Re: [Condor-users] Changing a working directory for a DAG within a DAG?

[Masakatsu Ito <mito@xxxxxxxxxxxxxx>]

I'm sorry for making typing errors in 'param[n]/inner.dag' files.
Now let me correct them.

Masakatsu Ito wrote:

> Thank you for you reply and offer to diagnose things.
>
> Well, actually I think I've resolved the problem by modifying
> the inner DAG description files like :

# param[n]/inner.dag
# [n] = 1, 2, 3 ..
#
Job Pre[n] param[n]/preparation.submit
Job Sim[n] param[n]/simulation.submit

PARENT Pre[n] CHILD Sim[n]

> Previously I had used the same description for all
> param[n]/inner.dag , but I've added 'param[n]/' assuming that
> DAGMan does not change the working directory to param[n]/ .
>
> ( I use the previous release version of CONDOR, 6.6.11. )
>
> R. Kent Wenger wrote:
>
> > Sorry for the long delay in responding...
> >
> > On Fri, 7 Jul 2006, Masakatsu Ito wrote:
> >
> >  
> >
> > > I am trying to use a DAG within a DAG to perform
> > > a number of simulations which is easier to construct
> > > hierarchically. I have to put these simulation outputs
> > > into hierarchical directory structure, which a post (analysis)
> > > script requires. But the problem is that I've failed to put
> > > them into the directory structure, because I cannot get
> > > condor_dagman to change a working directory.
> > >
> > > Basically I've tried to follow the way in the thread
> > > https://lists.cs.wisc.edu/archive/condor-users/2006-March/msg00004.shtml 
> > >
> > > which suggested using -UseDagDir . But this thread
> > > is not completely clear for me to follow.
> > >   
> >
> >
> > Yes, unfortunately -UseDagDir is somewhat complicated to explain.
> >
> >  
> >
> > > My goal is to sort the simulation outputs into the directories
> > > param1, param2, ... . The whole directory tree should be
> > > like :
> > >
> > > --+-- outer.dag
> > > |
> > > +-- param1 -+-- inner.dag
> > > | +-- preparation.submit
> > > | +-- simulation.submit
> > > | +-- preparation.inp
> > > | +-- ( simulation.out )
> > > |
> > > +-- param2 -+-- inner.dag
> > > | +-- preparation.submit
> > > | +-- simulation.submit
> > > | +-- preparation.inp
> > > | +-- ( simulation.out )
> > > ;
> > >
> > > I used two nested DAGs. The inner DAG describes the
> > > dependency between preparation and simulation programs.
> > >
> > > # inner.dag
> > > #
> > > Job Pre preparation.submit
> > > Job Sim simulation.submit
> > >
> > > PARENT Pre CHILD Sim
> > >
> > > I wrote the outer DAG description file to sort
> > > the simulation results into the directories.
> > >
> > > # outer.dag
> > > #
> > > Job Param1 param1/inner.dag.condor.sub
> > > Job Param2 param2/inner.dag.condor.sub
> > >
> > > I created these "inner.dag.condor.sub" using :
> > >
> > > % condor_submit_dag -no_submit -UseDagDir param1/inner.dag
> > >
> > > as the thread
> > > https://lists.cs.wisc.edu/archive/condor-users/2006-March/msg00004.shtml 
> > >
> > > suggested.
> > >
> > > But, for my case, -UseDagDir does not cause DAGMan to change
> > > a working directory when invoking condor_submit_dag.
> > > After I submitted the outer DAG like
> > >
> > > % condor_submit_dag outer.dag
> > >
> > > the job of the inner DAG, Pre, failed because
> > > the preparation program was not able to find
> > > its input, preparation.inp . Moreover, the log file of
> > > this job was put into the parent directory.
> > > It seems that despite of using -UseDagDir, DAGMan kept
> > > using the parent directory, which I'm trying to avoid.
> > >
> > > Probably I miss some points in the thread.
> > > So I'd be very grateful if anyone gives me any hint,
> > > advice, or comments.
> > >   
> >
> >
> > Hmm -- I made a test case that's essentially exactly like your case, and
> > it all worked fine.
> >
> > Could you send all of the condor.sub files, and also the submit files
> > for the node jobs?  That might help diagnose things.
> >
> > Kent Wenger
> > Condor Team
> >  

--
Masakatsu Ito
Nanotechnology Research Center
FUJITSU LABORATORIES LTD.