Crash of OMIP2 simulation using TL319_tn14 grid and noresm2_5_alpha04_v3

[matsbn]

Using branch feature/noresm2_5_alpha04_v3 of https://github.com/mvertens/NorESM.git, compset 2000_DATM%JRA_SLND_CICE_BLOM_DROF%JRA_SGLC_SWAV_SESP and grid combination TL319_tn14, the simulation crashed at what seemed the first attempt to write CICE diagnostics.

The error message in cesm.log.* was:
[b4167:12257] *** An error occurred in MPI_Gather
[b4167:12257] *** reported by process [47501952548864,123]
[b4167:12257] *** on communicator MPI COMMUNICATOR 49 SPLIT FROM 44
[b4167:12257] *** MPI_ERR_TRUNCATE: message truncated
[b4167:12257] *** MPI_ERRORS_ARE_FATAL (processes in this communicator will now abort,
[b4167:12257] *** and potentially your MPI job)

I had a feeling it could have something to do with LFS stripes and I saw that in env_run.xml, PIO_STRIDE was set to $MAX_MPITASKS_PER_NODE. This is 128 on Betzy, but I was running with fewer processors that this for CICE (96 processors). When manually setting PIO_STRIDE to 8 for all components (a bit random), the simulation ran fine. Not sure this is the reason for the crash, but if it is, maybe PIO_STRIDE should be set to the minimum of $MAX_MPITASKS_PER_NODE and processors per component?