[torqueusers] cpu-time limit specified, but wall-time enforced
narora at Princeton.EDU
Fri Jun 2 13:55:29 MDT 2006
We have torque-2.0.0p2 and maui-3.2.6p13 based setup. There are 2
execution queues, distinguished by different cpu-time limits. There is
one route queue which routes the jobs based on the requested cpu-time.
The queue definitions have appropriate resources_max.cput and
resources_min.cput declarations. And the users are required to specify
-l cput=<time> option to qsub.
The jobs get submitted to the right queues, based on the cput parameter
value. But then, Torque/Maui seem to be enforcing wall-time instead and
a job is killed if the resources_used.walltime exceeds Resource_List.cput !!
For example, I submit a job with qsub -l cput=1:0:0 pbs-script. And the
job takes more than 1hr in wallclock time, while the cpu-time usage is
still less than 1hr. This job would be killed with a "MOAB_INFO: job
exceeded wallclock limit" message.
We have not specified any walltime parameters in queue/server
definitions or during job submission.
While the job is running, qstat reports both cputime and walltime usage.
Whereas, checkjob only reports walltime usage:
Job Id: 37414
resources_used.cput = 00:00:00
resources_used.mem = 2568kb
resources_used.vmem = 9264kb
resources_used.walltime = 00:37:10
checking job 37414
Creds: user:narora group:staff class:medium qos:DEFAULT
WallTime: 00:37:48 of 1:00:00
where, clearly Maui has set the max allowed walltime to be the same as
the cput value I specified to qsub !!
Can someone please suggest what's going wrong here?
More information about the torqueusers