Growing a Lab for Automated Upstream Testing: Challenges and Lessons Learned
summary
Collabora runs a LAVA lab of about 158 devices (now growing past 170) across ~50 racks serving KernelCI and MesaCI with thousands of jobs daily and ~100k jobs/month. The stack uses LAVA, lavacli and a lava-gate GitLab runner plus tftp and Chromebook tools (vcq cable + cr50) to allow remote power, serial access and arbitrary kernel+DTB boots. New device onboarding runs 1,000 health checks per device, staging vs production separation, automated infra-error detection, DB pagination optimizations and a DB-load generator; main failures are cables, batteries, network and heat.
AI-written summary · about
chapters (5)
- 0:00Start
- 8:36LAVA test job example (1/4)
- 14:22Chromebook Bring-Up in LAVA
- 27:39Scaling Challenges
- 28:42Performance Improvements
related talks
40mIntroduction to LAVA, the Linaro Automated Validation Architecture
22mLAVA community enabled testing
35mManaging CI Across Multiple Farms for GPU Testing
26mLAVA + OpenQA = Automated, Continuous Full System Testing
07mOniro CI/Testing integration with LAVA
45mUpdate on LAVA Testing for bare metal systems
34mFrom Zero to First Test in Your Own LAVA Laboratory in less than 45 minutes)
41mBeyond unit tests