Skip to main content
Live usage comes from the Kubernetes metrics API, the same place kubectl top reads from. It’s served by metrics-server, which most managed clusters run already. KubeStacks asks for it every 5 seconds while its window is open, and needs nothing else installed.
A cluster's overview: nodes, pods and workloads, CPU and memory with their last hour, and what needs attention.A cluster's overview: nodes, pods and workloads, CPU and memory with their last hour, and what needs attention.

The overview's CPU and memory cards: what's in use, against requests, limits and allocatable.

​
Where you see it

The overview's CPU and Memory cards

Each card leads with how much of the cluster is in use, as a percentage and as “12.2 cores of 27.5 cores in use”. The meter below it marks where requests reach, and the row underneath lists:Pods that have finished (Succeeded or Failed) don’t count, since they no longer hold anything.The small chart beside the number is a trend. With usage history, it covers the last hour, and History → opens the Metrics page. Without it, it covers the time Since you opened it.
The overview’s Nodes card has a CPU and a memory meter for each node, against what that node can allocate. Top CPU and Top memory list the five pods using the most.The Nodes list has the same meters in its CPU and Memory columns, and you can sort by them.
The Pods list has CPU and Memory columns, sortable, so the hungriest pods are a click away. In Workloads, each workload’s usage is added up over its pods. A pod’s detail panel shows each container’s usage next to its state.

​
Without metrics-server

Everything else keeps working. Where usage would be, KubeStacks shows what it can work out from the cluster itself:
  • The overview’s cards show what’s requested instead of what’s used (“6.8 cores of 27.5 cores requested”), with the note Live usage needs metrics-server.
  • The Nodes list shows each node’s allocatable CPU and memory instead of a meter.
  • Pods’ CPU and Memory columns stay empty, and Top CPU and Top memory don’t appear.
A node that reports nothing while metrics-server runs (one that’s NotReady, say) shows No metrics.

​
Install metrics-server

If your cluster doesn’t have it, metrics-server’s own chart or manifest installs it in a minute. It goes in kube-system.
On a local cluster like kind, kubelets serve self-signed certificates, and metrics-server needs --kubelet-insecure-tls to read them. With the chart, pass --set 'args={--kubelet-insecure-tls}'. Don’t use it in clusters that matter.
Usage shows up in KubeStacks within a minute or so, once metrics-server has its first samples. You don’t need to restart anything.

​
Live usage and history

Live usage is a snapshot, a few seconds old. To see how usage changed over the last hour, day or week, and compare namespaces, workloads, pods or nodes, KubeStacks charts history from a Prometheus or VictoriaMetrics in the cluster.

Usage history

The Metrics page, and a Metrics tab on pods, workloads and nodes.

The overview

Everything else on a cluster’s first page.