Definition of the SLO SLI SLA triad
The SLO SLI SLA triad brings together three notions that are defined in terms of one another. The “Service Level Objectives” chapter of the book “Site Reliability Engineering”, published online by Google (sre.google/sre-book), sets them out as follows:
- the SLI (service level indicator): a quantitative measure of some aspect of the level of service provided;
- the SLO (service level objective): a target value, or range of values, for a service level measured by an SLI;
- the SLA (service level agreement): an explicit or implicit contract with users that includes consequences of meeting, or missing, the SLOs it contains.
The same chapter notes that, in everyday usage, the word SLA covers several meanings depending on the context. Its test: what happens if the objective is missed? Without an explicit consequence, it is almost certainly an SLO.
From measurement to contract: how the three notions fit together
| SLI | SLO | SLA | |
|---|---|---|---|
| Nature | A measure | A target set on that measure | A contract |
| Expressed as | A measured value, for example a ratio | A threshold and a time window | Objectives and their consequences |
| When service degrades | A finding | A decision by the team, with no contractual effect | The consequence written into the contract, for example a credit |
The “Site Reliability Workbook”, a second book published on the same site, recommends giving the SLI the form of a ratio: good events divided by total events. For latency, the first book prefers percentiles to the average; our article on load testing explains why.
The SLO adds a threshold and a measurement window, rolling or calendar, to the indicator. The first book suggests keeping a safety margin: an internal SLO stricter than the level advertised to users.
What a written objective makes it possible to decide
The chapter describes a loop: measure the SLIs, compare them with the SLOs, decide whether action is needed, then act. Without an objective, it says, one would know neither whether to intervene nor when. A published SLO also sets users’ expectations.
The gap between 100% and the SLO forms the error budget, which is used to arbitrate between changes and stability.
The limitations of the exercise
An SLI is sometimes only an approximation: the book cites latency, measured at the server because it cannot be measured at the user’s end. It advises choosing few objectives, not basing the target on current performance and being willing to revise it.
SLO, SLA, RTO, monitoring threshold: not to be confused
The service level agreement also features among the notions of the ITIL framework. The SLO remains a management tool: as long as no contract takes it up, missing it has no agreed consequence.
An availability SLO adds up the failures over a period; an RTO caps the duration of an interruption after a disaster, as our RTO RPO entry explains.
Nor is an SLI a monitoring threshold: the book calls for starting from what matters to users. Our SRE entry presents the discipline that uses these three notions.
Example: bookings on a carpooling app
Suppose a carpooling app. Its team chooses as its SLI the share of booking requests that succeed, measured over a rolling four weeks. Over the last window, 1,200,000 requests were received and 3,600 failed: the SLI is 1,196,400 out of 1,200,000, or 99.7%.
The SLO is set at 99.5%, a figure taken for the example and not a recommended value: it is met. The contract with business customers guarantees 99% and provides for a credit below that. This is the SLA, set below the SLO to leave a margin.
Choosing the journeys to measure, the indicators and the objectives with product owners: this is the first strand of our SRE and observability page.