Container Stats
Live CPU, memory, and disk I/O charts for any pod or node — with Prometheus-backed history when available.
Open stats
Right-click a pod or node → Stats. A charts window opens.
For pods: one line per container (multiple charts if the pod has several containers).
For nodes: node-wide aggregates.
Charts
Each metric renders as a live line chart (via egui_plot):
| Metric | Source |
|---|---|
| CPU (cores) | metrics.k8s.io (metrics-server) or Prometheus |
| Memory (bytes) | metrics.k8s.io or Prometheus |
| Disk read / write (bytes/s) | Prometheus only |
| Network rx / tx (bytes/s) | Prometheus only (when present) |
Live sampling
When Prometheus is not available, Kubezilla polls metrics.k8s.io every few seconds and builds up the series in memory. Close the Stats window and the live series resets.
Prometheus integration
Kubezilla auto-discovers Prometheus in your cluster by probing a short list of common Service + namespace combinations:
| Namespace | Service | Port |
|---|---|---|
monitoring | prometheus-k8s | 9090 |
monitoring | prometheus-operated | 9090 |
monitoring | prometheus | 9090 |
prometheus | prometheus-server | 80 |
prometheus | prometheus-kube-prometheus-prometheus | 9090 |
If your Prometheus lives at a different Service/namespace, auto-discovery won't find it.
When discovered:
- Historical data back-fills the chart (range depends on your Prometheus retention).
- You can pick a time range — the chart re-queries on change.
- Instance metrics include disk and network I/O (where cAdvisor and node-exporter emit them).
If auto-discovery picks the wrong Prometheus (or fails), you can provide a manual URL in Settings (TBD).
Chart controls
- Zoom — scroll to zoom time axis.
- Pan — drag to pan.
- Double-click — reset to full range.
- Hover — shows the exact timestamped value.
Tips
- No chart showing? —
metrics-serveris not installed, and Prometheus wasn't auto-discovered. Install one of the two. - Chart looks flat at zero — the pod/container is idle or metrics-server is stale (its pods crash surprisingly often). Check
kubectl top pod. - I want longer history — configure Prometheus retention, or use Thanos/Mimir and point Kubezilla at the federated endpoint.
- Stats + Logs side-by-side is a strong debugging combo. Open Stats, then right-click → Logs on the same pod — the two windows run independently.