I recently did some work on a somewhat neglected C# code base; the basic function of the application was to pull down some custom reports from a third-party web API, parse the results, and store them in a database. There were a few obvious design problems, and the application hadn’t been built or deployed in years, but it did include a small test suite–around 50 unit tests, written with Microsoft’s testing framework.

I was able to get the project to compile without too much trouble, but when I ran the test suite, I was shocked–the run time for those 50 tests was around ten minutes! At first I thought that the tests might be integrated tests in disguise–hitting the production API and database–but all the tests seemed to be mocking out everything appropriately, using the Moq framework.

The answer turned out to be some throttling logic, based on default configuration values. Included in the configuration settings was a value indicating how long the application should wait between requesting customer reports–30 seconds. In production this didn’t matter a great deal–the reports were pulled down nightly, as a scheduled task. The test suite was another matter, though–this setting didn’t affect every test, but it pushed several tests into the minute range for run time.

The Solution

The project was using a settings file to specify configuration values; the configuration file gets wrapped in a class automatically, and the values can be changed at runtime. By using the AssemblyInitialize annotation, I was able to set the timeout value to zero during the test runs:

[TestClass] public class TestInitialization { [AssemblyInitialize] public static void BeforeEach(TestContext testContext) { Properties.Settings.Default.CustomerReportAccessDelayInSeconds = 0; } }

Doing this brought the total runtime for the test suite down to 20 seconds–not great, but not bad for C#!

The Takeaway: Keep things Fast!

This might seem like an obvious fix, but the original developers lived with a ten minute test suite, and thought that was normal. Twenty seconds seems fast in comparison, but for a typical Ruby or Python test suite, I’d expect to do at least an order of magnitude better. Faster is always preferable, but the context is obviously going to make a difference.

In general, when it comes to TDD, anything longer than a few seconds will have me looking to optimize–either looking at my design, checking my tooling, or for very large suites, by running a subset of tests.

comments powered by Disqus