Top Ten Worst Repositories to host on GitHub
summary
GitHub recounts real failure modes: built Lariat caching layer to collapse thousands of identical fetches, developed Spokes ref-update bottleneck mitigation for merge-button storms, and throttled a hackathon that produced ~7,000 pushes in an hour. Surprising numbers include a 365 GB repo filled with repeated images, 260k commits and constant fetches from CocoaPods (broken by shallow-clone workflow), and repositories with hundreds of thousands of tiny files or 26,000 distinct commit authors that force expensive maintenance.
AI-written summary · about
chapters (14)
- 0:00Intro
- 0:35Git Infrastructure Team
- 1:12Long tail of unusual interactions
- 2:23github/github
- 4:46kubernetes/kubernetes
- 7:37commitwars/commitwars
- 11:11whosonfirst-data/whosonfirst-data-venue-us-ny
- 12:09gambuzzi/gambuzzi github.io
- 13:50Cocoa Pods/Specs
- 19:586.5 hubblestack/hubblestack_data
- 22:31barryclark/jekyll now
- 30:07Very big networks
- 34:09enkidevs/comunit
- 37:10Conclusions
related talks
47mHow to Avoid Creating a GitHub Junkyard
01h 06mHow GitHub delivers GitHub using GitHub
43mMy First Year of Pull Requests
31mGitHub as a self-sustaining product for large-scale environments
17mNo More Walled Gardens: Removing Barriers with SCM...
26mBuilding an enterprise git repository (the hard bits!)
32mHacker Traction Through GitHub Actions: Is Your (Open Source) Project Safe?
54mAbusing GitHub for fun and profit: Actions and Codespaces Security
