Service level indicator

Measure a behavior that matters

An SLI turns a service question into a measurement. Can users complete a request? How long does it take? How often does the service return the correct result?

For an illustrative search API, an availability SLI might be the fraction of well-formed requests that succeed. A latency SLI might use the time observed by the client, which can include delays that server-side timing misses.

SLI, SLO, and SLA

An SLI is the measurement. A service level objective, or SLO, is the target for that measurement. A service level agreement, or SLA, is an agreement that may attach consequences to missing a target.

The terms are related but interchangeable wording can cause confusion. A dashboard showing an error rate is not itself an objective.

Choose a small useful set

Collecting every available metric does not create a better SLI. Select a few indicators that represent the user journeys and system behaviors the team needs to manage. Document the aggregation window, population, exclusions, and source.

Inspect distributions where averages hide tail behavior. A service can have a good mean latency while a meaningful group of users experiences very slow responses. An SLI should support action when it changes, so connect it to a response and an owner.