If GitHub Actions fails once, start by asking what changed. A rerun is useful when you want another attempt at the original event. If you fixed the code or workflow in a new commit, you need a run that picks up that commit. GitHub reuses the original event’s commit SHA and ref on a rerun.
Choose the smallest useful retry
Rerun failed jobs if one job failed because a dependency service briefly went offline. GitHub also reruns jobs that depend on it. Jobs outside the retry can supply outputs from the earlier attempt. If the suspect is a successful upstream job or you want to observe every step again, rerun all jobs.
Example: a failed test versus a fixed test
Suppose an iOS test job timed out while fetching a package, but the checkout and build steps completed. Rerunning the failed jobs lets you test the transient-failure theory against the same commit. If you changed the package configuration to fix the problem, trigger a new run on the updated commit instead. Compare the commit SHA before reading the result.
Verify the retry
Open the run summary, confirm the attempt and read the first meaningful error in the failing step. GitHub documents a 30-day window for reruns and requires repository write access. For a release workflow, review the jobs that may run again before triggering it.
GitHub documentation: https://docs.github.com/en/actions/how-tos/manage-workflow-runs/re-run-workflows-and-jobs