https://docs.docker.com/engine/reference/commandline/docker/
1.基础镜像选择
https://docs.nvidia.com/datacenter/cloud-native/container-toolkit/user-guide.html
选择Nvidia cuda支持gpu：https://hub.docker.com/r/nvidia/cuda/tags?page=1&ordering=last_updated&name=10.0
runtime版：用于服务部署， 大小1.7G；devel版：用于开发，功能完整，大小3G。
docker pull nvidia/cuda:10.0-cudnn7-runtime-centos7
docker pull nvidia/cuda:10.0-cudnn7-devel-centos7
2.yum换源
启动容器：docker run -itd —gpus=all —net=host —name=temp -v /docker_tmp:/server_api 1ba4e7500fa3 /bin/bash
进入容器：docker exec -it temp /bin/bash
备份yum原始的源：mv /etc/yum.repos.d /etc/yum.repos.d.bk
创建yum源目录：mkdir /etc/yum.repos.d
下载阿里云yum源：wget -O /etc/yum.repos.d/CentOS-Base.repo http://mirrors.aliyun.com/repo/Centos-7.repo
重建缓存：yum clean all && yum makecache
3.安装python3
安装python环境编译工具
yum -y install openssl-devel bzip2-devel expat-devel gdbm-devel readline-devel sqlite-devel
yum -y install gcc automake autoconf libtool make wget vim
移除centos自带python软连接（软链到python2的）
cd /usr/bin && mv python python.bk
下载python3源码
mkdir -p /usr/local/python/python3
cd /usr/local
wget https://www.python.org/ftp/python/3.6.8/Python-3.6.8.tgz
tar - xzvf Python-3.6.8.tgz
编译python3源码
cd /usr/local/Python-3.6.8
./configure —prefix=/usr/local/python/python3
make && make install
直接make && make install会报错：zipimport.ZipImportError: can’t decompress data; zlib not available，先如下解决：
vim Modules/Setup # 把#zlib zlibmodule.c -I\((prefix)/include -L\)(exec_prefix)/lib -lz 注释去掉
cd /usr/local/Python-3.6.8/Modules/zlib && ./configure && make install
cd /usr/local/Python-3.6.8
设置软链接
ln -s /usr/local/python/python3/bin/python3 /usr/bin/python
验证安装
python -V
修改yum
注意:由于centos7的yum要使用到python2.7.5的环境，我们要指定yum使用的python的版本，不然使用了3.6.8可能会导致yum命令无法使用，修改头部设置为使用python2.7
vim /usr/bin/yum
#!/usr/bin/python -→ #!/usr/bin/python2.7
4.安装pip工具
cd /usr/local
wget http://bootstrap.pypa.io/get-pip.py
python get-pip.py
ln -s /usr/local/python/python3/bin/pip3 /usr/bin/pip
pip -V
5.中文支持
echo “export LC_ALL=en_US.UTF-8” >> /etc/profile
source /etc/profile
echo “export LC_ALL=en_US.UTF-8” >> ~/.bashrc
source ~/.bashrc
6.制作镜像
docker export -o gpu_python3.tar 8020a9753211(容器号)
docker import gpu_python3.tar nvidia/cuda:1.9.0-cuda10.2-cudnn7-runtime-python3
docker tag nvidia/cuda:1.9.0-cuda10.2-cudnn7-runtime-python3 image:tag