A service-level indicator (SLI) is a quantitative measure of an aspect of a service’s health or performance. Latency, server error rate, throughput, and uptime can serve as SLIs. Teams agree on these measures to determine how well a service is performing and use them to guide monitoring and alerting. section 11.2 section 11.2.5
Separate the measurement from the policy built around it:
- An SLI measures an aspect of service health, such as request latency or server error rate. section 11.2.5
- A service-level objective (SLO) sets the target or threshold for an SLI. section 11.2.5
- A service-level agreement (SLA) formalizes commitments to customers about meeting SLI targets. section 11.2.5
- An error budget describes the server errors permitted during a period before users become dissatisfied; detailed treatment is deferred. section 11.2.5
Latency SLI and SLO
Measure endpoint latency as the SLI, then define an SLO requiring latency of 500 milliseconds or less for 99% of requests over seven days. The SLI is the observed measure; the SLO is the target applied to it. section 11.2.5
Choose SLIs from the customer’s perspective, then use them as the basis for monitoring and alerts. Evolve them as the API product and user experience change. Avoid collecting metrics without a clear strategy: infrastructure metrics describe resources such as CPU, memory, queues, and database connections, whereas SLIs describe individual services. section 11.2 section 11.2.5
