enmyj / gpu_docker

dockerfiles for our dc office gpu development machine

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

gpu instance image dockerfiles

each directory in this repository is a separate context for a docker image. These images allow users of our gpu box to launch isolated docker containers, with GPUs attached, for the ultimate experience in computationTM. The images are regularly re-built and live on the gpu box (and hopefully dockerhub, soon).

layers

there are several images that are simple layers on top of other images, so here's a brief rundown of the ones we have defined so far:

  1. tensorflow
  • The official tensorflow Docker images, as defined here
  1. eri_python
  • starts with the tensorflow base image and installs the most commonly used python libraries
  1. eri_dev
  • a development environment baesd on eri_dev with a jupyter lab server running on an exposed port, as well as basic volume mounting for shared data
  1. eri_python_r
  • installs Rstudio and the most commonly used R libraries on top of the eri_python image
  1. eri_dev_p_r
  • a development environment based on eri_python_r with a jupyter lab server and an Rstudio server running on exposed ports, as well as basic volume mounting for shared data

making updates without automation

basically, run andrew's build script:

python3 build.py

the build script has a few optional arguments to configure different build parameters. access the help menu for more info:

python3 build.py --help

rotating build logs are saved to /var/log/gpu_docker/build.log (with fallback ./logs/build.log) to aid in debugging failed builds.

old instructions

until we have set up a nightly or automated build, please take care to increment versions on images and tag things appropriately. we should be able to rebuild all images based on some overall git version tag someday, but not today!

for now the process should be roughly as follows: for each image in the dependency chain of the "innermost" docker image you have updated,

  1. docker build --no-cache -t IMAGE_TAG_NAME .
  2. docker tag NEWSHANUMBER IMAGE_TAG_NAME:vX.Y.Z

About

dockerfiles for our dc office gpu development machine


Languages

Language:Dockerfile 58.5%Language:Python 29.6%Language:R 8.8%Language:Shell 3.2%