Singularity lets users run applications and jobs inside a Linux container. This allows for applications that were built to run on a different distribution of Linux or have specific software dependencies that are not compatible with our cluster systems to run. Applications from the Singularity Hub or Docker Hub can run without needing to install anything.
Singularity is available on ARC and is installed on all nodes. The most current version of Singularity will be installed at every major upgrade. Currently, Singularity 3.8.0 is installed on all nodes. Older versions can be used from modules (Eg.
module load singularity/2.4.1).
Singularity can be used interactively. If you wish to enter a specific environment provided by a container, such as Ubuntu:
singularity shell docker://ubuntu
Adding Packages to a Docker Image
- Build a "Sandbox" of the dockerhub image -- we'll use the docker centos:7.6.1810 image as an example:
singularity build --sandbox /tmp/centos/ docker://centos:7.6.1810
- Modify the image -- add the emacs editor into the image:
- Note the "-f" which means to use fakeroot which gives root privilege inside the conainer (so you can use yum install)
singularity run -f -w /tmp/centos/ yum -y install emacs
- Package up the sandbox into a singularity 1 file image:
singularity build -f centoswithemacs.simg /tmp/centos/
- Now you can use the container image in a job (replacing <your computation> with the program inside the container that you want to run):
singularity run centoswithemacs.simg <your computation>
Converting a Docker image into a Singularity image
You can convert a Docker image that has already been pulled on your system into a Singularity image file with the following command:
# singularity build /tmp/output.sif docker-daemon://input:latest