What it is for
A job that runs at 3am produces no error when it fails to run. It simply does not happen, and nobody finds out until the Monday report is missing. A scheduled job flips that around: the job says it ran, and meulog alerts on the absence of the signal.
How to wire it
- At the end of your job, one HTTP call to meulog with its name. One line, no library.
- The job registers itself on the first heartbeat: name, project, and from then on it exists in the list.
- Adjust the interval (how often to expect the heartbeat) and the tolerance (how much delay is normal before it becomes an alert).
What you should see
- The list of jobs, with the last heartbeat and delay for each.
- The overdue ones highlighted at the top: that is the information that justifies the screen existing.
- The heartbeat history per day, which shows the gap: the week it stopped running appears as a break in the strip.
The event
An overdue job fires batida_perdida. It is the only alert source in
meulog born from nothing happening, which is why its
alert rule usually looks different from
the others: there is no volume to set a threshold on.