The Testing Pyramid Series
What exactly is the testing pyramid? What is the difference between an integration test and a system test? Are system tests integration tests? When should I be writing unit tests and when should I be writing integration tests? Can I write too many tests? Are all tests useful? Why even test at all?
About the Series
This is a first part of a multi-part series where we explore the traditional testing pyramid, challenge it, and try and answer common questions about testing.
What is the Testing Pyramid?
The testing pyramid is a concept originally described by Mike Cohn in his book Succeeding with Agile. Today there are many forms of the testing pyramid. The traditional pyramid consists of three tiers: UI (at the top), Service/API (in the middle) and Unit (at the bottom). The idea being that the majority of your tests are unit tests and your UI test suites being the smallest.
There have been other versions of the testing pyramid. For our purposes, let’s consider these three forms of testing:
We don’t have to think of these types as of testing in the form of a traditional pyramid - maybe something more configurable and that you can tailor to your needs.
That being said, I think we can all agree that we need less system tests. They are the slowest, least reliable, and most expensive. More on that later.
Regardless, I still call it the testing pyramid - but maybe with air-quotes. But if you want… we can call it something else. How about the Testing Stack or Testing Pyramid-like? Meh… Let’s stick with testing pyramid for now. And for the record: sometimes more unit tests aren’t a good thing. More on that later.
Unit tests are tests that test individual functions.
test x ↦ f (x) produces y
Unit tests are great for many reasons:
But if you’ve ever written a lot of unit tests you may have run into some of the problems listed below:
- coupling of tests with implementation (resulting in a lot of refactoring as requirements change)
- hard to test from the user perspective
- hard to test complicated workflows
- provides fake confidence in functionality (I wrote a bunch of unit tests!?! … but they didn’t really test the thing you did in the database)
I’ve found that unit tests are great in certain situtations. When you want to test individual functions and algorithms. They let you quickly answer questions such as does
my_cool_function(foo: any) actually return
bar: Thing, or is my algorithm working? It doesn’t make much sense to write a complete system test for this. But at the same time it depends on what
my_cool_function is doing and where it’s being used in your application’s stack. This leads me to my next, and favorite, form of testing.
Integration testing is the act of testing multiple units of code at once. What does this really mean? Well it depends. It really is up to you. I would argue anything that isn’t a unit test can be considered a form of integration testing. Modern software is complicated and consists of countless number of nested and interrelated functions. So it becomes a point of further classification, for example: api testing, greybox testing, system testings. Some examples of integration tests that come to mind:
- testing a function that depends on many more functions
- end-to-end testing (those are a lot of functions 😄)
- testing your backend’s internal service layer mocking out a database (or not)
This ambigiouty is a double edged sword. Ultimately it’s up to you, the software practitioner, to define what they mean to you and your team.
I tend to treat integration tests as slimmed down system tests where my tests function as codified contracts at api boundries. I design my test’s “plumbing” in such a way where I swap out unreliable and non-determistic components with ones that adhere to those component’s contract’s and are test-friendly implementations that allow my teammates to more efficienly test functionality and workflows without having to always test the non-deterministic and complex nature of the systems. Those unreliable and non-deterministic components are tested in isolation and their interesting behaviors are understood and documented. But we don’t allow their interesting behavior to adjust our approach to testing the rest of our apps.
I’ve noticed that this sentiment isn’t shared by a lot of folks. But I’ve found if you start breaking down your system into smaller digestible chunks and the interactions between them as codifiable contracts and test those contracts - testing and developing becomes much easier, faster, and maybe a little more fun.
- Allows for expressing complex user scenarios
- Much faster then system tests
- Can mix-and-match mocks for testing complex workflows that involve external systems (more on that later in this series)
- Allows you to do black box, white box, or grey box testing depending on your needs
- Allows you to test from the user’s perspective
- Biggest bang for your buck
- Integration tests vary between systems, tech stacks, and companies. Some companies write integration tests that mock out large part of systemts largely invalidating the whole point of testing
- Can be slow if written inefficiently (e.g. reprovisioning systems for each test)
- coupling of tests with implementation
These are your slowest but most thorough type of tests. System tests are a form of black box testing where you interface with a running and fully provisioned system. This is great since system tests can test everything. But at the same time it’s not ideal since you are testing everything at the same time everytime. They are also non-deterministic in nature. Things fail in environments that don’t get the same TLC as production and not understood by stakeholders other than those that provisioned them.
Why does email not work? Is it broken?!? … No we just didn’t configure it…
I’ve seen great amount of emphasis on system tests when many verifications can be placed lower down the stack as integration tests or unit tests. Before you write that (unnecessarily) complicated system test (e.g. a UI test or testing a RESTful api against an external service) ask yourself or the developer writing the feature if they wrote tests and what kind of tests. These sort of conversations will be covered in later articles.
- Allows for expressing complex user scenarios
- Allows you to test as the end user
- Allows you to do testing with external systems
- Very Slow
- Hard to maintain
- Tests external systems (external systems are already tested by the companies that provide those systems - why do it again?)
Later articles in this series will go into this concept - focusing more on unit and integration testing and less system testing.