Changing Oil for a Fast Running (Side) Car Quickly and Safely
summary
Pinterest runs a centralized tower control plane that pushes typed, templated configs via a host beacon agent to Envoy across VMs, containers and k8s, with Jenkins pipelines, unit tests and staged rollouts that reach full production in under 25 minutes across thousands of clusters. A near‑real‑time xDS feedback loop and per‑cluster health checks detect bad configs within a minute, enabling automated canary gating and rollbacks; migration to xDS v3 finishes with zero incidents, while global RTDS changes prove too risky and get disabled.
AI-written summary · about
chapters (16)
- 0:00Intro
- 0:26Agenda
- 0:53Pinterest
- 3:35Clusters
- 5:21The Scenario
- 5:51Three Perspectives
- 6:31Configuration
- 8:13Mesh Perspective
- 10:18Who owns Mesh
- 16:10Negative feedback loop
- 18:19Failures
- 21:21Confidence
- 22:45Visibility
- 24:03Summary
- 25:21Best Practice
- 26:31MTRS
related talks
11mxDSnap: A Lightweight Tool to Debug Envoy and Consul Dataplanes with Confidence
01h 04mOpening Keynote [Full]
32mHow Lyft Used Envoy to Rethink Microservice Development
13mAutomate an Observable Runtime with HashiCorp and Ansible
39mHow we learned to love the Data Center Operating System
46mDeploying and maintaining software for today's infrastructure
28mDevOps Road Blocks
31mEnvironments as a Service: improve DX