feat(nvidia_device_plugin): add initial NVIDIA device plugin role skeleton

roles/nvidia_device_plugin/README.md

@@ -0,0 +1,63 @@
+# 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
+
+```yaml
+- hosts: k8s_manager
+  become: false
+  roles:
+    - nvidia_device_plugin
+```
+
+## Configuration Examples
+
+### MIG Strategy
+
+```yaml
+nvidia_device_plugin_mig_strategy: "mixed"
+```
+
+### ConfigMap
+
+```yaml
+nvidia_device_plugin_config:
+  map:
+    default: |
+      version: v1
+      flags:
+        migStrategy: none
+    mig-mixed: |
+      version: v1
+      flags:
+        migStrategy: mixed
+  default: "default"
+```
+
+## References
+
+- [NVIDIA k8s-device-plugin GitHub](https://github.com/NVIDIA/k8s-device-plugin)
+- [Helm Chart Values](https://github.com/NVIDIA/k8s-device-plugin/blob/v0.17.1/deployments/helm/nvidia-device-plugin/values.yaml)

roles/nvidia_device_plugin/defaults/main.yml

@@ -0,0 +1,97 @@
+---
+nvidia_device_plugin_chart_version: "0.17.1"
+nvidia_device_plugin_namespace: nvidia-device-plugin
+
+nvidia_device_plugin_config:
+  map: {}
+  default: ""
+  name: ""
+
+nvidia_device_plugin_compat_with_cpu_manager: false
+nvidia_device_plugin_mig_strategy: "none"
+nvidia_device_plugin_fail_on_init_error: true
+nvidia_device_plugin_device_list_strategy: "envvar"
+nvidia_device_plugin_device_id_strategy: "uuid"
+nvidia_device_plugin_nvidia_driver_root: "/"
+
+nvidia_device_plugin_image:
+  repository: nvcr.io/nvidia/k8s-device-plugin
+  tag: ""
+  pull_policy: IfNotPresent
+
+nvidia_device_plugin_resources: {}
+
+nvidia_device_plugin_node_selector: {}
+
+nvidia_device_plugin_affinity:
+  node_affinity:
+    required_during_scheduling_ignored_during_execution:
+      node_selector_terms:
+        - match_expressions:
+            - key: feature.node.kubernetes.io/pci-10de.present
+              operator: In
+              values:
+                - "true"
+        - match_expressions:
+            - key: feature.node.kubernetes.io/cpu-model.vendor_id
+              operator: In
+              values:
+                - "NVIDIA"
+        - match_expressions:
+            - key: nvidia.com/gpu.present
+              operator: In
+              values:
+                - "true"
+
+nvidia_device_plugin_tolerations:
+  - key: CriticalAddonsOnly
+    operator: Exists
+  - key: nvidia.com/gpu
+    operator: Exists
+    effect: NoSchedule
+
+nvidia_device_plugin_priority_class_name: "system-node-critical"
+
+nvidia_device_plugin_runtime_class_name: null
+
+nvidia_device_plugin_gfd:
+  enabled: true
+  name_override: gpu-feature-discovery
+  no_timestamp: null
+  sleep_interval: null
+  security_context:
+    privileged: true
+
+nvidia_device_plugin_nfd:
+  name_override: node-feature-discovery
+  enable_node_feature_api: false
+  master:
+    service_account:
+      name: node-feature-discovery
+      create: true
+    config:
+      extra_label_ns:
+        - nvidia.com
+  worker:
+    tolerations:
+      - key: node-role.kubernetes.io/master
+        operator: Equal
+        value: ""
+        effect: NoSchedule
+      - key: nvidia.com/gpu
+        operator: Equal
+        value: present
+        effect: NoSchedule
+    config:
+      sources:
+        pci:
+          device_class_whitelist:
+            - "02"
+            - "03"
+          device_label_fields:
+            - vendor
+
+nvidia_device_plugin_mps:
+  root: "/run/nvidia/mps"
+
+nvidia_device_plugin_kubeconfig: "/home/{{ ansible_user }}/.kube/config"

roles/nvidia_device_plugin/meta/main.yml

@@ -0,0 +1,12 @@
+---
+galaxy_info:
+  role_name: nvidia_device_plugin
+  author: ops
+  description: Installs NVIDIA Kubernetes Device Plugin via Helm with GPU Feature Discovery support
+  license: MIT
+  min_ansible_version: "2.14"
+  platforms:
+    - name: EL
+      versions:
+        - "9"
+  dependencies: []

