Crusoe Managed Kubernetes uses a shared-responsibility model. Crusoe operates the underlying cloud and control plane, while your platform team still makes explicit choices for node pools, access, add-ons, storage use, workload policy, and some network configuration.

See the Crusoe Managed Kubernetes for the upstream configuration and requirements.

Prerequisites

• A Crusoe Managed Kubernetes cluster with the required GPU node pool and NVIDIA components.
• Reviewed cluster access, network, storage class, registry, DNS, and workload identity configuration.
• The cluster connected to Polyaxon by installing Polyaxon Agent (compute cluster), with namespace permissions and a queue or preset selecting that node pool.

Connect the cluster

Follow Connect Kubernetes to Polyaxon to install Polyaxon Agent (compute cluster), configure storage and queues, and submit the shared GPU visibility example. Use the same namespace and queue that will run your workloads.

The example only checks GPU allocation and driver visibility. Validate storage, identity, and multi-node networking separately using the provider requirements below.

Provider requirements

• Record the owner for each item in Crusoe's shared-responsibility model before production rollout.
• Keep cluster, node-pool, add-on, compute-cluster connection, queue, and preset configuration in version control.
• Revalidate GPU, storage, identity, DNS, and network behavior after cluster or node-pool changes.
• Decide whether scaling is managed through fixed node pools or an AutoClusters configuration before setting queue expectations.
• Run a collective communication test on the exact nodes and interfaces used by distributed training.

For distributed training, continue with NCCL and RCCL validation.

For help choosing a provider, see Choosing GPU infrastructure for Polyaxon.

Troubleshooting

The GPU resource is missing

Check node-pool readiness, NVIDIA component status, kubelet capacity, and whether the queue selector and tolerations match the GPU nodes.

The operation cannot mount its volume

Verify the requested storage class, CSI driver state, availability-zone constraints, permissions, and whether the workload expects persistent or local storage.

The connected compute cluster cannot submit or collect logs

Verify the Polyaxon Agent deployment, cluster access, namespace RBAC, network policy, registry credentials, queue routing, and control-plane connectivity.

References

• Crusoe Managed Kubernetes — Provider documentation for creating and operating Crusoe Kubernetes clusters.
• NVIDIA CUDA container — Published CUDA container tags used for the GPU visibility check.
• Crusoe shared responsibility — The ownership boundary for infrastructure, control plane, nodes, access, networking, storage, add-ons, and workloads.
• Polyaxon job runtime — Job containers, resources, connections, initialization, and lifecycle settings.
• Polyaxon resource scheduling — Kubernetes requests, limits, custom resources, queues, and placement controls.

---