​
Overview

Kubernetes clusters run containerised workloads on managed control-plane and worker nodes, rather than on machines you operate yourself.
Kubernetes is an internal preview. It is visible only to platform administrators, carries a DEV badge in the sidebar, and several of its detail tabs are unfinished. Do not plan production workloads on it yet. This page documents what exists so administrators evaluating it know where the edges are.
Navigation: Compute → Kubernetes. The list is titled Kubernetes Clusters.
This screen uses its own toolbar rather than the shared one every other Compute screen uses. The visible consequence is that it has no description line beneath the title and its typography differs slightly. Nothing is missing — it’s just built differently.

​
Columns

The show/hide-columns menu lists internal column identifiers rather than the human labels you see in the header, so the entries there won’t read the same as the table. A known rough edge of the preview.

​
States

An unrecognised state displays its raw value with a warning icon rather than being hidden.
The code computes a per-cluster health grade (excellent, good, degraded, critical) that is never displayed anywhere. If you’re looking for a health indicator, it doesn’t exist in the interface yet.

​
Filtering and bulk actions

The filter offers All Clusters, Running, Stopped, Starting, Stopping, and Error. Selecting rows shows {n} clusters selected with bulk Start, Stop, and Delete.

​
Row actions

Upgrade Version in the row menu does nothing except tell you to go elsewhere: it shows “Use the details page to upgrade the cluster version”. Use the cluster’s detail page instead.
Start Cluster and Stop Cluster appear only in the Running and Stopped states respectively. In a transitional state — Creating, Starting, Stopping — neither is offered, so there’s no way to cancel an in-flight transition from this menu. Wait for it to settle.

​
The detail page

Breadcrumb Compute / Kubernetes / {cluster}. Tabs in order: Details, Access, Instances, Networks, Firewall, Port Forwarding, Load Balancing, Events, Comments. Header actions: Refresh, Start Cluster, Stop Cluster, Manage Nodes, Scale Cluster, Upgrade Version, Download Kubeconfig, Delete Cluster — all disabled while the page is loading.
Port Forwarding and Load Balancing tabs exist on this page, but both underlying capabilities are switched off platform-wide in the current phase — see Public IPs. Expect them to be empty or inert regardless of the cluster’s state.

​
Connecting to a cluster

Download Kubeconfig or Copy Kubeconfig gives you the credentials kubectl needs.
A kubeconfig contains cluster credentials. Treat the downloaded file as a secret: don’t commit it, don’t paste it into chat, and delete it when you’re done evaluating. Anyone holding it has whatever access it grants.
Always run kubectl config current-context before a destructive command. Merging a new cluster’s config can silently change your active context, and kubectl delete doesn’t ask which cluster you meant.

​
What isn’t finished

Being specific, since this is a preview:
Those four placeholder tabs aren’t wired into the detail page’s tab list, so you won’t encounter them as empty tabs — they simply don’t appear. The nine tabs listed above are what exists. Manage workloads with kubectl in the meantime.

​
Empty and error states

Virtual Machines

The generally available compute surface.

Networking

The VPC clusters run inside.