Choose whether an Azure Monitor SLI counts requests or time windows

Source facts

Azure Monitor request-based SLIs compare good requests with total requests. Window-based SLIs instead compare intervals that meet a quality threshold with all intervals in the compliance period. Microsoft describes request-based evaluation as reflecting per-request success despite uneven traffic, while the window model can smooth brief poor-performance bursts. Setup requires a service group; the prerequisite links to preview service-group creation. Microsoft Learn.

Applicability

Use this decision when defining what a reliability percentage should represent. The target and time horizon still need an explicit service-level decision; choosing the same percentage does not make two different evaluation populations equivalent.

DSE recommendation

Choose the evaluation population before setting the target percentage: individual requests or qualifying time intervals. Ask the service owner how a short, heavily loaded incident should influence the measure compared with a quiet interval. Document the intended numerator and denominator for request evaluation, or the qualifying threshold for windows. Keep the rationale beside the SLI so later reviewers do not interpret a window percentage as a request-success percentage.

Verification

Use representative historical telemetry containing both busy and quiet intervals. Compare the two proposed interpretations and check which answers the approved service question. Inspect the preview for the selected signals, filters and formulas before creating the SLI. Retain the chosen method and the comparison evidence; a target being met under one method should not be reported as a successful result under the other or as proof of every user’s experience.

Official references

Microsoft Learn: Service-level indicators. Source reviewed September 9, 2026.