December 13, 2014 #rails #test

TL;DR

# in config/initializers/delayed_job_config.rb
Delayed::Worker.delay_jobs = true

# in your spec
Delayed::Worker.new.work_off

Common Setup of DelayedJob

Assume you follow DelayedJob readme example to configure it like this: Delayed::Worker.delay_jobs = !Rails.env.test?, what it does is in test env it doesn’t delay the job, meaning DelayedJob is being transparent, the job you put will be executed in “real time”. In most cases you don’t even need to worry about it, your test should be just fine, but recently it caught me…

It Fails When…

To give some background, I’m working on a API centric rails project. In order to authenticate with API we pass in access token for every request, and that’s done in the middleware layer. Since access token is stored in cookie, and in middleware we can’t access browser cookie directly, so another tool called RequestStore is used. If in the same request, what you stored in RequestStore you can access it later no matter the context, a unrealistic example would be you store a cookie value to RequestStore then use it in model later. Don’t do that :).

The code below is a simplified version to illustrate the flow.

class ApplicationController < ActionController::Base
  before_action :set_api_access_token

  def set_api_access_token
    RequestStore.store[:access_token] = cookies.signed[:access_token]
  end
end

class Authentication < Faraday::Middleware
  def call(env)
    env[:request_headers]['Authorization'] = RequestStore[:access_token] if RequestStore[:access_token]
    @app.call(env)
  end
end

Every api request happened inside the rails ApplicationController stack should have the access token being set, but what would happen in a different context like rake task or DelayedJob where you need to send request to the API? The before_action is not gonna be executed there so RequestStore[:access_token] would be nil. This is an easy-to-spot issue if you try it once, but if you follow the TDD work flow and write test for it first, then it’ll fail you.

With Delayed::Worker.delay_jobs set to false in test env, the job will be executed immediately in the same request, so the RequestStore[:access_token] still contains the value and will pass to the Authorization header in the middleware, spec passed but but in real world env it failed. Typical false positive result.

To Run It Manually

# in config/initializers/delayed_job_config.rb
Delayed::Worker.delay_jobs = true

# in your spec
# here is the code to enqueue a job to DelayedJob queue
visit post_path(post)
# run it manually
Delayed::Worker.new.work_off
# expectation
expect(api_endpoint).to have_been_requested
end

Delayed::Worker.new.work_off returns an Array like [1, 0] indicating succeeded job counts and failed job counts. I’ve also seen some people testing against that like expect(Delayed::Worker.new.work_off).to eq([1, 0]), personally I don’t think it’s necessary.

  1. You have your own expectation right after that and that should be the main concern of the spec. If the job failed, your spec should be failed too.
  2. What if multiple jobs are enqueued while you’re only focusing one of them in the spec? Update the value to [2, 0]? That’s just noise.

I guess what I encountered is a rare case, but definitely an interesting case. I kinda prefer this way to mimic real world environment to prevent any possible regressions.

Commercial time: If you’re about to build a API centric rails app, be sure to check out the awesome gem called spyke made by @balvig, the slogan is “Interact with remote REST services in an ActiveRecord-like manner.”