TDD for APIs @ Europython 2015 by Michael Kuehne

It is always tough to test a complex API comprehensively. The additional level of complexity brings us to the question “How can we validate that our API is working as intended?”

In this talk I will explain how to use test driven development for APIs to solve this problem and even further how TDD can drive an API Design towards a more usable design. I will outline my practical approach with an implementation example based on django. And finally I will give you a brief summary of my lessons learned using this approach in customer projects.

Transcript

1. TDD for APIs Michael Kuehne
2. Michael Kuehne •  Hamburg, Germany •  Independent Software Engineer •  Likes Python & APIs •  Clients projects mostly API- related
3. "How can we validate that our API is working as intended?“ - Steve Klabnik
4. TDD Cycle Inside-Out Write a failing Unit Test Make the Test pass Refactor
5. API   Business  Logic   Task  Management   ORM   Message  Queue   No9fica9on   System   Mobile  Device   Single  Page  App   Service   API   Models   Database   API   Example Architecture Clients   External   Services  
6. Defined  Input   POST /mailboxes HTTP/1.1 Host: http://api.mkuehne.com/ Content-Type: application/json { ‘username’: ‘michael’ ‘mail_address’: ‘michael@mkuehne.com’ } Defined  Output   HTTP/1.1 201 CREATED Date: Wed, 22 Jul 2015 11:45 Content-Type: application/hal +json Location: /mailboxes/1a3b3s24bd
7. „Unit Tests confirm we have written the code right“ – Unknown „Acceptance Tests confirm that we built the right software.“ – Ian Cooper
9. Acceptance Tests on all Levels
10. BDD Cycle Outside Inside Write failing Acceptance Tests Write failing Unit Tests Make Tests pass Refactor
11. Numbers from Clients Project •  ~3.000 Tests •  ~2.000 Unit Tests •  ~1.000 Acceptance Tests •  > 300 PyTest Fixtures
12. Test Architecture •  Integration tests => different test project •  Use your own client sdk •  Functional & Unit tests => main test project •  Debian packages •  Mock only external services
13. How to design API tests? •  Always test to the end •  F.I.R.S.T. Principle •  Reusable helper classes •  Deterministic test environment •  PyTest: Fixtures, fixtures, fixtures §  Clean database & ORM §  Test data §  Versioned Endpoints
14. Testing Versioned endpoints with Django Rest Framework and PyTest
15. API   Business  Logic   Task  Management   ORM   Message  Queue   No9fica9on   System   Service   API   Models   Database   API   Asynchronous Architecture Clients   External   Services  
16. Tests for asynchronous behaviors •  Celery (task management) caches! •  Validate the input parameters synchronous •  Test Deltas (validation <delta> processing) •  Make aync results accessable •  Wait for all async tasks in each test
17. Wait for all task in celery_taskstate
18. Async Testing: Outside to Inside API   Business  Logic   Celery  Task   Task  without  Celery  
19. "TDD doesn't drive good design. TDD gives you immediate feedback about what is likely to be bad design." - Kent Beck
20. How can TDD help improve API design or avoid bad design? •  Inverse Conway Maneuver •  API design evolves from the outside to inside •  Refactoring without changing behavior •  Endpoint Versioning with confidence
21. Lessons learned •  Great API specifications required •  Best API design come from outside the team •  Don‘t forget your versioned Endpoints •  Don‘t underestimated the test effort for API with asynchronous behaviors •  Complexity vs. deterministic environments
22. Thank you and use TDD for your APIs Michael Kuehne michael@mkuehne.com