How a Service Mesh Helped Us Build Production Cloud-Native Services
summary
A complete rewrite replaces a fragile PHP monolith with 70 microservices across 5 languages and 5 databases and uses OpenResty + Lua to reduce access token latency from 500ms to 40ms. Linkerd is deployed for sidecar proxying, health aware load balancing, retries, tracing and metrics, and is up and running in under a day with mostly default settings. The mesh cuts tail latencies, lets teams do canary and traffic mirroring, and proves effective in production.
AI-written summary · about
chapters (8)
- 0:00Start
- 3:00The most used and slowest performing part
- 3:41How did we approach a solution?
- 9:08Development
- 9:50Circuit Breaking problem
- 15:05Canary deployment
- 15:58Shadowing traffic
- 16:40Dark Launches
related talks
39mGrow with Less Pains - Meshing From Monolith to Microservices
35mNavigating the Service Mesh Ecosystem
04mInterview
37mIntro: Linkerd
30mService Mesh Is Still Hard. The Things we all Did Wrong & Tales of Woe
34mService Mess to Service Mesh
34mFrom Monolith to Microservices with Kubernetes and Linkerd
43mWTF Do We Need a Service Mesh?