Breaking up Lyft’s Development Monolith

At the beginning of 2020, a Lyft developer trying to ship changes could spend an hour or more a week provisioning and maintaining a remote dev machine that ran an isolated copy of Lyft. Developers needed a way to focus locally on their service while sharing a stable testing environment.

This talk will dive deep into how we use request context propagation to enable developers to safely share our staging environment. You will learn how we attach data to requests and use that to override mesh routing to enable developers to iterate on untested code without thrashing each other. We will also get into the details of how we ensure context propagation throughout both the network and systems like queues.

As part of that I’ll also enumerate our struggles to quantify the good and bad of our new development environment. We will discuss some of our failures with surveys and adoption data and our success with second order metrics such as build times.