The drain time for a farm is too long to be considered for anything but
the most dire interventions.
You don't need to drain the queues, just temporarily stop the jobs. For
PBS/torque:
qsig -s STOP `qselect -q whetever -s R`
reboot, and restart the jobs
qsig -s CONT `qselect -q whatever -s R`
Martin.
> -----Original Message-----
> From: LHC Computer Grid - Rollout
> [mailto:[log in to unmask]] On Behalf Of
> Maarten Litmaath
> Sent: 27 June 2005 16:17
> To: [log in to unmask]
> Subject: Re: [LCG-ROLLOUT] rebooting a CE
>
>
> Jeff Templon wrote:
>
> > Hi *,
> >
> > We need to reboot our CE soon (kernel upgrade). Used to be if you
> > rebooted a CE machine, condor-G on the WMS would decide
> that your jobs
> > must all be dead, and restart them elsewhere.
>
> That is a long time ago. These days jobs in steady state are not
> affected by a reboot of the CE or the RB. Jobs in transit (e.g.
> just finishing) will fail.
>
> > What is the situation now? Do we need to drain queues
> before rebooting?
>
> Draining the queues is always a good idea.
>
|