Dependency health versus user impact
A queue, a worker, and a website expose different failure surfaces. Queue depth can rise during normal throughput bursts; errors can increase while fallback preserves availability. Combining expected ownership, queue age, terminal outcomes, and public synthetic checks prevents misleading incident conclusions.
Existing telemetry policy
NutsNews already defines a signal budget and permitted label taxonomy. Importing dashboards does not authorize unlimited new telemetry. Existing policy defers some trace and profile export. The portal exposes trace availability honestly and should not infer completeness merely from the presence of a Tempo datasource.
Source ownership
The central portal reads existing metrics rather than taking over every NutsNews provisioning pipeline. This avoids two repositories fighting over the same Grafana or server resource. New ownership is explicit; migrations require a single writer and a documented handoff.
Separate authentication from telemetry
A 401 means the browser needs a new owner session. Treat it independently from a source query failure or an empty series. Google ID verification and authorization happen on the portal server; Grafana tokens never enter the browser.
Application versus host views
The application spans multiple machines. Separate host views prevent an average from obscuring a saturated machine, while the combined NutsNews library preserves cross-service investigation. These additions reuse existing Grafana collection and do not install another agent or restart production workers.