ufs-community / ufs-weather-model

UFS Weather Model
Other
141 stars 248 forks source link

CICE PIO w/ HDF5 fails to write output via link #2260

Open aerorahul opened 6 months ago

aerorahul commented 6 months ago

Description

The UFS-weather-model PR #2145 recently added control for PIO namelist in CICE. The update to CICE, specifically in ice_pio.F90 exposed an issue writing the output via links when using hdf5 for CICE history and restart formats. This was discovered in the global-workflow on WCOSS2, when the hash of the ufs-weather-model was attempted to be updated. See discussion in https://github.com/NOAA-EMC/global-workflow/pull/2505

Thanks to @DeniseWorthen for tracing the source of the issue and working with developers to provide an update that will re-enable writing CICE output (and restarts) via links when hdf5 is used.

HDF5 is being used on WCOSS2, because it does not have the correct version of the pnetcdf library. An update is expected soon for WCOSS2 and pnetcdf. This error was reproduced on Hera by choosing hdf5.

To Reproduce:

An error such as below will be experienced:

21:  (abort_ice)ABORTED:
21:  (abort_ice) called from ice_pio.F90
21:  (abort_ice) line number          223
21:  (abort_ice) error =
21:  (ice_pio_check)Permission denied, (ice_pio_init) ERROR: Failed to create file .
21:  /CICE_OUTPUT/iceh_ic.2021-03-23-43200.nc

Additional context

See global-workflow PR https://github.com/NOAA-EMC/global-workflow/pull/2505

Output

https://gist.github.com/WalterKolczynski-NOAA/54d183f100ffcbac2786d393766c490c

DeniseWorthen commented 6 months ago

The rebuilt pio is now available on Cactus (#2232). If it resolves the issue w/ using pnetcdf, a UWM hotfix will be required to revert the WCOSS2 work-around in default_vars. The need to use hdf5 on WCOSS2 is breaking the G-W. @BrianCurtis-NOAA @jkbk2004

DeniseWorthen commented 6 months ago

@aerorahul I was provided an interesting link from one of our CICE colleagues https://github.com/open-mpi/ompi/issues/12141. I don't follow all the details in that issue or how relevant it is to our case.