Create an alert definition when a trace-derived metric has a stable baseline and a threshold breach requires action. Common candidates are service latency, error rate, and throughput.

Define the condition

1. In the trace service view, select the service, operation, and representative time range.
2. Identify the metric and normal operating range.
3. Open the alert-definition workflow available for metrics in your deployment.
4. Select the corresponding metric and scope it to the intended client, service, and operation labels.
5. Set the threshold, evaluation duration, severity, and recovery behavior.
6. Attach the appropriate notification or escalation policy.
7. Save the definition and verify it with controlled traffic or a known historical interval.

Avoid noisy definitions

Use a sustained evaluation window instead of alerting on a single point. Add a minimum-traffic condition when rates are unstable at low volume. Keep label filters bounded; highly specific or ephemeral labels can create many independent alert series.
Document the dashboard, trace search, and response procedure that responders should use. If the required trace metric does not appear in the metric selector, confirm that span-metric generation is enabled and that the service has reported during the selected window.