Problem

• We always run all CI GitHub Actions, even for components which are not changed
• E.g. when you change some Ruby code we run also the unit tests for the Rust code which is actually not needed
• That wastes resources and the CI takes longer time than needed
• You might get a false alarm if tests in the other part fail e.g. because of some network problem

Solution

• Split the single GitHub CI workflow into smaller pieces
• Use the paths option so the tests are run only when relevant files are modified, otherwise the tests are skipped
• For the code coverage it uses a new coveralls.io feature which allows inheriting the previous code coverage for the parts which are not tested in the current run
• See the coveralls documentation:
  • https://coveralls.io/better-monorepo-support
  • https://docs.coveralls.io/parallel-builds
  • https://github.com/marketplace/actions/coveralls-github-action

Notes

• The result is that when you edit a file which is not actively used by any part (like updating README.md file) then no unit test will run and there will be no green check mark ✅ in the pull request. Just be prepared for that... 😉
• The paths matching is set for any change in the directory besides some listed exceptions. If there are some more files which do not influence the tests they can be added later. But it is safer to run one more unnecessary build than to skip one important build.

Testing

• Tested manually
• See test runs in my fork: web frontend, Ruby service, Rust part