When to NOT Set SLOs: Lots of Strangers Are Running My Software!
summary
Argues against shipping fixed SLOs to customers and instead provides built-in blackbox SLIs (deploy-test and app-availability), guidance, and a CRE-style onboarding to help each ops team set their own thresholds. Demonstrates two real thresholds: 99.5% => ~3.5 hours downtime/month for a public PaaS and 90% => ~3 days/month for an internal dev platform, shows using a risk analysis matrix to compute error-budget impact, and says the vendor helps with observability but lets customers choose SLOs.
AI-written summary · about