq workers with RSpec is smooth and wonderful, and there’s even a gem for it. However, this post is about Sidekiq, and Minitest is the default, so we’ll stick with it for the examples to keep the cognitive overhead to a minimum.

Unit Testing Workers — Start at the Bottom

Unit testing workers is simple and fast. Always start at this most granular level for testing branching logic, edge cases, or complex logic within workers.

A worker’s initializer doesn’t take any arguments, so the object isn’t instantiated with any state. This nudges your worker’s methods toward a functional style of passing objects and arguments around. Wonderfully, testing these types of pure input/output methods is simple and straightforward.

Since jobs, along with their arguments, are persisted to Redis before they are dequeued and executed, a best practice is to pass simple identifiers to the #perform_async method call instead of heavier instances of ActiveModel::Record. Instead, they’re often fetched from the database within a worker’s #perform method. This can make unit testing workers more expensive and less unit-like, so it is best to have additional methods defined so #perform can delegate down and you can unit-test with test doubles or non-persisted models.

Here is an example of a worker responsible for indexing new events so they are searchable by end users. The #perform method calls .find to load the model and then hands it off to #index_event where the real work of indexing takes place.

class IndexWorker 
  include Sidekiq::Worker

  def perform(event_id) 
    event = Event.find(event_id)


  def index_event(event) 

In our test case, we can avoid persisting any events to the database and instead pass them directly to the index_event method.

class IndexWorkerTest < Minitest::Test 
  def test_indexing_event 
    worker = 
    event =

    assert event.indexed?

With complex workers that coordinate multiple tasks or rely on helper methods, this style of isolated unit testing really pays off. Tests can be faster and more focused.

Testing Worker Queuing — When Testing the Boundaries

Sidekiq’s fake testing mode operates similarly to the ActionMailer testing API, in which jobs are queued up in a .deliveries array rather than being executed immediately. Jobs within the queue can be queried, inspected, and optionally “drained” to process enqueued jobs. This mode is activated simply with the fake! directive:


Testing with fake mode is the first level of integration testing with other objects, one step beyond unit testing workers. Testing this way promotes decoupled and faster tests, as the worker doesn’t have to perform any actual work.

It isn’t appropriate for full integration testing or situations where you want to process jobs during a test. For example, indexing data in a background job when you want to make assertions on finding the data during the test. The inline testing mode, discussed next, is best for these cases.

Worker queues are global and will therefore persist between tests. Unchecked, your tests will bleed state, and your test suite will become order dependent. To combat this, be sure to clear jobs between tests:

def teardown

Our event indexing application allows users to post new events, which need to be indexed immediately. The collaboration of those two objects is wrapped up in a form object that can be tested separately.

class EventForm 
  attr_reader :worker

  def initialize(worker: IndexWorker) 
    @worker = worker 

  def create(params) do |event| 
      worker.perform_async( if event.valid? 

The form object triggers the event indexing worker, but the test isn’t concerned with what work is being done in the worker — it only needs to verify that the job was enqueued.

class EventFormTest < Minitest::Test 
  def test_enqueuing_index 
    form = IndexWorker) 
    event = form.create(id: 123, name: 'New Event')

    assert_equal 1,
    assert_equal [123],['args']

The job queue is simply an array filled with very simple hash structures representing the job:

{ "class"      => "IndexWorker",
  "args"       => [1],
  "retry"      => true,
  "queue"      => "default",
  "jid"        => "c7b41a718032908180a1de9c",
  "created_at" => 1440171298.594218 }

Making assertions against a job is extremely straight forward and doesn’t require a domain specific testing API. This style of testing can also be accomplished with stubs, but that’s more invasive and less flexible.

Testing Inline Processing — When You Need Results

Inline testing mode performs enqueued jobs synchronously within the same process, rather than asynchronously in a separate, dedicated process. This closely mirrors production behavior, but without the difficulty of multiple processes and race conditions.


Inline mode bypasses Redis integration entirely. Jobs are pushed into queues and then immediately popped off and executed.

Inline processing is ideal for feature tests involving separate communicating processes, for example full stack tests that use capybara. By avoiding asynchronous job processing, you gain more predictable test runs, ones where you aren’t plagued by periodic race conditions.

Now we want to add a new integration test to our event management app. The new test simulates the happy path of a user posting a new event and then trying to find that event through search:

class SubmitAndSearchTest < ActionDispatch::IntegrationTest

  # Note that inline mode must be configured during test setup, running inside 
  # of a testing block won't propagate between client (browser) and server 
  # (rails) processes.

  def setup 

  def test_search_after_event_submission 

    visit '/events'
    post_an_event(name: 'Meetup')
    assert_equal '/search', page.current_path
    assert_contains 'Meetup', page.body

The test not only requires an event to be saved to the database, but the event must also be indexed to show up in search results. Using the inline mode for testing means that the indexing job is processed immediately and the results will include the new event. If the job were processed asynchronously, there is a good chance that the controller would respond before anything were indexed, a textbook race condition causing flickering tests.

Sidekiq Testing Disabled — When You Must Know Everything Works

Disabling testing altogether reverts Sidekiq back to asynchronous processing. This is the default mode that Sidekiq runs in. It relies on separate clients and servers and a real Redis instance for queuing.

