Make sure a PSCW epoch is active on MPI_Win_complete calls

[codymann-cornelisnetworks]

The test case in mpich test suite v4.2.1 is failing due to MPI_Win_Complete returning with no failures when it should be failing with MPI_ERR_RMA_SYNC. Simple fix included in this PR to make sure that a check is done for an active pscw epoch before further actions are taken.

config.log

Script to generate the failure:

#! /usr/bin/env bash
export MPI_PATH="/home/cmann/builds/ompi.debug"
export MPI_LIB_PATH="$MPI_PATH/lib"
export MPI_BIN_PATH="$MPI_PATH/bin"

export LIBFAB_PATH="/home/cmann/builds/libfabric.debug"
export LIBFAB_LIB_PATH="$LIBFAB_PATH/lib"
export LIBFAB_BIN_PATH="$LIBFAB_PATH/bin"

./configure --enable-strictmpi --with-mpi=$MPI_PATH --disable-dtpools

 if [[ $? -ne 0 ]]
 then
    echo "Configure failure"
    exit 1
fi


HOSTS="phwtstl005,phwtstl006"
export RUNTESTS_SHOWPROGRESS=1
# export MPITEST_PPNARG="-N %d"
export VERBOSE=1
export MPITEST_PPNMAX=36
export MPITEST_TIMEOUT=300
export MPITEST_PROGRAM_WRAPPER="--host $HOSTS --mca btl_ofi_disable_sep 1 --mca btl_ofi_disable_hmem 1  --mca osc ^ucx --mca pml ^ucx --mca btl ofi --mca mtl_ofi_enable_sep 0 --mca mtl ofi --map-by :OVERSUBSCRIBE -x FI_OPX_UUID=$RANDOM -x FI_PROVIDER=opx -x LD_PRELOAD=/home/cmann/code/libfabric-devel/debug/handlers/segfault_abort_handler.so -x LD_LIBRARY_PATH=$LIBFAB_LIB_PATH:$MPI_LIB_PATH:$LD_LIBRARY_PATH -x PATH=$MPI_BIN_PATH:$PATH"
make testing

Test output:

/home/cmann/builds/ompi.debug/bin/mpiexec -n 2   --host phwtstl005,phwtstl006 --mca btl_ofi_disable_sep 1 --mca btl_ofi_disable_hmem 1  --mca osc ^ucx --mca pml ^ucx --mca btl ofi --mca mtl_ofi_enable_sep 0 --mca mtl ofi --map-by :OVERSUBSCRIBE -x FI_OPX_UUID=7553 -x FI_PROVIDER=opx -x LD_PRELOAD=/home/cmann/code/libfabric-devel/debug/handlers/segfault_abort_handler.so -x LD_LIBRARY_PATH=/home/cmann/builds/libfabric.debug/lib:/home/cmann/builds/ompi.debug/lib: -x PATH=/home/cmann/builds/ompi.debug/bin:/home/cmann/.vscode-server/cli/servers/Stable-fee1edb8d6d72a0ddff41e5f71a671c23ed924b9/server/bin/remote-cli:/home/cmann/.local/bin:/usr/share/Modules/bin:/usr/local/bin:/usr/bin:/usr/local/sbin:/usr/sbin ./win_sync_complete 
0: Operation succeeded, when it should have failed
1: Operation succeeded, when it should have failed
 Found 2 errors

[devreal]

LGTM, thanks for the patch!

[devreal]

@janjust Seems the Nvidia CI ran out of space.

[janjust]

@janjust Seems the Nvidia CI ran out of space.

Thanks - I notified our admin, he'll clean it up

[codymann-cornelisnetworks]

Just refreshed to the latest main. I'm new to contributing to Open MPI so please feel free to let me know if I'm missing any steps here for the contribution process.