Since GitLab 17 CI jobs with a Windows Runner can no longer be canceled

Status update - 2024-11-14

The following MRs were merged to fix the cancelling issue on Windows Runner:

• shell executor: Fix graceful termination of jobs on Windows (gitlab-runner!4808 - merged) • Nils Gladitz • 17.5: Implement graceful termination of Windows Runner using pwsh shell. Unfortunately, it doesn't gracefully terminate powershell jobs but they end up being forcefully terminated after 10 min
• kubernetes executor: All the known issues were fixed by the MRs below:
  • Fix graceful termination of jobs on Windows (gitlab-runner!4808 - merged) • Nils Gladitz • 17.5
  • Cancel stage script upon job cancellation in at... (gitlab-runner!4813 - merged) • Romuald Atchadé • 17.5
  • [k8s] Terminate PowerShell Script children proc... (gitlab-runner!5081 - merged) • Romuald Atchadé • 17.6

The issue will be closed. If anything else arises, please feel free to create a separate issue. 🙇🏿

Overview

@allison.browne @ajwalker I've upgraded my self-hosted instance and runners to version 17 now but the issue seems to remain. I manually (via the UI) cancel one of my build jobs and it goes to the canceling state but the spawned build process does not get killed. I continue to see new build sub-processes spawning on the system with the runner and the job log keeps filling up. Minutes later after the job has successfully run to completion I see:

2024-05-16T18:32:23.135801Z 00O section_end:1715884343:step_script
2024-05-16T18:32:23.135801Z 00O+WARNING: script canceled externally (UI, API)
2024-05-16T18:32:23.136924Z 00O section_start:1715884343:after_script
2024-05-16T18:32:23.136924Z 00O+Running after_script
2024-05-16T18:32:23.897287Z 01O Running after script...
...

Status

• Fix for pwsh was merged in 17.5
• Fix for powershell scheduled for 17.6