ERROR using intra-comm:sockets in one participant

Using preCICE
, , ,
1

Hello everyone,

I have a rather daunting problem in establishing communication between my two solvers, namely OpenFOAM and a FEM commercial solver, while running on a cluster.

I am currently using preCICE v2.5.0 and OpenMPI for OpenFOAM.

The FEM solver, instead, has its own pre-compiled preCICE library (also 2.5.0). In order to run the solid partecipant in parallel, I have to use the intra-comm:sockets option in the precice-config.xml. Otherwise, this error shows up:

—[precice] ^[[31mERROR: ^[[0m Implicit intra-participant communications for parallel participants
are only available if preCICE was built with MPI. Either explicitly define an intra-participant
communication for each parallel participant or rebuild preCICE with
“PRECICE_MPICommunication=ON”.

Everything works perfectly fine running both partecipants in parallel IF OpenFOAM is running on only one node.
When I switch to multiple nodes, OpenFOAM hangs during initialization, as it cannot connect the primary rank to the others:

Screenshot 2025-09-30 124904
Screenshot 2025-09-30 124904808×202 115 KB

I should also mention that if I run OpenFOAM on multiple nodes, without the key intra-comm:sockets on the other participant (hence running it in serial) it works.

To communicate between the two participants I’m using:

<m2n:sockets from="Fluid" to="Solid" exchange-directory="/complete/path/to/run-folder/" network="ib0" />

Have you got any idea why this happens? I have to run huge simulations and this is a big bottleneck for me.

Thank you in advance,
Alice

2

Hi @alicezanella,

We back-ported some important fixes to the version 2 branch as the 2.5.1 release. I highly recommend you to at least upgrade to that.
In general, sticking to a recent preCICE version doesn’t only bring new shiny features, but gives you the latest bug fixes, better error checks and debugging options. Often the latter save more time than the former.

This is indeed a bit puzzling and I have some questions and suggestions:

  • Could you please upload the full configuration(s) of your case? If you aren’t allowed:
    • Do you use the intra-comm:sockets option only in the solid participant?
    • Did you configure an interface? (by default it is loopback which restricts you to a single node)
  • Is the preCICE build used by OpenFOAM configured with MPI or without? Or is it the same as used by your commercial solid code?
  • Could you compile preCICE with debug -DPRECICE_RELEASE_WITH_DEBUG_LOG=ON and then enable the debug log in your configuration? This should give you more information on where exactly the communication buildup fails.
  • Could you make sure that the exchange directory of the <m2n:sockets/> and the <intra-comm:sockets/> point to different locations and that they are empty before you start the simulation?
  • Could you try swapping the to and from in the configured m2n?