Scaling Your Engineering System
summary
An engineering system runs a monorepo with 1,500 packages holding ~10 million lines of code and ~1,000 developers (≈800 unique committers/month). It builds custom infra: a topological task runner called laga build system with cloud backfill caching to avoid rebuilding unchanged work, a midgard yarn strict install that cuts installs from ~30 to ~8 minutes, Git/Scalar performance fixes, a beachball changelog flow with a separate version branch, RFCs for changes, and visual diff testing; these tools work at scale.
AI-written summary · about
chapters (24)
- 0:00Introduction and background
- 1:04What is an engineering system
- 3:51Front-end devops and team roles
- 5:14Microsoft’s engineering system scale
- 7:20Principles for scaling engineering systems
- 10:05Generalizing solutions at scale
- 12:15Measuring everything for improvement
- 14:00Sharing components across teams
- 15:45Benefits of a monorepo
- 18:13Fixing Git for monorepos
- 20:18Laga: monorepo build tool
- 22:47Build graph and task scheduling
- 23:52Caching with backfill and build Excel
- 25:14Installing dependencies at scale
- 27:02Midgard yarn strict and pnpm
- 29:54Strict dependency management
- 30:36Semver challenges at scale
- 34:10Dependency duplication issues
- 36:17Merging multiple monorepos
- 37:20Managing common dependencies
- 38:44CI/CD at massive scale
- 40:27Testing strategies at scale
- 41:51Visual regression testing
- 42:11Conclusion and Q&A
related talks
43mThe Engineering Playbook for Growing Android Teams
53mScaling Engineering by Hacking Conway's Law
28mFrom Startup to Scale Up: How to Scale Engineering teams
51mScaling Sustainable Innovation through Team Autonomy
40mWhy your best engineers should look after the worst systems
53mVery Large Development: How to Run Code Review for 1000 Developers
01h 28mApps at Scale: An Industry Discussion
45mPeople and Culture