Capybara for your API frontend

May 21, 2016

The problem

Recently, I was faced with the challenge of feature testing a Rails-API's AngularJS front-end.

Immediately, I thought of using Angular's feature testing tool, Protractor. While Protractor is a completely valid solution, I found it to be less than convenient. I wanted something that integrated nicely into my existing Rails test suite.

Keeping your test suite DRY

Protractor tests are written in JavaScript. I was already writing factories and tests for my API in Ruby. Committing to Protractor would mean rewriting every factory in JavaScript and maintaining a second test suite.

Testing your code shouldn't be a pain. When your codebase grows to a certain point, tests are the best way to prove your app works. There had to be a better solution!

The solution

After some research, I found that Capybara did exactly what I wanted. Capybara (like most e2e testing tools), can test any webpage regardless of its origin (localhost, on the web, etc.). Here's how:

Capybara.javascript_driver = :poltergeist # or selenium

Capybara.server_port = 3000        # API test server port
Capybara.server_host = 'localhost' # API test server host
Capybara.app_host    = 'http://localhost:4567' # Front-end location

Add the code snippet above to your rails_helper.rb and you're ready to start writing feature specs!


Wait, where's all my data?

Users of the database_cleaner gem may want to take note of this gotcha. If data created in your specs isn't being served to your app, you're likely using the transaction cleaning strategy for your feature specs. While wrapping your database calls in transactions is great for unit tests, your feature specs and test server manipulate the database with separate connections.

Take a look at the database_cleaner README on GitHub for a configuration that works with Capybara. Alternatively, there is also a hack for ActiveRecord that lets your share a single database connection between instances.

Capybara wont wait for my app!

An unfortunate drawback to Capybara is its lack of awareness of asynchronous processes. This makes it incredibly difficult to effectively test apps that are dependent on asynchronous data. Fortunately, Capybara is flexible enough to fix this problem yourself (or with a gem or two).

There are several workarounds for jQuery and Angular that help with waiting on asynchronous data. Alternatively, you can use thread sleeps to pause test execution. Remember to use thread sleeps responsibly (ideally never).

My tests raise circular dependency errors

This is probably the most interesting gotcha of the three mentioned. These dependency exceptions are raised when too many requests hit your test server at once. As you can imagine, this is a very common practice for API-oriented apps.

Thoughtbot wrote an article talking about this issue in greater depth. Including the Rack::Lock middleware in your server's configuration fixes the problem. You can add Rack::Lock to your test environment with this line of code:

# config/environments/test.rb

config.allow_concurrency = false