Dear Root experts,
my colleagues and myself are trying to use proof to parallelize our analysis code (basically selecting events and filling histograms) and I see a strange behavior that I don’t understand. I have to say that I’m quite new in using Proof.
The code is executed for different datasets with a piece of code similar to :
string macroName = “ProofSelectorMyCutFlow.C+”; // inheriting from TSelector
for(unsigned int i=0;i<datasets.size();i++){
proof->Process(datasets[i].Name().c_str(),macroName.c_str());
}
Using a multi-CPU unix machine, at the first iteration, everything worked fine and the datasets was processed quickly on several nodes.
At the second iteration (for the second dataset) some of the jobs failed with the error :
/a/path/proofserv.exe: symbol lookup error: /an/other/path/session-sbgui5.in2p3.fr-1316513974-28220/worker-0.3/./ProofSelectorMyCutFlow_C.so: undefined symbol: etc…
This error appears on 3/10 jobs. Other jobs for the same datasets are just fine.
Finally, after the second iteration, all “submissions” failed with the error :
Info in TProofLite::Process: not idle: cannot accept queries.
I would really appreciate if you can help me to understand what’s going on.
Best Regards,
Jeremy