Error Output When Running in Parallel (Mpirun)

[anandanandan86]

I am getting the following error message repeatedly output (line after line, endlessly) when running OpenLB in parallel using mpirun:

[############:#####] Read -1, expected ######, errno = 1

Does any one know if there is any way to stop this error message from appearing? The calculation runs fine without this error in serial mode. Also, even with the error, contour images and results appear to be output correctly in parallel mode, but I would still like to remove the error if possible. Thanks in advance.

[anandanandan86]

Note: the #s I included in the error message stand in for alphanumeric characters that occasionally change depending on the line

[Adrian]

It is hard to tell you anything here without knowing how exactly you executed this and what exactly your program implements.

Are you sure that you compiled in MPI mode using the correct compiler settings?

[anandanandan86]

I think the correct compiler settings should have been used as there is no error during compilation, and the calculation is exporting contours and result files as expected (at least below a certain core count). Is there any file that I should share to shed more light on the issue?

The program is a modification of the cylinder3d example case that has been changed to model flow within a 3d room with a few obstructions inside. The Smagorinsky turbulence model that comes with OpenLB is also implemented in this case (in contrast to the original cylinder3d case, which was laminar).

[Adrian]

Did you add any custom IO? Do these messages appear during initialization or e.g. for every getResults / post processing call?

I could imagine such messages originating from e.g. faulty gnuplot execution. Do they still appear if you only export VTK files?

[anandanandan86]

Hello, I checked and this error seems to appear during every iteration. In this calculation setBoundaryValues is being called outside the loop (only at the start of the calculation as boundary conditions do not change during the calculation). I also commented out getResults temporarily, but even so, the error is output repeatedly during each iteration (with only collideAndStream being called for each iteration).

[Adrian]

I just stumbled upon an issue where these same error messages are generated for another MPI-based simulation tool. Are you perchance running this in a Docker container / some oder abstraction layer?

[anandanandan86]

Hello, this is quite a late response but yes, we are running in a Docker container. Is there any way to suppress the output in that case?

[Adrian]

As per the referenced issue: If you only care about suppressing the error, setting OMPI_MCA_btl_vader_single_copy_mechanism=none seems to work.

[Author]

Posts