This is an example of using schema.org to label a Dockerfile. Specifically, we will walk through examples that use already defined schema.org specifications, and other examples that use specifications for containers that are under development. The goal is to demonstrate utility in being able to label a Dockerfile (a container recipe) and then different ideas for how this could be done.
Labeling our containers and container recipes with schema.org metadata means that we can serve it alongside the containers, and have the properties indexed by Google Search. This means that regardless of where your recipe is hosted (a registry or Github pages) your container and recipes will (finally) be discoverable. Note that the official definition for a container "recipe" has not been integrated into schema.org, and we have had good discussion with the OCI community about a proper name. Any of the following would be a good contender:
Thing > CreativeWork > SoftwareSourceCode > BuildDefinition
Thing > CreativeWork > SoftwareSourceCode > BuildInstructions
Thing > CreativeWork > SoftwareSourceCode > BuildPlan
Thing > CreativeWork > SoftwareSourceCode > BuildRecipe
Thing > CreativeWork > SoftwareSourceCode > Configuration
Thing > CreativeWork > SoftwareSourceCode > ContainerConfig
Thing > CreativeWork > SoftwareSourceCode > ImageDefinitionJust to provide you with more examples (for the same position in the hierarchy) I have used several of the names above in the examples below. The only significantly different is the "SoftwareSourceCode" that cuts away some of the additional recipe metadata. This is also the only specification that is production schema.org.
Toward this goal, we are going to walk through several simple examples representing a Dockerfile as each of:
- ImageDefinition extraction shown here.
- ContainerRecipe extraction shown here.
- SoftwareSourceCode extraction shown here. This is a specification already defined in schema.org, but does little to describe a Dockerfile in detail.
For each of the above, the metadata shown is also embedded in the page as json-ld (when you "View Source.") The examples are minimal in that I didn't do any special parsing of the containers to extract more meaningful tags for software, this would be valuable to do to improve the search.
Each folder above includes a script to extract (extract.py),
a recipe to follow (recipe.yml), and the specification in yaml format (in the
case of a specification not served by production schema.org). When you view
the github pages of this
repository served by the master branch, you will see that the index.html
in each folder serves a simple page to show the extracted metadata. The
Dockerfile is the recipe that we aim to describe using schema.org,
and is for the container toasterlint/storjshare-cli on Docker Hub. I chose it pretty randomly because it started with
"toast."
Before running these examples, make sure you have installed the module (and note this module is under development, contributions are welcome!)
pip install schemaorgTo extract a recipe for a particular datatype, just run the extract.py file
in the corresponding folder. You can look at any of the extractors to get a gist
of what we do to generate the final metadata for Github pages. Generally we:
- Read in a specific version of the schemaorg definitions provided by the library
- Read in a recipe for a template that we want to populate (e.g., google/dataset)
- Use helper functions provided by the template (or our own) to extract
- Extract, validate, and generate the final dataset
The goal of the software is to provide enough structure to help the user (typically a developer) but not so much as to be annoying to use generally.
If I am a provider of a service and want my users to label their data for my service,
I need to tell them how to do this. I do this by way of a recipe file, in each
example folder there is a file called recipe.yml that is a simple listing of required fields defined for the entities that are needed. For example, the recipe.yml in the "SoftwareSourceCode" folder tells the parser that we need to define
properties for "SoftwareSourceCode" and an Organization or Person. For example.
with the schemaorg Python module
I can learn that the "SoftwareSourceCode" definition has 121 properties,
but the recipe tells us that we only need a subset of those
properties for a valid extraction.
The specification.yml file is only provided in the ContainerRecipe folder because this isn't a production specification provided by schema.org. For those that are published (e.g., SoftwareSourceCode and Dataset) the definitions are provided in the python module.
This is the code snippet that shows how you extract metadata and use the schemaorg Python module to generate the final template page. This file could be run in multiple places!
- In a continuous integration setup so that each change to master updates the Github Pages metadata.
- Using a tool like datalad that allows for version control of such metadata, and definition of extractors (also in Python).
- As a Github hook (or action) that is run at any stage in the development process.
- Rendered by a web server that provides Container Recipes for users that should be indexed with Google Search (e.g., Singularity Hub).
Check out any of the subfolders for: