Maity: Board

process: lead or ops may cancel a hanging CI run after a fixed limit

tsk-kr6f8w50r0cb open

Oliver adopted this at the 2026-11-03 jour fixe (req-gec2x1vgxrpx). Until now, every cancel needed a human.

Record the rule in context/process.md (next to the job-timeout entry from #1253):

• Who: the team lead or ops.
• When: a job runs past a fixed limit without finishing. Proposed limit: 15 minutes over the longest green run in the history (today about 76 min for hub-postgresql, 54 for hub-sqlite, 44 for hub-mariadb), so before the 90-minute timeout-minutes fires.
• How: gh run cancel <run> on that run only, then gh run rerun --failed once the cause is known or as a retry. Never on a run that is still making progress.
• Report: the run id, job and elapsed time go to the team chat in the same minute.
• Alternatives considered: a cancel always by a human (status quo, rejected by Oliver), relying on the 90-minute timeout alone (keeps main blocked for up to 90 minutes).

Notes

1. Lotse (lotse) · 2026-11-03

   Same context/process.md PR, also adopted on 2026-11-03 (req-8m0ja218geax): after the lower PR of a stack is squash-merged, the upper PR merges origin/main in instead of rebasing, then retargets to main. Rejected alternative: rebase (replays the old commits as conflicts).