Container clusters
The cluster register: what earns a row, why a pod never does, and where a worker node lives.
A row in the register is a cluster. A node pool is a count on the cluster. A pod is never a row, and there will be no pod column.
A single data center may run thousands of pods whose lifetime is measured in hours. That is telemetry, not an inventory, and there is no action in the product that can be taken against a pod — you cannot schedule it a window, you cannot assign it an owner, and you cannot ask a question about it that will still be true in the morning. Workloads is a count on the cluster, and that is the whole measure kept.
A worker node is a server, and it is in the server list. It is an operating system on a DNS name, patched on the distribution vendor's cycle, and the server inventory is already the authority for it. The cluster is the entity; its nodes are servers.
The columns
| Column | What it says |
|---|---|
| Cluster name | The cluster's identity, and its record key |
| Distribution | OpenShift · Amazon EKS · Azure AKS · Google GKE · Rancher · Kubernetes |
| Version | The cluster version |
| Support ends | The recorded date; turns red once it has passed |
| Status | The five status values shared across infrastructure |
| Location | A site name, or the name of the cloud account the cluster runs in |
| Nodes | How many nodes are in the cluster |
| Node pools | How many node pools |
| Namespaces | A count |
| Workloads | A count — the Deployment / StatefulSet grain, not the pod |
| vCPU utilization | A meter over vCPU used and total |
| Memory utilization | A meter over memory used and total |
| Owning team | Who is responsible for the cluster |
Memory utilization, Namespaces and Owning team are not in the default set, and neither are the four numeric halves the two meters are built from. All of them are added through the column picker, and all of them are in the export.
The distribution name is normalized, not translated
A source reports eks or openshift in lowercase, because that is what the cluster calls itself through its own interface. The screen shows the product's published name — Amazon EKS, OpenShift — and the same name in both languages. A Kubernetes distribution has one name worldwide, and inventing a Hebrew one would be a mistake, not a service.
A value the product does not recognize is shown as it stands. It is neither dropped nor guessed at.
A cluster appears as a node on the connection map. The edge from a node to it is Member of, and the edge from a business system running on it is Runs on.
Updated
This page is the file content/docs/en/v1/infra/containers.mdx