[MPICH2-dev] Two mpich2-1.0.3/mpiexec issues

[Rusty Lusk]

Thanks, Dmitry.  We have tested this and it is indeed fixed in our cvs
tree, and will therefore be fixed in the next release.

Regards,
Rusty

From: "Ezhov, Dmitry" <dmitry.ezhov at intel.com>
Subject: RE: [MPICH2-dev] Two mpich2-1.0.3/mpiexec issues
Date: Thu, 20 Apr 2006 20:18:12 +0400

> Dear Rusty,
> 
> Here is the example illustrating the second issue:
> 
> export TMP_ENV_VAR=1
> mpiexec -genvlist TMP_ENV_VAR -n 1 -host host1 ./a.out : -n 1 -host
> host2 ./a.out
> 
> or the same using configfile:
> 
> export TMP_ENV_VAR=1
> mpiexec -configfile a.out.cfgfile
> and a.out.cfgfile consists of following lines:
> -genvlist TMP_ENV_VAR
> -n 1 -host host1 ./a.out
> -n 1 -host host2 ./a.out
> 
> a.out can be any MPI application, "Hello, World", for example.
> 
> Thank you.
> 
> -- 
> Best regards.
> 
> Dmitry Ezhov
> Software Engeneer,
> 
> Intel SSG/DPD/Parallel Solution Lab,
> Russia, Sarov.
> 
> -----Original Message-----
> From: Rusty Lusk [mailto:lusk at mcs.anl.gov] 
> Sent: Thursday, April 20, 2006 6:47 PM
> To: Ezhov, Dmitry
> Cc: mpich2-dev at mcs.anl.gov; Supalov, Alexander; Voronov, German;
> rbutler at mcs.anl.gov; rbutler at mtsu.edu
> Subject: Re: [MPICH2-dev] Two mpich2-1.0.3/mpiexec issues
> 
> Hi Dmitry,
> 
>     Thanks for pointing these problems out.  The first one was already
> fixed in our development version.  It may be that we have already fixed
> the second one as well, but I would like to make sure.  Can you send me
> an example invocation of mpiexec that illustrates the problem, so that
> we can attempt to reproduce it?  
> 
> Thanks,
> Rusty Lusk
> 
> From: "Ezhov, Dmitry" <dmitry.ezhov at intel.com>
> Subject: [MPICH2-dev] Two mpich2-1.0.3/mpiexec issues
> Date: Wed, 19 Apr 2006 15:46:55 +0400
> 
> > Dear Sirs,
> > 
> >  
> > 
> > The first issue is that mpiexec fails when it is used with -genv
> option:
> > 
> >  
> > 
> > mpd_uncaught_except_tb handling:
> > 
> > exceptions.KeyError: key genv not found in parm db
> > 
> >  
> > 
> > I guess the failure occurs because of another key value is used in
> line
> > #535 of the mpiexec.py module:
> > 
> > parmdb['genv'][args[argidx+1]] = args[argidx+2]
> > 
> >  
> > 
> > I tried to correct this issue by changing 'genv' to '-genv' and after
> it
> > mpiexec works properly with -genv option.
> > 
> > What do you thing about this correction?
> > 
> >  
> > 
> > The second issue is that mpiexec fails when -genvlist option is used
> > with two or more -host options (for example, using configfile):
> > 
> >  
> > 
> > mpd_uncaught_except_tb handling:
> > 
> > exceptions.AttributeError: 'list' object has no attribute 'split'
> > 
> >  
> > 
> > I guess this failure happens due to -genvlist parameters parsing
> section
> > (mpiexec.py, start line #656) executes each time the
> handle_local_argset
> > procedure is called. The first call (i.e. the first parsing section
> > execution) converts -genvlist option parameters from string into list,
> > but the second one tries to convert obtained list into list again
> using
> > string splitting method. And this operation results in raising
> > AttributeError exception.
> > 
> > How do you think, whether I'm right in assumption that the issue could
> > be probably fixed by moving parameters parsing section before calling
> > the aforementioned procedure in order to execute this section only
> once,
> > for example, to line #269? At least, I tried to do it and got the
> > workable mpiexec.py module properly working with -genvlist option
> > together with configfile or many -host options. 
> > 
> >  
> > 
> > What is your opinion about these two issues?
> > 
> >  
> > 
> > Thank you in advance.
> > 
> >  
> > 
> > -- 
> > 
> > Best regards.
> > 
> >  
> > 
> > Dmitry Ezhov
> > 
> > Software Engeneer,
> > 
> >  
> > 
> > Intel SSG/DPD/Parallel Solution Lab,
> > 
> > Russia, Sarov.
> > 
> >  
> > 
>