Podman

Podman is a container platform focused on security. It is very robust and is supported by Red Hat. Podman supports images in the Open Container Initiative (OCI) format – these are also supported by Docker, Kubernetes and Singularity among others.

To get more information about Singularity, the following manual pages are available:

man podman
man podman-run
man podman-build

Using Containers with Podman

To use a container non-interactively with Podman, the basic syntax is:

podman run --rm CONTAINER_NAME ARGS

To use a non-interactively with Podman, the basic syntax is:

podman run -it --rm CONTAINER_NAME ARGS

To use a container non-interactivel with a GPU with Podman, the basic syntax is:

TODO

To make a folder on the host available within the container environment, you must additionally use a bind-mount. For example, to make your home folder available at the same place inside a non-interactive coontainer:

podman run --volume "$HOME:$HOME" --rm CONTAINER_NAME ARGS

The containers will be pulled automatically if they available from DockerHub and not available locally.

Building Containers with Podman

With Podman, containers can be built by using a Dockerfile. See Dockerfile reference for the full syntax.

To build a container, create a folder for the Dockerfile, say CONTAINER_FOLDER. Then store or write a Dockerfile at CONTAINER_FOLDER/Dockerfile. Then, to build the container, use the basic syntax is:

podman build -t CONTAINER_NAME CONTAINER_FOLDER

Examples of Building and Using Podman Containers

Some examples of building and using containers are given below:

Build and Use a Container with Tensorflow and Various Python Packages (Podman)

To extend the tensorflow/tensorflow image on DockerHub) by installing various Python packages inside it, you could use the dockerfile below:

From tensorflow/tensorflow:2.17.0-gpu
RUN pip install pandas matplotlib scikit-learn pyyaml keras biopython numba viennarna keras_tuner

Assuming the above is stored as a file called Dockerfile in a folder called tensorflow-build, it can be built with:

podman build -t tensorflow-plus tensorflow-build

The container can then be executed with:

podman run -it --rm tensorflow-plus

If you have a python script in your home folder called myscript.py, you could create (and submit – see Accessing Compute Resources) a job file such as the following to run the python script in the container on a CPU node.

#!/bin/bash
#SBATCH -p cpu

cd
podman run --volume "$HOME:$HOME" --rm tensorflow-plus "$HOME/myscript.py"

To run the same python script on a GPU node with a GPU available, you could use the following: