Right Grid
  • Overview
  • Transcript
Trajectory

Production as an Experiment Lab

Ramin Keene Fuzzbox

How do we progress from testing in prod to experimentation in prod? Progressive delivery is a transformative cultural shift that holds great appeal to the enterprise, for whom any notion of continuous implied a loss of control that was far too scary. Yet, testing in prod, be it via feature flags, chaos engineering, or canary rollouts, makes assumptions that would invalidate any field experiment: That any team forming a hypothesis for a test possesses a sophisticated amount of knowledge and control over their systems, ie: what they are, and how they run, how to change them, and how they might break. That any existing knowledge the team possesses of the system is free from bias, contamination, or tampering with experiment behavior. So we ask, what might testing, and thus experimenting, in prod look like for a team that has inherited a production system that no one has ever worked on or seen before? In this talk, we’ll explore how to bring techniques from the world of data science, personalization, and field experimentation into the world of distributed systems, using a mix of distributed trace data, metrics, fault injection, and multivariate a/b testing as a foundation for ongoing analysis, discovery, and proactive experimentation with how systems bend, twist, and ultimately break, safely. You’ll learn modern techniques being adopted by cutting edge organizations to support teams shipping changes safely at speed, as well as here of some wildly experimental delivery techniques that your boss will NEVER go for (at least for a few more years). You’ll leave understanding how adopting safety as a primary concern over caution and correctness can lead to safer, more reliable software, and ultimately, happy users and operators on call!

Downloads slides

Ramin Keene

Ramin has spent the last 5 years working with data teams and large enterprises to put machine learning, a/b testing, and data science products into production. He’s made ALL the mistakes and then some, helping companies lose thousands, if not millions, of dollars along the way. He is currently based in Seattle and spends his time working on adversarial experimentation tools that target infrastructure and release artifacts to help teams inspect and learn about their software AFTER it has been baked and released.

We were unfortunately not able to record this session. It's too bad you can't feature flag life! In lieu of the recording we've provided the original speaker deck with highlights noted by audience members on the day-of.

You can find it here