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

- name: Verify NVIDIA driver installation
  shell: nvidia-smi
  register: nvidia_verification
  failed_when: false
  
- name: Display NVIDIA driver status
  debug:
    msg: "NVIDIA driver status: {{ nvidia_verification.stdout_lines }}"
  when: nvidia_verification.rc == 0
  
- name: Verify CUDA installation
  shell: /usr/local/cuda/bin/nvcc --version
  register: cuda_verification
  failed_when: false
  
- name: Display CUDA status
  debug:
    msg: "CUDA status: {{ cuda_verification.stdout_lines }}"
  when: cuda_verification.rc == 0
  
- name: Verify Container Toolkit installation
  shell: nvidia-ctk --version
  register: toolkit_verification
  failed_when: false
  
- name: Display Container Toolkit status
  debug:
    msg: "Container Toolkit status: {{ toolkit_verification.stdout_lines }}"
  when: toolkit_verification.rc == 0

- name: Restart host service by killing /opt/yunion/bin/host process
  shell: pgrep -xf '/opt/yunion/bin/host --common-config-file.*' | xargs -r kill || true
  become: true
  args:
    executable: /bin/bash