Singularity

Container provide an executable environment comprised of pieces of software to facilitate portability and reproducibility of execution. You may build a container on your personal laptop, copy it to the high-performance cluster (like MOGON 2) and execute it there. The next day you may decide to run the same container at another computational facility. Instead of compiling code for some different cluster environment (different modules, different filesystems, etc.) from scratch, you just take the same container you've created the other day on your laptop.

It is an open-source project initiated at the Lawrence Berkeley National Laboratory. It is focused on:

• Integration over isolation by default. Easily make use of GPUs, high speed networks, parallel filesystems on a cluster.

• Portability and reproducibility of computations. The single file SIF container format is easy to transport and share.

• A simple, effective security model. You are the same user inside a container as outside, and cannot gain additional privilege on the host system by default. Where other container solutions allow the escalation of privileges onto the host system, this is prevented by Singularity.

We provide Singularity as an environment module:

$ module load tools/Singularity

You can use Singularity in your interactive job. After starting the job, ssh to the dedicated nodes and change into the folder with your Singularity container. Most of the time, it has extension sif, so your container may be called my-fluffy-cow.sif. To start the shell inside this container, simply execute

$ singularity shell my-fluffy-cow.sif

Cross your fingers and if everything goes right, your prompt will tell you that you are inside the container now.

You want just to check your container quickly? Or maybe execute a specific command as a part of your data analysis? In this case, you don't need to start a shell inside of the container. You can rather execute a specific command inside of the container. Let's take our example my-fluffy-cow.sif container:

$ singularity exec my-fluffy-cow.sif pwd

The part of the command following the name of the container will be executed in the container (in this case the working directory of the container will be printed onto the screen). After its successful execution your prompt will return to the cluster shell.

Other use cases are described in the official documentation here.

• It is possible to directly download a Docker image from the docker hub and convert it into Singularity image:

  $ singularity build my-fluffy-cow.sif docker://namespace/my-fluffy-cow

  More details can be found on official Singularity website here.

• If you have already your own image, which is not on the docker hub, you still can convert it. Here are the steps you have to do on your local machine, where you have a Docker image called my-fluffy-cow:

1. Find out the Docker image ID with docker images. In our example it is 362df94cb548:

   $ docker images
   REPOSITORY                  TAG       IMAGE ID       CREATED        SIZE
   my-fluffy-cow               latest    362df94cb548   1 hour ago     1.72GB

2. Create a tarball using Docker save command:

   $ docker save 362df94cb548 -o my-fluffy-cow.tar

3. Copy the tarball to MOGON 2 via scp command (here into the $HOME folder assuming you have your config setup as suggested here):

   $ scp my-fluffy-cow.tar mogon2:~/

After this, you'll have to wait couple of moments until the archive is copied. When it is done, login to the MOGON 2 and:

1. Load Singularity module with:

   $ module load tools/Singularity

2. Convert Docker tarball to Singularity image:

   $ singularity build my-fluffy-cow.sif docker-archive://my-fluffy-cow.tar