symbol lookup error about pyslurm

[ralphbean]

This is great. I haven't been able to get it to work yet -- didn't know the LD_PRELOAD trick.

Looking forward to phantez's response!

[phantez]

Hi,

The github repo is not up2date at all and the problem you rise is
known and not solve on this version of the code.
Mark Roberts who is the main dev. of pyslurm
(http://www.gingergeeks.co.uk/pyslurm/) has a working version of the
pyslurm with slurm 2. You should get in touch with him for more
precise informations.

Stéphan Gorget

On Thu, Oct 20, 2011 at 4:25 PM, pjodrr
[email protected]
wrote:

Hello,

When I run:

   python -c "import pyslurm; a, b = pyslurm.slurm_load_jobs('',2); print a"

I get:

   python: symbol lookup error: /usr/lib/slurm/select_cons_res.so: undefined symbol: slurm_get_debug_flags

and indeed that symbol is undefined:

   # nm --dynamic /usr/lib/slurm/select_cons_res.so|grep slurm_get_debug_flags
                U slurm_get_debug_flags

but included in /usr/lib/libslurm.so.22:

   # nm --dynamic /usr/lib/libslurm.so.22|grep slurm_get_debug_flags
   00000000000553b0 T slurm_get_debug_flags

which is interesting since pyslrum.so depends on libslurm:

   # ldd /usr/lib/python2.6/dist-packages/pyslurm/pyslurm.so
   ...
       libslurm.so.22 => /usr/lib/libslurm.so.22 (0x00007f9fb37e9000)
   ...

I don't know why it cannot find the symbol.
If I do:

   # LD_PRELOAD=/usr/lib/libslurm.so.22 python -c "import pyslurm; a, b = pyslurm.slurm_load_jobs('',2); print a"

then it works.  Can you solve the mystery?

Thanks,

 Peter

Reply to this email directly or view it on GitHub:
#1

Stéphan

[pjodrr]

Hi,

I cannot find any contact information on the website you are referring to. In fact the link to the git repository points to "gitbug.com". So how can I get in touch with Mark Roberts?

thanks,

Peter

[gingergeeks]

Guys,
My apologies to all especially Stephan because I meant to send this info to him but work and family take a lot out of my available time. It took much tracing of the loader to work this out via LD_DEBUG and discussed it with Moe and Danny from schedmd (SLURM dev) some time ago. Anyway, without boring you further the pyslurm init script (from dev) needs to be modified to perform the correct loader method :

u'''
    PySLURM

    Here we modify the dlopen settings to RTLD_GLOBAL
    otherwise we cannot see the symbols correctly.

'''

import os, sys

__version__ = "2.2.7-1"

old_dlopen_flags = ''
if hasattr(sys, "setdlopenflags"):
    old_dlopen_flags = sys.getdlopenflags()
    import DLFCN
    sys.setdlopenflags(old_dlopen_flags | DLFCN.RTLD_GLOBAL)

from pyslurm import *

if old_dlopen_flags:
    if hasattr(sys, "setdlopenflags"):
            sys.setdlopenflags(old_dlopen_flags)

Please let Stephan and myself know if this works for you guys so we can close the issue, it is in my dev branch and is in current use on various clusters.

Mark

[pjodrr]

Hi,

yes, that solves it for me,
Looking forward to the official release!
Or, is it possible to have read access to your dev branch?

Regards,

Peter