Thinking like a site reliability engineer to improve CI
summary
OpenShift CI applies SRE practices to a billion test runs per quarter, using a diff SLO and a bespoke bisect tool that diffs 100+ repos plus the OS to catch regressions within hours. Statistical alerts use the Fisher exact test to flag drops (example: 99.98% confidence that a component fell from 99.9% to 94.9%), triggering code reverts or bugs; feature rollout uses feature gates to collect integration data; oncall shifts target under 50% toil.
AI-written summary · about
chapters (13)
- 0:00Introduction
- 3:34What is SRE
- 7:27SRE for CI
- 10:02From angry messages to trust
- 16:22Bisecting CI problems
- 18:09Test trends with stats
- 22:50Reducing CI toil
- 24:55Speed versus slowdowns
- 26:59Feature gates for innovation
- 29:09SRE’s promise
- 30:12Q&A on CI data
- 33:07Team structure for on-call
- 34:11Pre-merge vs regression testing
related talks
27mApplying SRE Principles to CI/CD
26mIncreasing the collaboration between SREs and Developers
07mIGNITE: Metrics, Margins, Mutiny – How to make your SREs (not) run away
35mOpenShift CI: What if we stopped retesting everything all the time?
24mHow to Destroy an SRE
30mDifferences in SRE Implementations across Companies
26mTechnology and Cultural Transformation using Site Reliability Engineering
42mSite Reliability Engineering as Systems Thinking