ctest failures when setting multiple MPI_ARGS

[David Davies]

For a particular platform I am running on I need to run mpiexec with -launcher ssh options. To do this I set -DMPI_ARGS="-launcher;ssh" when running cmake. This allows e.g. eckit_test_mpi to succeed but ECKIT-221.sh and ECKIT-166.sh fails with:

    Start 118: ECKIT-166.sh

118/118 Test #118: ECKIT-166.sh ...............................***Failed 0.01 sec

• cd /var/tmp/frwd-8923939/eckit-build/regressions
• exe=./ECKIT-166.x
• '[' 1 == 1 ']'
• '[' /home/h01/frwd/extraction/lfric-bundle/Scriptify/installs/mpich/bin/mpiexec '!=' MPIEXEC-NOTFOUND ']'
• /home/h01/frwd/extraction/lfric-bundle/Scriptify/installs/mpich/bin/mpiexec -launcher
  [mpiexec@expspicesrv030] HYDU_set_str (utils/args/args.c:198): cannot assign NULL object
  [mpiexec@expspicesrv030] launcher_fn (ui/mpich/utils.c:797): error setting launcher
  [mpiexec@expspicesrv030] match_arg (utils/args/args.c:156): match handler returned error
  [mpiexec@expspicesrv030] HYDU_parse_array (utils/args/args.c:178): argument matching returned error
  [mpiexec@expspicesrv030] parse_args (ui/mpich/utils.c:1642): error parsing input array
  [mpiexec@expspicesrv030] HYD_uii_mpx_get_parameters (ui/mpich/utils.c:1694): unable to parse user arguments
  [mpiexec@expspicesrv030] main (ui/mpich/mpiexec.c:148): error parsing parameters

On the other hand running cmake with -DMPI_ARGS="-launcher ssh" results in e.g. test_mpi_parallel failing like this:

Start 56: eckit_test_mpi_parallel

56: Test command: /home/h01/frwd/extraction/lfric-bundle/Scriptify/installs/mpich/bin/mpiexec "-launcher ssh" "-n" "4" "/var/tmp/frwd-8923939/eckit-build/tests/mpi/eckit_test_mpi_parallel"
56: Environment variables:
56: OMP_NUM_THREADS=1
56: Test timeout computed to be: 1500
56: [mpiexec@expspicesrv030] match_arg (utils/args/args.c:163): unrecognized argument launcher ssh
56: [mpiexec@expspicesrv030] HYDU_parse_array (utils/args/args.c:178): argument matching returned error
56: [mpiexec@expspicesrv030] parse_args (ui/mpich/utils.c:1642): error parsing input array
56: [mpiexec@expspicesrv030] HYD_uii_mpx_get_parameters (ui/mpich/utils.c:1694): unable to parse user arguments
56: [mpiexec@expspicesrv030] main (ui/mpich/mpiexec.c:148): error parsing parameters
1/1 Test #56: eckit_test_mpi_parallel ..........***Failed 0.03 sec
[mpiexec@expspicesrv030] match_arg (utils/args/args.c:163): unrecognized argument launcher ssh
[mpiexec@expspicesrv030] HYDU_parse_array (utils/args/args.c:178): argument matching returned error
[mpiexec@expspicesrv030] parse_args (ui/mpich/utils.c:1642): error parsing input array
[mpiexec@expspicesrv030] HYD_uii_mpx_get_parameters (ui/mpich/utils.c:1694): unable to parse user arguments
[mpiexec@expspicesrv030] main (ui/mpich/mpiexec.c:148): error parsing parameters

whereas the ECKIT* tests pass.

[Willem Deconinck]

Thank you for the bug report.
Perhaps the expected behaviour is to use -DMPI_ARGS="-launcher ssh" and not -DMPI_ARGS="-launcher;ssh"
In that case it should be changed in ecbuild. I will get on with that now.

[David Davies]

Okay, thanks. It doesn't matter which way is preferred so long as it works for all the tests.

[Willem Deconinck]

ecbuild develop branch now has this fix. You should now be able to use -DMPI_ARGS="-launcher ssh"
Please close the issue when you verified it works for you.
Thank you for discovering this bug.

[David Davies]

Thanks, this works for me.