Re: [Condor-users] Getting Failed Jobs to Restart

[Jaime Frey <jfrey@xxxxxxxxxxx>]

SIGQUIT and SIGKILL are treated specially for standard universe.  
Condor interprets them as meaning the job was kicked off a machine  
with or without a checkpoint, respectively. The other signals result  
in the job leaving the queue, and all signals result in the job  
leaving the queue in the vanilla universe.

With on_exit_remove, you can control when Condor lets the job leave  
the queue. For example, you can Condor to leave the job in the queue  
if it terminates with exit code 3. The expression is limited to  
information available in the job ad.

 -- Jaime

On Aug 26, 2005, at 10:30 AM, Avi Flamholz wrote:

> Yes, but what if the job fails do to some internal exception - ie
> uncaught exception in c++. Can Condor do anything there? I noticed
> that in the standard universe, if I sent a running process sigkill
> condor would register it as idle and then restart it in the next
> negotiation cycle.
>
> Can this be done in the standard universe?
>
> Thanks for the response Jamie.
> -Avi
>
> On 8/26/05, Jaime Frey <jfrey@xxxxxxxxxxx> wrote:
>
> > On Aug 25, 2005, at 6:11 PM, Avi Flamholz wrote:
> >
> >
> > > I am running a simple python script to test my condor  
> > > configuration -
> > > obviously in the vanilla universe. It simply computes the value  
> > > of pi
> > > for a while, times itself, and prints what machine it's on. I  
> > > made it
> > > run for a while so that I would have a chance to monitor it on the
> > > remote machines.
> > >
> > > The desired functionality is this - If a job fails (dies due to some
> > > exception or failure) Condor should restart it from scratch. I have
> > > notices that in the standard universe, condor can do this. Can it  
> > > also
> > > be done in the vanilla universe? What are the limitations. For  
> > > the end
> > > task it is unlikely that I will be able to relink the code, as it is
> > > legacy material and there are few people around who know enough  
> > > pascal
> > > to know what it's doing. So I would like to be able to support this
> > > functionality in the vanilla universe.
> >
> > Condor automatically restarts all jobs that don't complete due to a
> > Condor failure or being kicked off a machine. The standard universe
> > does one better by restarting the jobs where they left off, instead
> > of at the beginning.
> >
> > You can also tell Condor to restart jobs that complete on their own
> > with the on_exit_remove expression.
> >
> > +---------------------------------- 
> > +---------------------------------+
> > |            Jaime Frey            |  Public Split on  
> > Whether        |
> > |        jfrey@xxxxxxxxxxx         |  Bush Is a  
> > Divider              |
> > |  http://www.cs.wisc.edu/~jfrey/  |         -- CNN Scrolling  
> > Banner |
> > +---------------------------------- 
> > +---------------------------------+
> >
> >
> > _______________________________________________
> > Condor-users mailing list
> > Condor-users@xxxxxxxxxxx
> > https://lists.cs.wisc.edu/mailman/listinfo/condor-users
>
> _______________________________________________
> Condor-users mailing list
> Condor-users@xxxxxxxxxxx
> https://lists.cs.wisc.edu/mailman/listinfo/condor-users

+----------------------------------+---------------------------------+
|            Jaime Frey            |  Public Split on Whether        |
|        jfrey@xxxxxxxxxxx         |  Bush Is a Divider              |
|  http://www.cs.wisc.edu/~jfrey/  |         -- CNN Scrolling Banner |
+----------------------------------+---------------------------------+