[torquedev] Setting extra_resc causes problem

Joshua Bernstein jbernstein at penguincomputing.com
Thu Jun 18 12:10:37 MDT 2009

Al Taufer wrote:
> The man page for pbs_server_attributes has the following description for the extra_resc:
> "Add additional string-type job resources. They have no effect within TORQUE and are only advisible to the scheduler. They can not be used for resources_default/min/max."
> The feature has apparently been in for awhile. It was put into the 2.3 branch in Feb 2008.

Maybe its been forgotten about because its only documented in the man page and 
not in the somewhat more obvious online documentation. Perhaps we can have it 
added to the online docs? Shall I open a bug for this?



> Al Taufer
> Cluster Resources, Inc.
> ----- "Joshua Bernstein" <jbernstein at penguincomputing.com> wrote:
>> What is extra_resc supposed to do? Is this a new feature?
>> -Joshua Bernstein
>> Senior Software Engineer
>> Penguin Computing
>> Al Taufer wrote:
>>> There seems to be a problem with both the 2.3 and trunk branches. 
>> It seems like the resource_list can get messed up for jobs. The
>> scenario is as follows:
>>> echo "sleep 30" | qsub -l walltime=33:33:33
>>> qmgr -c 's s extra_resc = garbage'
>>> echo "sleep 30" | qsub -l walltime=33:33:33
>>> The first job gets the correct walltime while the second job does
>> not.  The problem seems to be that setting the extra_resc calls
>> init_resc_defs() which recreates the svr_resc_def values. This
>> prevents chk_svr_resc_limit() from matching up the resources.
>>> The fix seems to be in attr_fn_resc.c function find_resc_entry(). 
>> If we change the following line from:
>>>     if (pr->rs_defin == rscdf)
>>> to:
>>>     if (!strcmp(pr->rs_defin->rs_name, rscdf->rs_name))
>>> then it seems to work fine.
>>> Is there some reason we are checking to see if the rs_defin points
>> to the same place as rscdf instead of just comparing the rs_name?
>>> Al Taufer
>>> Cluster Resources, Inc.
>>> _______________________________________________
>>> torquedev mailing list
>>> torquedev at supercluster.org
>>> http://www.supercluster.org/mailman/listinfo/torquedev

More information about the torquedev mailing list