Cloudpods/ocboot/onecloud/roles/utils/ai-env/tasks/install-cuda.yml

- name: Check if CUDA is already installed
  shell: /usr/local/cuda/bin/nvcc --version
  register: cuda_check
  ignore_errors: true
  failed_when: false

- name: Set CUDA installation flag
  set_fact:
    cuda_installed: "{{ cuda_check.rc == 0 }}"

- block:
  - name: Extract CUDA installer filename
    set_fact:
      cuda_installer: "{{ cuda_installer_path | basename }}"

  - name: Ensure rsync is installed on remote host
    package:
      name: rsync
      state: present
    become: true

  - name: Copy CUDA installer to remote host using synchronize
    ansible.builtin.synchronize:
      src: "{{ cuda_installer_path }}"
      dest: "/opt/nvidia/{{ cuda_installer }}"
      mode: push
    delegate_to: localhost
    become: false
    when: cuda_installer_path is defined

  - name: Check CUDA installer help
    shell: |
      cd /opt/nvidia
      ./{{ cuda_installer }} --help
    become: true
    args:
      executable: /bin/bash
    register: cuda_help
    ignore_errors: true

  - name: Display CUDA installer help
    debug: var=cuda_help.stdout_lines

  - name: Create temporary directory for CUDA extraction
    file:
      path: /opt/nvidia/tmp
      state: directory
      mode: '0755'
    become: true

  - name: Verify CUDA installer file integrity
    shell: |
      cd /opt/nvidia
      file ./{{ cuda_installer }}
      ls -lh ./{{ cuda_installer }}
    become: true
    register: cuda_file_check
    changed_when: false

  - name: Display CUDA installer file info
    debug: var=cuda_file_check.stdout_lines

  - name: Install CUDA
    shell: |
      cd /opt/nvidia
      export TMPDIR=/opt/nvidia/tmp
      export TEMP=/opt/nvidia/tmp
      export TMP=/opt/nvidia/tmp
      chmod +x {{ cuda_installer }}
      setsid bash -c "./{{ cuda_installer }} --silent --toolkit" < /dev/null > /tmp/cuda_install.log 2>&1 || {
        EXIT_CODE=$?
        cat /tmp/cuda_install.log
        exit $EXIT_CODE
      }
      cat /tmp/cuda_install.log
    become: true
    args:
      executable: /bin/bash
    register: cuda_install_result
    environment:
      DEBIAN_FRONTEND: noninteractive
      TERM: dumb
      TMPDIR: /opt/nvidia/tmp
      TEMP: /opt/nvidia/tmp
      TMP: /opt/nvidia/tmp

  - name: Check CUDA installation result
    shell: |
      if [ -f /usr/local/cuda/bin/nvcc ]; then
        /usr/local/cuda/bin/nvcc --version
        exit 0
      else
        echo "CUDA installation failed: nvcc not found"
        exit 1
      fi
    become: true
    args:
      executable: /bin/bash
    register: cuda_verify_result
    failed_when: cuda_verify_result.rc != 0

  - debug: var=cuda_install_result.stdout_lines
  - debug: var=cuda_verify_result.stdout_lines

  when: not cuda_installed | default(false)