artemsokolovdh/docker-test repository overview

⁠docker-test: Automatic building of parallel Docker images

The example demonstrates how to set up parallel Docker image builds from Dockerhub. In the overall setup, we have two versions of the code: 1) a fixed version tagged v1.0 and 2) the master branch. We would like to derive two Docker images for each of these codebases: an image based on tensorflow and an image based on tensorflow-gpu.

To solve this, we create hooks/build which augments the default image name provided by Dockerhub with a "-gpu" suffix. The parallel build is then constructed and pushed to Dockerhub by hand. On the Dockerhub side, we create two automatic build rules: one for v1.0 and the other for master. Each build rule will execute hooks/build, which in turn will correctly generate two parallel Docker images.

⁠On the GitHub side:

1. Dockerfile: Define the base image variable TF_IMAGE and give it a default value.
2. hooks/build: Use --build-arg to override the default value of TF_IMAGE in parallel image builds.
3. hooks/build: Use $IMAGE_NAME, which is provided by Dockerhub, to dynamically generate a tag for parallel image builds.
4. hooks/build: Build and push parallel images by hand. End the script by building the default image.
5. Create a v1.0 release. In this example, the difference between v1.0 and most recent master is the version of Python.

⁠On the Dockerhub side:

Create two automatic build rules:

• master -> latest
• v1.0 -> 1.0

⁠Testing

docker pull all four images. We expect that 1.0 and 1.0-gpu will have python2, while latest and latest-gpu will have python 3. The *-gpu images should have GPU support.

docker run -it --rm --runtime=nvidia artemsokolovdh/docker-test:1.0 bash
root@0449637fe38e:/# python --version
Python 2.7.15+
root@0449637fe38e:/# ls /dev/nvidia*
ls: cannot access '/dev/nvidia*': No such file or directory

docker run -it --rm --runtime=nvidia artemsokolovdh/docker-test:1.0-gpu bash
root@ae32b88f5cc6:/# python --version
Python 2.7.15+
root@ae32b88f5cc6:/# ls /dev/nvidia*
/dev/nvidia-uvm  /dev/nvidia-uvm-tools  /dev/nvidia0  /dev/nvidiactl

docker run -it --rm --runtime=nvidia artemsokolovdh/docker-test:latest bash
root@c02020c6d02c:/# python --version
Python 3.6.8
root@c02020c6d02c:/# ls /dev/nvidia*
ls: cannot access '/dev/nvidia*': No such file or directory

docker run -it --rm --runtime=nvidia artemsokolovdh/docker-test:latest-gpu bash
root@0e652fb49598:/# python --version
Python 3.6.8
root@0e652fb49598:/# ls /dev/nvidia*
/dev/nvidia-uvm  /dev/nvidia-uvm-tools  /dev/nvidia0  /dev/nvidiactl
root@0e652fb49598:/# exit