Re: [condor-users] Unexplained status=128

[Belay T Beshah <belay.beshah@xxxxxxxxxxxxxxxxxxxxxxxx>]

Colin Stolley wrote:

Any body has any ideas on why this happens and how to tackle this 
problem? Or is there any way I could know that the GUI is running on the 
node? or that the user is actively logged in even though the keyboard 
and mouse are not moving?
    

This has been a reoccuring problem. Have you tried running depends as a
Condor job to profile the executable in the problematic environment? I've
found running

depends /c /pa:1 /ot:profile_output.txt my_prog.exe 

Gives a nice (and quite verbose) listing of dll loads and other
information. Perhaps it could give you some additional clues?

Yes, that is what I mean when I said used  "depends" to write the dependency to file. I have analysed  that file and could see that all the DLL's are found correctly.  Looked into the starter logs but nothing, if I am still logged on the node but exit the GUI program everything works fine again.

Also, keep in mind that since the condor-reuse-vmX user is created
as a non-interactive user, it has no user profile or customized
HKEY_CURRENT_USER registry hive (just the defaults). An application
like RegMon (see http://www.sysinternals.com/ntw2k/source/regmon.shtml)
might expose these dependencies.

That is one thing I haven't checked. The worker application were specifically designed not to use any registry and relay on environment variables since those could be passed to them easily using the *.sub files. I will have to check if any of the DLL's expects a registry setting. What would have been nice is if windows gave a better error message which DLL it was that it couldn't load.

Thanks
BTB