Skip to main content
Edit this page

Analytics Pages

Four read-only pages for trends and patterns.

Alerts Pattern​

An embedded Looker dashboard over the raw alert stream — alert volumes by type, cluster, and time. Use it to answer "which alerts are noisiest?" and "when did this start?" without writing queries. (This is raw alert data; for actionable groupings, that's AI Triage.)

Issues Pattern (superadmin)​

An embedded Looker dashboard over the AI-grouped issues — issue volumes, severities, and lifecycle trends over time. The filters at the top (Date, Customer, Cluster Name, Issue Name, Severity) apply to every tile, and clicking a slice or bar (a severity, a cluster, a category, a day) filters the rest of the dashboard to it. Click it again, or remove its chip from the filter bar, to clear it.

Breaking Thresholds​

Clusters approaching their limits: compaction score backlogs, disk, memory, query and transaction errors, FE journal latency and QPS, each card listing the clusters over its threshold. This is the "who's going to break next week" page — worth a scan during capacity planning. The cards come from the latest health-score run (the same run as Fleet Sentinel), so they are as current as that run.

Usage Patterns​

Fleet utilization over the last 24 hours: the health distribution from the latest health-score run, and the ten busiest or most loaded clusters by compaction score, p99 query latency and queries per second. The metrics come from the hourly fleet metrics, which refresh about every 6 hours, so the newest few hours are not in the charts yet.

LLM Usage (superadmin)​

Where the AI budget goes: daily token/cost breakdowns by source (chat, triage grouping, investigations, patrol) and by model, with a monthly projection. If AI features feel slow or expensive, this page tells you which one and why.

note

The two Looker pages need a working Looker session — if an embed shows an error instead of a dashboard, see the Admin Guide's troubleshooting section (the usual cause is the Looker-to-database tunnel needing an IP refresh).