On 03/03/2014 10:11 AM, Rhys Ulerich wrote:
Lastly, am I nuts to be running atop OpenMPI 1.6.5 in the first place? I ask as http://www.hdfgroup.org/HDF5/release/platforms5.html shows no OpenMPI coverage in the test matrix.
I would be interested to hear if this problem occurs with MPICH-3.1: the backtraces indicate "stuff" is going on in the attribute cleanup code, and we (the MPICH/ROMIO folks) have had some back and forth with the OpenMPI folks about order of destruction of attributes. I *thought* we had that cleared up several years ago. If you can test with MPICH that will help us understand the problem a bit better.
You can see in one set of backtraces a subset of HDF5 processs are trying to resize the file at close, while meanwhile another set of MPI processes are trying to close the file as part of a "destructor".
==rob -- Rob Latham Mathematics and Computer Science Division Argonne National Lab, IL USA _______________________________________________ Hdf-forum is for HDF software users discussion. [email protected] http://mail.lists.hdfgroup.org/mailman/listinfo/hdf-forum_lists.hdfgroup.org
