add nvidia related

defaults/main.yml

@@ -22,3 +22,4 @@ china_mirror_omf: false
 cpufreq_policy: none
 vm_agents_status: present
 customization: true
+nvidia_driver_version: null

tasks/main.yml

@@ -35,6 +35,9 @@
   when: ansible_os_family == 'RedHat'
 - name: pip
   include_tasks: pip.yml
+- name: nvidia
+  include_tasks: nvidia.yml
+  when: nvidia_driver_version
 - name: docker
   include_tasks: docker.yml
   when: install_docker

tasks/nvidia.yml

+- name: nvidia apt key
+  become: true
+  apt_key:
+    url: 'https://nvidia.github.io/nvidia-container-runtime/gpgkey'
+- name: set distribution string
+  set_fact:
+    container_runtime_distribution_string: '{{ ansible_distribution | lower }}{{ ansible_distribution_version }}'
+  when: ansible_distribution_release != 'focal'
+- name: set distribution string for Ubuntu 20
+  set_fact:
+    container_runtime_distribution_string: 'ubuntu18.04'
+  when: ansible_distribution_release == 'focal'
+- name: container runtime apt repo
+  become: true
+  apt_repository:
+    repo: 'deb https://nvidia.github.io/{{item}}/stable/{{container_runtime_distribution_string}}/$(ARCH) /'
+    filename: nvidia-container-runtime
+  with_items:
+    - nvidia-container-runtime
+    - libnvidia-container
+- name: install nvidia and related packages
+  become: true
+  apt:
+    update_cache: true
+    name: 'nvidia-driver-{{ nvidia_driver_version }},nvidia-cuda-dev,nvidia-cuda-doc,nvidia-cuda-gdb,nvidia-cuda-toolkit,nvidia-cuda-toolkit-gcc,nvidia-container-runtime'
+- name: check install of cudnn
+  find:
+    paths:
+      - /usr/lib/x86_64-linux-gnu
+    patterns: 'libcudnn.so.7*'
+  register: cudnn_result
+- name: install cudnn
+  become: true
+  apt:
+    deb: 'https://minio.mycard.moe:9000/nanahira/cudnn/{{item}}.deb'
+  with_items:
+    - libcudnn7_7.6.5.32-1+cuda10.1_amd64
+    - libcudnn7-dev_7.6.5.32-1+cuda10.1_amd64
+    - libcudnn7-doc_7.6.5.32-1+cuda10.1_amd64
+  when: not cudnn_result.files[0] is defined and ansible_distribution == "Ubuntu" and ansible_distribution_major_version|int >= 18

tasks/upgrade.yml

@@ -22,6 +22,10 @@
   apt:
     name: python-is-python3
   when: ansible_distribution == 'Ubuntu' and upgrade
+- name: install nvidia driver before reboot
+  apt:
+    name: nvidia-driver-{{ nvidia_driver_version }}
+  when: nvidia_driver_version
 - name: 系统源 重启
   become: true
   shell: reboot

templates/daemon.json.j2

 
 {
-{% if ansible_distribution == "Fedora" %}
 
 {% if registry %}
   "registry-mirrors": [
     "{{registry}}"
-  ]
+  ],
 {% endif %}
 
-{% else %}
-
-{% if registry %}
-  "registry-mirrors": [
-    "{{registry}}"
-  ],
+{% if nvidia_driver_version %}
+  "runtimes": {
+    "nvidia": {
+      "path": "/usr/bin/nvidia-container-runtime",
+      "runtimeArgs": []
+    }
+  },
 {% endif %}
-  "live-restore": true
 
+{% if ansible_distribution != "Fedora" %}
+  "live-restore": true,
 {% endif %}
+
+  "log-driver": "json-file",
+  "log-opts": {
+    "max-size": "1g",
+    "max-file": "1"
+  }
 }