Running Containers on HPC
Last updated on 2025-08-21 | Edit this page
Overview
Questions
- How is singularity different to Docker?
- How do I use my Docker images on a shared HPC?
Objectives
- Learn how to convert Docker images to SIF
- Distinguish the
runandexecsubcommands
You can find more detail about using Singularity in the singularity-introduction Carpentries workshop.
Singularity is a container engine, like Docker. However, unlike
Docker, container images are stored as single files called
.sif (Singularity Image Format). For a number of reasons,
Singularity suits shared High Performance Computing (HPC) environments
much better than Docker, so is valuable to learn if you work in these
environments. A related tool called singularity is a fork of Singularity
that generally has the same command line interface.
Singularity Command Line Interface
Like we did with Docker, try to work out what commands Singularity
has. Which one do you think is the equivalent of
docker run?
singularity run behaves similarly to
docker run, but as we will see, the arguments are somewhat
different.
Running Docker Containers
Since Singularity containers have their own file format, if we have a
Docker image we want to run, it first has to be converted. We can do
this using singularity pull. For example, we can pull the
container we previously pushed to Docker Hub:
This creates a file called alpine_python.sif in our
working directory. To run this container, we then use
singularity run:
Singularity Exec
If we want to modify the command run in the container, we have to use
singularity exec. For example, to make Python add numbers
like in our sum example, we could do: