Crash with EnergyRescalerTool

ROOT
1

Dear experts, I’ve update the version of the EnergyRescalerTool (twiki.cern.ch/twiki/bin/view/At … gyRescaler). It’s a very very simple code, it read some numbers from a table in a file and return a number, nothing special. With the new version I got this in the slaves:

#0  0x000000384a89a115 in waitpid () from /lib64/libc.so.6
#1  0x000000384a83c481 in do_system () from /lib64/libc.so.6
#2  0x00002b2ddbb7ace0 in TUnixSystem::StackTrace ()
   from /gpfs/storage_4/users/home/proof/root/lib/libCore.so
#3  0x00002b2ddbb773db in TUnixSystem::DispatchSignals ()
   from /gpfs/storage_4/users/home/proof/root/lib/libCore.so
#4  <signal handler called>
#5  0x000000384d8b5fa2 in __gnu_cxx::__exchange_and_add ()
   from /usr/lib64/libstdc++.so.6
#6  0x000000384d89c819 in std::basic_string<char, std::char_traits<char>, std::allocator<char> >::~basic_string () from /usr/lib64/libstdc++.so.6
#7  0x00002b2de50095a7 in ~HiggsSelector (this=0x170e9700)
    at /gpfs/storage_4/users/home/turra/Higgs/higgs/./HiggsPar/HiggsBaseSelector/HiggsSelector.h:24
#8  0x00002b2de5009621 in ~HiggsAnalysis (this=0x170e9700)
    at /gpfs/storage_4/users/home/turra/Higgs/higgs/./HiggsAnalysis.h:19
#9  0x00002b2de4e73616 in TProofPlayer::~TProofPlayer ()
   from /gpfs/storage_4/users/home/proof/root/lib/libProofPlayer.so
#10 0x00002b2de4e84aa1 in TProofPlayerSlave::~TProofPlayerSlave ()

It seems that the slave crash in the destructor. In the master I got:


#0  0x000000384a89a115 in waitpid () from /lib64/libc.so.6
#1  0x000000384a83c481 in do_system () from /lib64/libc.so.6
#2  0x00002b2bcc9a9ce0 in TUnixSystem::StackTrace ()
   from /gpfs/storage_4/users/home/proof/root/lib/libCore.so
#3  0x00002b2bcc9a63db in TUnixSystem::DispatchSignals ()
   from /gpfs/storage_4/users/home/proof/root/lib/libCore.so
#4  <signal handler called>
#5  0x000000384d89b588 in std::basic_string<char, std::char_traits<char>, std::allocator<char> >::basic_string () from /usr/lib64/libstdc++.so.6
#6  0x00002b2bd7f4ed7f in HiggsAnalysis::Terminate (this=0x1e2a54d0)
    at /gpfs/storage_4/users/home/turra/Higgs/higgs/./HiggsAnalysis.C:323
#7  0x00002b2bd213b331 in TProofPlayerLite::Finalize ()
   from /gpfs/storage_4/users/home/proof/root/lib/libProofPlayer.so
#8  0x00002b2bd213c701 in TProofPlayerLite::Process ()

and

 *** Break *** write on a pipe with no one to read it
SysError in <TUnixSystem::UnixSend>: send (Broken pipe)

 *** Break *** write on a pipe with no one to read it
SysError in <TUnixSystem::UnixSend>: send (Broken pipe)

 *** Break *** write on a pipe with no one to read it
SysError in <TUnixSystem::UnixSend>: send (Broken pipe)

 *** Break *** write on a pipe with no one to read it
SysError in <TUnixSystem::UnixSend>: send (Broken pipe)

....

This tool is a class declared as class member (no pointer). It has a nohop destructor ({ }). If I go back to the previous version of the tool everythings work. Using prooflite 5.28c.

Do you have some hints?

2

Someone really to explain me why…

I’ve added in HiggsSelector.h this line:

~HiggsSelector() {cout << "INFO: ~HiggsSelector()" << endl; }

and now all work. This is completely magic.

3

Hi,

Adding something to the destructor implies to recompile it.
Your two crashes were not at the same place but they were both related to std::~basic_string . Are you sure you were using compatible binaries?
Are you able to reproduce the crash going back to the previous ~HiggsSelector ?

Gerri

4

[quote=“ganis”]Hi,

Adding something to the destructor implies to recompile it.
Your two crashes were not at the same place but they were both related to std::~basic_string . Are you sure you were using compatible binaries?
Are you able to reproduce the crash going back to the previous ~HiggsSelector ?

Gerri[/quote]

I’m really surprised. I’ve commented the HiggSelector destructor and it works… I’m sure I’m using compatible binaries, everythings is inside a par file and I’m using ClearPackages.

5

I didn’t look at the previous replies, but I will note that the version of the EnergyRescalerTool bundled in SUSYTools works fine with PROOF, so maybe you want to compare your code to that.