The following example shows how OpenNMS can be extended to detect missed executions for recurring activities like backups or scheduled jobs.
The core functionality is implemented in the following Drools program (commented below):
First we need to define (at least) two UEIs: one for uei.<yournamespace>/job/recurring/Warning and one for uei.<yournamespace>/job/recurring/Normal. The events must be configured so that a Normal event clears any previous Warning. At the moment I feed these events into OpenNMS using syslog, but I am planning to replace syslog with my sendevent web-hook.
Each event carries three additional params (visible in the screenshot above):
- job or backupset : carries the job or backupset name, because one host can execute multiple jobs. It must be used in the event reduction key to achieve correct warnings resolution
- every : the value 'every' means it is an externally submitted event while 'missed' is used with events generated internally by expired Drools timers (missed executions). Every can be used as varbind filter to implement different notifications for 'regular' failures and missed execution failures
- interval: positive integer value indicating the repeating interval in hours (24 for daily jobs, 1 for hourly jobs, and so on)
Note that with this setup a successful execution will also clear any missed execution alarm.
As for the drools program the relevant parts are: the definition of the Execution fact. Execution carries the data necessary to identify the node and job plus the timer set to the interval value of the event.
The 2 following rules define the handling of the initial and subsequent events while the third handles the expiration of an interval. The code should be self-explanatory, ask in the comments if you need help.