Uber’s Android CI Engine: Lean, Green, and Fast
summary
Uber’s merge queue Submit Q implements four core components—enumerator, profiler, prioritizer and selector—to enumerate speculative merge trees, run conflict analysis against build target hashes, and schedule only high‑probability builds using ML models (classification at 97% accuracy, build‑time regression MAPE 5–7%). At scale (4,500+ engineers, ~65,000 changes/month, ~130M LOC) it reduces CI work by ~53%, CPU consumption by ~37% and median wait times by ~45% by using probabilistic speculation and bypassing large diff.
AI-written summary · about
chapters (13)
- 0:00Uber’s CI engine intro
- 1:24Green mainline challenges
- 2:50Red mainline business impact
- 3:32Submit Q for green mainlines
- 4:13Submit Q core components
- 6:19Conflict analysis to reduce builds
- 8:31Probabilistic speculation efficiency
- 10:59Bypassing large diffs
- 12:46ML build prioritization
- 13:50Build success prediction
- 14:32Submit Q end-to-end
- 15:15Impact at Uber scale
- 16:18Q&A
