[torqueusers] dependencies on completed jobs

Alexander Saydakov saydakov at yahoo-inc.com
Tue Sep 5 16:31:10 MDT 2006


> -----Original Message-----
> From: torqueusers-bounces at supercluster.org [mailto:torqueusers-
> bounces at supercluster.org] On Behalf Of 'Garrick Staples'
> Sent: Tuesday, September 05, 2006 3:16 PM
> To: torqueusers at supercluster.org
> Subject: Re: [torqueusers] dependencies on completed jobs
> 
> On Tue, Sep 05, 2006 at 02:58:41PM -0700, Alexander Saydakov alleged:
> > > -----Original Message-----
> > > From: torqueusers-bounces at supercluster.org [mailto:torqueusers-
> > > bounces at supercluster.org] On Behalf Of Garrick Staples
> > > Sent: Tuesday, September 05, 2006 2:40 PM
> > > To: torqueusers at supercluster.org
> > > Subject: Re: [torqueusers] dependencies on completed jobs
> > >
> > > On Tue, Sep 05, 2006 at 12:45:27PM -0700, Sam Rash alleged:
> > > > So we've noticed that if we submit job A, then submit job B which
> > > depends on
> > > > A (-W depend=afterok:B_job_id), and A has already completed (we have
> > > > keep_completed set to at least 30 min), B get stuck in the hold
> state.
> > > Is
> > > > this intentional?  Or a bug?
> > > >
> > > >
> > > >
> > > > It seems like B should surely run.
> > > >
> > > > (maybe A updates its dependents when it completes and B won't check
> > > > explicitly?)
> > >
> > > The simple test works fine for me.
> >
> > I am afraid you did not get the point. Try submitting the second one
> after
> > the first one has finished.
> >
> > In general, you never know if it has finished of not. Even if you have
> just
> > checked a second ago.
> 
> I see, this does look like a bug.  Since the exit status is available in
> completed jobs, the correct dep calc could be done at that point.

Right, so we submit dependencies in the reverse order using on: and
beforeok: to be sure.




More information about the torqueusers mailing list