Skip to main content

Reading a cluster health card

Every cluster you add gets a card on the dashboard, Available Kubernetes Clusters. The card sums up ROZOOM's health checks for that cluster: an overall status, the most important problem, five areas, and two scores. Checks start on their own once the cluster is added.

The dashboard opens in Compact view. Detailed is a longer card with the same results plus the refresh state, notes on stored credentials and score breakdowns. Switch with the Compact / Detailed toggle at the top of the dashboard.

Overall status​

BadgeWhat it means
OKAll checks passed.
WARNINGAt least one check warns, for example a node under one pressure condition, or pods that have restarted.
CRITICALAt least one check is critical, for example a node reported Critical, a node under more than one pressure condition, or a critical result from the API server, etcd, certificates, admission webhooks, pod issues or warning events.
UNKNOWNThere is no result to show, for example the check could not run.
PAUSEDThe linter is off for this cluster, see Pausing the linter.
UNREACHABLEROZOOM cannot reach the cluster. The alert below says Cluster unreachable and why.
AUTH EXPIREDDetailed view only: the cluster rejected the credentials. In Compact view such a cluster shows UNKNOWN.

Under the badge the card names the most important problem with its own badge, for example API latency is high with Crit. In Detailed view this is the Primary Alert box, with a sentence on what it means.

The five areas​

Below that, the card shows five areas, each with its own badge: OK, Warn, Crit, or - when there is no data. The worst check in an area decides its badge.

AreaWhat it coversClick opens
HealthAPI server, etcd, pods with problems, warning events. Shows node count, crashes and restarts.Overview
CapacityResource requests and limits, autoscaling (HPA), pod QoS classes. Shows pod and workload counts.Pods
HAPod disruption budgets, topology spread, readiness and liveness probes.Overview
SecurityPod Security admission, network policies, secrets hygiene, security hardening.Compliance Hub
UpgradesCertificates, API priority and fairness, admission webhooks.API Deprecation Scan

The two scores​

Next to the badge the card shows two numbers from 0 to 100: the Health score first, then the Cluster score. Hover a number to see its name.

  • Health score weighs the control plane (35 points), nodes (25), workloads (20), observability (10) and platform hygiene (10). 85 and above is Healthy, 65 and above Degraded, below that Unhealthy.
  • Cluster score adds reliability (up to 50) and security (up to 50). 85 and above is Healthy, 60 and above At Risk, below that Unsafe.

In Detailed view each score expands to the problems that pull it down and how many points fixing them gives back, for example Fixing these gives +26.

Refresh​

The refresh icon on the card (Refresh cluster health) runs all checks for that cluster now. Between refreshes, checks repeat on their own; the selector next to the score sets how often: every 1, 5, 10, 15 or 30 minutes, 5 by default.

When a cluster cannot be reached​

  • Cannot connect: the card shows UNREACHABLE and the alert Cluster unreachable with the reason, for example "No route to the cluster address. Check the server address in your kubeconfig, then your VPN or network." Below it are Last successful contact with the time the cluster last answered and a Retry button. Detailed view shows the version from the last successful contact, for example K8s version: Last known: v1.33.1, or K8s version: Not available if the app has never reached the cluster.
  • Credentials rejected: Detailed view shows AUTH EXPIRED and Credentials expired: the token or credentials in the kubeconfig were rejected by the API server. Compact view shows UNKNOWN.

In both cases go to Manage Kubernetes Clusters, check that the cluster is up and that its credentials still work, and connect it again if they changed.

On Manage Kubernetes Clusters the same cluster shows Unreachable on its button instead of Online, the line "unreachable, last contact" with the time, and Fleet Health counts it as offline. Clicking the button still switches the cluster offline.