[FLASH-USERS] H5Screate_simple: zero sized dimension

[Christoper Daley]

Mateusz Ruszkowski wrote:
>
>
> Hi all,
>
> [apologies if you are getting this message more then once; we had some 
> trouble sending the e-mail]
>
> We are getting a runtime error related to hdf5. The crash occurs the 
> moment the code is attempting to write the initial conditions. The 
> standard output includes the following error messages:
>
> .
> .
> .
> Initial dt verified
> Driver_abortC called
> Error: negative return from memspace H5Screate_simple
> Calling MPI_Abort for immediate shutdown
>
> and:
>
> .
> .
> .
> HDF5-DIAG: Error detected in HDF5 library version 1.6.6-snap21 
> MPI-process 0. Back trace follows.
> #000: H5S.c line 1595 in H5Screate_simple(): zero sized dimension for 
> non-unlimited dimension
> major(01): Function arguments
> minor(05): Bad value
> MPI: On host columbia1, Program 
> /u/.realmounts/home7/nieusma/Flash/Planet_IR, Rank 0, Process 8697
> 1 called MPI_Abort(<communicator>, 1)
>
>
> A similar error was reported on the FLASH mailing list in the past and 
> a suggestion was made that this could be due to (min refinement) > 
> (max refinement) in flash.par. We checked the refinement levels and 
> this does not appear to be a problem in the above run so the cause of 
> this problem is probably different in our case. Did anybody encounter 
> similar issues?
>
>
> thanks,
> Mateusz & Justin
>

Hi Mateusz, Justin,

I think this is happening because there are zero blocks on at least
one processor. You can check whether this is the case by looking in
the flash log file from your run. Look for the last
amr_refine_derefine message of the form:

[GRID amr_refine_derefine] min blks 17 max blks 18 tot blks 69

I expect there is a min blks 0 in your message. A temporary
workaround is to run your application on fewer processors.

Regards,
Chris