Docker_tutorial

Docker Tutorial for creating a GPU-enabled docker container from a git project without an existing Docker Image.

Downloading a base Image (Optional)

$ sudo docker pull nvidia/cuda:11.1.1-devel-ubuntu20.04

Use 'sudo' every time if you need a GPU enabled container. Without sudo the final command won't work and the container may not be accessible from root.

See the nvidia gitlab to get the right version for you.

You can also directly do the next step, but downloading the base image for your project first ensures it works before creating a new image.

Creating a Dockerfile

Now that the image exists locally, we can make a Dockerfile to create a new Docker image, adding for example Python, Pip, wget, and mini-conda.

Create a Dockerfile in a folder in your documents.

$ cd ~/Documents
$ mkdir Docker
$ cd Docker
$ gedit Dockerfile

Once this is done, let's add things to our Dockerfile :

# Pulling cuda image
FROM nvidia/cuda:11.1.1-devel-ubuntu20.04

# Making sure the following command are started from the image's root
WORKDIR /

# Installing base utilities (python3, pip, wget)
RUN apt-get update \
    && apt-get install -y build-essential \
    && apt-get install -y python3-pip python3-dev \
    && apt-get install -y wget \
    && apt-get clean \
    && rm -rf /var/lib/apt/lists/*

# Install miniconda
ENV CONDA_DIR /opt/conda
RUN wget --quiet https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh -O ~/miniconda.sh && \
    /bin/bash ~/miniconda.sh -b -p /opt/conda

# Put conda in path so we can use conda activate
ENV PATH=$CONDA_DIR/bin:$PATH

# Moving to /Project directory
WORKDIR /home/Project

The final line will create the 'Project' directory in the Docker image and the command line of the containers will start in this folder.

Building the image

To use the Dockerfile to create an image, from the ~/Documents/Docker folder:

$ sudo docker build -t <image_name>:<image_tag> .

You now have an image from your Dockerfile.

We can make sure by running:

$ sudo docker image ls

You should see your docker image, with its tag, ID, date, and size.

Creating a container mounted on a git repository

If this hasn't been done before, clone your git repository

$ cd ~/Documents
$ git clone <YOUR_GIT_REPO>

To create a container mounted on this git repository (that we'll call 'Project'):

$ sudo docker run --name <container_name> -v /home/user/Documents/Project:/home/Project --gpus all --shm-size 5G -it <image_name>:<image_tag>

The '/home/user/Documents/Project' path corresponds to your git repository.

The '/home/Project' is your container working environment.

Every change made to one folder will affect the other.

The '--gpus all' argument allows the container to use every GPUs available.

The -shm-size 5G argument is usefull to increase shared memory size, which is set at 64 MB by default and can be easily overflowed by neural network applications. In this example we'd have 5 GB of shared memory.

The -it will put you in a shell environment once the container is running. From there you can use ls and see your git repository's files and folders to make sure the container is mounted properly.

To see your containers:

$ sudo docker ps -a

-a will show every (running and stopped) containers.

Note that the conda environment may require you to exit and restart your container. You may want to try this if you get an error like: "CondaError: Run 'conda init' before 'conda activate'"

Start and stop a container

To exit the shell environment, type exit or do CTRL+D. This will automatically stop the container.

To start a container and get directly in a shell:

$ sudo docker start <container_name> -i

To stop a container:

$ sudo docker stop <container_name>

---

Congratulations, you reached the end of this tutorial !

---

Some useful links:

nvidia container toolkit ¹

nvidia container toolkit install ²

nvidia sample workload ³

Docker volumes ⁴

Containerize an app⁵

Dockerfile reference⁶

run ubuntu as a container⁷

run a docker imaage⁸

nvidia/cuda⁹

Index of /compute/cuda/opensource/image¹⁰

cuda supported tags¹¹

bind mounts¹²

docker logs¹³

remove all unactive containers¹⁴

start container¹⁵

install conda in docker¹⁶

install python in docker¹⁷

step-by-step guide to Dockerfiles¹⁸

seeing shm occupation¹⁹

Footnotes

1. https://github.com/NVIDIA/nvidia-container-toolkit ↩

2. https://docs.nvidia.com/datacenter/cloud-native/container-toolkit/latest/install-guide.html ↩

3. https://docs.nvidia.com/datacenter/cloud-native/container-toolkit/latest/sample-workload.html ↩

4. https://docs.docker.com/storage/volumes/ ↩

5. https://docs.docker.com/get-started/02_our_app/ ↩

6. https://docs.docker.com/reference/dockerfile/ ↩

7. https://www.makeuseof.com/run-ubuntu-as-docker-container/ ↩

8. https://docs.docker.com/reference/cli/docker/container/run/ ↩

9. https://hub.docker.com/r/nvidia/cuda?uuid=e5799b04-7e76-4ac6-8cc7-744741cd42fd%0A ↩

10. https://developer.download.nvidia.com/compute/cuda/opensource/image/ ↩

11. https://gitlab.com/nvidia/container-images/cuda/blob/master/doc/supported-tags.md ↩

12. https://docs.docker.com/storage/bind-mounts/ ↩

13. https://docs.docker.com/reference/cli/docker/container/logs/ ↩

14. https://docs.docker.com/reference/cli/docker/container/prune/ ↩

15. https://docs.docker.com/reference/cli/docker/container/start/ ↩

16. https://fabiorosado.dev/blog/install-conda-in-docker/ ↩

17. https://stackoverflow.com/questions/76340960/cuda-to-docker-container ↩

18. https://medium.com/@anshita.bhasin/a-step-by-step-guide-to-create-dockerfile-9e3744d38d11 ↩

19. https://github.com/pytorch/pytorch/issues/5040 ↩