acm-uiuc/nvdocker

nvdocker

nvdocker is library built on top of the docker-py python sdk to build and run docker containers using nvidia-docker.

Targets nvidia-docker2

Installation

• Install nvidia-docker

  https://github.com/NVIDIA/nvidia-docker#quickstart

• Install nvdocker

pip install nvdocker

Usage

These variables are already set in NVIDIA's official CUDA images.

visible_devices

This variable controls which GPUs will be made accessible inside the container.

• Possible values:
  • 0,1,2, GPU-fef8089b …: a comma-separated list of GPU UUID(s) or index(es),
  • all: all GPUs will be accessible, this is the default value in our container images,
  • none: no GPU will be accessible, but driver capabilities will be enabled.
  • void or empty or unset: nvidia-container-runtime will have the same behavior as runc.

driver_capabilites

This option controls which driver libraries/binaries will be mounted inside the container.

• Possible values
  • compute,video,graphics,utility …: a comma-separated list of driver features the container needs,
  • all: enable all available driver capabilities.
  • empty or unset: use default driver capability: utility.
• Supported driver capabilities
  • compute: required for CUDA and OpenCL applications,
  • compat32: required for running 32-bit applications,
  • graphics: required for running OpenGL and Vulkan applications,
  • utility: required for using nvidia-smi and NVML,
  • video: required for using the Video Codec SDK.

require

A logical expression to define constraints on the configurations supported by the container.

• Supported constraints
  • cuda: constraint on the CUDA driver version,
  • driver: constraint on the driver version,
  • arch: constraint on the compute architectures of the selected GPUs.

Expressions

Multiple constraints can be expressed in a single environment variable: space-separated constraints are ORed, comma-separated constraints are ANDed.

Multiple environment variables of the form rew are ANDed together.

cuda

The version of the CUDA toolkit used by the container. If the version of the NVIDIA driver is insufficient to run this version of CUDA, the container will not be started.

Possible values

• cuda>=7.5, cuda>=8.0, cuda>=9.0 …: any valid CUDA version in the form major.minor.

cuda_vesion

Similar to NVIDIA_REQUIRE_CUDA, for legacy CUDA images.

In addition, if NVIDIA_REQUIRE_CUDA is not set, NVIDIA_VISIBLE_DEVICES and NVIDIA_DRIVER_CAPABILITIES will default to all.

disable_require

Single switch to disable all the constraints of the form

Copyright and License

This project is released under the UIUC/NCSA License.

docker-py is licensed under the Apache License 2.0.

nvidia-docker and nvidia-container-runtime are licensed under the BSD 3-clause license.

nvdocker's Issues

Look at Luda

https://github.com/ryanolson/luda

Pull non local container images

acm-uiuc/gpu-cluster-backend#9

Pull hard coded paths into settings in the class

Probably some sort of setting struct

nvdocker/nvdocker/nvdocker.py

Line 18 in fe737c6

home_dir = "/vault/"

GPU Allocation as a Parameter

Set NVIDIA_VISIBLE_DEVICES

Verify nvidia-docker2 compatablity

Upload nvdocker to PyPI

Here are instruction on how to do it: https://packaging.python.org/tutorials/distributing-packages/#uploading-your-project-to-pypi

Update the interface

They changed the API again as GPUs have been integrated into docker upstream, either this package isnt needed anymore (check if docker-py supports gpus) or this package needs to be updated.

This needs to be more dynamic to reflect driver on host

nvdocker/nvdocker/nvdocker.py

Line 9 in fe737c6

volumes = {'nvidia_driver_387.12':{'bind':'/usr/local/nvidia', 'mode':'ro'},

Update README.md with better info

Add better documentation on how to use nvdocker.

Move nvidia docker environment variables to a user editable setting

nvdocker/nvdocker/nvdocker.py

Lines 7 to 9 in dd79257

	gpu_devices = ['/dev/nvidiactl', '/dev/nvidia-uvm', '/dev/nvidia1', '/dev/nvidia0']
	nvidia_driver = 'nvidia-docker'
	volumes = {'nvidia_driver_387.12':{'bind':'/usr/local/nvidia', 'mode':'ro'},

These were hardcoded in for the GPU cluster.

Integrate pygpuinfo to get system status in the rest API

https://github.com/TimZaman/pygpuinfo/blob/master/pygpuinfo.py

Add NCSA/UIUC Licenses

Use the standard NCSA/UIUC license ACM uses.

Example Usage

How to use nvdocker

Conda package

Both the docket module and the python api

Docstrings and documentation

We need doc strings and documentation generated from them for the package