Beyond Distributed Tracing

on YouTube ↗14m

" summary

Introduces CUI attribution as a propagated metadata string (e.g. youtube/watch) built on Google Census’ baggage mechanism to aggregate latency and error metrics by user interaction across services. Shows a real YouTube outage traced in 15 minutes from product-area alerts down to a root service handling millions qps by following one qps needle in millions, and demos production-safe experiments that replay or fail only specific CUIs. Implementation is simple, interoperable with tracing systems, and effective in practice.

AI-written summary · about

" chapters (7)

1. 0:00Start
2. 3:45Server metrics
3. 5:15CUI (Critical User Interaction) attribution
4. 7:04Night and day difference
5. 8:09Potential root-causing workflow?
6. 11:20Safer Chaos engineering
7. 13:11Beyond distributed tracing

" related talks

• 37mObservability, Distributed Tracing & the Complex WorldDave McAllister · GOTO Chicago 2019
• 16mPragmatic execution tracingNeven Miculinic · WebCamp Zagreb 2018
• 25mTracing GoJaana B. Dogan · Gopherpalooza 2018
• 59mAn introduction to distributed tracingThomas Watson · HolyJS 2018 Moscow
• 51mNavigating Distributed TracesBill Westlin · QCon São Paulo 2018
• 57mDistributed Tracing: How the Pros Debug Concurrent and Distributed SystemsAaron Stannard · NDC London 2019
• 20mDistributed Request Tracing for the Streaming Video EcosystemJosh Evans · Demuxed 2021
• 42mAn introduction to distributed tracing and ZipkinAdrian Cole · Full Stack Fest 2017