Kubernetes Installation

NVIDIA Mission Control uses separate Kubernetes clusters for administrative services and user workloads. Run the BCM setup tools on the active BCM head node to deploy these clusters onto their Kubernetes nodes. The BCM head node is distinct from the Kubernetes admin and user control plane nodes.

For the cluster names and BCM node categories used in these instructions, refer to Names Used in This Section.

Administrative Cluster (k8s-admin)

The k8s-admin control plane is created during initial deployment. This cluster is the primary management layer. It hosts core infrastructure services and operational components, including:

• Autonomous Job Recovery

• Autonomous Hardware Recovery

• Domain Power Service

• Observability stack

User Cluster (k8s-user)

The k8s-user cluster hosts user GPU workloads. Create its Kubernetes control plane on the k8s-system-user nodes as part of Run:ai Installation, then deploy Run:ai onto that cluster. The administrative cluster created in this section does not replace the user cluster.