Dashboard and Metrics

Business analytics, system health, calendar ranges, how downloads are counted, and what to do about gaps.

Separate two data domains

/admin/overview displays business analytics and system health. Their administrator APIs are /api/v1/admin/dashboard/business and /api/v1/admin/dashboard/system. The dashboard is a read model, not the write authority for identity, files, quota, or objects.

Analytics Worker projects events into that model, so results may lag. Refreshing a chart doesn't prove background processing finished, and a temporarily missing curve doesn't prove business data was lost.

Calendar ranges

range accepts today, 7d, and 30d, interpreted as calendar dates in Asia/Shanghai, ending when the response is generated. today starts at local midnight, 7d includes today and the preceding 6 calendar dates, and 30d includes today and the preceding 29.

These aren't fixed rolling 24-, 168-, or 720-hour windows. Compare actual start/end times, timezone, and coverage—not just the selected label.

Two download measurements

A Successful Download is recorded only after Storage completes the full object-byte response. Issuing or consuming a Download Ticket isn't a complete successful download.

A share's downloadCount, however, increases at successful ticket issuance. A share allowance can decrease while successful downloads remain zero if transfer didn't finish. The two facts aren't necessarily inconsistent. See Sharing.

Storage and health

Committed Storage Bytes are charged by Core after version finalization. Physical Object Bytes measure stored objects, potentially including historical versions and temporarily unreconciled objects. Don't expect them always to match.

System Health uses available, partial, and unavailable for data-source availability. Fields retain null when only some sources can answer. Missing Prometheus data isn't evidence of zero CPU usage, a healthy service, or a resolved incident.

Freshness and coverage

Freshness describes the latest processed fact and its lag. Coverage describes when reliable observation began and where gaps remain. Fresh data can still have incomplete historical coverage; a long coverage period can still have current processing lag. Preserve null or incomplete status rather than filling charts with zeros.

Diagnose gaps or delays

Distinguish 401/403 permission failures, 503 service failures, and partial/null fields in a successful response. The operator can then inspect analytics-worker, database, Prometheus, and Core. A useful report includes range, time, availability, and a redacted error. Don't change file or quota facts based only on a chart.