EAI_5684 rancher disk

[woojae-siloai]

Summary
Add complete support for RANCHER_DISK parameter to dedicate separate storage for /var/lib/rancher directory, designed primarily for GPU worker nodes with intensive workloads.

What's New
RANCHER_DISK Parameter:

• Dedicates separate disk for /var/lib/rancher (kubelet and container runtime data)
• Handles raw device paths like CLUSTER_DISKS (automatic formatting and mounting)
• Highly recommended for GPU worker nodes with heavy workloads
• Optional for control plane nodes and CPU worker nodes
• Can be combined with CLUSTER_DISKS and CLUSTER_PREMOUNTED_DISKS

Key Features

• Automatic Setup: Device formatting, mounting at /var/lib/rancher, fstab persistence
• Complete Cleanup: Unmounting, fstab cleanup, clean directory restoration
• Auto-Discovery: Cleanup command automatically finds all bloom storage without config
• Consistent Previews: Same cleanup information whether using config file or not

Implementation
Core Components:

• Configuration schema validation and Go config parsing
• Ansible playbook for device setup (rancher_storage.yaml)
• Cleanup functions with proper device management
• Auto-discovery system for consistent cleanup previews

Documentation:

• Complete parameter documentation and examples
• GPU worker node guidance and recommendations

Example Usage

GPU Worker Node (Recommended)

CLUSTER_DISKS: /dev/nvme0n1          # Application storage
RANCHER_DISK: /dev/nvme2n1           # Dedicated /var/lib/rancher

Control Plane (Optional)

RANCHER_DISK: /dev/nvme1n1           # Optional dedicated RKE2 storage

[Q-Dub]

LGTM!

[woojae-siloai]

@Q-Dub
I updated a bit to cover a legacy node case that which is bloomed with mounting /var/lib/rancher and updating fstab manually configured. Tested with medium and large size including mock legacy case on pd3.

[brownzebra]

LGTM