copying here the conclusion:
user job processed two input files, adding pile up, to add pile up
the same two minbias files have been used, opening and closing
them for each of the two inputs (let's not care here if that
makes a statistical biased sample or not).
Somehow the FJR produced by cmsRun has sort of a cronological list
of all input files which have been opened, including duplicates in
a case like this where opening same file twice makes sense
This duplication is currently fed through all software layers to
the DBS3 server, since it is present in the list of parents
parsed from FJR by old (ancient ?) ProdCommon.FwkJobRep.ReportParser
(I think written by Dave Evans - fnal years ago ) and from
there on always passed as a list.
At first sight, if I simply put a set() sooner or later,
everyghing will be fine.
see: https://hypernews.cern.ch/HyperNews/CMS/get/dmDevelopment/1573/1/1.htm
copying here the conclusion: user job processed two input files, adding pile up, to add pile up the same two minbias files have been used, opening and closing them for each of the two inputs (let's not care here if that makes a statistical biased sample or not). Somehow the FJR produced by cmsRun has sort of a cronological list of all input files which have been opened, including duplicates in a case like this where opening same file twice makes sense
This duplication is currently fed through all software layers to the DBS3 server, since it is present in the list of parents parsed from FJR by old (ancient ?) ProdCommon.FwkJobRep.ReportParser (I think written by Dave Evans - fnal years ago ) and from there on always passed as a list.
At first sight, if I simply put a set() sooner or later, everyghing will be fine.