Segmentation Fault

[ecole41]

I had a working Pinefarm environment and I now get this error whenever trying to run pinefarm:

zsh: segmentation fault

This has started after trying to produce a T3 positivity FK-table by running pinefarm run T3 /Users/ellacole/codes/nnpdf/nnpdfgit/nnpdf/nnpdf_data/nnpdf_data/theory_cards/40000000.yaml

where the T3 directory is a copy of theNNPDF_POS_ANTI_DOWNdirectory with pid: -1 changed to flavour: 103.

I now get this segmentation error even when running pinefarm run using an NNPDF pinecard such as ATLAS_WP_7TEV:

zsh: segmentation fault  pinefarm run ATLAS_WP_7TEV 
/Users/ellacole/miniconda3/envs/pinefarm_2/lib/python3.9/multiprocessing/resource_tracker.py:216: UserWarning: resource_tracker: There appear to be 1 leaked semaphore objects to clean up at shutdown

I have tried to reinstall pinefarm in a new environment and still get this.

Please let me know if this is something that you have seen before or might know how to help. Thank you

[felixhekhorn]

concerning T3: mmm, you can not just change the pid, I think - also because this convention of $T_3 \sim 103$, is just an invention of us NNPDF theory people. (I assume you meant pid: 103 and not flavour: 103, right? else this should have already failed reading the pinecard). Take a look to the positivity grid generator: https://github.com/NNPDF/pinefarm/blob/main/src/pinefarm/external/positivity.py and you can see immediately that this was not written for other then flavour space PDF - BUT of course you can just adjust it 😇 I'm not 100% sure yet on what the best strategy is (also think for yourself 🙃 ), but maybe we can reuse some of the logic for the integrability grid generator as there we can deal with evolution flavour elements

concerning the seg fault: mmm, I've never seen this and I don't understand on how this can happen in a fresh environment ... is your global environment somehow mixed in (for mg5 or something)? can you get some more debug information? e.g. add some random print statements here and there in the code to see which point you reach and where you fail. Also let's try to ping @scarlehoff

[ecole41]

I think you're right that it's an issue with mg5, as this now isn't working for me. I'll try to get mg5 working and then hopefully this issue will be fixed. Thank you

[ecole41]

I've found that I can temporarily fix this segmentation fault error by removing the ./.prefix directory (where lhapdf was installed). After removing that directory, running a positivity grid would work once, but on the next run I'd get the same segmentation fault until I removed it again.
To try to fix this permanently, I have switched to using a working lhapdf installed in my conda environment by pointing to this is the pinefarm.toml file but I then get this error:
FileExistsError: [Errno 17] File exists: '/Users/ellacole/miniconda3/envs/pinefarm_mg5/bin/lhapdf'

It looks as though whenever I run a positivity grid, pinefarm tries to install a new lhapdf.

[felixhekhorn]

It looks as though whenever I run a positivity grid, pinefarm tries to install a new lhapdf.

okay, if that's the case (which might well be), maybe there is something we should do. So, from here

pinefarm/src/pinefarm/external/interface.py

Line 91 in cae7fdf

_ = install.lhapdf()

we jump to this function:

pinefarm/src/pinefarm/install.py

Line 298 in cae7fdf

def lhapdf():

now, if pinefarm wants to install a new version, this means that the installed() function there does not detect the existing one. Can you please check this independently? i.e. inside your environment run import lhapdf - if that does not work, then you didn't install the python module correctly; if it does work, pinefarm should not try to reinstall - and there might be other problems lurking around. E.g. you can then try to actually use a PDF and not just import the module.

I can understand why an attempted reinstall would trigger a FileExistsError, but I don't understand how it would cause the segmentation fault (which is where we started)

Also if it is LHAPDF then everybody (all generators) should have the problem as they all need LHAPDF (i.e. I don't think this is related to positivity specifically)