What I Wish I Had Known Before Scaling Uber to 1000 Services
recording made private on YouTubehttps://youtu.be/kb-m2fasdDY
summary
Over 1,000 production services grow into a 2,000‑engineer org (200 a year and a half earlier) and >7,000 repos, fragmenting languages (Node, Python, Go, Java) and tooling. Hidden costs: RPC semantics and JSON brittleness, P99 fan‑out failures where a 1% slow case multiplies, migration debt instead of immutability, and divergent dashboards. Concrete fixes: Zipkin/OpenTracing with ~1% sampling for distributed traces, flame graphs for cross‑language profiling, zap structured logging, chaos/failure testing and testing load against production.
AI-written summary · about
related talks
21mThe great SOA migration
07mMonolith to over 60 Services
01h 01mMicroservices Gone Wrong
51mMicroservices and Microfrontends: Scaling, Learning, Growing
23mMicroservices – Diptanu Choudhury
53mScaling Engineering by Hacking Conway's Law
26mFighting the Good Fight at the Hot Gates of Micro-Services
33mPros/Cons of Highly Scalable Microservice Systems