slo-sli-definition

Start by choosing a target like 99.9% availability, meaning the service succeeds 99.9% of the time. Calculate the SLI as successful requests divided by total requests, then derive the monthly error budget from the gap between 100% and your target.

The SLO is the promised target, such as 99.9% availability, while the SLI is the actual measured value, such as 99.95%. The SLI tells you whether you are meeting the SLO commitment.

Use histogram_quantile with the rate of your request duration bucket metric, for example histogram_quantile(0.95, rate(http_request_duration_seconds_bucket[1h])). This returns the p95 latency to compare against your target.

A 99.9% SLO allows 0.1% of the month as downtime, which equals about 2,592 seconds or 43 minutes. Higher targets shrink this: 99.99% allows roughly 4 minutes per month.

Internal tools typically use 99.0%, SaaS platforms 99.9%, financial systems 99.99%, and safety-critical infrastructure 99.999%. Choose based on user impact and the infrastructure cost of higher redundancy.

When the budget drops below roughly 25%, stop deployments, postpone experiments, and focus on stability work. This prevents exhausting the allowance and breaching the SLO before the month ends.