[torqueusers] Dependencies among several jobs

nathaniel.x.woody at gsk.com nathaniel.x.woody at gsk.com
Mon Aug 28 12:12:09 MDT 2006


Vincenzo,

In what is probably a bad idea, when we want to use job dependencies, we 
creat all jobs in the held state so that we could collect the jobids, add 
those to the appropriate jobs and then removed the hold on them.  It was a 
lot of work, but it seemed to work allright.  We typically do all of our 
submissions using the pbs_python toolkit, so this is _relatively_ easy to 
do this way.  Don't know if that's any help to you our not.  In general, 
we had some problems with the job dependencies because as near as we could 
tell, the jobid that is passed as a dependency must be a jobid currently 
known to the system in order for that dependency to be enforced properly.  
If, for instance, you started a job with id; 42.yourcluster, and wanted to 
submit a number of jobs that should execute only when that finished, you 
needed to make sure that all jobs with that depenency are submitted BEFORE 
42.yourcluster completes.  I think the only moral there is to watch your 
synchronization, which may or may not be very controllable from a script 
file.

Best,
Nate





"Vincenzo Di Nicola" <vdinicol at hotmail.com> 
Sent by: torqueusers-bounces at supercluster.org
25-Aug-2006 18:44
 
To
torqueusers at supercluster.org
cc

Subject
[torqueusers] Dependencies among several jobs






Dear Torque users,

I have recently started to use Torque/Maui, but I have some doubts about 
job 
dependencies I hope you can clarify. It looks to me that the way to 
enforce 
them is by using "-W depend=beforeany:jobid" (or similar) in the qsub 
command or in the #PBS directive. But how can I know beforehand the job id 

of something that hasn't been submitted yet?

I would like to submit at once, say, 30 jobs with dependencies between 
each 
other. The only solution I have thought so far is to avoid "before" 
dependencies and create a script file: there I submit each single job, 
read 
its jobid, and use the "-W depend=afterany" option with regard to the 
previous job ids.
But is there a nicer way to enforce dependencies, maybe without having to 
rely on a jobid (but perhaps a job name)?

Thank you in advance,
Vincenzo Di Nicola

_________________________________________________________________
Call friends with PC-to-PC calling -- FREE 
http://imagine-msn.com/messenger/launch80/default.aspx?locale=en-us&source=wlmailtagline


_______________________________________________
torqueusers mailing list
torqueusers at supercluster.org
http://www.supercluster.org/mailman/listinfo/torqueusers


-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://www.supercluster.org/pipermail/torqueusers/attachments/20060828/8ba48112/attachment.html


More information about the torqueusers mailing list