luda

ludicrously awesome [w]rapper for nvidia-docker

Free software: MIT license
Documentation: [coming shortly]

pip install luda

Features

Opinionated wrapper for docker/nvidia-docker designed to provide Singularity-like functionality to Docker images.
Best used for container images that run DL/HPC-like jobs, not suited for long-running daemons or services that require root.
Volume mounts a /bootstrap volume and overrides the container image ENTRYPOINT to map the host USER, UID and GID [future work] into the container. Current the docker commandline is echo'ed to the terminal on container startup.
Automounts the current working directory on the host to /work inside the container. /work becomes the current working directory inside the running container. --work can be used to specify an alternative default working directory for use inside the container; it will be mounted to /work.
Automounts $HOME on the host to /home/$USER inside the container. --home option.

Quickstart

luda nvidia/cuda:8.0-devel

This the equivalent of the following docker command:

nvidia-docker run --rm -t -i  \
  -v /Users/ryan/Projects/luda/luda/bootstrap:/bootstrap:ro \
  --entrypoint /bootstrap/init.sh \
  --env HOST_USER_ID=501 \
  --env HOST_GROUP_ID=20 \
  --env HOST_USER=ryan \
  --env HOST_GROUP=staff \
  -v /Users/ryan:/home/ryan \
  -v /Users/ryan/Projects/luda:/work \
  --workdir /work \
  nvidia/cuda:8.0-devel /bin/bash

This launches a new container based on the nvidia/cuda:8.0-devel image; however, the magic happens in the bootstrapping, where the host user that launched the container is created inside the container on launch (entrypoint).

This is exceptional convenient for development as your current working directory is mapped into /work which then becomes the active working directory inside the contanier. Edits made inside the container are written as the USER/UID of the host user.

Volumes

luda intercepts the -v/--volume option and provides convenience methods similar to docker-compose in that relative paths are supported. If no

# absolute path readonly
--volume /path/data:/data:ro

# relative path readonly
--volume /path/data:/data:ro

# relative path, no internal path --> mount internal at `/{{ basename(hostpath) }}`
# mounts $PWD/data --> /data inside the container
# --volume data

# same as above, but readonly
# --volume data::ro

Home Directory

The user's home directory is a special case which mounts the user's home directory on the host to /home/$USER in the container. This option is enabled by default, but can be disabled by passing --no-home on the commandline.

Current Working Directory

luda will map the current working directory from which the luda command was executed on the host to /work in the container and override the container's working directory to /work. This behavior can be overridden by passing a volume mount or disabled by passing None to to the --work option.

Examples:

# mounts the current working directory on the host to `/my-working-dir`
# in the container; `/my-working-dir` become the default working directory
--work .:/my-working-dir

# mounts `~/other-dir` to `/other-dir` in the container; `/other-dir`
# becomes the default working directory in the container
--work ~/other-dir

# use the working directory as specified by the container image
--work None
--work none

Abbreviations

You can set up abbreviations for commonly used URLs by including an abbreviations key in the yaml config file. By default, luda includes the nv: which expands to nvcr.io/nvidia/{0}, where {0} is the remainding portion of the image name after the abbreviation.

in config.yml

abbreviations:
  nv: nvcr.io/nvidia/{1}

Usage nv:tensorflow:17.04 expands to nvcr.io/nvidia/tensorflow:17.04:

luda nv:tensorflow:17.04

Displays

luda --with-display nvidia/cuda:8.0-devel

todo: show opengl containers

Docker

luda --with-docker nvidia/cuda:8.0-devel

Templates

Templates provide an easy way to extend container images with pre-defined content. Assume I have the following Dockerfile defined in ~/.config/luda/templates/dev.

RUN apt-get update && apt-get install -y --no-install-recommends \
        vim sudo python-dev python-pip && \
    rm -rf /var/lib/apt/lists/*

RUN pip install luda

The developer option --dev is a special case of --template dev. Running the following commands performs a one-time extensions of the nvidia/cuda:8.0-devel image with the Dockerfile above. The new images generated will be luda/nvidia-cuda-8.0-devel:dev or luda/{{ base_image }}:{{ template }} where base_image has all / and : replaced with -.

luda --dev nvidia/cuda:8.0-devel

luda --template dev nvidia/cuda:8.0-devel

The first time this command is invoked luda/nvidia-cuda-8.0-devel:dev will be created. Subsequent invocation will either update the image if either the base image (nvidia/cuda:8.0-devel) or the template directory (~/.config/luda/templates/dev) has detected changes.

Acknowledgements

Thanks to Deni Bertovic's blog. luda provides very similar functionality but does not require specially crafted base images and wraps the details of the docker command-line.

The project contains a copy of su-exec. Copyright reproduced below.

The MIT License (MIT)

Copyright (c) 2015 ncopa

Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all
copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
SOFTWARE.