Open MPI logo

MTT Devel Mailing List Archives

  |   Home   |   Support   |   FAQ   |   all MTT Devel mailing list

Subject: Re: [MTT devel] mtt questions
From: Mike Dubman (mike.ompi_at_[hidden])
Date: 2011-01-06 07:24:39


Thanks a lot!

On Wed, Jan 5, 2011 at 4:01 PM, Joshua Hursey <jjhursey_at_[hidden]> wrote:

> Since I used Open MPI, I didn't have to change the MPI Install phase much
> at all. But if you are running something non-MPI, you may have to update
> this section.
>
> For the MPI Details I have:
> ------------------------------
> [MPI Details: OMPI]
> exec = ./run-correctness.pl -test &test_executable() \
> -storage-dir /nfs/odinnfs1/san36-1-0/jjhursey/tmp/snapshot/ \
> -np &test_np() \
> -mca "@fixes@ -am ft-enable-cr @pml@" \
> --prefix &test_prefix() -v 2
>
> fixes = -mca orte_base_help_aggregate 0
> ft_options = -am ft-enable-cr
> -------------------------------
>
> For the Test Get/Build/Run phases I have:
> -------------------------------
> [Test get: iu_ft_cr]
> module = SVN
> svn_url = https://svn.open-mpi.org/svn/ompi-tests/trunk/iu/ft/correctness
>
>
> [Test build: iu_ft_cr]
> test_get = iu_ft_cr
> save_stdout_on_success = 1
> merge_stdout_stderr = 1
> stderr_save_lines = -1
>
> module = Shell
> shell_build_command = <<EOT
> make clean all ADD_FLAGS=""
> EOT
>
>
> [Test run: iu_ft_cr]
> include_section = Defaults Test run
>
> test_build = iu_ft_cr
> timeout = &max(240, &multiply(30, &test_np()))
> pass = &and(&test_wifexited(), &eq(&test_wexitstatus(), 0))
> np = &env_max_procs()
>
> simple_pass:tests = &enumerate( "noop", "quiet", "simple", "progress",
> "inflight", "barrier" )
> -------------------------------
>
>
> -- Josh
>
>
> On Jan 5, 2011, at 7:35 AM, Mike Dubman wrote:
>
> > Thanks for info!
> > Josh, can you please send me mtt.ini file you used with C/R tests as a
> reference for using non-MPI application?
> >
> > Also, found following inside RunEngine.pm code:
> >
> > # Set some thresholds for an early exit
> > my $break_threshold;
> > $break_threshold->{MTT::Values::PASS} = Value($ini, $section,
> "break_threshold_pass");
> > $break_threshold->{MTT::Values::FAIL} = Value($ini, $section,
> "break_threshold_fail");
> > $break_threshold->{MTT::Values::TIMED_OUT} = Value($ini, $section,
> "break_threshold_timeout");
> > $break_threshold->{MTT::Values::SKIPPED} = Value($ini, $section,
> "break_threshold_skipped");
> >
> > but it seems like "break rules" are available per single test section
> instead of whole mtt session.
> > Regards
> > Mike
> >
> >
> > On Tue, Jan 4, 2011 at 3:42 PM, Joshua Hursey <jjhursey_at_[hidden]>
> wrote:
> >
> > On Jan 4, 2011, at 8:04 AM, Jeff Squyres wrote:
> >
> > > On Jan 4, 2011, at 7:06 AM, Mike Dubman wrote:
> > >
> > >> Do you know if there is mtt option to stop mtt execution if job`s
> failure ratio succeeds specified value, something like:
> > >>
> > >> [mtt]
> > >> stop_on_test_failures=1%
> > >
> > > I don't recall an option like that, offhand.
> >
> > I don't think so either. This would be neat to have for those rare
> scenarios when something fundamental in Open MPI breaks and we would prefer
> to stop testing early instead of continuing to run tests even if they are
> all going to fail.
> >
> > >
> > >> Also, are there any ini files examples/successes of how to use mtt
> with non-MPI based applications?
> > >
> > > No. It *shouldn't* be hard, though -- just make the MPI Details exec
> statement not use "mpirun"...? (of course, our phase names are biased
> towards "MPI", but they could just as well be "Middleware", for example)
> >
> > I successfully did this when testing the C/R functionality in Open MPI,
> and a couple other groups are looking at supporting non-MPI middleware as
> well. The C/R tests are all wrapped up in their own perl script (to do
> start/checkpoint/restart/...), so were essentially a non-MPI application. As
> Jeff mentioned, it was just enough to change the MPI Details section to use
> your own launcher (or you may be able to even go without a launcher if you
> just want to run the tests directly). Then you will need to define new Test
> Build/Run phases for the non-MPI test suite that you are using. And you
> should be all set.
> >
> > If for some reason that does not work, let us know since it is something
> that needs to be fixed since the MTT client should be flexible enough to
> support a wide manner of middleware.
> >
> > -- Josh
> >
> > >
> > > --
> > > Jeff Squyres
> > > jsquyres_at_[hidden]
> > > For corporate legal information go to:
> > > http://www.cisco.com/web/about/doing_business/legal/cri/
> > >
> > >
> > > _______________________________________________
> > > mtt-devel mailing list
> > > mtt-devel_at_[hidden]
> > > http://www.open-mpi.org/mailman/listinfo.cgi/mtt-devel
> > >
> >
> > ------------------------------------
> > Joshua Hursey
> > Postdoctoral Research Associate
> > Oak Ridge National Laboratory
> > http://users.nccs.gov/~jjhursey <http://users.nccs.gov/%7Ejjhursey>
> >
> >
> > _______________________________________________
> > mtt-devel mailing list
> > mtt-devel_at_[hidden]
> > http://www.open-mpi.org/mailman/listinfo.cgi/mtt-devel
> >
> > _______________________________________________
> > mtt-devel mailing list
> > mtt-devel_at_[hidden]
> > http://www.open-mpi.org/mailman/listinfo.cgi/mtt-devel
>
> ------------------------------------
> Joshua Hursey
> Postdoctoral Research Associate
> Oak Ridge National Laboratory
> http://users.nccs.gov/~jjhursey <http://users.nccs.gov/%7Ejjhursey>
>
>
> _______________________________________________
> mtt-devel mailing list
> mtt-devel_at_[hidden]
> http://www.open-mpi.org/mailman/listinfo.cgi/mtt-devel
>