roles/nvidia_device_plugin/tasks/deploy.yml

@@ -0,0 +1,36 @@
+---
+- name: Helm | Add NVIDIA device plugin chart repository
+  ansible.builtin.command: helm repo add nvdp https://nvidia.github.io/k8s-device-plugin
+  environment:
+    KUBECONFIG: "{{ nvidia_device_plugin_kubeconfig }}"
+    PATH: "/usr/local/bin:/usr/bin:/bin"
+  register: _helm_repo_add
+  changed_when: "'already exists' not in _helm_repo_add.stdout"
+  failed_when: _helm_repo_add.rc != 0 and 'already exists' not in _helm_repo_add.stdout
+
+- name: Helm | Update chart repositories
+  ansible.builtin.command: helm repo update
+  environment:
+    KUBECONFIG: "{{ nvidia_device_plugin_kubeconfig }}"
+    PATH: "/usr/local/bin:/usr/bin:/bin"
+  changed_when: false
+
+- name: NVIDIA Device Plugin | Render values
+  ansible.builtin.template:
+    src: values.yml.j2
+    dest: /tmp/nvidia-device-plugin-values.yml
+    mode: "0600"
+
+- name: NVIDIA Device Plugin | Install or upgrade
+  ansible.builtin.command: >
+    helm upgrade --install nvidia-device-plugin nvdp/nvidia-device-plugin
+    --namespace {{ nvidia_device_plugin_namespace }}
+    --version {{ nvidia_device_plugin_chart_version }}
+    --values /tmp/nvidia-device-plugin-values.yml
+    --timeout 15m0s
+    --wait
+  environment:
+    KUBECONFIG: "{{ nvidia_device_plugin_kubeconfig }}"
+    PATH: "/usr/local/bin:/usr/bin:/bin"
+  register: _helm_install
+  changed_when: "'STATUS: deployed' in _helm_install.stdout or 'has been upgraded' in _helm_install.stdout"

roles/nvidia_device_plugin/tasks/gfd.yml

@@ -0,0 +1,16 @@
+---
+- name: GPU Feature Discovery | Check if already deployed
+  ansible.builtin.kubectl:
+    name: "gpu-feature-discovery"
+    kind: "DaemonSet"
+    namespace: "{{ nvidia_device_plugin_namespace }}"
+    state: "list"
+  environment:
+    KUBECONFIG: "{{ nvidia_device_plugin_kubeconfig }}"
+    PATH: "/usr/local/bin:/usr/bin:/bin"
+  register: _gfd_check
+  changed_when: false
+
+- name: GPU Feature Discovery | Deploy via Helm (if enabled in values)
+  ansible.builtin.include_tasks: deploy.yml
+  when: nvidia_device_plugin_gfd.enabled and (_gfd_check.stdout | length == 0)

roles/nvidia_device_plugin/tasks/main.yml

@@ -0,0 +1,9 @@
+---
+- name: Create namespace
+  ansible.builtin.include_tasks: namespace.yml
+
+- name: Deploy NVIDIA Device Plugin via Helm
+  ansible.builtin.include_tasks: deploy.yml
+
+- name: Deploy GPU Feature Discovery
+  ansible.builtin.include_tasks: gfd.yml

roles/nvidia_device_plugin/tasks/namespace.yml

@@ -0,0 +1,10 @@
+---
+- name: Create namespace
+  ansible.builtin.kubectl:
+    name: "{{ nvidia_device_plugin_namespace }}"
+    api_version: "v1"
+    kind: "Namespace"
+    state: "present"
+  environment:
+    KUBECONFIG: "{{ nvidia_device_plugin_kubeconfig }}"
+    PATH: "/usr/local/bin:/usr/bin:/bin"

