[GE users] Job dependency and job_name

Iwona Sakrejda isakrejda at lbl.gov
Thu Apr 8 01:14:21 BST 2004



Ron Chen wrote:

> I got the answers below by submitting test jobs to my
> cluster --
> 

Thanks a lot for your reply, you motivated me to do my
own tests and  I also realized that my questions were not quite clear.


> 
>>A job name is not unique, so for how long does a
>>memory of a job name live in the SGE system? 
>>
> 
> SGE won't allow you to depend on non-unique job names:
> 
> % qsub -hold_jid sleep sleep2
> denied: non-ambiguous jobnet predecessor "sleep"
> (found jobs 6 and 8)


I was able to reproduce that and it is a part of my problem.



>>But what if that name was used in the
>>past? How far back into the past will SGE look?
>>
> 
> As long as the jobname does not match those in the job
> list, SGE will assume that it is done.
> 


This SGE "list" is a problem for me, because after the job is
done it still lingers on the list (and you can see it with
the -s z option of qstat). If I submit the next job with the
same name and then try to submit a third one that would
depend on that name I get a message:
pdsflx005 113% qsub -hold_jid apple testit.csh
denied: non-ambiguous jobnet predecessor "apple" (found jobs 114175 and 114176)
The first job is already done, but still on the list.
I could try to run some more experiments but there are too many
parameters I can think of. So the question is for how long
do job linger on this list after completion?

I am not trying to design a circular dependency. I even do not
know why one of my users wanted to know that. I already contacted him
and asked what he was trying to acomplish. But then got curious
about the initial question...

Thanks a lot,

Iwona




---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscribe at gridengine.sunsource.net
For additional commands, e-mail: users-help at gridengine.sunsource.net




More information about the gridengine-users mailing list