k8s/roles/nvidia_device_plugin/README.md

1.9 KiB

NVIDIA Device Plugin

Installs NVIDIA Kubernetes Device Plugin via Helm with GPU Feature Discovery support.

Requirements

  • Kubernetes cluster with NVIDIA GPU workers
  • NVIDIA drivers installed on GPU nodes
  • nvidia-container-toolkit configured

Role Variables

Variable Default Description
nvidia_device_plugin_chart_version 0.17.1 NVIDIA device plugin Helm chart version
nvidia_device_plugin_namespace nvidia-device-plugin Kubernetes namespace
nvidia_device_plugin_config {} ConfigMap configuration
nvidia_device_plugin_compat_with_cpu_manager false Enable CPUManager compatibility
nvidia_device_plugin_mig_strategy none MIG strategy: none, single, mixed
nvidia_device_plugin_fail_on_init_error true Fail on init error
nvidia_device_plugin_device_list_strategy envvar Device list strategy
nvidia_device_plugin_device_id_strategy uuid Device ID strategy
nvidia_device_plugin_nvidia_driver_root / NVIDIA driver root path
nvidia_device_plugin_gfd.enabled true Deploy GPU Feature Discovery
nvidia_device_plugin_nfd see defaults Node Feature Discovery config

Example Playbook

- hosts: k8s_manager
  become: false
  roles:
    - nvidia_device_plugin

Configuration Examples

MIG Strategy

nvidia_device_plugin_mig_strategy: "mixed"

ConfigMap

nvidia_device_plugin_config:
  map:
    default: |
      version: v1
      flags:
        migStrategy: none      
    mig-mixed: |
      version: v1
      flags:
        migStrategy: mixed      
  default: "default"

References