Github Actions: Re-intentar automáticamente los jobs de CI que un reclamo de spot instance mató

GitHub Actions jobs of CI that were killed by a spot instance reclaim can be automatically retried, according to the source material. The issue occurs when AWS takes away the instance used for the CI runner mid-task, causing GitHub to display a red X on any running step. This is identical to a test failure. The workflow includes a listener that triggers after any CI or deployment run finishes.

If a failed run is identified, it fetches the logs of the failed jobs and looks for specific lines produced only by the runner's host. If a match is found, it calls the rerun-failed-jobs function. However, there are three cases where this action is not taken - if there's a more recent run of the same workflow and branch, if re-running an older attempt would only exacerbate the concurrency group, or if a third attempt still shows a shutdown signal, indicating the pool has fallen and a human should examine the Auto Scaling Group.

The source material also discusses the importance of checking the ASG for potential outages and the need to adjust maxCapacity to ensure a replacement can be added when needed.