Quarterly is the right default for service and component dashboards. The ones leadership opens during a customer call deserve monthly; the ones built during an incident deserve a decision at incident close rather than a cadence. Set the tier on the dashboard, derive the cadence from the tier, and let this checklist be the pass you run when the cadence comes due.
The scope is a set of dashboards, not one. Run the per-dashboard commands in a
loop over the search results for the tier under review; the UID=svc-overview
prefix on each command is where that loop variable goes. Substitute your own
Grafana address and a token with read access to search and dashboards.
What the review is looking for
Four outputs, and they are different kinds of finding. Broken panels are a correctness bug in something the team relies on. Ownership gaps are a process finding, and usually the reason the correctness bug survived. Metric drift is a bill arriving from a change somebody made elsewhere. Tier mismatch — a dashboard whose declared importance does not match how often anybody has actually looked at it — is the one that tells you which dashboards to delete.
Access this needs
A Grafana service-account token with read access, and read access to whichever data sources the panels query so the metric-existence checks can run. The comparison against the provisioning repository needs a checkout of it. Nothing here writes; archiving and deleting are the follow-up change, made deliberately and reviewed like any other.
Sign-off
- Reviewer: ________________ Date: ___________
- Dashboard owner: __________ Date: ___________
- On-call representative: ____ Date: ___________