Edward Bueler <[email protected]> writes:
One more thing, to further clarify that I am not doing something obviously wrong. Here is what I see about my mpiexec.
I can reproduce this. I don't know if the change was intentional, so I'm cross-posting. MPICH developers: a shell script containing multiple mpiexec commands no longer makes it past the first one. Apparently one now needs to set stdin to /dev/null to continue. $ cat t.sh #!/bin/sh $MPIEXEC -n 1 hostname $MPIEXEC -n 1 date ## MPICH-3.1b1 (also 3.0 series), "date" output not printed $ MPIEXEC=/opt/mpich/bin/mpiexec ./t.sh & [5] 29761 $ batura [5]+ Stopped MPIEXEC=/opt/mpich/bin/mpiexec ./t.sh ## Open MPI - note that "date" output is printed $ MPIEXEC=/usr/bin/mpiexec ./t.sh & [6] 29780 $ batura Sun Feb 9 14:12:35 MST 2014 [6] Done MPIEXEC=/usr/bin/mpiexec ./t.sh