[GE users] Jobs being suspended incorrectly

opoplawski orion at cora.nwra.com
Fri Apr 16 22:41:07 BST 2010


On 04/16/2010 03:30 PM, opoplawski wrote:
> On 04/16/2010 03:23 PM, opoplawski wrote:
>> On 04/07/2010 08:29 AM, reuti wrote:
>>>
>>> can you please post the queue definitions.
>>
>> Okay, here we go:
>>
>> $ qstat -u \* | grep apapane
>>      16680 0.56000 run_cora.c dombroski    S     04/16/2010 15:01:34
>> mpi at apapane.cora.nwra.com          8
>>
>> $ qstat -f | grep apapane
>> admin.q at apapane.cora.nwra.com  BIPC  0/0/1          0.03     lx26-amd64
>> ivm.q at apapane.cora.nwra.com    BIPC  0/0/4          0.03     lx26-amd64
>> compute.q at apapane.cora.nwra.co BIPC  0/4/4          0.03     lx26-amd64    S
>> mpi at apapane.cora.nwra.com      PC    0/4/4          0.03     lx26-amd64
>>
>> Why does compute.q at apapane show 4 slots used?
>> Why is the job in S when it is in the mpi queue?
>
> Okay, I figured this out - this is an 8 cpu job - it put 4 in
> compute.q at mpi and 4 in mpi at apapane.  Hmmm, how to I avoid this?
>

qmaster messages:

04/16/2010 15:01:34|schedu|earth|W|Jobs 16680 & 16680 dispatched to 
master/subordinated queues 
"mpi at apapane.cora.nwra.com"/"compute.q at apapane.cora.nwra.com". Suspend 
on subordinate to occur in same scheduling interval. Policy conflict!

This: http://gridengine.sunsource.net/issues/show_bug.cgi?id=437
seems to indicate that I need to use "urgency" resource requests, which 
sounds like a pain.  Maybe I need to just remove the mpi PE completely 
from the compute.q.

-- 
Orion Poplawski
Technical Manager                     303-415-9701 x222
NWRA/CoRA Division                    FAX: 303-415-9702
3380 Mitchell Lane                  orion at cora.nwra.com
Boulder, CO 80301              http://www.cora.nwra.com

------------------------------------------------------
http://gridengine.sunsource.net/ds/viewMessage.do?dsForumId=38&dsMessageId=253750

To unsubscribe from this discussion, e-mail: [users-unsubscribe at gridengine.sunsource.net].



More information about the gridengine-users mailing list