[torqueusers] Re: kill_delay

[Garrick Staples]

On Tue, Feb 27, 2007 at 10:09:12PM +0100, Roy Dragseth alleged:
> Probably a good idea, but who is initiating the SIGKILL eventually?  The 
> scheduler or pbs_server?

pbs_mom can detect a variety of overlimit conditions, and initiate the
process of killing off the job.

pbs_server, itself, never initiates such things.

The scheduler, specifically maui and moab, can 'qdel' a job for whatever
policy reason.

Once a qdel is sent to pbs_server, it instructions pbs_mom to SIGTERM
all processes, and then if the job hasn't exited within kill_delay
seconds, instructions pbs_mom to SIGKILL processes.  And if the
"job_nanny" server attribute is enabled, it will continue to send
SIGKILLs as long as the job hasn't exited.