1. 下载Ubuntu20.04.6镜像

登录阿里云官方镜像站：阿里巴巴开源镜像站-OPSX镜像站-阿里云开发者社区

2. 测试环境

Server OS：Ubuntu 20.04.6 LTS
Kernel: Linux 5.4.0-155-generic x86-64
Docker Version：24.0.5, build ced0996
docker-compose version：1.25.0

Docker OS：Ubuntu 20.04.5 LTS
Nvidia GPU Version：NVIDIA-SMI 470.161.03
CUDA Version: 12.1
TensorFlow Version：1.15.1
python Version：3.8.10

3. Ubuntu下安装pip3 python3

Ubuntu下用apt命令安装

apt install python3-pip

4. Ubuntu下安装docker

curl https://get.docker.com | sh && sudo systemctl --now enable docker
测试
sudo docker run hello-world
提示：显示以下结果，表示安装成功

5.启动
systemctl start docker

6.停止
systemctl stop docker

7.重启
systemctl restart docker

8.设置开机启动
sudo systemctl enable docker

5. Ubuntu下安装Docker Compose

一个使用Docker容器的应用，通常由多个容器组成。使用Docker Compose不再需要使用shell脚本来启动容器。Compose 通过一个配置文件来管理多个Docker容器，在配置文件中，所有的容器通过services来定义，然后使用docker-compose脚本来启动，停止和重启应用，和应用中的服务以及所有依赖服务的容器，非常适合组合使用多个容器进行开发的场景

1. 卸载旧版本Docker Compose
如果之前安装过Docker Compose的旧版本，可以先卸载它们：

sudo rm /usr/local/bin/docker-compose

2. 下载Docker Compose最新版
从Docker官方网站下载Docker Compose最新版本的二进制文件：

sudo curl -L "https://github.com/docker/compose/releases/latest/download/docker-compose-$(uname -s)-$(uname -m)" -o /usr/local/bin/docker-compose

3. 授权Docker Compose二进制文
授予Docker Compose二进制文件执行权限

sudo chmod +x /usr/local/bin/docker-compose

4. 检查Docker Compose版本
docker-compose --version

docker-compose version 1.25.0, build unknown

6. Ubuntu下安装NVIDIA Docker

官网地址搜索Installing on Ubuntu and DebianInstalling on Ubuntu and Debian — container-toolkit 1.13.5 documentation

错误处理

Troubleshooting — container-toolkit 1.13.5 documentation

$ curl https://get.docker.com | sh \ && sudo systemctl --now enable docker

$ distribution=$(. /etc/os-release;echo $ID$VERSION_ID) \
&& curl -fsSL https://nvidia.github.io/libnvidia-container/gpgkey | sudo gpg --dearmor -o /usr/share/keyrings/nvidia-container-toolkit-keyring.gpg \
&& curl -s -L https://nvidia.github.io/libnvidia-container/$distribution/libnvidia-container.list | \
sed 's#deb https://#deb [signed-by=/usr/share/keyrings/nvidia-container-toolkit-keyring.gpg] https://#g' | \
sudo tee /etc/apt/sources.list.d/nvidia-container-toolkit.list

$ distribution=$(. /etc/os-release;echo $ID$VERSION_ID) \
&& curl -fsSL https://nvidia.github.io/libnvidia-container/gpgkey | sudo gpg --dearmor -o /usr/share/keyrings/nvidia-container-toolkit-keyring.gpg \
&& curl -s -L https://nvidia.github.io/libnvidia-container/experimental/$distribution/libnvidia-container.list | \
sed 's#deb https://#deb [signed-by=/usr/share/keyrings/nvidia-container-toolkit-keyring.gpg] https://#g' | \
sudo tee /etc/apt/sources.list.d/nvidia-container-toolkit.list

$ sudo apt-get update

执行sudo apt-get update -y 报错如下

E: Conflicting values set for option Signed-By regarding source https://nvidia.github.io/libnvidia-container/stable/ubuntu18.04/amd64/ /: /usr/share/keyrings/nvidia-container-toolkit-keyring.gpg !=
E: The list of sources could not be read.

