jack7z
May 29, 2023, 9:12pm
#1
Hey guys,
I’m trying to run a desorption model and for some reason I kept getting this error that says mpi can’t find the mfixsolver. Do you know what is wrong with my code?
Thanks,Stack_Desorp.mfx (15.6 KB)usr_rates (4).f (2.8 KB)
==========================================================================
mpirun was unable to launch the specified application as it could not access
Executable: ./mfixsolver
16 total processes failed to start
The script that I used to upload my task was the following.
#!/bin/bash -l
#$ -l h_rt=48:00:00 # Max run time
#$ -pe omp 16 # Number of cores/nodes
module load openmpi/3.1.4_gnu-10.2.0
build_mfixsolver --batch --dmp
mpirun -np 16 ./mfixsolver -f Stack_Desorp.mfx NODESI=4 NODESJ=4
cgw
May 31, 2023, 11:43am
#2
I don’t see the problem right away.
I would add a pwd, env and ls in the submit script, between the build_mfixsolver and the mpirun lines, just to see what’s going on.
build_mfixsolver --batch --dmp
env
pwd
ls
mpirun -np 16 ./mfixsolver -f Stack_Desorp.mfx NODESI=4 NODESJ=4
jack7z
June 5, 2023, 1:35am
#3
Thank you Charles! I ran env, pwd, and ls. It seems the mfx.sh was created. Is mfx.sh the same thing as ./mfixsolver?
I pasted the run result below:
SCC_OPENMPI_DIR=/share/pkg.7/openmpi/3.1.4_gnu-10.2.0/installModuleTable003 =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 =cmFyaWVzOi9zaGFyZS9tb2R1bGUuNy9tYWNoaW5lLWxlYXJuaW5nOi9zaGFyZS9tb2R1bGUuNy9tYXRoLWVuZzovc2hhcmUvbW9kdWxlLjcvcHJvZ3JhbW1pbmc6L3NoYXJlL21vZHVsZS43L3N0YXRpc3RpY3M6L3NoYXJlL21vZHVsZS43L3V0aWxpdGllczovc2hhcmUvbW9kdWxlLjcvdmlzdWFsaXphdGlvbiIsfQ==LMOD_REF_COUNT__LMFILES =/share/module.7/libraries/openmpi/3.1.4_gnu-10.2.0.lua:1;/share/module.7/programming/gcc/10.2.0.lua:1;/share/module.7/programming/miniconda/23.1.0.lua:1ModuleTable001 =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 =/share/module.7/libraries/openmpi/3.1.4_gnu-10.2.0.lua:/share/module.7/programming/gcc/10.2.0.lua:/share/module.7/programming/miniconda/23.1.0.luaModuleTable_Sz =4ModuleTable002 =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
cgw
June 5, 2023, 7:15pm
#4
Ok, I think I see the problem.
Something has gone wrong with the build_mfixsolver script, in particular it looks like the --batch flag is not working correctly.
I triedbuild_mfixsolver --batchmfixsolver, it built the interactive (Python-enabled) version, called mfixsolver.sh, but because the --batch flag was specified, it did not create the symlink to mfixsolver.
Install the project...
-- Install configuration: "RelWithDebInfo"
-- Up-to-date: /tmp/xyz/mfixsolver.so
-- Installing: /tmp/xyz/mfixsolver.sh
For batch-mode there should not be a .sh or .so file, just the plain binary `mfixsolver.
(mfix-23.1.1) ./mfixsolver.sh --print-flags
dmp python
The python in the output means that the solver was built with Python (interactive) support, the batch-mode solver should just say dmp (or nothing, if dmp was not enabled).
I’m not sure when this stopped working but we will issue a fix as soon as possible.
In the meanwhile:
Is there a reason for not using the interactive version? Can you just dobuild-mfixsolver --dmp and use that?
The --batch flag is supposed to be synonymous with --server=none, but --server=none actually works while --batch does not.
I’d add -j 8 to the build command to make the build go faster (parallel compilation)
build_mfixsolver --server=none --dmp -j 8
mfix-23.1.1)$ ls
bld.log build mfixsolver
(mfix-23.1.1)$ file mfixsolver
mfixsolver: ELF 64-bit LSB pie executable, x86-64, version 1 (SYSV), dynamically linked, interpreter /lib64/ld-linux-x86-64.so.2, for GNU/Linux 3.2.0, with debug_info, not stripped
mfix-23.1.1)$ ./mfixsolver --print-flags
dmp
Sorry for the trouble!
– Charles
jack7z
June 8, 2023, 8:59pm
#5
Thank you Charles! The major reason why I don’t use GUI is that I need to run MFiX on the school cluster, and it seems the school cluster doesn’t support GUI. Also I need to run 100 simulations at the same time on the cluster, so it’s easier submitting the job through the terminal instead of the GUI.
I tried “build-mfixsolver --dmp” but got the error saying “pymfix.py: error: unrecognized arguments: NODESI=4 NODESJ=4”. Did I miss something?
Thanks,
cgw
June 8, 2023, 9:42pm
#6
The interactive solver does not allow setting keywords on the command line.
You need to put the nodesi, nodesj, nodesk in the file.
Or use the batch solver, build with build_mfixsolver --dmp --server=none (to work around bug in build script). If you use the batch solver you can put the nodes* definitions on the command line.
jack7z
July 31, 2023, 10:03pm
#7
Sorry for the late reply. Thank you Charles!