Service modeling
summary
Service modeling documents real endpoint behavior at GitHub and measures CPU, memory, MySQL queries, cache hits/misses and unicorn worker usage to prioritize fixes. Key findings: permissions table accounts for >40% of rows fetched, six controllers serve ~50% of traffic, and a create-issue request traces to 3.56s total with 0.9s spent creating the issue. Process uses tracing and flame graphs, the large unicorn collider performance testing environment that can run live-sampled traffic, dashboards and targeted 1–2 page reports.
AI-written summary · about
chapters (18)
- 0:00Introduction and welcome
- 1:26About the speaker
- 2:47In memory of Chris NOA
- 4:12What is service modeling
- 5:14Why GitHub needs this
- 7:22Initial findings and surprises
- 9:09Issue creation workflow analysis
- 10:33Impact of React migration
- 12:17Steps to implement service modeling
- 13:45Drawing request diagrams
- 14:27Analyzing downstream effects
- 15:30Using the USE P method
- 16:53Choosing metrics to monitor
- 17:13Large Unicorn Collider demo
- 17:55Creating monitoring dashboards
- 18:38Identifying optimization opportunities
- 19:41Final report and collaboration
- 20:01Measuring success
