[Mauiusers] Maui not honouring standing reservations

Dave Jackson jacksond at clusterresources.com
Thu Jul 7 15:14:04 MDT 2005


Drew,

  Can you force this failure if you restart Maui while a job of one type
or another is on one of the reserved nodes?

  Let us know and we can try to reproduce it.

Dave

On Thu, 2005-07-07 at 12:47 -0700, Drew Leske wrote:
> Hi again,
> 
> Anybody have any ideas at all?  Possible problems with my config?  Am I 
> misusing the directive or the concept?
> 
> I've seen other threads on this subject but no resolution.
> 
> Drew.
> 
> Drew Leske wrote:
> > Hi all,
> > 
> > I've been having problems with standing reservations.  Under certain 
> > conditions--it appears to happen only when Maui is restarted, and so 
> > only if jobs are queued at the time--Maui does not honour the standing 
> > reservations I have configured.
> > 
> > I am using Maui 3.2.6p11 with Torque 1.2.0p3.  (On an older cluster I am 
> > and have been using standing reservations with Maui 3.2.5p2 with OpenPBS 
> > without any troubles.)
> > 
> > The following snippet is relevant:
> > 
> > ----
> > 
> > SRCFG[express]         CLASSLIST=express,test
> > SRCFG[express]         HOSTLIST=c06b05,c06b06,c06b07,c06b08,c06b09
> > SRCFG[express]         PERIOD=INFINITY
> > 
> > SRCFG[mars]            CLASSLIST=mars
> > SRCFG[mars] HOSTLIST=c08b01,c08b02,c08b03,c08b04,c08b05,c08b06,c08b0
> > 7,c08b08,c08b09,c08b10,c08b11,c08b12,c08b13,c08b14,c09b01,c09b02,c09b03,c09b04, 
> > 
> > c09b05,c09b06,c09b07,c09b08,c09b09,c09b10,c09b11,c09b12,c09b13,c09b14
> > SRCFG[mars]            PERIOD=INFINITY
> > 
> > ----
> > 
> > Since implementing this I have been finding non-mars jobs on hosts in 
> > this list.  (This is a problem because these nodes are "politically" 
> > distinct, and we have an agreement with the owners that only their jobs 
> > will run here.)  The problem seems to happen when Maui is restarted.  If 
> > I have Maui report the reservations via "mdiag -r" it seems to recognise 
> > the reservations.
> > 
> > I have looked into using partitions as well, but the documentation seems 
> > to suggest that this is not the way to go.
> > 
> > Thoughts?  This has been driving me nuts for a while.
> > 
> > Cheers,
> > Drew.
> > 
> 



More information about the mauiusers mailing list