MDAnalysis API leads to Internal error (bug) / segfault about plumed2 HOT 14 CLOSED

yes it may be, even if I think that the problem should have been fixed in plumed 2.8.2 (sorry I am editing my former reply). Did you patched gromacs with an earlier version of plumed?

from plumed2.

Hi,
Thanks for the quick reply!

I did patch it with 2.7.6 before, but actually I reverted the patch (plumed patch -R) before I reapplied the 2.8.2 patch.

I will try to revert the 2.8.2 patch and then patch and recompile with the current master branch.
If that doesn't work, I might actually recompile gromacs from scratch and apply the 2.8.2 patch.
I know that in terms of debugging, it would probably be good to do it the other way around, but I don't want to be too invasive on this machine. (Currently, I expect an issue with my installation rather than an actual bug, so I think for you it does not make a difference for now, or does it?)
After all that, I will report on the result. This might take a while, because the machine is currently still running with other jobs that I don't want to interrupt...

from plumed2.

Update on this matter:
I tried a fresh installation of gromacs-2022.5 with the plumed patch version 2.8.2 (on a different machine) and I have the problem there, too:

• simulation without plumed gmx mdrun -s npt.tpr : no visible problems
• simulation with plumed gmx mdrun -s npt.tpr -plumed plumed.dat: either segfaults immediately, or runs for a few steps, but the protein blows up.

Content of plumed.dat:

MOLINFO STRUCTURE=start.pdb PYTHON_BIN=python3 MOLTYPE=protein
OW: GROUP ATOMS={@mda:{name OW}}

I did not try the current master branch, yet. This is what I will do next.

from plumed2.

@Clownshift two probably silly questions:

• Did you try without the OW: GROUP ATOMS={@mda:{name OW}} line?
• Did you try with an empty plumed.dat file?

Your input should not do anything. However, the second line will launch a separate process with a python interpreter. It should not interfere with GROMACS, but who knows. The first line instead should just read a file. I don't know how it could interfere with GROMACS.

Thanks!

from plumed2.

Thanks for the super quick reply! I just tried both your suggestions. (I only ran the simulations for a few picoseconds, but the above described issues happen basically immediately...)

• Did you try without the OW: GROUP ATOMS={@mda:{name OW}} line?

If I comment this out, it runs with no errors or visible problems whatsoever.

• Did you try with an empty plumed.dat file?

Also that works without any issues.

from plumed2.

I will still go ahead and apply a patch from the current master branch and give you an update.

from plumed2.

By the way, I did not describe that very precisely before: If I start the simulation repeatedly, sometimes it actually runs for a few steps. If I look at this trajectory the protein unfolds immediately. (This is completely unexpected, since it is a stable globular protein actually).

I am not sure, if that helps to understand the issue...

from plumed2.

I think it's independent of the patch. There's something wrong in launching a separate process from plumed. I think there's no obvious solution to this, I guess it depends on the exact configuration.

It might be useful if you post for future reference as much information as possible about your system. E.g.:

• how you installed MDAnalysis
• exact compilers and MPI version used for gromacs and plumed.
• operating system

As a practical solution, I suggest to use python to create a comma separated list of atom numbers and then insert it in the plumed input. Be careful that numbering starts from 1 (not from 0)

from plumed2.

Also notice that the plumed selectors can be used as well. It's less flexible than MDAnalysis though

from plumed2.

Alright, some specs about the machine I am currently working on.

OS: Ubuntu 20.04.
Python installation: Anaconda3 (conda version 22.9) I use a conda environment in which I installed MDAnalysis.
gcc version: 9.4
CUDA version: 11.5
GMX version: 2022.5 (with CUDA)
Both GMX and plumed have been compiled without MPI. I used the same compiler versions for GMX and for plumed.
Also, to mention that explicitly, I compiled plumed from the downloaded tar-archive. I did not try to install it via conda.

from plumed2.

Also notice that the plumed selectors can be used as well. It's less flexible than MDAnalysis though

Okay, thanks for the alternative solutions.

from plumed2.

Just for completeness:
I patched gromacs 2022.5 with plumed version 2.9b and the problem still persists...

branch: https://github.com/plumed/plumed2/tree/v2.9
last commit: 610cb32

from plumed2.

I think it's independent of the patch. There's something wrong in launching a separate process from plumed. I think there's no obvious solution to this, I guess it depends on the exact configuration.

Based on your comment, I will give up on trying to reinstall, but will workaround this issue, as you suggested above.

If I can do some additional tests that would help you to figure out this issue, feel free to ask.

from plumed2.

For reference:
I solved the issue with a gromacs ndx-file that I generated with MDAnalysis. Very solid workaround.

I assume that the issue is caused by my particular installation. Therefore, I close this issue.

from plumed2.