Problem with mode installation

[yukai]

Hello, I found this problem when installing CICE mode. What is the reason?
cli_25]: write_line error; fd=16 buf=:cmd=init pmi_version=1 pmi_subversion=1
:
system msg for write_line failure : Bad file descriptor
[cli_25]: Unable to write to PMI_fd
[cli_25]: write_line error; fd=16 buf=:cmd=get_appnum
:
system msg for write_line failure : Bad file descriptor
Fatal error in MPI_Init: Other MPI error, error stack:
MPIR_Init_thread(474):
MPID_Init(152).......: channel initialization failed
MPID_Init(426).......: PMI_Get_appnum returned -1
[cli_26]: write_line error; fd=18 buf=:cmd=abort exitcode=1094415
:
system msg for write_line failure : Bad file descriptor
How to solve this problem？

 

[duvivier]

Hi Yukai! Great to see you've registered successfully! :)

 

[tcraig]

This is almost certainly a problem on your system. Are you able to try a different compiler or different version of the MPI library? The model needs a consistent installation of compilers and related software like MPI. A quick google of the error message reveals a few ideas,

dask-mpi fails to start · Issue #4214 · dask/dask

I'm using dask-mpi on a managed cluster with slurm. As a test I request two nodes with eight tasks as srun --nodes=2 --ntasks-per-node=8 --pty bash I then run mpirun --n 16 dask-mpi --scheduler-fil...

github.com

Intel® Software - Clusters and HPC Technology

clusters60.rssing.com

but I cannot say whether any are helpful. We are unlikely to be able to provide much assistance in resolving this issue. You should discuss the problem with your computer system administrators, confirm you are using the appropriate versions of the compiler and MPI lib, and see whether you can make some progress that way.

Separately, you could try running with a pe count of "1x1". This should "turn off" the MPI library in the code. It would be interesting to know if this helps, but would not be a long term solution.

 

[yukai]

Thank you!
"The model needs a consistent installation of compilers and related software like MPI."
What version of the compiler and mpi is consistent with the model?

 

[tcraig]

We test the model with various versions of the intel, pgi, cray, gnu, and nag compilers. And we test with different versions of MPI. A consistent version of compiler and MPI depends on your system installation. For instance, you generally want MPI and netcdf to be built with the same compiler that you're compiling the CICE model with. Normally, as CICE modelers, we are not involved in these steps. System administrators install the compilers and build the required libraries on the hardware, and we use them. If we have problems, we consult with the system administrators to help us resolve the problems. What kind of hardware are you working on? Is it a large shared system? Are you responsible for installing the compilers and other software?

 

[yukai]

I just installed the cice mode, but this problem occurred during the installation, I asked the administrator and he said it might be the mode problem.