Skip to content

Latest commit

 

History

History
315 lines (174 loc) · 4.44 KB

README.md

File metadata and controls

315 lines (174 loc) · 4.44 KB

Home

Docker usage

For the beginner

Docker architecture

docker architecture

Docker run

docker run을 통해 컨테이너를 실행

basic run command

$ docker run [OPTIONS] IMAGE [COMMAND] [ARG...]

basic options

Options:
  -d, --detach                         Run container in background and print container ID
  -e, --env list                       Set environment variables
  -i, --interactive                    Keep STDIN open even if not attached
  -p, --publish list                   Publish a container's port(s) to the host
      --rm                             Automatically remove the container when it exits
  -t, --tty                            Allocate a pseudo-TTY
  -v, --volume list                    Bind mount a volume
  -w, --workdir string                 Working directory inside the container

examples

simple run

$ docker run -it \
	{{base image}} \
	python train.py

background (detached)

$ docker run -dt \
	zironycho/pytorch:1.1.0-slim \
	python train.py

volume mount & working directory

$ docker run -it \
	-v $(pwd):/workspace \
	-w /workspace \
	zironycho/pytorch:1.1.0-slim \
	python train.py

remove container when it exits

$ docker run -it \
	--rm \
	zironycho/pytorch:1.1.0-slim \
	python train.py

publish port

$ docker run -it \
	-p 9999:8888 \
	zironycho/pytorch-jupyter:1.1.0-slim \
	jupyter notebook  --ip 0.0.0.0 --port 8888 --no-browser --allow-root
  • ip에 주의

env

$ docker run -it \
	-e MAX_EPOCH=100 \
	-v $(pwd)/tacotron:/workspace \
	-v /database:/database/data \
	-w /workspace \
	zironycho/pytorch:1.1.0-slim \
	python train.py

container exit이후

  • volume으로 mount하지 않는 이상 날아감

ngc(nvidia gpu cloud)

registry for nvidia optimized image

  • nvcr.io/nvidia/pytorch:19.05-py3
  • ...

Docker exec, logs, ps, stop, images

그 밖의 자주 사용하는 commands

docker exec를 컨테이너에 사용하여 접속

$ docker exec -it {{container_id}} bash

docker logs -f 를 사용해여 컨테이너 로그출력 & follow

$ docker logs -f {{container_id}}

docker ps를 사용하여 상태출력

$ docker ps
$ docker ps -a

docker stop을 사용하여 container exit

$ docker stop {{container_id}}
  • or rm

docker images를 사용하여 이미지 목록보기

$ docker images

Dockerfile

docker image를 만들기 위한 description

File

FROM zironycho/pytorch:1.1.0-slim

WORKDIR /workspace

COPY requirements.txt .
RUN pip install -r requirements.txt && rm -rf /root/.cache/pip
RUN apt get update
  • cache: file status

build image

$ docker build -t my-pytorch:v1 -f docker/Dockerfile .

Tips

  • ignore file/dir when it build

    • .dockerignore 를 이용하여 image를 생성할 때 layer에 포함시키지 않기
**/__pycache__
**/.idea
**/generated
.pytest_cache
**/*.pyc
**/*.wav
**/*.png
**/*.t7

  • 필요한 파일만 copy하기.

    • union file system
    • example) copy again, again, again

docker-compose

config file for multi-container applications

File

version: "3.7"

services:
  train-1:
		build:
			context: .
			dockerfile: ./docker/Dockerfile
		command: python train.py
    volumes:
    - .:/workspace
    - my-database:/database
    environment:
    - MAX_EPOCH=100
    - NVIDIA_VISIBLE_DEVICES=0

	train-2:
    image: zironycho/pytorch-jupyter:1.1.0-slim
    environment:
    - HELLO=WORLD
    - NVIDIA_VISIBLE_DEVICES=1,2
    ports:
    - "8888:8888"



volumes:
  my-database:
  	driver: local

docker and nvidia-docker

nvidia gpu를 사용하기 위해서는 nvidia-docker를 사용함.

$ docker run --runtime=nvidia \
	--rm \
	nvidia/cuda:9.0-base nvidia-smi

Refs: