Project Thoth Naming Convention Schema for Images

• Status: proposed
• Date: 2021-Jun-17

Technical Story: As Thoth goal to provide curated stack and images, it would be nice to have a convention for naming of the images.

Context and Problem Statement

Image names are important for branding and let others identify easily a specific image they need. For example “I want to work on computer vision project with Tensorflow, what stack and image should I use?” Having a trusted well maintained source of images with clean naming convention can help on that.

Considered Options

• s2i-{application} standard name currently, but not everyone knows what S2I is.
• odh-{application} for branding/marketing having ODH in front seems to be the best solution, in this way the name will be shorter as well.
• ps-{application} as it shows what our intention is: we want to provide a curated/predictable software stack, it might be used by ODH or RHODS or others, it might use S2I or other technology. Moreover helps from pipeline creation point of view, because the length of repo name on quay can crate issues.

These names are for the core repository name, then overlays will allow for combinations of libraries based on other criteria, for example ml_framework and/or hardware:

`ps-nlp`
├── overlays                    # Overlays structure for builds
│   ├── ps-nlp-tensorflow          # NLP image with TensorFlow ML framework
│   ├── ps-nlp-tensorflow-gpu      # NLP image with TensorFlow ML framework for GPU
│   ├── ps-nlp-pytorch             # NLP image with Pytorch ML framework
│   ├── ps-nlp-pytorch-gpu         # NLP image with Pytorch ML framework for GPU
│   └── ps-nlp-scikit-learn        # NLP image with Scikit-learn ML framework
└── ...

Decision Outcome

Selected option: ps-{application} as it shows what our intention is: we want to provide a curated/predictable software stack, it might be used by ODH or RHODS or others, it might use S2I or other technology. Moreover helps from pipeline creation point of view, because the length of repo name on quay can crate issues.

Positive Consequences

• users can immediately select an image based on the application they want.
• using overlays we can have a variety of combination, not just for ml_framework

Negative Consequences

• N/A