Envoy Project Intro
summary
Envoy is deployed as a sidecar across Lyft’s infra — about 20,000 hosts handling peaks of 5 million RPS — and configuration is moved from hand-edited JSON into a control plane that translates service manifests into xDS gRPC streams. The system stores configs in S3/GCS, generates XDS artifacts via a data-processing layer, caches them, and serves them with an XDS server; changes are treated as code (Git PRs, CI, staging/canary), most config edits avoid binary restarts, and dashboards trace config versions through rollout.
AI-written summary · about
chapters (14)
- 0:00Intro
- 0:27My time at Lyft
- 2:07The Story
- 3:18Frame of Reference. The Control Plane
- 4:16Envoy Design Goals
- 8:11Envoy Rollout Service Sidecar
- 8:40Envoy Rollout Service Mesh
- 11:45Pain points
- 12:48Configuration Management - The Present
- 17:01Documentation
- 17:33Making a Change
- 18:01Deployment
- 19:11Versioning
- 22:01The Future
related talks
34mIntro: Envoy
33mGet to know Envoy - the modern proxy for cloud-native infrastructure - Baptiste Collard
35mEnvoy Maintainer Q+A
11mA decade of Envoy: From Lyft's Debugging Nightmare to the AI Infra...
11mWelcome + Introduction
27mEnvoy - Overview and Update: Senior Envoy Maintainer
33mIntro: Envoy
21mDeep Dive: Envoy