How to answer tricky questions the SRE way
summary
Defines SLI, SLO and error budgets and argues for choosing reliability targets that save cost and speed feature delivery. Recommends using a 99.95% SLO example, stopping launches when the error budget is exhausted, and alerting on error budget burn rate rather than CPU or disk noise. Cites Google Chubby as a cautionary case of hidden hard dependencies and shows a concrete performance mismatch: halving pipeline latency (1s to 0.5s) is pointless when a third party adds a 2 minute delay.
AI-written summary · about
related talks
38mImplementing SRE practices: SLI/SLO deep dive
27mGetting Started with SRE
30mImproving Reliability w/ Error Budgets & SRE
31mSite Reliability Engineering (SRE) and...
30mTechniques for SLOs and Error Budgets at Scale
30mMetrics for good, not evil
21mSRE, what is it all about?!
29mMeasuring Reliability: What Got Us Here Won't Get Us There