GPU: pr update

roles/gpu/tasks/driver.yml

@@ -0,0 +1,63 @@
+---
+- name: Check if system needs to be restarted
+  ansible.builtin.command: '/bin/needs-restarting -r'
+  register: needs_restarting
+  failed_when: 'needs_restarting.rc > 1'
+  changed_when: 'needs_restarting.rc == 1'
+  become: true
+  notify: reboot_server
+
+- name: Reboot system if needed
+  ansible.builtin.meta: flush_handlers
+
+- name: Install yum requirements for gpu driver installation
+  ansible.builtin.yum:
+    state: 'installed'
+    update_cache: true
+    name:
+      - tar
+      - bzip2
+      - make
+      - automake
+      - gcc
+      - gcc-c++
+      - pciutils
+      - elfutils-libelf-devel
+      - libglvnd-devel
+      - bind-utils
+      - wget
+  become: true
+
+- name: Gather facts to get the latest kernel version
+  ansible.builtin.setup:
+  become: true
+
+- name: Install kernel developement package matching running kernel version
+  ansible.builtin.yum:
+    name: 'kernel-devel-{{ ansible_kernel }}'
+  register: yum_result
+  failed_when: yum_result.rc != 0
+  when: true
+  become: true
+
+- name: Download a driver installation file from NVidia
+  ansible.builtin.get_url:
+    url: '{{ gpu_url_dir }}/{{ gpu_runfile }}'
+    dest: '/root/{{ gpu_runfile }}'
+    mode: '0700'
+  become: true
+
+- name: Install driver from .run file
+  ansible.builtin.command: '/root/{{ gpu_runfile }} --silent --driver'
+  register: install_result
+  failed_when: install_result.rc != 0
+  when: true
+  become: true
+
+- name: Remove installation file
+  ansible.builtin.file:
+    path: '/root/{{ gpu_runfile }}'
+    state: absent
+  become: true
+
+...

roles/gpu/templates/nvidia-persistenced.service

@@ -0,0 +1,14 @@
+[Unit]
+Description=Initialize GPU at the startup of the system
+Before=slurmd.service
+
+[Service]
+ExecStart=/usr/bin/nvidia-persistenced --verbose --user {{ nvidia_user }}
+ExecStopPost=/bin/rm -rf /var/run/nvidia-persistenced
+Type=forking
+PIDFile=/var/run/nvidia-persistenced/nvidia-persistenced.pid
+RestartSec=15
+Restart=always
+
+[Install]
+WantedBy=multi-user.target