Kubernetes

Create a Kubernetes Cluster

Deploy a managed cluster with node pools.

Creating a Kubernetes cluster in flex.plane requires three things: a name, a Kubernetes version, and at least one node pool. The platform handles everything else: provisioning VMs, bootstrapping the control plane, joining worker nodes, and setting up Konnectivity tunnels.

The cluster creation form lets you configure the cluster name, Kubernetes version, and node pools with compute profiles and zone placement.

Choose a Kubernetes version

The cluster creation form shows all available Kubernetes versions in a dropdown. The platform maintains a curated set of tested versions.

Unless you have a specific compatibility requirement, choose the latest available version. It will have the most recent security patches and features.

Define node pools

Node pools are groups of worker nodes with identical configuration. Each node pool specifies:

  • name: a unique name for the pool (e.g. "general", "gpu-workers", "batch").
  • region and zone: where the worker VMs will be provisioned.
  • size: the number of worker nodes (replicas). Minimum 0, maximum 10.
  • computeProfile: the CPU and memory configuration for each node in the pool.
  • taints: optional Kubernetes taints to control pod scheduling on these nodes.

You can create a cluster with multiple node pools, each in a different zone or with different compute profiles. This lets you build heterogeneous clusters, for example a pool of small general-purpose nodes and a pool of large memory-optimized nodes.

Each node pool can have at most 10 nodes. If you need more capacity, create additional node pools.

Launch the cluster

Fill in the cluster name, select a Kubernetes version, and configure your node pools in the creation form. Click Create to start provisioning.

After creation, the cluster goes through several states:

  1. Provisioning: the control plane is being deployed on the management cluster and worker VMs are being created.
  2. Ready: the control plane is up and the API server is reachable.
  3. Running: all requested worker nodes have joined and are ready to schedule pods.

You can monitor progress from the Kubernetes cluster list and detail pages.

The Kubernetes list shows all clusters with their version, status, and node pool health.

The control plane starts in seconds (it runs as pods on the management cluster). Worker nodes take a few minutes because VMs need to boot, install Kubernetes components, and join the cluster.
Cluster creation is an asynchronous operation. The portal shows real-time status updates as the control plane and worker nodes come online.