Job scripts submitted on Glenn, Oakley, or Ruby all go a submit filter before reaching the resource manager, Torque. A bug has been discovered in our submit filter which prevents jobs with the substring "-A" in their PBS directives from being parsed and interpreted correctly. This behavior is only observed in jobs where the "-A" substring is specified in the job script -- If an argument with the "-A" substring is passed to qsub from the command line this behavior is not present.
Take the following job script:
#PBS -lnodes=1:ppn=12 #PBS -Nthisisajobname-Andsoisthis #PBS -lwalltime=1:00:00 sleep 10
This job script defines a 1 node job for 1 hour of walltime that is named "thisisajobname-Andsoisthis".
If one attempts to submit the above job the following error will be given:
$ qsub qsub_filter_group_test.pbs qsub filter $Revision: 490 $, $Date: 2014-09-23 17:02:52 -0400 (Tue, 23 Sep 2014) $ Group: ndsoisthis is not valid. Please choose from: appl qsub: Your job has been administratively rejected by the queueing system. qsub: There may be a more detailed explanation prior to this notice.
This is occuring because the submit filter is interpreting the string after the "-A" substring to be the group to charge against, when in reality it is part of the job name. This behavior is present when the "-A" substring is present in any PBS directive.
Our temporary workaround is to not use the substring "-A" within the PBS directives unless it is being used to define the group to charge against.
We are working on a better long term solution, and will update this known issue once one is developed.