Building + operating a service mesh at a mid-size company
summary
Cookpad runs a homegrown Envoy-based service mesh for ~200 developers, ~100 services and 90M monthly users, deployed on AWS ECS with Envoy sidecars per ECS task. Configuration lives in a central GitHub repo; CI generates xDS/RDS and uploads to S3 so central GitHub + CI -> S3 xDS. A Rust SDS translates REST POST-to-GET for xDS; metrics use DogStatsD/StatsD exporters into Prometheus/Grafana. Results: fine-grained observability and host-independent retries/timeouts/circuit-breaking; challenges include xDS warmup and some EC2 edge cases.
AI-written summary · about
chapters (20)
- 0:00Start
- 0:38Agenda
- 0:58Our business
- 1:28Our scale and organization structure
- 2:18Technology stack
- 3:23Operational problems
- 3:35Library approach solutions
- 3:46GoPython Java apps?
- 4:27Timeline
- 4:55In-house control-plane
- 5:34Considerations
- 6:12Our service mesh components
- 9:45Configuration file
- 14:42Dashboards
- 17:07Integrate with in-house platform console
- 17:30Envoy on EC2 instances
- 18:47The hard points
- 19:05Observability
- 19:18Continuous development of app platform
- 19:42Next challenges
related talks
39mGrow with Less Pains - Meshing From Monolith to Microservices
26mService Mesh Failure Stories at Scale
38mAsk Me Anything About Service Mesh
19mDoorDash Diaries: Mastering Microservices at Mega Scale
20mThe Crazy Service Mesh Ecosystem
56mHow to get a grip on your microservices system using a service-mesh
35mUnlocking the Power of Kuma Service Mesh: Enhancing Reliability and D...
36mPanel: Ask Me Anything: Microservices and Service Mesh [B]