Proper shutdown of NEMO+OASIS if NEMO aborts...

[masif]

It has been observed that the coupled simulation of EC-Earth3 (having nemo-3.3.1) once/as failed due to NEMO blow-up (particularly the following case has been observed) keep on hanging in queue system of HPC until the assigned wall-clock time finishes.

===>>> : E R R O R
         ===========

  stpctl: the zonal velocity is larger than 20 m/s
  ======
 kt=  8505 max abs(U):  8.7432E+96, i j k:   967    3    1

           output of last fields in numwso

In consultation with coupled model developers/experts, the following options have been tested and technically now the coupled model works fine:

• option # 1

>  Index: sources/nemo-3.3.1/NEMO/OPA_SRC/nemogcm.F90
>  ===================================================================
>  --- sources/nemo-3.3.1/NEMO/OPA_SRC/nemogcm.F90    (revision 1862)
>  +++ sources/nemo-3.3.1/NEMO/OPA_SRC/nemogcm.F90    (working copy)
>  @@ -168,6 +168,9 @@
>         IF( nstop /= 0 .AND. lwp ) THEN   ! error print
>            WRITE(numout,cform_err)
>            WRITE(numout,*) nstop, ' error have been found' 
>  +         #if defined key_oasis3 || defined key_oasis4
>  +         CALL prism_abort_proto(0, 'nemo_gcm', 'ABORT because of error')
>  +         #endif
>         ENDIF
>         !
>         CALL nemo_closefile

• option # 2

Index: sources/nemo-3.3.1/NEMO/OPA_SRC/nemogcm.F90
===================================================================
--- sources/nemo-3.3.1/NEMO/OPA_SRC/nemogcm.F90 (revision 1862)
+++ sources/nemo-3.3.1/NEMO/OPA_SRC/nemogcm.F90 (working copy)
@@ -172,7 +172,11 @@
       !
       CALL nemo_closefile
 #if defined key_oasis3 || defined key_oasis4
+    IF (nstop==0) THEN
       CALL cpl_prism_finalize           ! end coupling and mpp communications with OASIS
+    ELSE
+      CALL prism_abort_proto(0, 'nemo_gcm', 'ABORT because of error')
+    ENDIF
 #else
       IF( lk_mpp )   CALL mppstop       ! end mpp communications
 #endif

May we request to bring the change into the latest nemo release.

Thanks a lot in advance.

[timgraham]

I agree that this has been a pain at earlier NEMO versions but from looking at the code this is already fixed in NEMO3.6 so should no longer be an issue.

In 3.6 the coupled model nemo_gcm will call cpl_finalize (in SBC/cpl_oasis3.F90) which in turn calls oasis_abort if nstop=/0.