Crash while running depletion with cluster geometry

User Support
1

Hello.
A run-time error is occurring when running a depletion problem with a CANDU type fuel bundle. The model makes use of the temperature interpolation setting, and the run time error occurs in nuclide.cpp, line 655, with the comparison:
…
const auto& grid {grid_[i_temp]};
const auto& xs {xs_[i_temp]};

if (p.E() < grid.energy.front()) {
…
Running with the debugger, it is looking like the grid object is incorrectly allocated in some manner.

The crash occurs after the full set of cross sections are loaded following the volume computation run.

Details below
OpenMC Version: 0.12.2
Git version: cbfcf908f8
Platform: 64-bit linux, CentOS
OpenMP: off.
MPI: OpenMPI 4.1.1
Python: 3.9.5, 3.9.7
Compiler: crash occurs when built using gcc-8.4.0, gcc-11.1, intel 19.1.3.304.
Cross section data: crash occurs with endfb7.1, lib80x.
Depletion chains tested: chain_endfb71_pwr_0.12.xml, chain_casl_pwr_0.12.xml.

Supplemental

  • The crash does not occur when chain_simple.xml , taken from the pincel depletion example, is used.
  • Crash does not occur if we fix all temperatures to 300K.
  • Crash does not occur for other problems, such as a molten salt reactor lattice cell.
  • Crash occurs even when diff_burnable_mats=False in the burnup operator call.
  • The crash does not occur on a separate laptop build, or on another cluster.
2

UPDATE
I have discovered that the crash does not occur when a different implementation of the HDF5 system library is used in the build. We had two: one with -parallel in the module title and one without. Interestingly, it is the one with “-parallel” that is causing the problem. I will consult with our sysadmins to better understand this.

3

@AlexandreTrottier sorry for the late response on this. Were you ever able to get to the bottom of what was happening? It can be difficult to get HDF5, MPI, and h5py to play nicely together (see notes in install instructions here).

4

OFFICIAL USE ONLY / À USAGE EXCLUSIF

Hi,

The issue was traced to the use of a parallel version of the HDF5 library. We have addressed temporarily by using a non-parallel HDF5.

Sincerely,

AT

image002.jpg