Docker issues

[javierfernandezfidalgo]

Hello everyone,

I am trying to compile and use the code which looks amazing.

The Dockerfile gives an error on the line that contains make -f Makefile_docker all. It can be corrected by replacing ARG BASE_IMAGE=ubuntu:20.04 with ARG BASE_IMAGE=ubuntu to use the latest version (maybe the changed something in Fossal Foca that prevents the code from working?).

However, after dealing with that and obtaining a correct compilation and exporting OMP_NUM_THREADS=4 I can't get the program to print anything in the terminal.

Concerning the taylor-green vortex test case, it yields a segmentation fault error somewhere so I may be doing something wrong.

Do I need to do anything else besides what's done in the dockerfile?

Thanks in advance!

[TakisCFD]

1. you need to export OMP_NUM_THREADS=N , n being the value that you want
2. you then need to run via mpirun -np M ./ucns3d_p
3. you must ensure that the grid file, UCNS3D.DAT, ucns3d_p are in the directory where you will execute the previous two commands

[javierfernandezfidalgo]

Thanks for the reply. I think that is precisely what I am doing.
I copied the .msh and .DAT from the Taylor-Green test folder and the executable from the folder it was compiled

I attach the Docker Desktop terminal output that I receive (because in VSCode I can't get any output when attaching it to the running container and that's weird)

[TakisCFD]

can you please attach your UCNS3D.DAT file?

[TakisCFD]

and your history.txt file since I can see when it crashes

[javierfernandezfidalgo]

Find attached the requested files. Thanks
files.zip

[TakisCFD]

UCNS3D.DAT.zip

Try with this UCNS3D.DAT and let me know

[javierfernandezfidalgo]

With the updated UCNS3D.DAT file keeping everything else as stated before, I get the following error now

[TakisCFD]

this is not related to the dat file, something else is probably the cause of this error.
I am not able to reproduce this error on a natively compiled (gnu compiler/mpich) system.
I have the suspicion that the mpi compiler is not properly setup in your docker environment.
The fact that you have two Parmetis Initiated and two Parmetis Operations completed because you are using two mpi processes is wrong (you should only be able to see one), so I am 100% sure that the mpi compiler in your environment is not setup correctly.

[javierfernandezfidalgo]

Thank you for the feedback Takis. I will try to manually install it on a full linux distro outside docker and see what the differences are. The docker ubuntu image is lacking a lot of important libraries and that may be the cause of the problem. I will post here if I get it sorted out.

[javierfernandezfidalgo]

I have found a solution (or maybe a workaround?). Setting up a full ubuntu distro gives the same error as in docker, meaning is not a docker issue so I went back to docker for simplicity.

I have the suspicion that the mpi compiler is not properly setup in your docker environment.

You are right. But I just followed the Dockerfile provided and this means mpich doesn't work out of the box. I would appreciate some guidance on how to set it up correctly.

However, if I ditch mpich completely and install openmpi (compiling ParMETIS with openmpi as well) it works. I was able to run the taylor-green test case with the default UCNS3D.DAT and grid.msh using gnu compiler + openmpi. I do not know if everything works as intended by changing the mpi implementation.

If you want, after I do some cleaning, I can provide a Dockerfile that builds the aforementioned configuration.

Thanks again

[tttllll00t]

I also have the same problem. Could you please attach your dockerfile so that I can operate it? Thank you very much

I have found a solution (or maybe a workaround?). Setting up a full ubuntu distro gives the same error as in docker, meaning is not a docker issue so I went back to docker for simplicity.

I have the suspicion that the mpi compiler is not properly setup in your docker environment.

You are right. But I just followed the Dockerfile provided and this means mpich doesn't work out of the box. I would appreciate some guidance on how to set it up correctly.

However, if I ditch mpich completely and install openmpi (compiling ParMETIS with openmpi as well) it works. I was able to run the taylor-green test case with the default UCNS3D.DAT and grid.msh using gnu compiler + openmpi. I do not know if everything works as intended by changing the mpi implementation.

If you want, after I do some cleaning, I can provide a Dockerfile that builds the aforementioned configuration.

Thanks again