roles/nvidia_device_plugin/templates/values.yml.j2

@@ -0,0 +1,128 @@
+{% if nvidia_device_plugin_config.name or nvidia_device_plugin_config.map %}
+config:
+{% if nvidia_device_plugin_config.name %}
+  name: "{{ nvidia_device_plugin_config.name }}"
+{% endif %}
+{% if nvidia_device_plugin_config.map %}
+  map:
+{% for key, value in nvidia_device_plugin_config.map.items() %}
+    {{ key }}: |
+{{ value | indent(6, false) }}
+{% endfor %}
+{% endif %}
+{% if nvidia_device_plugin_config.default %}
+  default: "{{ nvidia_device_plugin_config.default }}"
+{% endif %}
+{% if nvidia_device_plugin_config.fallback_strategies %}
+  fallbackStrategies: {{ nvidia_device_plugin_config.fallback_strategies }}
+{% endif %}
+{% endif %}
+
+compatWithCPUManager: {{ nvidia_device_plugin_compat_with_cpu_manager | lower }}
+
+{% if nvidia_device_plugin_mig_strategy is defined and nvidia_device_plugin_mig_strategy %}
+migStrategy: "{{ nvidia_device_plugin_mig_strategy }}"
+{% endif %}
+
+{% if nvidia_device_plugin_fail_on_init_error is defined %}
+failOnInitError: {{ nvidia_device_plugin_fail_on_init_error | lower }}
+{% endif %}
+
+{% if nvidia_device_plugin_device_list_strategy is defined and nvidia_device_plugin_device_list_strategy %}
+deviceListStrategy: "{{ nvidia_device_plugin_device_list_strategy }}"
+{% endif %}
+
+{% if nvidia_device_plugin_device_id_strategy is defined and nvidia_device_plugin_device_id_strategy %}
+deviceIDStrategy: "{{ nvidia_device_plugin_device_id_strategy }}"
+{% endif %}
+
+{% if nvidia_device_plugin_nvidia_driver_root is defined and nvidia_device_plugin_nvidia_driver_root %}
+nvidiaDriverRoot: "{{ nvidia_device_plugin_nvidia_driver_root }}"
+{% endif %}
+
+image:
+  repository: {{ nvidia_device_plugin_image.repository }}
+  tag: "{{ nvidia_device_plugin_image.tag }}"
+  pullPolicy: {{ nvidia_device_plugin_image.pull_policy }}
+
+resources: {{ nvidia_device_plugin_resources | default({}) | to_nice_yaml(indent=2) | indent(2) }}
+
+nodeSelector: {{ nvidia_device_plugin_node_selector | default({}) | to_nice_yaml(indent=2) | indent(2) }}
+
+affinity:
+  nodeAffinity:
+    requiredDuringSchedulingIgnoredDuringExecution:
+      nodeSelectorTerms:
+{% for term in nvidia_device_plugin_affinity.node_affinity.required_during_scheduling_ignored_during_execution.node_selector_terms %}
+      - matchExpressions:
+{% for expr in term.match_expressions %}
+        - key: {{ expr.key }}
+          operator: {{ expr.operator }}
+          values:
+{% for val in expr.values %}
+            - {{ val }}
+{% endfor %}
+{% endfor %}
+{% endfor %}
+
+tolerations: {{ nvidia_device_plugin_tolerations | to_nice_yaml(indent=2) | indent(2) }}
+
+priorityClassName: "{{ nvidia_device_plugin_priority_class_name }}"
+
+{% if nvidia_device_plugin_runtime_class_name is defined and nvidia_device_plugin_runtime_class_name %}
+runtimeClassName: "{{ nvidia_device_plugin_runtime_class_name }}"
+{% endif %}
+
+devicePlugin:
+  enabled: true
+
+{% if nvidia_device_plugin_gfd.enabled %}
+gfd:
+  enabled: true
+  nameOverride: "{{ nvidia_device_plugin_gfd.name_override }}"
+{% if nvidia_device_plugin_gfd.no_timestamp is defined %}
+  noTimestamp: {{ nvidia_device_plugin_gfd.no_timestamp | lower }}
+{% endif %}
+{% if nvidia_device_plugin_gfd.sleep_interval is defined %}
+  sleepInterval: "{{ nvidia_device_plugin_gfd.sleep_interval }}"
+{% endif %}
+  securityContext:
+    privileged: {{ nvidia_device_plugin_gfd.security_context.privileged | lower }}
+{% endif %}
+
+{% if nvidia_device_plugin_nfd %}
+nfd:
+  nameOverride: "{{ nvidia_device_plugin_nfd.name_override }}"
+  enableNodeFeatureApi: {{ nvidia_device_plugin_nfd.enable_node_feature_api | lower }}
+  master:
+    serviceAccount:
+      name: "{{ nvidia_device_plugin_nfd.master.service_account.name }}"
+      create: {{ nvidia_device_plugin_nfd.master.service_account.create | lower }}
+    config:
+      extraLabelNs:
+{% for ns in nvidia_device_plugin_nfd.master.config.extra_label_ns %}
+        - {{ ns }}
+{% endfor %}
+  worker:
+    tolerations:
+{% for tol in nvidia_device_plugin_nfd.worker.tolerations %}
+    - key: "{{ tol.key }}"
+      operator: "{{ tol.operator }}"
+      value: "{{ tol.value | default('') }}"
+      effect: "{{ tol.effect }}"
+{% endfor %}
+    config:
+      sources:
+        pci:
+          deviceClassWhitelist:
+{% for cls in nvidia_device_plugin_nfd.worker.config.sources.pci.device_class_whitelist %}
+            - {{ cls }}
+{% endfor %}
+          deviceLabelFields:
+{% for field in nvidia_device_plugin_nfd.worker.config.sources.pci.device_label_fields %}
+            - {{ field }}
+{% endfor %}
+{% endif %}
+
+mps:
+  root: "{{ nvidia_device_plugin_mps.root }}"