Re: [Condor-users] limit execution time for a job

[Todd Tannenbaum <tannenba@xxxxxxxxxxx>]

Dominik Pich wrote:
> Hi,
> I studied the manual/tutorials and found a lot to describe _where_  
> and when to run jobs.
> I know about requirements and rank and later I also found the  
> priority option to categorize jobs based on 'importance'.
>
> Now what I want to do is:
> I want to express a maximum execution time (processor time) for a  
> given priority
> e.g. a job with a -10 prio has only 10s processor time

I am curious to know why you would want such a policy.

Usually folks would setup Condor to preempt (interrupt) a lower priority 
job if there is a higher priority job waiting, and then give the machine 
to the higher priority job.

This sort of setup prevents the undesirable situation where you have 
plenty of idle/available machines but still kick off jobs because they 
hit some  "maximum execution time".

> Is that feasible with condor? 

Sure.  Although I really recommend you state *what you really want* at a 
higher level.  Condor is more flexible/powerful than many other 
schedulers where everything revolves around time slots and job execution 
limits.  Are you really looking for max execution time based on job 
priority as you say, or are you really looking for a good way to 
interrupt lower priority jobs with higher priority jobs if they are 
available?  If the latter, there are better ways to achieve it with Condor.

However, if this is indeed what you really want, there are two ways to 
do it - either in the job's policy or in the machine's policy.  If you 
"control"/own the machines, and want the machines to enforce this 
policy, put limits like this into the machine (aka condor_startd - the 
condor_startd is the daemon that enforces the policy of a machine's 
owner) policy.  To read how to do this, see
  http://www.cs.wisc.edu/condor/manual/v6.8/3_5Startd_Policy.html
in the manual, but the short answer is you could place something like 
following into the condor_config file on the machine(s) where you want 
such a limit enforced:

   START = True
   WANT_SUSPEND = False
   PREEMPT = IfThenElse((JobPrio == -10) && \
                        ((CurrentTime - JobStart) > 10), True, False)
   KILL            = $(ActivityTimer) > 10


The above machine policy says this machine is always available to start 
jobs, that it never wants to suspend jobs, that it should try to 
gracefully shutdown (preempt) jobs after 10 seconds of execution if the 
job priority is -10, and that it should hardkill the job if it doesn't 
go away in 10 seconds after being asked nicely.

If you want to enforce this in the job's policy, check out the 
PERIODIC_REMOVE expression in your condor_submit job description file.

Hope this helps point you in the right direction,
regards,
Todd



> Maybe some's already done something  
> like that!?
>
> Regards,
> Dominik
>
>
> _______________________________________________
> Condor-users mailing list
> To unsubscribe, send a message to condor-users-request@xxxxxxxxxxx with a
> subject: Unsubscribe
> You can also unsubscribe by visiting
> https://lists.cs.wisc.edu/mailman/listinfo/condor-users
>
> The archives can be found at either
> https://lists.cs.wisc.edu/archive/condor-users/
> http://www.opencondor.org/spaces/viewmailarchive.action?key=CONDOR