Wednesday, 24 August 2016

Test Your Live System using Live Service Tests

Traditionally there are three categories of functional tests.

  • Acceptance
  • Integration
  • Unit

This is often refereed to as the testing pyramid. Unit tests form the bulk of your suite, followed by a smaller subset of integration tests. Acceptance tests that cover features should be the tip of your testing strategy, few in number. These are great but there is a missing suite of tests - live service tests.

  • Live Service Tests
  • Acceptance
  • Integration
  • Unit

Live Service Tests.

The role of live service tests (LST) is to test the live system against the production environment and configuration. LST would be fewer in number than acceptance tests. Unlike acceptance tests, these should run constantly. Once a run has completed, kick of a new test run. This will require a dedicated machine or piece of infrastructure, but the value provided is well worth it.

LST should focus on journeys instead of functionality or features. In contrast to acceptance tests a user journey would be the core purpose of the system. For example, a LST suite to cover this blog would ensure the home page loads, an individual post can be loaded, and the archive is accessible. The rest of the site such as comments or social media interactions could be broken, but the core purpose of the system is working. Readers can read and browse the blog. If at any time the tests detect a failure in the core journey there is a big problem.

Why

LST offer the fastest feedback possible due to the fact they are constantly running. It is far more desirable to detect a problem before your users do. Naturally LST offer great protection after deploys. Deployment of new code is one of the times you are more likely to encounter issues, so a suite of tests triggered after a deployment is a natural fit. LST also protect against unplanned events. In my experience, exceeding disk space, DNS failure, third party issues and more have all be detected.

How To

Adding another suite of tests may sound like increased effort but the cost associated with LST is rather low. Sometimes acceptance tests can be run as LST, meaning no extra effort. Care must be taken here if the tests perform anything destructive or anything that interacts with third parties.

Alternatively writing LST is simpler than standard acceptance tests. The same tooling can be used such as Selenium, NUnit and so forth. As the tests themselves focus on journeys rather than functionality, the tests are often less complex to write.

The only difficulty LST introduce is the fact they are executing against the live system. Consider interactions with a third party. Using a real account on the real system may be problematic. One way to get around this issue is to embed test functionality within the system itself. For example you could set up a test account which the tests use. Instead of executing against the third party system, the dummy account is used. Likewise most third parties have test accounts which can be setup and used instead.

LST are a nice compliant to a diagnostic dashboard. If your dash is reporting no issues, and your tests are green, you can be confident the system is operating in a good enough state.

Lessons

  • Functional tests are not enough.
  • Use live service tests to test the real production system.
  • Run live service tests constantly for the earliest feedback possible.
  • Alter production code to introduce test functionality.
  • Make use of test accounts and anything else that third parties may offer.

No comments:

Post a Comment