解决办法：docker和nvidia-docker的安装以及错误记录_小白tb的博客-CSDN博客

# grep "nvidia.github.io" /etc/apt/sources.list.d/*

/etc/apt/sources.list.d/nvidia-container-toolkit.list:deb [signed-by=/usr/share/keyrings/nvidia-container-toolkit-keyring.gpg] https://nvidia.github.io/libnvidia-container/stable/ubuntu18.04/$(ARCH) /
/etc/apt/sources.list.d/nvidia-container-toolkit.list:#deb [signed-by=/usr/share/keyrings/nvidia-container-toolkit-keyring.gpg] https://nvidia.github.io/libnvidia-container/experimental/ubuntu18.04/$(ARCH) /

# cd /etc/apt/sources.list.d

# rm -rf *

# sudo apt-get install -y nvidia-container-toolkit

# sudo nvidia-ctk runtime configure --runtime=docker

# sudo systemctl restart docker

# sudo docker run --rm --runtime=nvidia --gpus all nvidia/cuda:11.6.2-base-ubuntu20.04 nvidia-smi

执行最后一个命令行遇到，解决“docker: Error response from daemon: Unknown runtime specified nvidia”问题
解决方法：
重启就行

sudo systemctl daemon-reload
sudo systemctl restart docker

如遇

docker: Error response from daemon: failed to create task for container: 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: nvml error: driver not loaded: unknown.
root@lenovo:/home/lenovo# sudo apt-get update

5. 查看所有docker所有进程

# docker ps -all
CONTAINER ID IMAGE COMMAND CREATED STATUS PORTS NAMES
79c75f07de8a nvidia/cuda:11.6.2-base-ubuntu20.04 "/bin/bash" 50 minutes ago Exited (0) 50 minutes ago wonderful_euler

docker ps 显示的是运行起来的进程，docker ps -all是全部进程

进程没有起来。。。。

查看日志发现无进程日志

# docker logs 79c75f07de8a

解决办法 # apt install nvidia-cuda-toolkit

安装速度较慢，大约2.5个小时。。。。。

6. 如果还是不行，参考官网报错分析（处理重大问题）

Troubleshooting — container-toolkit 1.13.5 documentation

# ausearch -c 'nvidia-docker' --raw | audit2allow -M my-nvidiadocker
# semodule -X 300 -i my-nvidiadocker.pp

# nvidia-docker run -d nvidia/cuda:12.2.0-base-ubuntu22.04

如果正常，再次查看docker进程发现有nvidia docker进程了

发现还是错误报错如下：

使用命令行 # nvidia-container-cli -k -d /dev/tty info 排查

I0729 14:45:17.678470 181027 nvc.c:395] dxcore initialization failed, continuing assuming a non-WSL environment
W0729 14:45:17.679863 181027 nvc.c:258] failed to detect NVIDIA devices

启动nvidia-docker

# systemctl start nvidia-docker.service

报错如下：

Job for nvidia-docker.service failed because the control process exited with error code.
See "systemctl status nvidia-docker.service" and "journalctl -xe" for details.

# systemctl status nvidia-docker.service

Process: 185483 ExecStart=/usr/bin/nvidia-docker-plugin -s $SOCK_DIR (code=exited, status=217/US>
Process: 185484 ExecStartPost=/bin/sh -c /bin/mkdir -p $( dirname $SPEC_FILE ) (code=exited, sta>
Process: 185485 ExecStopPost=/bin/rm -f $SPEC_FILE (code=exited, status=217/USER)

输入 # journalctl -xe

Jul 29 14:43:47 xx systemd[1]: Failed to start NVIDIA Docker plugin.
-- Subject: A start job for unit nvidia-docker.service has failed
-- Defined-By: systemd
-- Support: http://www.ubuntu.com/support
--
-- A start job for unit nvidia-docker.service has finished with a failure.
--
-- The job identifier is 5063 and the job result is failed.
Jul 29 15:00:52 xx su[171964]: pam_unix(su:session): session closed for user root
Jul 29 15:01:34 xx sudo[183078]: pam_unix(sudo:session): session opened for user root by xx(>
Jul 29 15:01:35 xx sudo[183078]: pam_unix(sudo:session): session closed for user root

