Good practices for Integration Tests for my rails app external endpoints?

I'll keep it short, I've got a rails app which communicate with other apps, some using SOAP (non-rails apps of course...) and others with REST. I'm making integration tests to ensure that my endpoint开发者_如何转开发 wrapper classes have correct mappings and setup. However, they are executed by default by rake test which makes it slow and fragile. I wish to run unit tests frequently and integration tests "on-demand" only. How do you do that?

What're your preferences wrt such integration testing?

• How deep do you unit test and/or mock?
• Do you replicate whole SOAP or REST xml responses in stubs?
• Do you create "external endpoint" integration tests at all?

Update Q: How to exclude a test-dir while running rake test ?

If you go by what the Rspec/Cucumber folks suggest, then the integration test level is an inappropriate place to mock your data, because in some respects, it defeats the purpose of the integration/acceptance test. However, you have to mock stuff like paypal transactions, right? In my current project, I am facing a lot of this, and here are some of the solutions I am implementing:

1. Tagging tests that wont work in certain contexts. In my example, lots of servers live behind firewalls and so my tests dont pass if I am at home and not using vpn. So, in cucumber I can tag these as @firewall and tell it to run tests that are not tagged firewall. I'm pretty sure Rspec 2.0 supports this feature as well.

2. Mocking service requests. Yah, its probably a bad idea, but I am at a loss on how to do it otherwise with any kind of predictability. I have a separate test suite to affirm that the services are running, and from my rails app, i am assuming they are working properly. An example of this would be LDAP. And yes, in these circumstances, I tend to use a real response and do something like. response = double('response') ; response.expects(:data).and_returns('my xml here')

3. I do think regardless of the complexity of the system that end point tests are really important. I am really enjoying cucumber, it provides me 95% of what I need to do in functional tests, and so I end up writing fewer of these tests and more of the entire workflow tests.

The solution is VCR.

Excluding endpoint integration tests from rake test and being able to isolate and run them with rake test:endpoints was solved with only a few lines of code. I have to admit though, I spent a whole lot of hours swearing and cursing. There should be more documentation and explanations in the railties source. Ruby code like that tend not to be very self-explanatory, IMO.

Well, here it goes: create your task: lib/tasks/slow_tests.rake

require 'rails/test_unit/railtie'
desc "Runs all endpoint integration tests."
namespace :test do

  #hooks on to the test task through the 'test:prepare'
  #For details, check the railties gem (v3.0+) lib/rails/test_unit/testing.rake,
  #look for "task :test" and "namespace :test"
  TestTaskWithoutDescription.new(:endpoints => 'test:prepare') do |t|
    t.libs << 'test'
    t.pattern = 'test/endpoints/**/*_test.rb'
  end
end

Now I may put my fragile endpoint integration tests in the test/enpoints directory, running them whenever I want (not often)

Note: this supposes test/unit or shoulda.

You should write a thin layer around external APIs (Facade/Wrapper) and use the vcr-gem to "stub" network calls. You can get more information from my article on rails test architecture.