A study of a large-scale micro component architecture
summary
Spotify extracts telemetry and declarative metadata into BigQuery, builds a dependency graph (rendered with Graphviz/Neato and queried with graph algorithms) across 10,000+ production components averaging ~3k LOC, owned by ~600 squads. Analysis surfaces a scale‑free network of central nodes, informs an automated Fleet Management push and shows 2,600 production deploys/day (mostly direct to prod). Targeted dependency refactors and build avoidance cut a mobile CI workflow by 10x build time, while decommissioning lags (more components added than removed).
AI-written summary · about
chapters (11)
- 0:00Micro component architecture intro
- 3:10Defining components and systems
- 8:34Component ownership and scale
- 10:43Deployment frequency insights
- 11:06Component dependencies overview
- 13:56Dependency graphs analysis
- 16:47Dependencies versus team collaboration
- 20:00Optimizing mobile build times
- 23:50Conway's law and modularity
- 27:25Data-driven improvements
- 29:10Q&A
related talks
19mHigh Performance Backstage for High Performing Organizations
23mScaling the Sound: Fleet Management at Spotify
27mShifting Spotify Engineering from Spreadsheets to Backstage
27mBalancing autonomy, technical alignment, & speed through strategic collaboration
01h 00mFrontend: Graphs. Build. Dead Code
41mMODULARIZATION HOW HARD CAN IT BE
36mModularization - How hard can it be?
24mScaling the Stage: Managing 200+ Plugins & Thousands of Engineers