Re: [Condor-users] Automate removal of inefficient jobs

[Sarah Williams <saewill@xxxxxxxxx>]

On 7/13/11 5:15 PM, Burt Holzman wrote:
> Ian Chesal wrote:
> > > I did a plain 'condor_reconfig' when I was testing yesterday. Today I
> > > tried condor_reconfig -full -schedd, with the same result.
> > >
> > > Adding PERIODIC_REMOVE to the job submission script did work. I'd
> > > rather use SYSTEM_PERIODIC_REMOVE because these jobs are submitted
> > > through globus gatekeeper, so if I want to add things to the submission
> > > scripts I have to patch that. I'll tinker with the SYTEM_ version some
> > > more, and if that continues to not work I'll bite the bullet and patch
> > > globus.
> > Occasionally you'll encounter settings in Condor that require a 
> > restart to take effect. I didn't think this was one of them, but 
> > perhaps. Try:
> >
> > condor_restart -fast -schedd
> >
> > Brute force, but it should take after that. You can confirm the 
> > scheduler is seeing the setting with:
> >
> > condor_config_val -verbose -schedd SYSTEM_PERIODIC_REMOVE
>
> Hi,
>
> It's actually the condor_shadow that enforces SYSTEM_PERIODIC_REMOVE. 
> If you don't have a lot of shadows, you can send them all SIGHUP and 
> they should re-read their configuration files.  Restarting the schedd 
> also works, since all the shadows will die with it and be restarted.
>
> - B

Hi Burt,

Thanks, that was it! Once I did a full reconfig and set SIGHUP to all 
the shadows, the inefficient jobs were removed, and I see remove 
messages in the ShadowLog.

I was confused because there are several places in the documentation 
where it says schedd evaluates SYSTEM_PERIODIC_REMOVE. Here is one such:
http://www.cs.wisc.edu/condor/manual/v7.6/3_3Configuration.html

--Sarah