The code should be tested, linted and formatted correctly to pass server-side checks (this repo's GitHub Actions).
The local testing invocation is
# provided that you have an appropriate Python environment
just testwhich currently runs some tests in the server folder,
including testing
that all models in server/models folder can be loaded
and perform inference on a simple example.
The code is expected to work for Python 3.11 and later versions
(they should be added to the tests as necessary as well as reflected in pyproject.toml).
The local formatting invocation is
# provided that you have formatters installed
just fmtFor Python files, this uses isort + black for formatting.
The local linting invocation is
# provided that you have linters installed
just lintThe linting checks are as follows:
- isort + black, as well as prettier report no diffs
- flake8 is happy for Python files
- markdownlint, yamllint and hadolint are happy for Markdown/YAML/Dockerfiles
The pylint command provides some statistics and refactoring suggestions, but is not used to fail the buid.
On a push, the GitHub action super-linter is automatically triggered. It runs a bunch of linters for many different languages.
If the local linting succeeds
while the remote linting fails,
it may be necessary to add new linters to
just lint or make sure all relevant files are covered by it.
Most of the tools can be configured with files in .github/linters folder.
The maximal line length in Python files is 88.
When the release tag is created, the server-side CI workflow should automatically build and push the image to the cloud registry.
The image is deployed to intents.cluster.megaver.se by updating the corresponding Kustomization which is then picked up by the Argo CD.
We only build the image for CPU which helps to keep its size manageable (about 300 megabytes).