Build Health and Velocity Score: How Pinterest tracks the state of builds
summary
Pinterest defines a Build Health Score by multiplying measurable signals: percentage of local builds under 10 minute local / 45 minute CI thresholds, CI uptime, and integration test pass rate calculated monthly, with extra weight on uptime. The implementation surfaces grades like Android ~85% iOS ~70% and triggers a code yellow when critical. Root causes include CI OOMs fixed via the Gradle task configuration avoidance API. Arm Graviton CI trials show regressions of 1–31% and only ~$$50K/year (~10%) savings.
AI-written summary · about
chapters (15)
- 0:00Introduction and agenda
- 1:11Why builds matter
- 3:46Developer frustration with builds
- 6:53Build score importance
- 9:07Measuring build health metrics
- 11:35Local and CI build time thresholds
- 14:03CI uptime and test pass rate
- 16:11Calculating the build score formula
- 17:38Adapting metrics for other platforms
- 19:04Learning from score iterations
- 20:10Build score trends and fixes
- 21:33Fixing CI memory issues
- 22:15CI cluster downgrade impact
- 22:56Testing Graviton for CI
- 23:38Q&A
related talks
45mUI Testing at Pinterest
12mEngineering Productivity, Infrastructure
26mEfficient Android Builds at Scale: Lessons from Pinterest
47mBuilding a Foundation for Performance
26mStarting a Dedicated Web Performance Team the First Ten Months
34mBuild Scoring: A Formula for Blocking SEV1s from Production
47mHow to win in software development with tighter feedback loops
30mAndroid Developer Productivity