CI/CD Refactor

QHub CI/CD Pipeline Improvements

General notes and ideas

• can we create custom image for the running
  • custom AMI for CI-Run?
• benchmark the size of docker images
  • reducing the size much more might be difficult given the required packages
  • initial tests with Docker Slim were unsuccessful
• multi-stage docker image builds
  • look at centos git branch for initial attempt
• only trigger workflow if they are touched

Test order, only proceed if previous step was successful

• lint
  • docker lint (hadolint ?)
  • python lint (black/flake8/etc ?)
  • terraform lint (find one ?)
• docker
  • build images / push to registry
  • is there a way to use cached/pre-existing docker images during the CI process (to save time)
    • diff the dockerfiles and related files
• python
  • pytests (stages)
  • test providers (qhub init, qhub render)
• kubernetes local test
• (docs)

Integration tests

• terraform infrastructure
  • full integration tests
    • review - why are they failing so often…
    • what are the actions we should take if these tests fail?

Next steps

• Review the action triggers
  • avoid duplicates (push and pull requests)
• Review Test Provider workflow
  • is there another way to implement?
  • add concurrency?

Resources

• https://docs.aws.amazon.com/toolkit-for-visual-studio/latest/user-guide/tkv-create-ami-from-instance.html
• https://github.com/wagoodman/dive
• https://docs.github.com/en/actions/using-jobs/using-concurrency

[magsol]

This is awesome, thank you so much for putting this together! Yeah let’s see what Gonzalo has to say when we meet with him today and go from there.

[viniciusdc]

I think the major points to get this working in a future release are:

• Decoupling docker images into base and custom configurations
• Custom testing classes suit
• Enhances to pytest during qhub deployment (tests between stages)