[mpich-discuss] MPI_Comm_Spawn + UCX error

Wed Apr 21 18:20:35 CDT 2021

Hello,

I have been working with the MPI_Comm_spawn function, which was working
fine for a simple compiled version of MPICH 3.4.1, where in the
*configure* step
it only has the "--prefix" argument.

However, when this function was called with another compiled version of
MPICH 3.4.1 which uses Infiniband, an error arised. Am I missing some
arguments in the compilation step of MPICH when using UCX?

Here is the output of *mpichversion*:
$ mpichversion
MPICH Version:     3.4.1
MPICH Release date: Fri Jan 22 14:17:48 CST 2021
MPICH Device:     ch4:ucx
MPICH configure:  --prefix=/soft/gnu/mpich-3.4.1-ucx --with-device=ch4:ucx
--with-ucx=/soft/gnu/ucx-1.11
MPICH CC:  gcc    -O2
MPICH CXX:  g++   -O2
MPICH F77:  gfortran   -O2
MPICH FC:  gfortran   -O2
MPICH Custom Information:

The following is the information about the minimal code which arises the
error

Source: https://www.rookiehpc.com/mpi/docs/mpi_comm_spawn.php

Compiling: mpicc mpi_spawn.c

Running: mpirun -np 2 ./a.out

We are processes spawned directly by you, we now spawn a new instance of an
MPI application.

Assertion failed in file src/mpid/ch4/src/ch4_init.c at line 651:
MPIR_Process.comm_parent != NULL

/soft/gnu/mpich-3.4.1-ucx/lib/libmpi.so.12(MPL_backtrace_show+0x39)
[0x7fe15d506d41]

/soft/gnu/mpich-3.4.1-ucx/lib/libmpi.so.12(+0x32eaa8) [0x7fe15d4a6aa8]

/soft/gnu/mpich-3.4.1-ucx/lib/libmpi.so.12(+0x3602f8) [0x7fe15d4d82f8]

/soft/gnu/mpich-3.4.1-ucx/lib/libmpi.so.12(+0x225803) [0x7fe15d39d803]

/soft/gnu/mpich-3.4.1-ucx/lib/libmpi.so.12(PMPI_Init+0xa8) [0x7fe15d39d598]

./a.out(+0x123e) [0x55ece110a23e]

/lib/x86_64-linux-gnu/libc.so.6(__libc_start_main+0xf3) [0x7fe15cfa10b3]

./a.out(+0x114e) [0x55ece110a14e]

Abort(1) on node 0: Internal error
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mpich.org/pipermail/discuss/attachments/20210422/483aa931/attachment.html>