[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Condor-users] Notification of Cluster Complete - notprocesscomplete

On Thu, Jul 29, 2004 at 07:20:17PM +0100, Matt Hope wrote:
> > -----Original Message-----
> > From: condor-users-bounces@xxxxxxxxxxx
> > [mailto:condor-users-bounces@xxxxxxxxxxx]On Behalf Of Zachary Miller
> > Sent: 29 July 2004 19:09
> > To: Condor-Users Mail List
> > Subject: Re: [Condor-users] Notification of Cluster Complete - not
> > processcomplete
> > > 
> > > Disable all email notification (apart from maybe errors)
> > > After submitting the jobs run condor_wait pointing at the resulting clusterid
> > > and have that launch some simple mail script to send the mail.
> > 
> > i think it would be slightly better to submit condor_wait as 
> > a condor job
> > itself, and then have that job email you when it's finished.  
> > no scripts
> > necessary, and if the machine goes down for some reason, your 
> > condor_wait
> > process will come back up when condor does.
> Though this would have the side effect of tacking up a slot either on your pool or on your local machine which may be problematic if you run more concurrent clusters

No, you would submit the job as a "scheduler universe" job. Scheduler universe
jobs run on the submit machine - they're not matched, so you can run as many of 
them as you want. DAGMan is the prime example of a scheduler universe job - 
run 'condor_submit_dag -nosubmit' to see what it would tell the schedd to

The reason for running a scheduler universe job is that Condor will automatically
restart it if the machine reboots - you don't have to write fancy cron scripts to
make sure that it is still running for you.