基于docker-compose和ngc cloud快速创建指定pytorch版本的jupyter-lab环境

简介

本博客以构建指定版本的pytorch运行环境为例，给出Dockerfile、docker-compose.yml以及启动容器的运行脚本。

找到想要的容器

从中 https://docs.nvidia.com/deeplearning/frameworks/pytorch-release-notes/overview.html#overview中找到想要的镜像版本。本文以torch 1.8为例，要安装的镜像版本为pytorch:20.12-py3
在https://catalog.ngc.nvidia.com/orgs/nvidia/containers/pytorch/tags中找到镜像的pull tag：

docker pull nvcr.io/nvidia/pytorch:20.12-py3

文件目录

文件目录如下

• torch_1_8
  • Dockerfile
  • run.sh
  • docker-compose.yml

Dockerfile

FROM nvcr.io/nvidia/pytorch:20.12-py3 # 此处修改为根镜像的tag
MAINTAINER yourname

RUN apt-get update
RUN apt-get install -y openssh-server curl screen
RUN curl -sL https://deb.nodesource.com/setup_12.x | bash - && apt install -y nodejs

RUN python -m pip install -i https://pypi.tuna.tsinghua.edu.cn/simple jupyterlab ipympl nodejs ptvsd
RUN jupyter labextension install @jupyterlab/debugger
RUN conda install xeus-python -c conda-forge

RUN sed -ri 's/^PermitRootLogin\s+.*/PermitRootLogin yes/' /etc/ssh/sshd_config
RUN sed -ri 's/^#PermitRootLogin\s+.*/PermitRootLogin yes/' /etc/ssh/sshd_config

COPY run.sh /run.sh
RUN chmod 777 /run.sh

ENV NOTEBOOK_APP_TOKEN=jupyter_passwd NOTEBOOK_DIR=/data # 此处设定jupyterlab的访问密码以及容器内的数据路径
RUN echo "root:ssh_passwd" | chpasswd

# ssh
EXPOSE 22
# jupyter
EXPOSE 8888
# tensorboard
EXPOSE 6006

docker-compose.yml

version: '3'
services:
  dl:
    build: .
    entrypoint: /run.sh
    ports:
      - "12222:22"   #对应ssh端口，注意填写前先查看端口号占用情况(可只更改x)
      - "18888:8888" #对应jupyter端口，注意填写前先查看端口号占用情况(可只更改x)
      - "16006:6006" #tensorboard
    volumes:
      - ${host_data}:/data # {要挂载的主机数据目录}:{容器内的数据目录}
    cap_add:
      - SYS_ADMIN
      - DAC_READ_SEARCH
    security_opt:
      - apparmor:unconfined
    deploy:
      restart_policy:
        condition: on-failure
      resources:
        reservations:
          devices:
            - driver: nvidia
              device_ids: ['2'] # gpu编号，可以写多个
              capabilities: [gpu]

run.sh

#!/bin/bash
service ssh start # 容器打开时打开ssh sever
jupyter lab  --no-browser --ip=0.0.0.0 --allow-root --NotebookApp.token=${NOTEBOOK_APP_TOKEN} --notebook-dir=${NOTEBOOK_DIR}

docker-compose构建镜像及启动容器

构建镜像同时启动容器

docker-compose up -d

关闭并删除容器

docker-compose down

构建镜像

docker-compose build

容器的访问

ssh访问

ssh访问机器的12222端口，账户为root，密码ssh_passwd

浏览器访问upyter lab

浏览器访问18888端口，密码为jupyter_passwd