Finding and Fixing Strange Builds
summary
BuildBuddy shows a drill-down heatmap backed by an OLAP (ClickHouse) backend that slices builds by pattern, repo, host, user and more; highlighting a week reveals that a bare-metal test workflow accounts for 75% of P90 slow runs while only 13% of builds overall, and digging into invocation timelines exposes a single flaking test sitting idle for 10–15 minutes. The tooling surfaces action execution, download/upload times, RBE queueing, and Prometheus metrics, lets teams tag builds with text labels, and proves effective at locating and fixing the regression.
AI-written summary · about
related talks
39mSPACE and Beyond - Measuring Developer Productivity
14mFlaky Tests, Slow Builds, Overloaded CI? You're Guessing at the Problem
11mAre You Ignoring Your Most Expensive Bazel Build Problems? (Develocity Can Help)
28mResolving Outages Faster With Better Debugging Strategies
06mDevops Behavioral Economics: Displayed...
25mImplementing Metrics-Driven DevOps: Why and How
25mTroubleshooting at Scale: What DORA Metrics Miss
30mBuilding RailsPerf, a toolkit to detect performance regressions