Monitor
The alerts that fire now, and the CPU, memory, disks and network of one monitored host, from Prometheus.
How to read it
- The first line says when Prometheus answered last: "from Prometheus, … ago, every 60 s".
- Alerts firing: one badge for each alert. Red is
severity=critical. Yellow is every other severity. Critical alerts come first. A greennonemeans that nothing fires. - CPU and Memory available: the last hour, one point each minute.
- Disks: yellow at 80 % used, red at 90 %.
- Network: rates only. Virtual interfaces (veth, cni, flannel, docker and others) are left out.
Turn it on
Set dash.monitor_context (the kubectl context), dash.kubeconfig if necessary, and
dash.profile.monitor_host (the host label of the node_exporter series). A node with this
configuration can also send the data. Only admins see this panel.
When it looks wrong
| You see | It means | Do this |
|---|---|---|
| "Monitor Prometheus not reachable yet." | No answer from Prometheus yet, or the last call failed with no data | Run the same kubectl --context <ctx> get --raw on the server's box. Check the context and the service name. |
| An error with kubectl text | kubectl failed. The text is the first 300 bytes of its error. | Fix the kubeconfig or the context |
| "prometheus error" | Prometheus refused the query | Check the Prometheus version and the metric names |
| All graphs are flat or empty | monitor_host matches no host label |
Set monitor_host to the label value in Prometheus |
| An alert is yellow, but it is important | Only critical is red |
Set the alert's severity in its rule |
This panel does not read Alertmanager. Code: fleet/internal/dash/monitor.go.