EnvoyCon North America 2022

To scale Lyft’s engineers' productivity, the Lyft developer experience team pivoted away from costly and hard to maintain custom individual environments, and rebuilt the development environment around a shared staging environment.

The goal was to enable Lyft developers to run multiple versions of the same service (both in Kubernetes and on their laptop) without stepping on each other's toes. To do this the team designed a scriptable ingress proxy as well as used custom Envoy filters within Lyft’s service mesh to route traffic to these isolated instances, inject and propagate custom metadata, and offload the traffic to custom developer tools. The end result gives engineers special networking debugging superpowers during development.

In this talk, you’ll follow Lyft’s progression designing and developing these different components. You’ll deep dive into the custom Envoy filters and how they combine tracing, Original DST clusters, a custom xDS control plane, and local tooling to build this new developer experience. We will also cover realized benefits to Lyft’s engineering productivity as well as problems encountered along the way.