New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Recurring MPI segmentation fault #2641
Comments
The exact test is:
but you can reduce it down to
and it will still segfault |
Would be helpful if you can give a (full) file that reproduce and the command you used. |
@Helveg : As Nico mentioned, we need help to reproduce the issue. In pramodk/patch/pull/1, you can see my failed attempt to reproduce the segfault. I thought, importing neuron at the top/beginning, the issue should be reproduced? By the way, only additional thing I added is either I am running |
I think it's the exact sequence in which Patch sets things up with or without
I'll be working on this again in a few days, I can get you a better reproducer if this doesn't do the trick |
I went ahead and created a reproducer: https://github.com/Helveg/nrn-segfault/tree/main The setup sequence of Patch with And using a debugger you should be able to follow the steps taken by the |
* switched to pyproject.toml * define public api * black * bump glia * fix workflow file * fix workflow file pt 2 * fix main workflow deps * drop 3.8 * run some tests separately to avoid segfault see neuronsimulator/nrn#2641 * fix parallel spike assertions * bump numpy dep for faster 3.11 tests
* dump * add multi sim and test * fix merged pyproject * fix numpy ints messing with NEURON * fix NeuronPopulation typo and numpy int lookup * improve arborized model type handling (still open issue) * add chunked test * fix multi CM/chunk transmapping + gid on instances * validate multichunk test * validate multi CM * add ci * fix ci * use fixed arborize version that works without installing arbor * bump deps * avoid neuronsimulator/nrn#2641
Context
In many of my packages NEURON and MPI have to interact, and unless I import
mpi4py
before NEURON tries any MPI init, I get stuck at the following point:If you need I can give you some repos and commits to look at. It happens on Ubuntu + OpenMPI installed from
apt
and any pip installable NEURON version (I've checked until8.0.0
)The simplest test under which it occurred is in
nrn-patch
:which goes through a bit of internals which auto detects if we're running under MPI and does a lot of the weird
ParallelContext
calls you'd need to do in NEURON before anything parallel works. If you look at:https://github.com/dbbs-lab/patch/actions/runs/7061832089/job/19224331875#step:7:76
https://github.com/dbbs-lab/patch/actions/runs/7061933373/job/19224647645#step:7:102
Then you have an exact reproducer of how importing
mpi4py
before NEURON resolves the issue.The text was updated successfully, but these errors were encountered: