 |
 |
Hi Stephen, I'm successfully draining our nodes down using the method you describe. Just over a week ago, I issued setres -s 17:00:00_05/18 -e 15:00:00_05/19 -n block1 "node0[0-2].|node03[0-5]" 'showres' confirms this reservation: block1.0 User - 7:11:43 1:05:11:43 22:00:00 35/140 Mon May 18 17:00:00 I see that the reserved nodes are empty (in contrast to the rest of the cluster) aside from the odd 6-hour job and one 3-day job that will complete shortly. Our queues are all still open, but longer jobs are not running on the reserved nodes. Have you checked your reservation exists (and is sane) with 'showres'? Cheers, Mike. 2009/5/18 Stephen Childs <[log in to unmask]>: > Hi all, > > We have queues with different lengths set up (e.g. one day, three day, 30 > mins, etc.) When it comes to scheduling downtime these need to start > draining at different times (e.g. three day queues three days before > downtime starts etc.) We have been manually disabling the relevant queues at > the right times (qmgr -c "s q threeday enabled=false") which then trickles > through to the info sys just fine. However, this is fiddly as you have to > remember to set the downtimes for the various queues. > > It seemed like it would be easy to handle this in maui by setting up a > reservation that corresponds to the downtime (e.g. setres -s 10:30:00_05/20 > -e 15:00:00_05/20 ALL). Maui will then not execute jobs that wouldn't finish > before downtime starts (but it will queue them). However, the queues stay > open and are advertised as such in the info sys so jobs keep coming in even > on the "long" queues. > > Perhaps it would be better after all to set up at/cron jobs to shut the > various queues at the appropriate times? > > Just wondering if anyone else had some experience with this? > > Stephen > -- > Dr. Stephen Childs, > Research Fellow, EGEE Project, phone: +353-1-8961797 > Computer Architecture Group, email: Stephen.Childs @ cs.tcd.ie > Trinity College Dublin, Ireland web: http://www.cs.tcd.ie/Stephen.Childs >