[GE users] using RQS to limit run time

prentice prentice at ias.edu
Mon Oct 18 21:45:32 BST 2010


    [ The following text is in the "utf-8" character set. ]
    [ Your display is set for the "ISO-8859-10" character set.  ]
    [ Some characters may be displayed incorrectly. ]

prentice wrote:
> reuti wrote:
>> Hi,
>>
>> Am 18.10.2010 um 21:51 schrieb prentice <prentice at ias.edu>:
>>
>>> I would like to limit job run time based on job size, similar to  
>>> what's
>>> already been discussed here:
>>>
>>> http://gridengine.sunsource.net/ds/viewMessage.do?dsForumId=38&dsMessageId=246571
>>>
>>> So count me among those who would like a 'where' or 'if' clause in the
>>> RQS syntax. While creating multiple queues with defined h_rt values  
>>> will
>>> work, that seems a bit kludgy and more work than should be necessary.
>>>
>>> Related to this, can h_rt be limited using RQS? I created the  
>>> following
>>> rule to limit runtime to only 10 seconds for user prentice, but it
>>> doesn't seem to work:
>>>
>>> {
>>>   name         time_test
>>>   description  test to limit h_rt per_user
>>>   enabled      TRUE
>>>   limit        users prentice to h_rt=10
>>> }
>>>
>>> When I wrote this rule, I set h_rt=00:00:10, and after saving, SGE
>>> changed it to rt=10. Does that mean it thinks h_rt is 10 hours instead
>>> of 10 seconds?
>> if you define h_rt=01:00:00, what is the limit then transformed to?
> 
> $ qconf -srqs time_test
> {
>    name         time_test
>    description  test to limit h_rt per_user
>    enabled      TRUE
>    limit        users prentice to h_rt=3600
> }
> 
> 
> So it looks like it displays the values in seconds. Thanks for helping
> me clarify that.
> 
>> Keep in mind that limits defined in an RQS will be checked, but not  
>> enforced (unless you request it with "-l" to ?qsub?).
> 
> So unless I do qsub -l h_rt=10:00:00, my RQS will be ignored? That's not
> very useful. Or at least not as useful as I was hoping.
> 
> Prentice
> 

So I did qsub -l h_rt=10:00:00 as above, and the job is in state 'qw',
and the sched_info shows this:

"cannot run because it exceeds limit "/////" in rule "time_test/1"

Is there any way to have it kill the job instead of queuing it forever?

Prentice

------------------------------------------------------
http://gridengine.sunsource.net/ds/viewMessage.do?dsForumId=38&dsMessageId=288240

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



More information about the gridengine-users mailing list