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).
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:
- The official
tensorflow
Docker images, as defined here
- starts with the
tensorflow
base image and installs the most commonly usedpython
libraries
- a development environment baesd on
eri_dev
with ajupyter lab
server running on an exposed port, as well as basic volume mounting for shared data
- installs
Rstudio
and the most commonly usedR
libraries on top of theeri_python
image
- a development environment based on
eri_python_r
with ajupyter lab
server and anRstudio
server running on exposed ports, as well as basic volume mounting for shared data
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.
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,
docker build --no-cache -t IMAGE_TAG_NAME .
docker tag NEWSHANUMBER IMAGE_TAG_NAME:vX.Y.Z