Comparison
Service level indicatorvsService level objective
Service level indicator
you pick the one number that actually represents whether users are having a good time, like the fraction of requests under 300ms.
A measured quantity representing service quality, usually a ratio of good events to total events. The choice is the hard part: it must correlate with user experience, not with system internals. CPU utilisation is not an SLI; successful checkout rate is.
Full entry →Service level objective
you commit internally to 99.9 percent of requests succeeding over 30 days, and you now have a number that says when to stop shipping features.
A target for an SLI over a time window. Its real function is decision-making: it converts "is reliability good enough" from an argument into arithmetic. An SLO nobody is willing to act on — by slowing releases or prioritising reliability work — is a dashboard, not an objective.
Full entry →