Testing Workers

Worker modules are the primary "unit" of an Oban system. You can (and should) test a worker's callback functions locally, in-process, without touching the database.

Most worker callback functions take a single argument: an Oban.Job struct. A job encapsulates arguments, metadata, and other options. Creating jobs, and verifying that they're built correctly, requires some boilerplate...that's where Oban.Testing.perform_job/3 comes in!

Testing Perform

The perform_job/3 helper reduces boilerplate when constructing jobs for unit tests and checks for common pitfalls. For example, it automatically converts args to string keys before calling perform/1, ensuring that perform clauses aren't erroneously trying to match on atom keys.

Let's work through test-driving a worker to demonstrate.

Start by defining a test that creates a user and then use perform_job to manually call an account activation worker. In this context "activation" could mean sending an email, notifying administrators, or any number of business-critical functions—what's important is how we're testing it.

elixir

defmodule MyApp.ActivationWorkerTest do
  use MyApp.Case, async: true

  test "activating a new user" do
    user = MyApp.User.create(email: "[email protected]")

    {:ok, _user} = perform_job(MyApp.ActivationWorker, %{id: user.id})
  end
end

Running the test at this point will raise an error that explains the module doesn't implement the Oban.Worker behaviour.

text

1) test activating a new account (MyApp.ActivationWorkerTest)

   Expected worker to be a module that implements the Oban.Worker behaviour, got:

   MyApp.ActivationWorker

   code: {:ok, user} = perform_job(MyApp.ActivationWorker, %{id: user.id})

To fix it, define a worker module with the appropriate signature and return value:

elixir

defmodule MyApp.ActivationWorker do
  use Oban.Worker

  @impl Worker
  def perform(%Job{args: %{"id" => user_id}}) do
    MyApp.Account.activate(user_id)
  end
end

The perform_job/3 helper's errors will guide you through implementing a complete worker with the following assertions:

That the worker implements the Oban.Worker behaviour
That args is encodable/decodable to JSON and always has string keys
That the options provided build a valid job
That the return value is expected, e.g. :ok, {:ok, value}, {:error, value} etc.
That a custom new/1,2 callback works properly

If all of the assertions pass, then you'll get the result of perform/1 for you to make additional assertions on.

Testing Other Callbacks

You may wish to test less-frequently used worker callbacks such as backoff/1 and timeout/1, but those callbacks don't have dedicated testing helpers. Never fear, it's adequate to build a job struct and test callbacks directly!

Here's a sample test that asserts the backoff value is simply two-times the job's attempt:

elixir

test "calculating custom backoff as a multiple of job attempts" do
  assert 2 == MyWorker.backoff(%Oban.Job{attempt: 1})
  assert 4 == MyWorker.backoff(%Oban.Job{attempt: 2})
  assert 6 == MyWorker.backoff(%Oban.Job{attempt: 3})
end

Similarly, here's a sample that verifies a timeout/1 callback always returns some number of milliseconds:

elixir

test "allowing a multiple of the attempt as job timeout" do
  assert 1000 == MyWorker.timeout(%Oban.Job{attempt: 1})
  assert 2000 == MyWorker.timeout(%Oban.Job{attempt: 2})
end

Jobs are Ecto schemas, and therefore structs. There isn't anything magical about them! Explore the Oban.Job documentation to see all of the types and fields available for testing.