Fleet Sentinel — the Fleet Overview
The landing page. One screen answering: is the fleet healthy right now, and which clusters need attention?
What You See​
- KPI cards — total/active clusters, Critical / Warning / Healthy counts, average health score
- Cluster table — every scored cluster, sorted worst-first, with its unified health score (0–100), status color, and key risk factors
- Critical clusters — the worst offenders are surfaced here directly (there is no separate Critical Clusters page — it was merged into this view)
How Scores Work (the short version)​
Each cluster gets a unified score refreshed every 15 minutes: 60% metrics over the last 7 days (15 dimensions: compaction, disk, data growth, query errors, timeouts, slow queries and internal errors, transaction failures and rejects, memory, FE heap, BE memory, FE journal latency, connections, file descriptors), 25% recent firing alerts, 15% customer-tier weighting (tier S and A problems count more). Healthy is 80+, Warning 50–79, Critical below 50.
The metrics come from hourly fleet metrics that refresh about every 6 hours, so a score reflects the cluster's last week up to a few hours ago, not this minute. If the red strip at the top of the app names the health scorer, the scores have stopped updating and show their last run.
Clicking a cluster row opens the score breakdown so you can see which dimensions are dragging it down.
What To Do From Here​
- A cluster is Critical or slipping → check AI Triage for its grouped issues, or ask AI Chat about it by name
- Want the "why" behind a score → open the row's dimension breakdown
- Suspect something fleet-wide (bad version, regional issue) → read the latest Fleet Patrol report
The scoring runs on a 15-minute cadence — a cluster that just recovered may stay red for up to 15 minutes. The Refresh action forces a recompute if you can't wait.