Give each query-based health signal one deliberate numeric result

Source facts

The Azure Monitor health-model preview requires a KQL log signal to return one numeric value; its PromQL signal has the same single-value requirement. The health-model identity needs permission to read the target telemetry, and the entity must map to a real data source. The configuration then assigns degraded and unhealthy thresholds to the signal. A saved signal definition can be reused for other entities within the same model. Microsoft Learn.

Applicability

Use this check when converting an investigative query into an approved preview health signal. First identify the exact entity and observation the number should describe. A useful multirow investigation result is not yet the required scalar signal.

DSE recommendation

Define the entity, time scope and numerical meaning of the signal before reducing its query to one value. Ask the service owner to explain what a threshold crossing means operationally. Keep the selected population visible in the query review so aggregation does not inadvertently combine unrelated entities. Where a definition will be reused, inspect each intended assignment rather than assuming the original query context fits every target.

Verification

Run the proposed query under the intended reading identity and inspect the actual result shape and value. Compare it with a separately understood sample of the underlying telemetry. Exercise representative normal and adverse observations in a controlled test, then inspect Graph and Timeline transitions against the chosen thresholds. Preserve the query, assignment and observed value together; successful saving alone does not establish that the number represents the intended entity.

Official references

Microsoft Learn: Configure health-model signals preview. Source reviewed September 9, 2026.