[GE users] qmaster memory leak

Andreas Haas Andreas.Haas at Sun.COM
Mon Nov 8 12:41:57 GMT 2004


It is planned for the next update. You can track progress
via issue #1289.

Andreas

On Fri, 5 Nov 2004, Eric Andresen wrote:

> I can confirm this behavior on 6.0u1 and the latest CVS checkout.
>
> This is unfortunate, as I liked the way that subordinate queues handled
> automatic suspension of lower priority queues.
>
> Are there any plans to rework the subordinate system to (significantly)
> reduce memory consumption? As it stands, having three queues, 'high',
> 'normal', and 'low', with low as a subordinate of both, and normal as a
> subordinate of high uses approximately 890MiB of memory when each queue
> has 50 nodes with two slots apiece.
>
> Thanks,
> --
>    Eric Andresen
>    Systems Administrator
>    Mars Space Flight Facility
>    Arizona State University
>    eandres at mars.asu.edu
>    (480) 727-8471
>
> On Tue, 2004-10-05 at 15:50, Mike Brown wrote:
> > Andreas,
> >
> > I've looked into the problem from another angle, and it doesn't seem to
> > be a memory leak.  I tried recreating the queue that caused the problem,
> > and noticed that when I put fewer cpus in it, the memory usage comes
> > close to exceeding that of the machine, but does not increase.
> >
> > I think the problem is with subordinate queues.  As I increase the
> > number of queues subordinated, memory increases rapidly.  If I remove
> > the subordinate restriction, I am able to declare many different cluster
> > queues.  My current (5.3p5) configuration is similar to below, with
> > higher level queues subordinating lower level ones:
> >
> > level3 (subordinates level2, level1)
> > level2 (subordinates level1)
> > level1
> >
> > As I try to repeat this in 6.0u1, I have approximately 100 hosts at each
> > level, and hit 2GB memory usage while populating the 3rd level.
> >
> > Thanks for your help,
> >
> > Mike
> >
> >
> > Andreas Haas wrote:
> >
> > >You might try starting it again and use 'strace' to trace into
> > >it when starting qmaster manually from within a shell. Make
> > >sure you do a
> > >
> > >   setenv SGE_ND
> > >
> > >otherwise strace is pointless since would daemonize ...
> > >
> > >Andreas
> > >
> > >
> > >On Mon, 4 Oct 2004, Mike Brown wrote:
> > >
> > >
> > >
> > >>Linux 2.4.18-27.7.x i686
> > >>
> > >>Rayson Ho wrote:
> > >>
> > >>
> > >>
> > >>>On which platform??
> > >>>
> > >>>Rayson
> > >>>
> > >>>
> > >>>
> > >>>
> > >>>
> > >>>
> > >>>>I'm trying to configure SGE 6.0u1 for the first time.  I configured
> > >>>>users and cluster queues and ran successful test jobs.  After this, I
> > >>>>created a clone of an existing cluster queue and suddenly qmaster was
> > >>>>unresponsive.  I left it alone for several minutes, did a ps listing,
> > >>>>and saw one of the qmaster processes in the Z state.
> > >>>>
> > >>>>
> > >>>>
> > >>>>
> > >>>---------------------------------------------------------
> > >>>Get your FREE E-mail account at http://www.eseenet.com !
> > >>>
> > >>>---------------------------------------------------------------------
> > >>>To unsubscribe, e-mail: users-unsubscribe at gridengine.sunsource.net
> > >>>For additional commands, e-mail: users-help at gridengine.sunsource.net
> > >>>
> > >>>
> > >>>
> > >>>
> > >>
> > >>
> > >
> > >---------------------------------------------------------------------
> > >To unsubscribe, e-mail: users-unsubscribe at gridengine.sunsource.net
> > >For additional commands, e-mail: users-help at gridengine.sunsource.net
> > >
> > >
> >
> >
> >
> > ---------------------------------------------------------------------
> > To unsubscribe, e-mail: users-unsubscribe at gridengine.sunsource.net
> > For additional commands, e-mail: users-help at gridengine.sunsource.net
>
>
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: users-unsubscribe at gridengine.sunsource.net
> For additional commands, e-mail: users-help at gridengine.sunsource.net
>
>

---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscribe at gridengine.sunsource.net
For additional commands, e-mail: users-help at gridengine.sunsource.net




More information about the gridengine-users mailing list