RankArray strange behavior on MPI branch about ttk

[m-s-will]

Hmm that is indeed weird, because apparently it gets the information from rank 0, otherwise there would be garbage data in there. Does it always shop up as rank 0? What do these globalids report on their vtkGhostType e.g. in spreadsheet view? I will look further into it tomorrow.

[eve-le-guillou]

Thank you for your reponse.

Yes, it does always show up as rank 0. I haven't been able to reproduce the problem when executing the pipeline in the graphical interface, so I don't know what value has the vtkGhostType in the Spreadsheet view. It can probably be accessed from within TTK, though.

[m-s-will]

I also am not able to reproduce the problem in the Paraview UI and for some reason the pipeline.py does not run on my setup, pvbatch says that TTKGhostCellPreconditioning is undefined even though the python module and the TTK VTK wrappers should be built and I can find it by using:

import topologytoolkit
ghostCellPrecond = topologytoolkit.ttkGhostCellPreconditioning()

Is there something separate that needs to be done so I can use it with import paraview.simple?

[eve-le-guillou]

Indeed, I forgot to add from paraview.simple import * at the beginning of the second pipeline. I don't see anything else needed to execute the Pipeline. I compiled TTK on my machine using the following tutorial: https://topology-tool-kit.github.io/installation-linux-sources.html.
I didn't change any of the default path variables.

[m-s-will]

I got the problem to reproduce, but for me it doesn't always show up as rank 0, as expected there is sometimes garbage data in there. That means that the cause is probably some messages from R3 getting lost, which would also be consistent with the fact that the problem occurs more often when the system is under heavy load. I will investigate more if it really is not arriving or if the data in it is somehow seen as the wrong datatype, and hopefully find a fix in the next few days.

[m-s-will]

It seems like the problem is that sometimes the first few values of the sent array are not received correctly, leading to a corruption of these specific vertices. I have changed some things and am currently testing if it still appears.

[m-s-will]

Hi Eve,
please check out the PR, this (hopefully) fixes the error. I am currently on ~400 pipeline runs without the error appearing. Let me know if this or anything similar still appears for you.
Greetings,
Michael

[eve-le-guillou]

Hello Michael,

Thank you for looking into this.

I have compiled again using your code, and it does appear that the issue is no longer there. I have executed it over 50 times, while under heavy load.

Thank you very much!
Eve

[eve-le-guillou]

Seems like this issue has been corrected and the associated PR as been merged.
Thanks,