Best Practices

Given the limits outlined previously, there are most definitely best practices that we can provide to both use Singularity Hub and build containers. Importantly, if you don’t follow these best practices it can mean slowing down your research if you exceed the Singularity Hub rate limits for API or other views.

Recipe Naming

Don’t store two equivalently named recipes in different subfolders of your repository. It would mean possibly building two different containers with the same unique resource identifier, and then mixing them up.

Container Building

The following are good practices for building containers.

Practice container hygiene

Keeping Singularity Hub open and free for use, without limitations, is dependent on having an affordable cost structure. We pay for every file in storage, and every build, so it is not ideal to keep files or collections that won’t be needed.

Don’t be root

Generally don’t install to the /root folder, as this might have issues when the container is executed in user space.

Write commands in post

It’s best to put all install commands directly in the %post section, as opposed to within scripts that are executed from it. Depending on how you wrote your bash script, it could be that an error in a script was not passed on to the build.

Choose a building strategy that matches your development

You should choose a build option (manual, automated on commit, or deployment) based on your development strategy. If you have testing set up, deployment would be logical, as containers will be built after your tests have passed. If you commit to your repository regularly, you might consider a manual build to only trigger when you have a version ready.

use tags instead of collections for image versions

Let’s say that I have an image called AnalysisGPU and I’ve built it for two versions. Instead of creating two repostories on Github, AnalysisGPU_version1 and AnalysisGPU_version2, if I have two recipes, Singularity.version1 and Singularity.version2 under one repository called AnalysisGPU, then my images will be available as AnalysisGPU:version1 and AnalysisGPU:version2`