Competition Docker Image

The competition docker image defines the docker environment in which the submissions of the competitions or benchmarks are run. Each competition can have a different docker environment, referred by its DockerHub name and tag.

Default competition docker image¶

The default competition docker image is codalab/codalab-legacy:py37. More information and base images are available here: https://github.com/codalab/codalab-dockers

Set up another image¶

You can select another docker image:

In the competition.yaml file, using docker_image: username/image:tag
In the editor field "Competition Docker image" as shown in the following screenshot:

If the default image does not suit your needs (missing libraries, etc.), you can either:

Select any existing image from DockerHub
Create your own image from scratch
Create a custom image based on the CodaLab image (more information below)

Custom image based on CodaLab image¶

If you wish to create a custom image based on the Codalab image, you can follow the steps below:

Preliminary steps¶

1) Install Docker 2) Sign up to DockerHub

Method 1: update the image from a container¶

1) Start a container using the base image:

docker run -itd -u root codalab/codalab-legacy:py39 /bin/bash

2) Identify the running container ID using docker ps 3) Enter inside the container:

docker  exec -it -u root <CONTAINER ID> bash

4) Install anything you want at the docker container shell (apt-get install, pip install, etc.)
5) Exit the shell with exit
6) Push the new version to your DockerHub account:

docker commit <CONTAINER ID> username/image:tag
docker login
docker push username/image:tag

Method 2: update the Dockerfile and re-build¶

1) Download the Dockerfile of the base image: https://github.com/codalab/codalab-dockers/blob/master/legacy-py39/Dockerfile

2) Edit the file to include any library or program you need

3) Build the image

docker build -t username/image:tag .

4) Push it to your DockerHub account

docker login
docker push username/image:tag