Skip to main content

Overview and Nodes

Both screens are for admins only and read only. They show the cluster selected in the header.

Overview​

You want to: see in ten seconds whether the cluster is healthy. Admins land here after sign-in; g o from anywhere.

The Overview of prod-eu: pods running, pods needing attention, restarts, deployments ready, CPU and memory with 24 h trends, cluster usage chart, the pods needing attention, most restarts, restarts over time and busiest namespacesThe Overview of prod-eu: pods running, pods needing attention, restarts, deployments ready, CPU and memory with 24 h trends, cluster usage chart, the pods needing attention, most restarts, restarts over time and busiest namespaces
  • Tiles: pods running, pods needing attention (failing / warning), restarts in 24 hours, Deployments ready, CPU and memory as % of allocatable, each with a 24-hour trend.
  • Cluster usage, 24 hours against allocatable.
  • Pods needing attention with their status and last exit; Most restarts with the usual exit code; Restarts over time; Busiest namespaces (pods closest to their limits).
  • Warnings from the last 15 minutes and a node strip; a node opens on the Nodes screen.
  • Live events opens for the whole cluster, on Warnings.

Nodes​

You want to: know which node is full, cordoned or under pressure, and what runs on it. g n, or the rail.

The Nodes screen: one node, Ready, control-plane, kubelet version, OS / architecture, age, pods against capacity, CPU and memory used, requested and limited as % of allocatableThe Nodes screen: one node, Ready, control-plane, kubelet version, OS / architecture, age, pods against capacity, CPU and memory used, requested and limited as % of allocatable

Status (Ready / NotReady / pressure, cordoned), roles, kubelet version, OS / architecture, age, pods against capacity, and CPU / memory used, requested and limited as % of allocatable. Search by name, role or label (key=value). Click a node for its drawer:

A node drawer: summary (OS image, kernel, container runtime, IPs, scheduling), capacity / allocatable / requests / limits / used for CPU, memory, pods and ephemeral storage, and the usage history chartA node drawer: summary (OS image, kernel, container runtime, IPs, scheduling), capacity / allocatable / requests / limits / used for CPU, memory, pods and ephemeral storage, and the usage history chart

Summary (OS image, kernel, runtime, IPs, cordoned), capacity / allocatable / requests / limits / used for CPU, memory, pods and ephemeral storage, usage history (1 h / 6 h / 24 h, 7 days with Prometheus), its busiest pods, every pod on it, conditions, taints, labels, annotations and events.

Where the numbers come from​

DataSourceNeeds in the ClusterRole
pods, workloads, eventsthe informer cachethe read rules of rbac.yaml
node list, capacity, conditionsthe informer cachenodes: get / list / watch
current CPU / memorymetrics-server, sampled every 30 smetrics.k8s.io pods and nodes: get / list
usage historyOpenConsole's memory (24 h, reset on restart), or Prometheus (up to 7 days)for Prometheus "in the cluster": services/proxy on its Service

Without the nodes rules the Overview still works with absolute CPU / memory and no node strip, and Nodes says the connection can't list nodes. Opening the Overview is audited (cluster.view_overview), the node screens as node.list / node.get; their own refreshes are not.