pykube-models

Pydantic models generated from the Kubernetes API, for the pykube k8s client.

Rationale

The official Python client for Kubernetes is based on the OpenAPI Client Generator. I find the auto-generated client to be quite un-ergonomic to work with, and it doesn't include any of the modern Python type-hinting features that make for easy to use IDE integrations.

Pykube provides a simple Python client with some nice Django-like utilities, which make it quite nice to work with for simple workflows. However it doesn't have any knowledge of the Kubernetes OpenAPI spec, just a few basic classes that are hand-coded like Deployment. These hand-coded classes take an obj dictionary and so there's also no type hinting.

Pydantic provides nice type-hint-enabled data models which can be generated from OpenAPI specs. It should be possible to generate a set of Pydantic models once per k8s version, and use these to render the spec dict that Pykube expects.

TODO

• Figure out Pydantic constructor type hinting
  • Should be able to do this in mypy like: https://pydantic-docs.helpmanual.io/mypy_plugin/#generate-a-signature-for-model__init__
    • Couldn't get this set up immediately
  • Can also get this done with PyCharm plugin: https://pydantic-docs.helpmanual.io/pycharm_plugin/
• Full mypy testing?
• Figure out / report datamodel-code-generator bug: no packages for some generated modules, just the leaf modules.
• Set up a pypi package?
  • https://flit.readthedocs.io/en/latest/
  • https://packaging.python.org/tutorials/packaging-projects/
• Define dependencies (poetry? setup.cfg?)
  • datamodel-code-generator
  • pytest
  • mypy
• Test strategy?
  • VCRPY?
  • GH Actions?
  • Revisit pytest - had to add conftest.py to src/models to get pytest to find the src/ modules.
• Investigate baking in the apiVersion/kind to the base models; shouldn't need to re-specify.
  • Maybe could use a different base class? In datamodel-code-generator: --base-class BASE_CLASS Base Class (default: pydantic.BaseModel)
• Strategy for versioning? Probably want to support different k8s API versions.