[Rivet] Rivet 2.6.0 integration problems in CMS sosftware

[Janssen Xavier]

Hi,

> you can rename some of the folders in "analyses/". If the name does not
> start with "plugin", the compile will be skipped.
> 
> Are you compiling Rivet in a threaded environment with explicit TLS
> choices?

Difficult to say as CMS software compilation tool is kind of a black box and I have asked as well our computing support. Also I can’t disable easily directories in that context as it automatically download the tarball and process it directly. I will try removing the plugins from the runtime environment either.

Xavier.

> Thanks,
> 
>  David
> 
> 
> On 11/01/18 15:41, Janssen Xavier wrote:
>> 
>> Hi Andy,
>> 
>> Is there a way I can compile only a few plugins for a quick test to be
>> sure this is a limit problem ?
>> 
>> Cheers, Xavier,
>> 
>> 
>>> On 11 Jan 2018, at 16:34, Andy Buckley <andy.buckley at cern.ch
>>> <mailto:andy.buckley at cern.ch>> wrote:
>>> 
>>> Hi Xavier,
>>> 
>>> *Very* strange. The analysis loading has not changed at all, but the
>>> analysis compilation now uses the rivet-buildplugin script for all the
>>> built-in analyses as well as any external ones. I wouldn't expect that
>>> to change anything, and it's worked fine on my machine and others...
>>> but it is a change.
>>> 
>>> I'm no expert on this static TLS thing -- a quick Google suggests that
>>> the system is running out of memory due to loading plugins each with
>>> their own static memory requirements. Maybe we just crossed some
>>> threshold? My recollection is that we load all the lightweight
>>> analysis *loader* objects into memory, but only instantiate the actual
>>> analysis classes which are requested by the user. Any experts on the
>>> list who can suggest diagnostics?
>>> 
>>> Andy
>>> 
>>> *Dr Andy Buckley, Lecturer / Royal Society University Research Fellow*
>>> Particle Physics Experiment Group, University of Glasgow
>>> 	
>>> 
>>> 
>>> On Jan 11 2018, at 3:16 pm, Janssen Xavier
>>> <xavier.janssen at uantwerpen.be <mailto:xavier.janssen at uantwerpen.be>>
>>> wrote:
>>> 
>>> 
>>>    Hi,
>>> 
>>>    I was testing the integration of the latest release in our CMS
>>>    software. The installation and compilation is working well.
>>>    However trying to run Rivet to produce some MC events for 1 test
>>>    analysis I use for validation is failing with the following error:
>>> 
>>>    unable to load
>>>    /afs/cern.ch/work/x/xjanssen/cms/Rivet/10X_gcc630_Rivet260/CMSSW_10_0_0_pre3/lib/slc6_amd64_gcc630/pluginGeneratorInterfaceRivetInterface_plugins.so
>>>    <http://cern.ch/work/x/xjanssen/cms/Rivet/10X_gcc630_Rivet260/CMSSW_10_0_0_pre3/lib/slc6_amd64_gcc630/pluginGeneratorInterfaceRivetInterface_plugins.so>
>>>    because dlopen: cannot load any more object with static TLS
>>> 
>>>    where pluginGeneratorInterfaceRivetInterface_plugins.so is our
>>>    standard interface to Rivet. Just to be sure I tested back Rivet
>>>    2.5.4 in the same CMS software release and I don’t get any such
>>>    problem. Not sure exactly what is happening but what did you
>>>    exactly changed in the compilation/loading of analysis in the
>>>    latest release ?
>>> 
>>>    Cheers, Xavier.
>>>    _______________________________________________
>>>    Rivet mailing list
>>>    Rivet at projects.hepforge.org <mailto:Rivet at projects.hepforge.org>
>>>    https://www.hepforge.org/lists/listinfo/rivet
>>> 
>> 
>> 
>> 
>> _______________________________________________
>> Rivet mailing list
>> Rivet at projects.hepforge.org
>> https://www.hepforge.org/lists/listinfo/rivet
>>