Experiments for SRE
summary
Experiments use feature flags and staged ramps (example ramps: 1%, 10%, 50%) to limit blast radius, attribute outages to a treatment experiment id, and roll the experiment back by ramping to zero percent instead of rebuilding binaries. The workflow includes automated health checks at each stage, codable ramp policies, tradeoffs between full confidence interval analysis and faster signals for SRE action, and onboarding tools like command line interfaces and templates to reduce configuration mistakes and lower MTTR.
AI-written summary · about
chapters (9)
- 0:00Introduction
- 0:50What are experiments
- 5:03How can SRE use experiments
- 11:02Rolling back
- 14:07Promoting reliability
- 14:24Experiments in stages
- 17:50Experiments in lifecycle
- 20:35Experiments as canonical tool
- 27:26Conclusion
