Finding problems quickly in a quickly evolving opensource project
summary
A zero‑day automated system monitors 400 git trees hourly, pulls new commits, compiles across configs and architectures in VMs/qemu/KVM, runs ~5,000 tests/day, records vmstat/profiles with a “flight recorder,” and auto‑bisects regressions. The build robot emails failures to patch authors and maintainers; the pipeline flags ~600 build bugs and ~30 boot bugs per month and ~10 performance regressions, and it reduced build failures from 6.5% to 0.1% by catching issues right after merge.
AI-written summary · about
chapters (11)
- 0:00Start
- 1:14Overworked Maintainer
- 2:32The Linux Kernel
- 10:41The Development Cycles of the Linux Kernels
- 22:46Goals
- 27:42Error Logs
- 28:04Build Error Report
- 28:45The Build Error
- 32:30Boot Error Examples
- 35:14Performance Regression
- 40:08Boot Time
related talks
53mWhere The Wild Things Are: The State Of Open Source Supply Chain Risk Management In Three Stories
47mUnder the Radar: How we found 0-days in the Build Pipeline of OSS Packages
16mPerformance improvements: exposing hidden security fixes in Open Source
37mHow to Serve Open Source Maintainers Without Annoying Them?
17mBuilding Secure Open Source Communities From the Ground Up
52mScaling the Security Researcher to Eliminate OSS Vulnerabilities Once and for...
30mBringing Continuous Delivery to Open Source
52mCollaborating to Improve Open Source Security: How the Ecosystem Is Stepping Up