Critical Mass of Tests for Continuous Deployment
Critical Mass of Tests for Continuous Deployment
What is the minimum level of test coverage necessary to trust automated tools with continuous deployment? Let’s find out.
Join the DZone community and get the full member experience.Join For Free
Do you need to strengthen the security of the mobile apps you build? Discover more than 50 secure mobile development coding practices to make your apps more secure.
Does this situation sound familiar?
An engineering team relies on manual testing for quality assurance and preventing regressions. The manual process is lengthy, prone to errors, and doesn’t handle edge cases, so there are frequent regressions that slip into production. The deployment pipeline has slowed to a crawl, causing features to pile up on top of each other. Getting anything into production keeps getting slower.
It’s not all so bad. The team has some tests and automatically runs them with continuous integration. However, they don’t trust the test coverage enough to allow an automated system to ship into production. But why? What’s missing? What is the minimum level of test coverage necessary to trust automated tools with continuous deployment? Let’s find out.
The Way Is Continuous Deployment
Continuous Integration refers to integrating, building, and testing code within the development environment. Continuous Delivery builds on this, dealing with the final stages required for production deployment.
– [Martin Fowler][mfcd]
Leveraging automated testing as much as possible reduces the costs of building and deploying a system. Those are financial, temporal, and communication costs being paid by the entire company. Continuous deployment goes a long way toward achieving these business goals:
- It takes all of the ceremony out of releases. When you release often, the iterations are small and focused. That means less can break between releases, and it’s easier to fix problems that appear.
- Releasing often maintains momentum and the appearance of progress.
- The earlier you get features in front of real users, the faster you can get feedback.
So, what is the minimum amount of testing necessary to be confident in continuous deployment? In typical engineering fashion, it depends on what your system does, but there are some guidelines. First, let’s talk about the irreplaceable smoke test.
The bare minimum amount of testing for any integrated system is a smoke test. A smoke test gives you basic confidence that a system can be built, booted, and serve simple requests. A bit more is needed before you let an automated system push your code out to the entire world, but the smoke test is a good place to start.
Unless your application is purely an API, a smoke test should be run within the context of a real browser environment. There are subtle differences between simulated request/response testing and rendering real HTML. A broken link or incorrect asset reference may not be caught in a simulated test, but within a browser it will raise an error and fail the build.
Happy Path Story Testing
In any application there are a few paths through the application that are essential to the experience. Paths that weave together multiple features, like “onboarding” or “initial signup” should have an integration test that covers the happy path.
The happy path is the journey between features where everything works as expected. It isn’t concerned with failure scenarios, only critical features and the intersections between them. Where smoke testing exercises the boundaries between layers of an application, happy path testing exercises the boundaries between features of an application.
An application needs a happy path integration test for every user story that is critical to the business. For example, an online store would have a story test that traced a new customer from the point where they added an item to their cart right on through checkout. The customer may only add one item to checkout, and their payment method would work on the first try, but you could rest assured that the complete story was functioning as expected.
Building With a Production-Like Environment
For us, “end-to-end” means more than just interacting with the system from the outside — that might be better called “edge-to-edge” testing. We prefer to have the end-to-end tests exercise both the system and the process by which it’s built and deployed. [Growing Object Oriented Software Guided By Tests][goos]
Building the final compiled, minified, and compressed production release is a critical part of the deployment process. Any application doing something interesting has a lot of moving parts, some of which are never utilized outside of production mode.
Your testing story should include a production test, in which code is built and deployed to a production-like environment. That environment may be staging or an ad-hoc environment that simply verifies that the app can boot. The important part is that it is running in production mode, with production code, and mimics a production environment as closely as possible.
Integration Testing Enables Continuous Deployment
Note that there hasn’t been any mention of unit testing, generative testing, mutation testing, or any other in-depth developer-centric testing. All of the components critical to building up a body of tests for continuous deployment are focused around integration. Other automated testing styles, unit testing in particular, are tools to help refine your code, explore edge cases, and allow confident refactoring. Integration testing is about verifying boundaries.
Boundaries are where friction builds up between layers of a system. Classes and modules will operate perfectly fine in a vacuum, but when you connect them to a database, a file system, and other disparate components things break. That places the initial burden of automated testing squarely on integration. Build up enough trust in your integration tests and the momentum will sustain itself between releases. Then you can begin to trust in continuous deployment.
Published at DZone with permission of Moritz Plassnig , DZone MVB. See the original article here.
Opinions expressed by DZone contributors are their own.