Ian Chesal wrote:
You can use: SYSTEM_PERIODIC_REMOVE = ((JobStatus == 2) && ((CurrentTime - EnteredCurrentStatus) > 123456789)) Where 123456789 is the maximum run time you want a job execute for in seconds. The JobStatus part makes certain it only removes jobs that are currently running. This works for in my 6.8.x system.
You might also consider configuring your cluster to have lax preemption requirements but a long MaxJobRetirementTime, like:
PREEMPTION_REQUIREMENTS = (((CurrentTime - EnteredCurrentState) > (1 * (10 * 60)) && RemoteUserPrio > SubmittorPrio * 1.2))
but MaxJobRetirementTime=(60*60*72)In this case, jobs would get preempted but not replaced unless they were running more than 3 days.
- B