Dashboards
Build shared views of service resources, metrics, and current operations.
Dashboards combine resource health, charts, capacity, and operation summaries for a project or organization. They reference live resources rather than copying their configuration.
Create a dashboard
Open Observe → Dashboards → Create dashboard. Choose its scope and owner, then add panels. A project dashboard cannot request organization fleet data unless its viewers also have permission to read that data.
Begin with the service question the dashboard must answer. For deployment readiness, combine backend health, initialization, and recent operations. For storage response, combine replica state, latency, repair backlog, and backup age.
Add a metric chart only when its source and unit answer that question. Resource health counts are current observations; journal panels show events; operation panels show accepted work. Arrange them so an operator can move from a symptom to the resource or execution record responsible for it.
A panel selector can use resource IDs or labels. Fixed IDs suit a specific resource investigation. Labels suit a service whose droplets are replaced during deployment. Preview the selector after changing ownership or environment labels.
Time and comparison
A dashboard time range applies to historical panels. Current-health and active operation panels show their present state and are labeled accordingly. Do not interpret a current status count as the status at the beginning of the chart range.
Keep compared charts on matching units and aggregations. A maximum node memory value and an average guest memory value answer different questions. Use Metrics to interpret the sources.
Sharing
Dashboards inherit their scope's read permissions. Sharing a dashboard link does not grant access to its resources. Restricted panels show unavailable data rather than exposing values through a shared layout.
Editors can change panel definitions and filters. Viewers can adjust their own time range without saving a new shared configuration. Organization-wide edits should use a named owner so changes remain attributable.
Deleted resources and stale data
A panel referencing a deleted ID shows a missing-resource state. A selector that matches no resources shows an empty result. Neither is interpreted as a healthy resource count of zero.
An unreachable source retains its latest timestamp and stale indicator. A refresh updates the query; it cannot create a new observation from an offline node.
Operating views
For a service, combine load balancer health, droplet metrics, volume latency, backup age, and recent operations. For a region, combine allocatable capacity, zone distribution, storage repair backlog, mesh health, and node maintenance.
Export the selected measurements and journal records when an incident requires a durable report. A dashboard link remains a live view and its data can expire under retention policy.