MadGraph5_aMC@NLO

Apparent memory leak during output diagrams in Pop!_OS: MG5 v3.4.1

Asked by Adrian

I am running MG5 v3.4.1 on Pop!_os linux distribution. When I try to generate diagrams for any simple process, MG5 hangs when I try output <foldername>;
$ ./MG5_aMC_v3_4_1/bin/mg5_aMC
...
[sm model loaded]
...
MG5_aMC>generate e+ e- > e+ e-
INFO: Checking for minimal orders which gives processes.
INFO: Please specify coupling orders to bypass this step.
INFO: Trying process: e+ e- > e+ e- WEIGHTED<=4 @1
INFO: Process has 4 diagrams
1 processes with 4 diagrams generated in 0.004 s
Total: 1 processes with 4 diagrams
MG5_aMC>output test

I notice rampant memory usage in the system monitor at this last step until the computer freezes. I'm not sure if this is a Pop!_os issue or an MG5 issue, but I cannot replicate the issue in MG5 v2.9.13.

However, in v3.4.1 if I instead just use 'output' with no specified folder name, things proceed smoothly;

MG5_aMC>output
INFO: initialize a new directory: PROC_sm_0
INFO: remove old information in PROC_sm_0
INFO: Organizing processes into subprocess groups
INFO: Generating Helas calls for process: e+ e- > e+ e- WEIGHTED<=4 @1
INFO: Processing color information for process: e+ e- > e+ e- @1
INFO: Creating files in directory P1_ll_ll
INFO: Some T-channel width have been set to zero [new since 2.8.0]
 if you want to keep this width please set "zerowidth_tchannel" to False
INFO: Generating Feynman diagrams for Process: e+ e- > e+ e- WEIGHTED<=4 @1
INFO: Finding symmetric diagrams for subprocess group ll_ll
Generated helas calls for 1 subprocesses (4 diagrams) in 0.004 s
Wrote files for 12 helas calls in 0.024 s
ALOHA: aloha starts to compute helicity amplitudes
ALOHA: aloha creates 3 routines in 0.631 s
save configuration file to /home/thompson/projects/madgraph/PROC_sm_0/Cards/me5_configuration.txt
INFO: Use Fortran compiler gfortran
INFO: Use c++ compiler g++
INFO: Generate jpeg diagrams
INFO: Generate web pages
Output to directory /home/thompson/projects/madgraph/PROC_sm_0 done.
Type "launch" to generate events from this process, or see
/home/thompson/projects/madgraph/PROC_sm_0/README
Run "open index.html" to see more information about this process.
MG5_aMC>

Any idea what the issue could be? Would love some help to debug! In the meantime, I will use 2.9.13 for now which seems to work fine.

Question information

Language:
English Edit question
Status:
Solved
For:
MadGraph5_aMC@NLO Edit question
Assignee:
No assignee Edit question
Solved by:
Olivier Mattelaer
Solved:
Last query:
Last reply:
Revision history for this message
Olivier Mattelaer (olivier-mattelaer) said : 		#1

HI,

I do not reproduce this at all, the only suggestion is that the "test" directory might be already existing.
But in that case you should see:

INFO: directory /Users/omattelaer/Documents/git_workspace/version3/test already exists.
If you continue this directory will be deleted and replaced.
Do you want to continue? [y, n][60s to answer]

As you see, the code is expected to sleep for 60s and then restart by overwritting the "test" directory which can be slow/... depending of the directory and underneath filesystem.

If you do have a debug file created it might have more information. So far I do not know what I can do.

Cheers,

Olivier

Revision history for this message
Adrian (adr-traq) said : 		#2

No debug file was produced (I'm actually not sure how to turn on logs from MG5) but maybe this message when I try to kill the process:

-------------------------
^CTraceback (most recent call last):
  File "/home/thompson/MG5_aMC_v3_4_1/bin/mg5_aMC", line 145, in <module>
    import madgraph.interface.master_interface as interface
  File "/home/thompson/MG5_aMC_v3_4_1/madgraph/interface/master_interface.py", line 44, in <module>
    import madgraph.interface.madgraph_interface as MGcmd
  File "/home/thompson/MG5_aMC_v3_4_1/madgraph/interface/madgraph_interface.py", line 72, in <module>
    import madgraph.iolibs.export_cpp as export_cpp
  File "/home/thompson/MG5_aMC_v3_4_1/madgraph/iolibs/export_cpp.py", line 43, in <module>
    from madgraph.iolibs.export_v4 import VirtualExporter
  File "/home/thompson/MG5_aMC_v3_4_1/madgraph/iolibs/export_v4.py", line 56, in <module>
    import madgraph.various.diagram_symmetry as diagram_symmetry
  File "/home/thompson/MG5_aMC_v3_4_1/madgraph/various/diagram_symmetry.py", line 51, in <module>
    import madgraph.various.process_checks as process_checks
  File "/home/thompson/MG5_aMC_v3_4_1/madgraph/various/process_checks.py", line 71, in <module>
    from madgraph.interface.madevent_interface import MadLoopInitializer
  File "/home/thompson/MG5_aMC_v3_4_1/madgraph/interface/madevent_interface.py", line 87, in <module>
    import madgraph.madevent.gen_ximprove as gen_ximprove
  File "/home/thompson/MG5_aMC_v3_4_1/madgraph/madevent/gen_ximprove.py", line 61, in <module>
    import madgraph.madevent.hel_recycle as hel_recycle
  File "/home/thompson/MG5_aMC_v3_4_1/madgraph/madevent/hel_recycle.py", line 3, in <module>
    import argparse
KeyboardInterrupt
Traceback (most recent call last):
  File "/home/thompson/MG5_aMC_v3_4_1/bin/mg5_aMC", line 160, in <module>
    __import__('PLUGIN.%s' % options.plugin)
  File "/home/thompson/MG5_aMC_v3_4_1/PLUGIN/maddump/__init__.py", line 12, in <module>
    import new_interface.maddump_interface as maddump_interface
ModuleNotFoundError: No module named 'new_interface'
-------------------------

The part under the traceback then repeats indefinitely until I suspend (ctrl+z). Looks like an issue with maddump plugin?

Revision history for this message
Best Olivier Mattelaer (olivier-mattelaer) said : 		#3

Ah so this seems to be an issue with maddm which is not python3 compatible and therefore not compatible with 3.4.1.
So you can remove the directory /home/thompson/MG5_aMC_v3_4_1/PLUGIN/maddump/ and that should solve your issue.

Will try to install that plugin to see if I can implement a nice catch for such PLUGIN issue.

Cheers,

Olivier

Revision history for this message
Adrian (adr-traq) said : 		#4

Thank you so much!

Revision history for this message
Adrian (adr-traq) said : 		#5

Thanks Olivier Mattelaer, that solved my question.