An SLI (service level indicator) is the actual measurement behind a reliability target.
If an SLO says "the API should be available 99.95% of the time," the SLI is the number that tells you whether you're , and by how much. Without an SLI, an SLO is just a wish written on a whiteboard.
Most SLIs boil down to a ratio: good events divided by total events over a time window. For availability, that's successful requests over total requests. For latency, it might be requests completed under 200ms over total requests. The ratio produces a percentage, and that percentage is what you compare against the SLO threshold.
Common SLI types and what they measure#
The four SLIs you'll run into most are availability (did the request succeed?), latency (how fast?), throughput (how many requests per second?), and error rate (what fraction failed?). A single service usually tracks , not all four. An internal batch processor cares about availability and error rate. A user-facing API cares more about latency and availability. Picking the right SLI matters more than tracking every possible metric.
SLIs and monitoring#
Uptime monitoring generates one of the simplest SLIs: the percentage of checks that returned a healthy response over a given period. Uptime reports turn that raw data into the actual number you compare against your SLO. When the SLI dips below the target, the error budget , and that's when alerting should fire, not when a single check fails.
Related terms: SLO, SLA, uptime, observability