Etienne,

Thank you. I just did what Timo advised (to follow your approach).

It might be out of Dumux hands. If Dune doesn't handle recovering from the exception and deadlocks before we get back to Dumux, we cannot really do anything. Here is a paper about this general problem with MPI and C++ exceptions from the Dune community (https://arxiv.org/abs/1804.04481).

In this case, the only thing that helps (without implementing some error propagation framework) is Etienne's suggestion of manually removing exception throwing where you consider it acceptable failures.


Best regards,

Dmitry



On 6/30/22 10:22, Etienne Ahusborde wrote:
Hi Dmitry ,

We already had the same problem.

We understood that it is due to some "DUNE_THROW" commands.

Our solution (probably not the best) was to remove several "DUNE_THROW" commands and to replace them with a simple std::cout to keep the information that something is wrong.

Regards

Etienne

Le 29/06/2022 à 19:36, Dmitry Pavlov a écrit :

Hello,

In one porous flow simulation, I get this message:

Newton iteration  6 done, residual = 3.1654e+13
Solve: M deltax^k = rNewton: Caught exception: "FMatrixError [luDecomposition:/home/dpavlov/DUMUX/dune-common/dune/common/densematrix.hh:939]: matrix is singular"

... and it is fine. Matrix can be singular, no objection here. After this, the Newton solver reports that it did not converge and restarts with a smaller time step, and eventually it proceeds.

However, when I run the same simulation with MPI, I get this

Newton iteration  8 done, residual = 7.2437e+13
Solve: M deltax^k =

and the program hangs.

So it seems that it misses the singularity in the MPI mode and tries to proceed with the current time step, to no avail.

How this can be cured?

Best regards,

Dmitry


_______________________________________________
DuMux mailing list
DuMux@listserv.uni-stuttgart.de
https://listserv.uni-stuttgart.de/mailman/listinfo/dumux
_______________________________________________
DuMux mailing list
DuMux@listserv.uni-stuttgart.de
https://listserv.uni-stuttgart.de/mailman/listinfo/dumux
_______________________________________________
DuMux mailing list
DuMux@listserv.uni-stuttgart.de
https://listserv.uni-stuttgart.de/mailman/listinfo/dumux

Reply via email to