benchmark-runner
benchmark-runner copied to clipboard
PR merge process is overly complex and non-atomic
Describe the bug
This is multiple issues, but they are closely related and should be addressed in one fix:
- Following a PR merge, the github-action bot runs e2e testing, then does a second merge to the mainline to increment the version (in .bumpversion.cfg and setup.py). This results in (at least) two merges per PR, making the history unnecessarily confusing.
- If the e2e testing fails, the github-action does not bump the version. This requires an additional PR to be opened to manually increment the version.
The proper method would be for the pre-merge CI to run all desired tests and either merge the PR, including the incremented version, or not. The e2e CI requires building a new container image; that image should only be promoted following successful e2e testing and be combined with the PR merge.