metricEngineering & architecture

Service-Level Indicator

A service-level indicator is a quantitative measure of service behavior that represents an aspect of reliability experienced by users.

Download Markdown (.md)

01What it means

A service-level indicator, or SLI, is a measured proportion or distribution that describes service performance from a user-relevant perspective. Common indicators cover successful requests, latency, freshness, durability, or correctness. An SLI needs a defined event population, a good-event rule, a measurement point, and a time window.[1][2]

User-relevant reliability measurement Define the population and the good-event rule before calculating the indicator
SLI = good eligible events ÷ all eligible events
Eligible events
Explicit population being measured
Good rule
Success, latency, freshness, correctness, or durability
Measurement point
Where the event is observed
Window
Period over which the proportion is calculated

The measurement point and time window determine whether the SLI represents actual user experience.

02Why it matters

Reliability discussions become vague when teams rely on impressions or infrastructure health alone. A well-designed SLI turns an important user experience into evidence that can support objectives, alerts, reviews, and trade-offs. It also exposes whether the telemetry actually observes what users depend on.

03When to use it

Define SLIs before setting reliability objectives or contractual commitments. Use them for critical journeys and service capabilities, especially where failure has several forms. Prefer a small set of meaningful indicators over a catalog that no one can interpret or act upon.

04How it works

  1. Identify the user or system behavior that must remain dependable.
  2. Choose eligible events, such as valid requests or scheduled data updates.
  3. Define which events count as good using explicit success, latency, or correctness criteria.
  4. Select a measurement point close enough to represent user experience.
  5. Calculate good events divided by eligible events, or use an appropriate distribution, over a stated window.

05A practical example

Example

For a search API, the team counts valid queries at the load balancer as eligible events. A query is good when it returns a non-error response within 800 milliseconds. The SLI is the percentage of eligible queries that meet both conditions during the rolling 28-day window.

06Where it breaks

An SLI is a model, not the experience itself. Server-side measurements may miss client failures, averages can hide tail latency, and exclusions can quietly make performance look better. Indicators also require stable instrumentation and periodic review as user journeys and architecture change.

Common mistakes

  • Measuring host health instead of a user-visible capability.
  • Leaving the eligible population or good-event threshold ambiguous.
  • Using averages that conceal severe experiences in the distribution.

07Sources

  1. Service Level ObjectivesGoogle Site Reliability Engineering · Foundational book chapterService Level Objectives
  2. Implementing SLOsGoogle Site Reliability Engineering · Institutional implementation guideImplementing SLOs
View saved topics

Start typing to search titles, definitions, categories, and frameworks.