解决办法：

vim模式下：:%s/\/sbin\/nologin/\/bin\/bash/g

解决办法：Ubuntu开启ssh服务及允许root登录

1)安装ssh服务器端
Ubuntu默认没有安装ssh的server，需要安装

apt-get install openssh-server
ssh客户端是默认安装的，连接其它ssh服务器用的，使用 apt install openssh-client安装

2)允许远程使用root账号ssh连接本机
修改/etc/ssh/sshd_config文件
vim /etc/ssh/sshd_config
修改如下：允许root账户登录
#PermitRootLogin prohibit-password
PermitRootLogin yes
3)需要重启系统或者sshd服务
sudo /etc/init.d/ssh stop
sudo /etc/init.d/ssh start
sudo service ssh restart
4)安装ssh服务后，系统默认开启系统sshd，查看sshd状态如果不是默认启动，修改服务为enable
sudo systemctl enable ssh

依然报错：

# journalctl -xeu docker.service

7. 基于NVIDIA-Docker安装Tensorflow2.13版本

1. 查看下载的镜像

# docker image ls

2. 下载tensorflow v2.13版本的镜像

官网地址：TensorFlow | NVIDIA NGC

# docker pull nvcr.io/nvidia/tensorflow:23.06-tf2-py3

3. 再次查看下载的镜像

# docker image ls

4. 进入tensorflow容器

nvidia-docker run --rm -it nvcr.io/nvidia/tensorflow:18.03-py3 （清除镜像）

# nvidia-docker run -it nvcr.io/nvidia/tensorflow:23.03-tf1-py3

格式：nvidia-docker run -it {REPOSITORY容器名称:TAG号}

7. docker和nvidia-docker的安装以及错误记录

错误一：sudo apt-get update出现
问题二：docker run --runtime=nvidia --rm nvidia/cuda:8.0-devel nvidia-smi出现
问题三：sudo docker run --rm --gpus all nvidia/cuda:11.0-base nvidia-smi出现
问题四 sudo docker run --runtime=nvidia --rm nvidia/cuda:10.0-base nvidia-smi 出现
最终安装成功啦！
参考链接：
nvidia-docker的安装

错误一：sudo apt-get update出现
参考链接
E: Conflicting values set for option Signed-By regarding source https://nvidia.github.io/libnvidia-container/stable/ubuntu18.04/amd64/ /: /usr/share/keyrings/nvidia-container-toolkit-keyring.gpg !=
E: The list of sources could not be read.
解决方法

grep "nvidia.github.io" /etc/apt/sources.list.d/*

会列出1个或者2个文件
然后进入/etc/apt/sources.list.d/文件夹中终端打开，将列出来的文件删除即可。

问题二：docker run --runtime=nvidia --rm nvidia/cuda:8.0-devel nvidia-smi出现
docker: Got permission denied while trying to connect to the Docker daemon socket at unix:///var/run/docker.sock: Post http://%2Fvar%2Frun%2Fdocker.sock/v1.24/containers/create: dial unix /var/run/docker.sock: connect: permission denied. code example
解决方法
docker前加sudo就行了

问题三：sudo docker run --rm --gpus all nvidia/cuda:11.0-base nvidia-smi出现
“–gpus” requires API version 1.40, but the Docker daemon API version is 1.39
解决方法
docker版本和nvidia-docker版本不匹配，将两个全删除了，再安装即可。
参考链接：

ubuntu中docker彻底卸载
ubuntu16.04离线安装与卸载docker和nvidia-docker
低版本Docker升级高版本Docker【详细教程、成功避坑】

问题四 sudo docker run --runtime=nvidia --rm nvidia/cuda:10.0-base nvidia-smi 出现
docker: Error response from daemon: unknown or invalid runtime name: nvidia.

解决“docker: Error response from daemon: Unknown runtime specified nvidia”问题
解决方法：
重启就行

sudo systemctl daemon-reload
sudo systemctl restart docker

最终安装成功啦！