Nathan's Question on Adapting to a New Scheduler via submitters

[Jonathan Clark]

I think there's a problem with the PBS monitoring script—even though
my job seems to have finished successfully, the task failed
(presumably because qstat was called after the job finished):

[nschneid@cab Baseline.baseline]$ cat ducttape_stdout.txt
[nschneid@cab Baseline.baseline]$ cat ducttape_stderr.txt
Job queued at Sun Dec 2 15:06:10 EST 2012
Job began running at Sun Dec 2 15:06:10 EST 2012
qstat: Unknown Job Id 4905.cab.ark.cs.cmu.edu
[nschneid@cab Baseline.baseline]$ cat ducttape_torque_jobid.txt
4905.cab.ark.cs.cmu.edu
[nschneid@cab Baseline.baseline]$ cat ducttape_exit_code.txt
153

I used mark_done to get it to proceed to the next task.

Also, my 'test' task jobs are being processed one at a time. Shouldn't
they all be sent to the scheduler in parallel?

-Nathan

---------


(Fixed the qstat problem—it involved checking exit status with
tracejob, and moving all the qstat calls after that.)

-Nathan

----------

Glad to hear you got the issue ironed out. Adapting to different
scheduler configurations can be rather annoying, but enough bash
hackery usually gets the job done.

-Jon