Inline metrics and rightsizing
When a Prometheus is reachable, resource detail pages gain live charts and recommendations. Radar auto-discovers Prometheus on first launch and silently reconnects across restarts; if none is found, a “Discover Prometheus” action appears. Every panel below degrades gracefully - each hides silently when its data source isn’t present, so you never see empty “no data” panels.- CPU and memory charts with request/limit overlay. Dashed reference lines mark the workload’s summed requests and limits (runtime containers x ready replicas); the Y axis auto-extends so a hot workload’s usage doesn’t clip the limit line. Each panel header shows a saturation chip (e.g.
12% of limit,2.6% of request), tone-ramped so a low-utilization workload still reads as healthy at a glance. - Rightsizing strip (Deployment / StatefulSet / DaemonSet, full-screen Metrics tab). Per-container CPU/memory request recommendations from P95-over-24h. The tone is deliberately mild - most workloads are 2-3x over-provisioned and that’s fine - and a compact health badge (“CPU throttling” / “OOM risk” / “Needs review”) summarizes the worst case, hidden in the steady state to avoid badge-blindness.
- HPA replicas chart - current vs desired replicas over time, with min/max reference lines.
- PVC usage gauge - used vs capacity from kubelet volume stats, with traffic-light tone at 75% / 90%. Hides when the CSI driver doesn’t report volume stats or kubelet isn’t scraped (e.g. GMP defaults).
- Restart event lane - restart spikes as vertical markers on a dedicated row beneath the chart, over the chart’s selected time range.
- Multi-chart grid - the full-screen Metrics tab lays CPU, Memory, Net RX, and Net TX side-by-side (Disk I/O collapsible) on one shared time range; the drawer keeps the single-chart tabbed view.
Rightsizing, the HPA chart, and the restart lane rely on kube-state-metrics. If KSM isn’t being scraped (some Datadog Agent / managed-Prometheus defaults), those panels hide while CPU/memory charts continue to work.
What Radar Shows
ServiceMonitor Detail View:- Status conditions
- Job label and scrape endpoint configuration (port, path, interval, scheme)
- Service selector (matchLabels)
- Namespace selector scope
- Rule group breakdown with per-group rule counts
- Alert rules vs recording rules summary
- Group evaluation intervals
- Pod metrics endpoint configuration (port, path, interval, scheme)
- Pod selector (matchLabels)
- Namespace selector scope
Supported CRDs
| CRD | Group | Topology | Detail View | AI Summary |
|---|---|---|---|---|
| ServiceMonitor | monitoring.coreos.com/v1 | - | Yes | - |
| PodMonitor | monitoring.coreos.com/v1 | - | Yes | - |
| PrometheusRule | monitoring.coreos.com/v1 | - | Yes | - |
| Alertmanager | monitoring.coreos.com/v1 | - | Generic | - |