« Back to Index

[SLI, SLO, SLA]

View original Gist on GitHub

Tags: #SLI #SLO #SLA #Process #Service

SLI, SLO, SLA.md

SLI, SLO, SLA ?

When building a service, we have a responsibility to define some baseline agreements as to the service’s expected uptime and performance. This document focuses on the various terminology that we use to define these values.

• Service Level Indicator (SLI)
  What the service owner has chosen to measure progress towards their goal.
  e.g. What is good/bad service for your users.

• Service Level Objective (SLO)
  What the service owner’s goal is for the given indicator.
  e.g. A percentage that tells how many SLI errors your service can have in a specific period of time.

• Service Level Agreement (SLA)
  What the service owner is promising its users (typically the SLO + wiggle room).

Example

• Indicator: request latency.
• Objective: 99.5% of requests will be completed in 5ms.
• Agreement: 99% of requests complete in 5ms or you get a refund.

Below is an alternative example where we state (via a Datadog graph) that if the average request latency over a 1hr period exceeds one second, then we have a ‘service’ issue and it’ll display with a red background. This will signify that we’ve failed our SLA which is defined as being less than one second.

If the average request latency over a 1hr period is greater than half a second, then we have a ‘team’ issue and it’ll display with an orange background. This will signify that we’ve failed our SLO which is defined as being less than half a second.

If the average request latency over a 1hr period is less than half a second, then we have no issues and it’ll display with a green background. This will signify that we’ve reached our SLO which is defined as being less than half a second.