Containers/GPUs Exercise 1.1: Building a Singularity Container

The objective of this exercise is to learn the very basics of building your first Singularity image.

Getting a Sylabs Cloud Account

In this exercise we will use Singularity's equivalent of DockerHub, called Sylabs Cloud, to build and publish our image. To get started, first sign up for a free account at https://cloud.sylabs.io/

Image Definition File

The image definition file contains the instructions on how to build the image. Usually there is a base image, and a set of commmands which will set up the environment as you want it. Examples of commands include:

• Package system tools (apt/yum/...) to install/upgrade specific packages
• Langauage package tools (pip/...) to install language specific modules
• Steps to build software from source

You can find a full description of the image defintion format in the Singularity manual.

In Sylabs Cloud, click on Remote Builder in the top navigation bar. Remove the default definition file, and paste the following in:

Bootstrap: docker
From: opensciencegrid/osgvo-ubuntu-20.04:latest

%post
    apt-get update -y
    apt-get install -y \
            python3-pip \
            python3-cowsay
    python3 -m pip install numpy

This definition file installs a few packages from apt (that is, Ubuntu prepared packages), and one package via pip. Using package from the vendor is great as those packages are generally well tested. However, not everything is packaged by the OS vendors, and sometimes you need specific versions of a package and then using language tools like pip, or building from source, can be good solutions.

The definition file above has two problems we have to address before building. The first problem is that there is no Ubuntu package for python3-cowsay. The second problem (albeit minor), is that numpy might be better installed with apt. The fix here is to swap and the names of those two. Replace python3-cowsay with python3-numpy and replace numpy with cowsay. The final definition file should look like:

Bootstrap: docker
From: opensciencegrid/osgvo-ubuntu-20.04:latest

%post
    apt-get update -y
    apt-get install -y \
            python3-pip \
            python3-numpy
    python3 -m pip install cowsay

The last step is to name the Repository. The naming should start with your project, a slash, and then a name of your choosing. For example, if your project field says bob, you could name the repository bob/first-image.

Now hit Build and watch the image build.

Once the build is complete, the repository should be changed to "public". This is important as it will allow us to download the image directly to login04/login05 in the next exercise. Click on Dashboard on the top, find your repository in the list, and click it. In the middle you will see a Project Visibility box. Click Make Public.