From: Jeff Squyres (jsquyres_at_[hidden])
Date: 2006-11-06 07:45:41
On Nov 3, 2006, at 2:25 PM, Josh Hursey wrote:
> I have an INI File that looks something like what is enclose at the
> end of this message.
>
> So I have multiple MPI Details sections. It seems like only the first
> one is running. Do I have to list them out somewhere?
Yes. In the MPI Get section, you have to list an "mpi_details" field
that says which MPI Details section applies to this MPI. We don't
currently allow a comma-delimited list of Details section names, but
that could be added if you want/need.
> As a side question:
> Instead of using multiple MPI Details sections, if I use a single MPI
> Details section and told MTT to iterate over the parameters to mpirun
> at what point in the aggregation chain does this happen?
> Meaning will I see something like:
We hadn't really anticipated having multiple MPI details sections
that were relevant for a single MPI -- we had intended the MPI
details section to provide the generic method of how to run with that
MPI (e.g., mpirun command line differences between different MPI
implementations). But that could be changed if it would be helpful.
Ethan made a relevant comment here as well.
> Building trunk
> Running trivial
> Runing mpirun -mca btl tcp,self
> Runing mpirun -mca btl mx,self
> Runing mpirun -mca btl mvapi,self
> Running intel
> Runing mpirun -mca btl tcp,self
> Runing mpirun -mca btl mx,self
> Runing mpirun -mca btl mvapi,self
You will see this ^^^. The way it works is that the Test Run phase
will iterate through the successful Test Builds (which depends on the
successful Test Gets and MPI Installs, which, in turn, depends on
successful MPI Gets). When it identifies a test to run, it looks up
the corresponding MPI Install section, gets the MPI Details section,
and creates one or more command lines to execute. So a single Test
Run corresponds to a single Test Build / MPI Install / MPI Get tuple,
and therefore results in a single functlet for the command line to
execute.
> or will I see
>
> Building trunk
> Runing mpirun -mca btl tcp,self
> Running trivial
> Running intel
> Runing mpirun -mca btl mx,self
> Running trivial
> Running intel
> Runing mpirun -mca btl mvapi,self
> Running trivial
> Running intel
I'm not quite sure I understand this above example ordering -- I see
the "mpirun" lines before the "Running <testname>" lines. Did you mean:
Building trunk
Running trivial
Runing mpirun -mca btl tcp,self
Running intel
Runing mpirun -mca btl tcp,self
Running trivial
Runing mpirun -mca btl mx,self
Running intel
Runing mpirun -mca btl mx,self
Running trivial
Runing mpirun -mca btl mvapi,self
Running intel
Runing mpirun -mca btl mvapi,self
> I would actually prefer the last because if something is broken with
> the IB interconnect it doesn't slow down the other tests from
> reporting. E.g. I can prioritize a bit in the INI file.
I could see that. Unfortunately, that's not currently how the run
engine is structured.
> Cheers,
> Josh
>
> mmmmmmmmmmmmmmmmmmmmmmmmmmmmmm
> #=====================================================================
> =
> # Overall configuration
> #=====================================================================
> =
>
> [MTT]
>
> hostfile =
> hostlist =
> max_np =
>
> #---------------------------------------------------------------------
> -
>
> #=====================================================================
> =
> # MPI get phase
> #=====================================================================
> =
>
> [MPI get: ompi-nightly-trunk]
> mpi_details = Open MPI
>
> module = OMPI_Snapshot
> ompi_snapshot_url = http://www.open-mpi.org/nightly/trunk
>
> #---------------------------------------------------------------------
> -
>
> #=====================================================================
> =
> # Install MPI phase
> #=====================================================================
> =
>
> [MPI install: odin 32 bit gcc]
> mpi_get = ompi-nightly-trunk
> save_stdout_on_success = 1
> merge_stdout_stderr = 1
> vpath_mode = none
>
> make_all_arguments = -j 6
> make_check = 1
>
> compiler_name = gnu
> compiler_version = &shell("gcc --version | head -n 1 | awk '{ print \
> $3 }'")
> configure_arguments = \
> FCFLAGS=-m32 FFLAGS=-m32 CFLAGS=-m32 CXXFLAGS=-m32 \
> --with-wrapper-cflags=-m32 --with-wrapper-cxxflags=-m32 --with-
> wrapper-fflags=-m32 --with-wrapper-fcflags=-m32
>
> module = OMPI
>
> #---------------------------------------------------------------------
> -
>
> #=====================================================================
> =
> # MPI run details
> #=====================================================================
> =
>
> [MPI Details: Open MPI tcp sm]
> exec = mpirun @hosts@ -mca btl tcp,sm,self -np &test_np() --prefix
> &test_prefix() &test_executable() &test_argv()
>
> # Yes, all these quotes are necessary. Don't mess with them!
