Qemu Performance Regression CI
summary
A CI pipeline runs QEMU system-level benchmarks with runperf driving tests and pbench producing machine-readable result.json files, stores about 2 GB per build for 14 days, and uses a Jenkins-based dashboard with HTML reports. Full profiles take ~12 hours and three samples to be stable; the workflow does automated bisection between commits to pinpoint regressions or improvements and delivers VM XML, metadata and logs to reproduce failures locally or on specified hosts.
AI-written summary · about
chapters (10)
- 0:00Intro
- 0:16Email: Proposal for a regular upstream performance testing
- 2:28Developer sends a patch
- 2:40Our pipeline detects a regression
- 2:51We trigger a bisection job
- 3:02Bisection pinpoints a comm
- 9:41First-time setup - a worker
- 14:00Additional data on demand
- 15:00Current coverage
- 16:34Plans
related talks
29mQEMU & KVM Automated Performance Benchmarking @ SUSE
23mStop Performance Regressions Before Release Valkey’s CI Benchmarking
35mBisection: It's Not Just for Git
41mReducing Friction in Testing Using QEMU and Labgrid for Yocto-based Products
47mWhat I Learned Building a Mobile Testing Infrastructure
24mSupercharging CI: Test Suite Optimization for Speed, Stability, and Savings
45mPerformance lab. The power of the perf tool!
45mHow to Build a Performance Testing Pipeline from Scratch