Objective of this paper

Introduce the basic knowledge of docker Learn to package a custom docker and let others use it successfully

reference material

• Introduction to Docker - ruanyifeng's Weblog (ruanyifeng.com)
• How to build mmdetection environment in docker and encapsulate it into image_ Kongyan 66 blog - CSDN blog
• Docker -- from introduction to practice
• docker_practice/SUMMARY.md at master · yeasy/docker_practice (github.com)
• How to play docker? This wave of docker actual combat summary (very comprehensive) finally caught up- Zhihu (zhihu.com)
• Docker image container basic operation - common commands - iDEAAM - blog Garden (cnblogs.com)
• How to play docker? This wave of docker actual combat summary (very comprehensive) finally caught up- Zhihu (zhihu.com) -Orientation and setup | Docker Documentation
• Check the running container in linux and talk about the common commands of Docker container_ weixin_39568597 blog - CSDN blog
• Build a docker you want - Zhihu (zhihu.com)
• What is Docker? It's enough to read this dry article- Zhihu (zhihu.com) (Getting Started Guide)
• Docker tutorial - MOOC (imooc.com)
• [introduction to docker] 10 minutes to quickly learn docker - Zhihu (zhihu.com) (introductory Practice Guide)
• Docker actual combat summary (very comprehensive) - Zhihu (zhihu.com)
• Check the running container in linux and talk about the common command Weixin of Docker container_ 39568597 blog - CSDN blog

Basic concepts of docker

The birth background of docker

docker was born to solve the problem of inconsistency between development environment and multiple software versions.

docker is not a virtual machine

docker installation

sudo wget -qO- https://get.docker.com | sh sudo usermod -aG docker usernamexxx

Basic concepts of docker

Overall architecture of docker: Note that docker daemon is responsible for controlling command execution

image.png

There are several concepts in docker:

• dockerfile:
• image:
• container:

In fact, you can simply understand image as an executable program, and container is a running process.

If you need source code to write a program, you need dockerfile to "write" an image. Dockerfile is the source code of the image and docker is the "compiler".

Therefore, we only need to specify what programs are required and what configurations depend on in the dockerfile, and then hand over the dockerfile to the "compiler" docker for "compilation", that is, the docker build command. The generated executable program is the image, and then we can run the image. This is the docker run command, and the image is the docker container when it runs.

The specific use methods are not described here. You can refer to the official document of docker, where there is a detailed explanation.

actual combat

docker common instructions

docker pull # Get docker image

docker run -p... #Run container port mapping

docker run -v xxx(local path):xxx(docker Internal directory)# Directory mount

docker run -it # -The t option allows Docker to assign a pseudo TTY and bind it to the standard input of the container, - i keeps the standard input of the container open.

docker commit # Save changes to a new image

docker build # Create docker image

docker create # 

docker attach # Enter container (not image)

docker rm [docker id] # Delete docker container

docker rmi [docker image id] # Delete docker image

docker ps # View running docker s

docker ps -a # View docker running history

docker images # View the existing local docker image

docker inspect # View image internal details

docker cp # Copy files between host and container

docker stop # Stop container

docker search # Search for images in the remote warehouse docker hub

docker login # Log in to docker hub

docker push # Upload the local image to the docker hub

build custom docker

In addition to using the docker commit command, you can also write a Dockerfile (commonly known as this writing method) to define and save your own docker.

FROM xxx # Set base mirror

MAINTAINER xxxx # Famous document author / maintainer

RUN xxx # Execute command

COPY xxx # Copy file

ADD xxx # Add file

ENTRYPOINT xxx # Specify the execution entry for the container

EXPOSE xxx # Exposed port

CMD xxx # Execute the command and specify the execution entry for the container

WORKDIR # Specifies the path to run the command

ENV # Set environment variables for the environment in the container

USER # Specifies the user who executes the container

VOLUME # Specifies the volume on which the container hangs when it executes

Each line in the Dockerfile generates a new layer, that is, each line of commands will generate a new docker id.

Docker image provides a convenient environment for the deep learning training model. Users can build their own arbitrary environment without affecting the system environment. Here is the basic docker environment construction process.

• Task requirements: for example, I want to work in torch 1 Based on the docker of 7 (hereinafter referred to as base docker), install some packages you want to install to build a new docker.
• General logic: first pull the desired base docker locally, enter the docker locally, and then install the package you want in the docker. After all the packages are installed, exit the current docker, commit the docker to a new docker (your own docker), and then push your docker to the dockerhub repo.
• Specific steps:

1) Go to the dockerhub and find a torch 1 7 docker, pull down, for example, from torch official docker tags Find the torch 1 we want in it 7 development version. Once found, run locally:

docker pull pytorch/pytorch:1.7.1-cuda11.0-cudnn8-devel

The above operation is to pull the docker locally first.

2) Enter docker:

docker run --runtime=nvidia -it --name=test --ipc=host --net=host pytorch/pytorch:1.7.1-cuda11.0-cudnn8-devel

3) After entering docker, you can configure your environment to your heart's content.

4) After configuration, "exit" exits docker.

5) Commit the configured docker to your own docker (equivalent to copying a copy to your own docker).

 docker ps -a    # View and find the id of the configured docker
docker commit [docker id] geekyutao/foresee-rl:v0    # Geekyutao / forestee RL: V0 is the repo I created on dockerhub myself

6) Log in to the dockerhub locally, and then push your local docker to the dockerhub.

docker login -u [username] -p [password]    # If an error is reported, it is likely that the machine uses a docker certificate by default
# If an error is reported: delete the docker certificate under / usr/local/bin / and log in again
# sudo rm /usr/local/bin/docker-credential-pass

docker push geekyutao/foresee-rl:v0

7) Done! Delete the docker and docker image left during the operation. (basic operation with docker)

image