Open itay-space opened 3 months ago
Hi @itay-space! Thank you very much for reporting this and for the simple working example. I'll dig into it as soon as I can.
Hi @pshriwise
I wanted to follow up on the issue I reported last week. Were you able to reproduce the problem?
Your assistance is greatly appreciated. Thank you for your help!
Hi @itay-space!
Thanks for your patience. I found the cause of the problem for the example you've provided above. The underlying issue was actually resolved in https://github.com/openmc-dev/openmc/pull/2811, which should be included in the latest OpenMC release. In this case, it was causing nan
values in a particle position that led to an infinite loop in the mesh filter routine.
This doesn't seem to be the cause of all the hanging simulations you linked above, however, so I'm going to be looking into those as well separately, but I wanted to update you on this one sooner rather than later.
Bug Description
There is a serious bug in the calculations involving a mesh tally, specifically with the track-length estimator. The simulation freezes without producing any additional statepoints. This issue is evident because when it occurs, the number of CPUs in use drops from utilizing all available OpenMP threads to just one. Similar behavior has been reported multiple times in the OpenMC discussion forum: openmc-freezes-when-simulating-batches simulation-freezing particle-lost-without-warning-error-or-being-killed-stalling-the-run
Steps to Reproduce
Attached is a straightforward demonstration of this problem. The input file can be generated using the following Python code. This code will create a wall made of the H1 isotope and a mesh filter surrounding it. Additionally, there is an attached figure illustrating the simple geometry, with an arrow indicating the beam direction. In this example (with seed=17) the simulation will crash around the 6th batch.
Environment
The tests were done using the latest docker image:
And some more details:
Data library: fendl-3.2-hdf5