Currently, most of the run-ci label checking is done via static checks in each workflow, for example:
|
if: github.event_name != 'pull_request' || contains(github.event.pull_request.labels.*.name, 'run-ci') |
This approach has some drawbacks:
- After we disable the
labeled trigger for CI workflows, adding a new run-ci label to a PR no longer updates the PR info for existing runs. To pick up the new label, we have to push a new commit. Using the “Re-run” button alone keeps the old (outdated) PR info.
- Other backends’ CI workflows cannot easily reuse the same fine-grained PR gate logic.
To resolve this, we’d like each CI to reuse the shared logic in ./.github/workflows/pr-gate.yml and align their trigger rules with it.
The inputs of pr-gate can be defined with #13491
Progress:
Better CI failure panel:
Currently, most of the
run-cilabel checking is done via static checks in each workflow, for example:sglang/.github/workflows/pr-test-rust.yml
Line 75 in 890368c
This approach has some drawbacks:
labeledtrigger for CI workflows, adding a newrun-cilabel to a PR no longer updates the PR info for existing runs. To pick up the new label, we have to push a new commit. Using the “Re-run” button alone keeps the old (outdated) PR info.To resolve this, we’d like each CI to reuse the shared logic in
./.github/workflows/pr-gate.ymland align their trigger rules with it.The inputs of pr-gate can be defined with #13491
Progress:
Better CI failure panel:
run-cilabel in all workflows (the label would not be updated when retriggering)pr-gateinto the finish check's needs.