Apptainer

Apptainer (formerly known as Singularity) is a containerization tool designed for High-Performance Computing (HPC). Unlike Docker, it allows running containers without requiring root privileges, making it ideal for secure environments like supercomputers and university clusters.

Official Resources

Website: Apptainer.org
GitHub Repository: Apptainer GitHub
Documentation: Official Docs
Community: Apptainer Slack

Video Tutorials

Basic Usage

If you have issues with using/creating an Apptainer image, unload the XALT module from your environment. To clear the entire environment, run: ml -f purge. To unload only the XALT module, use: ml -f -XALT.

Running a Container

apptainer run docker://ubuntu

This will pull an Ubuntu image from DockerHub and run it.

Shell Access

apptainer shell docker://ubuntu

Opens an interactive shell inside the container.

Execute a Command

apptainer exec docker://ubuntu echo "Hello from Apptainer!"

Building and Editing Images

Creating an Apptainer Definition File

Example my_container.def file:

Bootstrap: docker
From: ubuntu:20.04

%post
    apt update && apt install -y python3

%runscript
    echo "This is my custom Apptainer container!"

Building an Image

apptainer build my_container.sif my_container.def

Editing an Image

Apptainer images (.sif format) are immutable, so you need to create a writable sandbox:

apptainer build --sandbox my_container/ my_container.def

Modify it:

apptainer shell --writable my_container/

Rebuild it into a .sif file:

apptainer build my_container_final.sif my_container/

Using MPI in Apptainer

Running MPI Inside a Container

Your definition file should install MPI:

Bootstrap: docker
From: ubuntu:20.04

%post
    apt update && apt install -y mpich

Build and run with MPI:

mpirun -np 4 apptainer exec my_mpi_container.sif mpi_hello_world

Running MPI From Host Into Container

Ensure your host has an MPI runtime and mount it into the container:

mpirun -np 4 apptainer exec --bind /usr/lib/x86_64-linux-gnu/openmpi/ my_mpi_container.sif mpi_hello_world

GPU Support

Installing CUDA Inside a Container

For CUDA support, install the necessary drivers inside the image:

Bootstrap: docker
From: nvidia/cuda:12.0-runtime-ubuntu20.04

Run with GPU:

apptainer exec --nv my_cuda_container.sif nvidia-smi

Tips and Tricks

Use --fakeroot to build without sudo:

apptainer build --fakeroot my_container.sif my_container.def

Use --bind /host/path:/container/path to access host directories.
Convert Docker images to Apptainer format for offline use:
```
apptainer pull ubuntu.sif docker://ubuntu
```

Debug with increased verbosity:

SINGULARITY_DEBUG=1 apptainer run my_container.sif

Conclusion

Apptainer is a powerful tool for running containers in HPC environments. With proper setup, it can efficiently run MPI workloads, GPU-accelerated applications, and custom software stacks while maintaining security and performance.

Apptainer

On this page