MCT error: GlobalSegMap haloed
Up to Bugs and debugs
Posted by Anonymous at January 21 2020
Hi,
I am test running EC-Earth4 (OIFS+NEMO4) and I see an OASIS/MCT error that I do not recognise:
MCT::m_AttrVectComms::GSM_gather_: Input GlobalSegMap haloed--not allowed error, stat =1 03F.MCT(MPEU)::die.: from MCT::m_AttrVectComms::GSM_gather_() application called MPI_Abort(MPI_COMM_WORLD, 2) - process 63
This happens when I increase the number of cores for OIFS and NEMO. Not too many, just ca. 200+100. With less cores, the error doesn't happen. I do not know this error from our old model version. I use SCRIP/GAUSWGT interpolations. I have TREAT_OVERLAY activated, even though I think it is not needed for this mapping. I've heard that the error is known to other people using OASIS+NEMO, particularly when activating land domain removal. However, it never occurred to us in the old EC-Earth3 version, even with land domain removal activated. It has been suggested to comment out the test that triggers the error in lib/mct/mct/m_AttrVectComms.F90, and that works in my case.
Nevertheless, I would like to raise the problem here to encourage a more permanent solution. Thanks a lot in advance for your input!
Best regards, Uwe. -- Uwe Fladrich (SMHI, Rossby Centre)
Posted by Anonymous at January 30 2020
Hi Uwe,
Eric Maisonnave had exactly the same problem at the MetOffice and he did exactly what you did. There is a ticket opened on the problem #2472 and we work on it.
Best regards, Laure
Posted by Anonymous at March 4 2020
Hi Laure,
I have the same problem while coupling COSMO-CLM with NEMO4. Can you please tell me how to solve the problem?
Best regards, Ha Hagemann
Posted by Anonymous at March 5 2020
Hi,
In oasis3-mct/lib/mct/mct/m_AttrVectComms.F90 you must comment the following lines (from line 693):
if(GlobalSegMap_haloed(GSMap)) then
ierr = 1
call die(myname_,"Input GlobalSegMap haloed--not allowed",ierr)
endif
Best regards, Laure
Posted by Anonymous at March 6 2020
Thanks a lot, Laure!
The error has gone. Can you please explain me what happened with that CALL in MCT? Is that a bug? If yes, was it removed in the newer version of OASIS3-MCT? I'm using the version 3.0.
Best regards, Ha
Posted by Anonymous at March 7 2020
Hi,
Please note that the best way to solve this problem is to update your NEMO sources with a version more recent than 4.0.3 .
However, the work around indicated by Laure, i.e. commenting lines 693-696 in oasis3-mct/lib/mct/mct/m_AttrVectComms.F90 (line 695 is "call die(myname_,"Input GlobalSegMap haloed--not allowed",ierr)"), is also a valid option.
Regards, Sophie
Posted by Anonymous at March 8 2020
The problem with the call in MCT came from a bug in NEMO (defining partitions with overlapping haloes) which is not allowed in OASIS3-MCT !
With best regards, Sophie
Posted by Anonymous at December 2 2022
I still have this error with NEMO 4.0.4
Posted by Anonymous at December 2 2022
I still have the same problem with NEMO 4.0.4 if I use land suppression and EXPOUT in the OASIS namelist file.
Posted by Anonymous at December 2 2022
Can you confirm that if you don't use land suppression, you don't have the problem? The EXPOUT specification most probably has nothing to see with this problem.
Sophie
