"John Urbanic" <urbanic@xxxxxxx> writes:
> Ed:
>
> After building with --enable-logging (I cannot figure the Fortran API for
> nc_set_log_level) I do indeed get meaningful error logging. I get the below
> sequence every time the put_var fails. This error is sporadic in both the
> variable affected as well as the file; about 90% of the files during this
> particular run were just fine, and specific failure points will vary from
> run to run. All PE's reported this same error here (I have only included
> one below), but often PEs succeed in writing even when others fail.
>
> As the trace terminates with the fairly discouraging "major: Internal error
> (too specific to document in detail)", I am at a loss. I did take Hernan's
> implied advice and built a 4.1.1 version, but it throws similar errors
> (involving "decrementing file ID failed").
>
> At this point, our netcdf conversion is at the mercy of your expert insight.
> Actually, the conversion is done - now it is a question of whether we have
> wasted our time...
>
> Hopeful,
> John
Howdy John!
What this is telling me is that this is not a HDF5 or netCDF problem,
but that your MPI layer is throwing the error at the HDF5 library.
What MPI layer are you using? On what platform? Is there an alternative
available?
For example, I use MPICH2...
Thanks,
Ed
--
Ed Hartnett -- ed@xxxxxxxxxxxxxxxx
