SLI (Service Level Indicator)
Also known as: service level indicator
An SLI is a quantitative measure of one aspect of service behaviour, expressed as the ratio of good events to valid events — successful requests over total requests, say. It is the measurement an SLO sets a target for, and it must reflect what users experience, not what is easiest to instrument.
Last reviewed · Part of the Architecture Glossary
In practice
A well-formed SLI is a ratio with both terms defined precisely:
availability = count(status < 500 AND status != 429_from_us) / count(valid requests)
latency = count(requests served < 300 ms) / count(valid requests)Note the second form. Latency as an SLI is a ratio of fast requests, not a percentile value — that makes it composable with an error budget in the same units as availability.
The definitions that decide whether the number is honest:
- Valid events. Client-side 4xx from malformed input are usually excluded; your own 429s are usually not.
- Measurement point. Load balancer logs capture what the user got, including your gateway's failures. Application metrics do not — a service that never received the request records nothing.
- Aggregation window. A 28-day rolling window is the common choice; calendar months create a budget cliff on the 1st.
When it matters
Before any SLO conversation. An SLO on a badly chosen SLI is a target on the wrong thing, precisely enforced.
Common mistake
Picking uptime of the process as the availability SLI. A pod that is running and returning 500s is 100% "up" and 0% useful. Measure the response, not the process.
See also
- SLO (Service Level Objective)An SLO is a target value for an SLI over a window — for example, 99.9% of requests succeed over 28 days.
- SLA (Service Level Agreement)An SLA is a contract with a customer that specifies a service level and the consequence of missing it — usually a service credit.
- Error BudgetAn error budget is the amount of unreliability an SLO permits — the complement of the target.
- p99 (99th Percentile)p99 is the latency value below which 99% of requests complete — one request in a hundred is slower.