---
# Provisions a dedicated local RAID1 partition for large model files (Ollama)
# on GPU worker nodes, carved out of unpartitioned space on the same mirrored
# OS disks. Run any time after the node's disks are in place; independent of
# kubeadm join order.
- name: GPU | Provision dedicated RAID1 storage for model files
  hosts: gpu_workers
  become: true
  roles:
    - gpu_model_storage