> hosts = &if(&have_hostfile(), "&join("--hostfile ", "&hostfile()")", \
> "&if(&have_hostlist(), "&join("--host ", "&hostlist
> ()")", "")")
>
> after_each_exec = <<EOT
> if test "$MTT_TEST_HOSTFILE" != ""; then
> args="--hostfile $MTT_TEST_HOSTFILE"
> elif test "$MTT_TEST_HOSTLIST" != ""; then
> args="--host $MTT_TEST_HOSTLIST"
> fi
> orterun $args -np $MTT_TEST_NP --prefix $MTT_TEST_PREFIX
> mtt_ompi_cleanup.pl
> EOT
>
> #---------------------------------------------------------------------
> -
>
> [MPI Details: Open MPI mx sm]
> exec = mpirun @hosts@ -mca btl mx,sm,self -np &test_np() --prefix
> &test_prefix() &test_executable() &test_argv()
>
> # Yes, all these quotes are necessary. Don't mess with them!
> hosts = &if(&have_hostfile(), "&join("--hostfile ", "&hostfile()")", \
> "&if(&have_hostlist(), "&join("--host ", "&hostlist
> ()")", "")")
>
> after_each_exec = <<EOT
> if test "$MTT_TEST_HOSTFILE" != ""; then
> args="--hostfile $MTT_TEST_HOSTFILE"
> elif test "$MTT_TEST_HOSTLIST" != ""; then
> args="--host $MTT_TEST_HOSTLIST"
> fi
> orterun $args -np $MTT_TEST_NP --prefix $MTT_TEST_PREFIX
> mtt_ompi_cleanup.pl
> EOT
>
> #---------------------------------------------------------------------
> -
>
> [MPI Details: Open MPI mvapi sm]
> exec = mpirun @hosts@ -mca btl mvapi,sm,self -np &test_np() --prefix
> &test_prefix() &test_executable() &test_argv()
>
> # Yes, all these quotes are necessary. Don't mess with them!
> hosts = &if(&have_hostfile(), "&join("--hostfile ", "&hostfile()")", \
> "&if(&have_hostlist(), "&join("--host ", "&hostlist
> ()")", "")")
>
> after_each_exec = <<EOT
> if test "$MTT_TEST_HOSTFILE" != ""; then
> args="--hostfile $MTT_TEST_HOSTFILE"
> elif test "$MTT_TEST_HOSTLIST" != ""; then
> args="--host $MTT_TEST_HOSTLIST"
> fi
> orterun $args -np $MTT_TEST_NP --prefix $MTT_TEST_PREFIX
> mtt_ompi_cleanup.pl
> EOT
>
> #---------------------------------------------------------------------
> -
>
> #=====================================================================
> =
> # Test get phase
> #=====================================================================
> =
>
> [Test get: trivial]
> module = Trivial
>
> #---------------------------------------------------------------------
> -
>
> #=====================================================================
> =
> # Test build phase
> #=====================================================================
> =
>
> [Test build: trivial]
> test_get = trivial
> save_stdout_on_success = 1
> merge_stdout_stderr = 1
> stderr_save_lines = -1
>
> module = Trivial
>
> #---------------------------------------------------------------------
> -
>
> #=====================================================================
> =
> # Test Run phase
> #=====================================================================
> =
>
> [Test run: trivial]
> test_build = trivial
> pass = &eq(&test_exit_status(), 0)
> timeout = &multiply(2, test_np())
> save_stdout_on_pass = 1
> merge_stdout_stderr = 1
> stdout_save_lines = 100
> np = &env_max_procs()
>
> module = Simple
> simple_only:tests = &find_executables(".")
>
> #---------------------------------------------------------------------
> -
>
> #=====================================================================
> =
> # Reporter phase
> #=====================================================================
> =
>
> [Reporter: IU database]
> module = MTTDatabase
>
> mttdatabase_realm = OMPI
> mttdatabase_url = https://www.open-mpi.org/mtt/submit/
> # OMPI Core: Change this to be the username and password for your
> # submit user. Get this from the OMPI MTT administrator.
> mttdatabase_username = XX
> mttdatabase_password = XX
> # OMPI Core: Change this to be some short string identifying your
> # cluster.
> mttdatabase_platform = IU - Thor - TESTING
>
> #---------------------------------------------------------------------
> -
>
> # This is a backup for while debugging MTT; it also writes results to
> # a local text file
> [Reporter: text file backup]
> module = TextFile
>
> textfile_filename = mtt-debug-report-2-$phase-$section-$mpi_name-
> $mpi_version.txt
> textfile_separator =
>>>>> ----------------------------------------------------------<<<<
>
>
> #---------------------------------------------------------------------
> -
>
> _______________________________________________
> mtt-users mailing list
> mtt-users_at_[hidden]
> http://www.open-mpi.org/mailman/listinfo.cgi/mtt-users
-- Jeff Squyres Server Virtualization Business Unit Cisco Systems