Disabled mode isn’t meant for running tests; it’s called “disabled” for a reason! It is, however, an excellent way to verify that Sidekiq has the correct configuration for Redis.

class SidekiqConfigurationTest < Minitest::Test 
  def setup 
    configure = -> (config) do 
      config.redis = { url: ENV.fetch('REDIS_URL') } 


  def test_server_enqueing_works 
    Sidekiq::Testing.disable! do 
      event = Event.create(name: 'Reception')


Exercise Your Rights

During everyday application testing, you always test a model at the unit level. You then test how a model coordinates with other models at the boundary level. Finally, you test how it integrates with the database and other services at the functional level. Treat Sidekiq the exact same way! Follow these simple guidelines, and you’ll be off to a great start:

  • Use fast, finely grained, unit tests to work through edge cases and refine worker behavior.
  • Test the boundaries of your jobs when you’re focusing on collaboration with other objects.
  • Test workers inline when you need to test the output of the jobs in the context of the rest of the system.

Subscribe via Email

Over 60,000 people from companies like Netflix, Apple, Spotify and O'Reilly are reading our articles.
Subscribe to receive a weekly newsletter with articles around Continuous Integration, Docker, and software development best practices.

We promise that we won't spam you. You can unsubscribe any time.

Join the Discussion

Leave us some comments on what you think about this topic or if you like to add something.

  • Steven Webb

    Thanks for the post Parker. There’s a lot of interesting stuff in there. There were a couple of things I don’t agree with though and wanted to get your thoughts.

    In your unit test section you make the “index_event” method public. Making this method public just to make your tests easier to write seems like a bad idea to me. I can easily see another developer coming along and adding another public method because its easier to test (say we want to update the indexes in both the database and say solr). Before you know it you’ve got 10 public methods, no indication that they shouldn’t be called (except via perform), and quite likely a heap of dependencies on other parts of your system. Instead, if you were to stub out the call to Event.find you could keep “index_event” private. As changes were made to the worker you would quickly see what its dependencies were because you would have to stub them all out. This would give you feedback on your design and help you minimise your dependencies.

    In the section on “Testing Worker Queuing” you say that “This style of testing can also be accomplished with stubs, but that’s more invasive and less flexible”. I’m not sure what you mean by this. In your example you are testing based on the hash pushed onto the job queue. AFAIK (and I might be 100% wrong) the job queue format isn’t a public api and isn’t standardised. As a result your test is dependent on the internal working of Sidekiq that can change at any point without notice. I would hate to do a minor version bump of Sidekiq and have to fix every worker test in the app. Wouldn’t it be better to mock out the public method calls that we know won’t be be broken between minor gem versions?

    Curious to hear your thoughts.

    • Thanks for your thoughtful feedback.

      Regarding public methods on Sidekiq workers, the only method that will ever be called on a worker instance by Sidekiq is `perform`. There isn’t a concrete notion of an interface in Ruby, but the `perform` method is an implicit interface. By that reasoning any other methods defined on the worker would be relegated to private. Having a single public method that relies on scores of internal private methods, all of which can only be tested via the single public method, is the classic iceberg class.

      Consider a worker that is formatting payloads or composing event keys. Those are very functional methods and can be tested cheaply and extensively on their own. You could do an Extract Class refactoring into a class specifically for working with them and then rely on that within the worker, if you prefer.

      The job queue format was borrowed from Resque originally and hasn’t changed since its release in 2012. It has been an extremely stable aspect of Sidekiq. Considering how many jobs may be scheduled or in a queue at any one time it would be disastrous for Sidekiq to change the job format between versions.

      I’m always extremely hesitant to mock out calls or rely on stubs. If any library APIs change between versions your test suite will still pass and you won’t find out until production.

      I lean toward testing values and boundaries whenever possible.

  • Fangorn

    Really useful tutorial.
    It is plain that in disabled mode Sidekiq jobs would write to the development database.
    What I do not understand is whether tests in inline mode would also write to the Rails development database or would instead save the event to the test database as usual done by the unit tests.

    • Whether Sidekiq jobs are run inline or run manually (drained), they are still running in the same test environment as Rails. That means they’ll use the test database, and under most circumstances they’re ran within a transaction. The point of running the tests in isolation is to avoid the slowdown incurred by hitting the database, not to prevent polluting the development database.

  • Fangorn

    I wonder whether is it necessary to configure Redis client and server to use REDIS_URL in the example for the disable mode. By default the server accepts connections on port 6379 of localhost, and clients too like redis-rb, if no configuration is explicitely specified, will use localhost and port 6379. Moreover, it is important to notice that by default REDIS_URL is unset. Therefore, as far as I have understood, the example relies on the assumption that REDIS_URL was previously set to a value different from redis://localhost:6379, in case the Redis server is in a machine different from the machine where tests are made.
    Surely setting REDIS_URL = redis://localhost:6379 would not make much sense unless explicitly setting REDIS_URL was required by Sidekiq in disable mode tests, and the official testing documentation does not mention it.

    • The REDIS_URL can specify more than just a server and port, it can also specify the database. By default Redis has 16 databases, which you can specify with a trailing number, e.g. redis://localhost:6379/0 or redis://localhost:6379/1. It is helpful, and a common practice, to use a different database in test mode vs development mode. Additionally, it isn’t unusual to have multiple Redis installs running with different ports, so your app may not want to use the default 6379 port.