Programme launch offer - Save 20% Book Now

Track Talk W12

The Anomalous Tester

James Bornefelt Westfall

15:00 - 15:45 CEST Wednesday 4th June

Meteorites struck Oslo, Norway in 2012. One of them crashed through the roof of someone’s garden shack. They did not make any automatic tests fail at my workplace. But they could have.

With so many variables potentially affecting the outcomes of automated tests, how do we see clearly? The answer is data: time series of test result data over time. How do we bring in machine learning to see even more clearly? The answer once again is data: well-structured historical test data for training the models.

This talk will cover an “Aha” insight that came to me while testing the Norwegian Ministry of Education’s system for national examinations. I was shocked by the random things happening when we started collecting logs during exam periods when up to 70 000 students were simultaneously connected to the system.

This experience impressed upon the need to capture data and build systems that cut through test result data noise and isolate meaningful, actionable trends. Trends that point to either latent problems in test infrastructure or to real problems in the system under test.

While the majority of test failures when taken individually are simple and unambiguous, reliably catching all of the insights you could have and should have captured from your test runs before a release is an entirely different problem.

I will demonstrate in detail how we calculate meaningful trends to give actionable insights. This will cover how we work strategically to fill holes in the data we collect to uncover latent problems that frustrate our customers. Finally, I will demonstrate how we use machine learning with homegrown models to solve the thorny problem of what to do with tests that we expect to fail.

It will be a wild ride. But I promise practical recipes you can apply to tame the chaos (somewhat). But still no solution for the meteorites.