Netflix + Gradle A Journey in Developer Productivity
summary
Netflix runs a 180‑module Android repo with ~30 developers and 10 variants and cuts iteration time with concrete changes: an engineering build (no dynamic buildDate, fewer plugins, fewer variants) drops studio assemble-debug runs from ~7m to ~2m across 939 builds; CI average falls ~14→8 minutes (≈46%); PR checks drop from ~39→25 and later to ~5m after AGP 7.1. Work uses Gradle Enterprise build scans, Gradle Profiler, a remote shared cache, modular feature modules + Dagger Hilt, version catalog and dependency locking; fixes include disabling PNG generation, normalizing JDK/manifest diffs and patching Apollo codegen ordering.
AI-written summary · about
chapters (17)
- 0:00Introduction and meetup story
- 1:09Netflix Android build overview
- 3:41Improving build incrementality
- 7:57Modularization journey
- 11:30Monitoring build regressions
- 13:38Build caching fundamentals
- 17:05Analyzing build scans
- 20:39Fixing cache misses
- 22:02Performance improvements results
- 23:04Build regression investigation
- 26:14AGP 7.1 performance boost
- 28:02Ongoing build health maintenance
- 29:05Dependency management with version catalogs
- 31:53Controlling transitive dependencies
- 33:58Build speed challenge and resources
- 35:00Quick recap and best practices
- 36:05Closing and hiring
related talks
45mNetflix’s Secret to Safe Automation at Scale
42mBuilding and Deploying Netflix
54mFrom Laptop Builds to Advanced CI
40mNail your Gradle build time
54mDeveloper Productivity Engineering Deep Dive
36mImproving Gradle Builds
40mSimple Recipes to Enhance Developer Productivity
32mFaster Feedback Cut Gradle Build and Pipeline Times in Half