非sudo运行GPU版Docker容器时出现libnvidia-ml.so.1错误求解决

尝试调用GPU运行Docker Hub的nvidia/cuda镜像，执行以下命令：

docker run -it --gpus all -v --name my-gpu nvidia/cuda:11.7.0-cudnn8-devel-ubuntu22.04

出现如下错误：

Unable to find image 'nvidia/cuda:11.7.0-cudnn8-devel-ubuntu22.04' locally

11.7.0-cudnn8-devel-ubuntu22.04: Pulling from nvidia/cuda
d19f32bd9e41: Already exists 
292e5e4dcc78: Already exists 
f027458ef473: Already exists 
ad9cd0a3350e: Already exists 
4c0e748dfb24: Already exists 
e40f2cbf6f5e: Pull complete 
3ac150f167fe: Pull complete 
dd80ebac36de: Pull complete 
fd2716719ab6: Pull complete 
e5ff1925518e: Pull complete 
Digest: sha256:1055a2fa47b063336f578f390131efa4bb02fbfe095608481fd32b6fca9b8b32
Status: Downloaded newer image for nvidia/cuda:11.7.0-cudnn8-devel-ubuntu22.04
docker: Error response from daemon: failed to create shim task: OCI runtime create failed: runc create failed: unable to start container process: error during container init: error running hook #0: error running hook: exit status 1, stdout: , stderr: Auto-detected mode as 'legacy'
nvidia-container-cli: initialization error: load library failed: libnvidia-ml.so.1: cannot open shared object file: no such file or directory: unknown.
ERRO[0465] error waiting for container: context canceled 

使用sudo执行相同命令则可正常运行：

sudo docker run -it --gpus all --name my-container-03  nvidia/cuda:11.7.0-cudnn8-devel-ubuntu22.04

需实现无需sudo即可正常运行该容器。

1. 将当前用户加入docker用户组 #

默认Docker命令需root权限执行，将用户加入docker组可免sudo操作：

• 若docker组不存在，先创建组：

  sudo groupadd docker
  

• 将当前用户添加到docker组：

  sudo usermod -aG docker $USER
  

• 重新登录系统（或执行newgrp docker临时生效），使权限变更生效。

2. 修复nvidia-container-toolkit权限 #

错误提示libnvidia-ml.so.1无法访问，是因为普通用户缺少调用nvidia-container-cli的权限，需设置setuid：

• 查找nvidia-container-cli的路径：

  which nvidia-container-cli
  

• 通常路径为/usr/bin/nvidia-container-cli，执行以下命令：

  sudo chmod u+s /usr/bin/nvidia-container-cli
  

3. 验证配置 #

完成上述步骤后，重新执行命令（注意原命令中-v参数缺少挂载路径，需补充或移除）：

docker run -it --gpus all --name my-gpu nvidia/cuda:11.7.0-cudnn8-devel-ubuntu22.04

内容的提问来源于stack exchange，提问作者S. Jay