[torquedev] keep completed jobs

David B Jackson jacksond at clusterresources.com
Sat Jan 7 01:23:18 MST 2006


Garrick,

  Awesome!  Nice job!

Dave



> On Sat, Jan 07, 2006 at 12:36:41AM -0700, David B Jackson alleged:
>> Garrick,
>>
>>   Did you fix the 'KEEPCOMPLETED' code?  It was in before from another
>> developer but would SEGV on occasion (once per 1000 jobs/once per day?)
>> We never got back to isolating the failure and backed it out.
>
> I think so.  Their were a few problems.  The work task created wasn't
> tied to the job, so it wasn't getting cleaned up correctly.
> set_statechar() didn't know about the new state, so *statechar was being
> read of the end of the string.  PJobState[] didn't have an entry for the
> new state.  And PJobSubState[] was missing a comma at number 29, so
> logged messages for all higher substates were off by one.
>
>
>>   I will roll another snaphsot now and we will start testing in house.
>
> I'll throw a 1000 jobs at this and see what happens.
>
>>
>> Thanks!
>> Dave
>>
>> > Dave, can you roll a snapshot please?
>> >
>> > I've just checked in the "keep completed jobs" support.  This keeps
>> jobs
>> > around for a bit after they've exited in state "C".  Set
>> > TORQUEKEEPCOMPLETED in $PBSHOME/pbs_environment, restart pbs_server,
>> and
>> > watch the magic!
>> >
>> > 'qdel' for a completed job is not allowed, but an admin can 'qdel -p'
>> > it.
>> >
>> > Currently this is hardwired at 300 seconds, but if everything seems to
>> > work,
>> > I'll throw in a server attribute.
>> >
>> >
>> > --
>> > Garrick Staples, Linux/HPCC Administrator
>> > University of Southern California
>> > _______________________________________________
>> > torquedev mailing list
>> > torquedev at supercluster.org
>> > http://www.supercluster.org/mailman/listinfo/torquedev
>> >
>>
>
> --
> Garrick Staples, Linux/HPCC Administrator
> University of Southern California
> _______________________________________________
> torquedev mailing list
> torquedev at supercluster.org
> http://www.supercluster.org/mailman/listinfo/torquedev
>



More information about the torquedev mailing list