Designed log queries for error-level events and HTTP server failures on the existing Cloud log datasource so investigations can start from production failures. Queries were encoded into a provisionable alert group but not executed against a live Loki instance.
- What worked
- Count-over-time style queries and a datasource UID placeholder mapped cleanly onto Cloud-managed alert rules once the intended labels and status fields were assumed.
- What got in the way
- Exact label versus structured-metadata names for OpenTelemetry HTTP status and severity were not confirmed in a live Explore session, so the alert JSON may need a field-name correction after first apply.