Tomawac problem when writing spectra file in parallel
Running Towawac on cluster (on 64 cores) with 451 spectra output points could leads to crash during simulation in ecrspe.f
The crash is not systematic and could occur at different time in the simulation...
It seems it occur for the same points in my case (probably linked to the domain decomposition)
The report of the crash is
Click to expand
forrtl: File exists
forrtl: severe (10): cannot overwrite existing file, unit 26, file /scratch/ccoulet/335064/00451-00316
Image PC Routine Line Source
out_user_fortran 00000000004430AA for__io_return Unknown Unknown
out_user_fortran 000000000045CB7D for_open Unknown Unknown
libtomawac4telema 000014AF289AF95C ecrspe_ Unknown Unknown
libtomawac4telema 000014AF28A4683F wac_ Unknown Unknown
libtelemac2d.so 000014AF29C69A32 telemac2d_ Unknown Unknown
libtelemac2d.so 000014AF29CBF267 MAIN__ Unknown Unknown
out_user_fortran 0000000000417A52 Unknown Unknown Unknown
libc-2.26.so 000014AF2586334A __libc_start_main Unknown Unknown
out_user_fortran 000000000041796A Unknown Unknown Unknown
According to ecrspe.f code, it seems each core write it's own points. So if a point belongs 2 two subdomains, I imagine we could have this kind of problem. Nevertheless, I unable to explain why sometimes it works and sometimes not...