Re: TFile memory leak?

  Hi,

If you know that a file has a canvas which needs to own its content simply do:

     TFile f("tmp.root");
     TCanvas* c = (TCanvas*)f.Get("can");
     TIter next( c->GetListOfPrimitives() );
     TObject *obj;
     while( (obj = next() ) {
       obj->SetBit(TObject::kCanDelete()); // Add missing ownership delegation
     }

Cheers,
Philippe.

On 9/9/10 12:57 PM, OKUMURA, Akira wrote:
> Hello Philippe,
>
> Thank you for your detailed explanation. I now understand that it was my mistake and how I should save canvases in ROOT files.
>
> In addition, I would like to know how I can delete the allocated memory with the bad files which have been already created in my HDD. I have many files which were created without caring about ownership delegation. Of course, I can recreate the files with small modification, but do you have any easy solution?
>
> Regards,
>
> --
> OKUMURA, Akira oxon@{astro.isas.jaxa.jp,stanford.edu}
> Institute of Space and Astronautical Science (ISAS/JAXA)
> Now at KIPAC/SLAC/Stanford
> Varian Physics #306
> 382 Via Pueblo Mall, MC 406 Stanford, CA94305
> TEL 650-736-0971/FAX 650-724-5065
> Skype : okumura.akira
>
> On 2010/09/09, at 10:32, Philippe Canal wrote:
>
>> Hi Akira,
>>
>> The issue is the ownership of the histogram object. The simplest fix is to use:
>>
>> TCanvas can("can", "can");
>> TH2D h("h", "h", 100, -3, 3, 100, -3, 3);
>> h.FillRandom("gaus", 100000);
>> h.DrawCopy("col");
>> can.SaveAs("tmp.root");
>>
>> The issue is that when you do h.Draw, it does associate the histogram with
>> the Canvas but it __correctly__ (especially in your case) does not delegate
>> ownership of the histogram to the Canvas. Since you did not delegate
>> ownership of the histogram to the Canvas, the Canvas __must__ assume
>> that you will store the histogram independently and thus upon reading
>> the file, the Canvas does not take ownership of the histogram and do
>> not delete it when you delete the Canvas.
>>
>> By using DrawCopy you create a new histogram and delegate its
>> ownership to the Canvas.
>>
>> Note that you of course need to delete the Canvas in the reading
>> loop in order to really get rid of the leak (this is because Canvas
>> are __not__ owned by the TFile they are read from).
>>
>> Cheers,
>> Philippe.
>>
>> PS. A slightly more cpu efficient option is:
>>
>> TCanvas can("can", "can");
>> TH2D *h = new TH2D("h", "h", 100, -3, 3, 100, -3, 3);
>> h->FillRandom("gaus", 100000);
>> h->Draw("col");
>> h->SetBit(TObject::kCanDelete); // Delegate ownership to the Canvas.
>> can.SaveAs("tmp.root");
>>
>> On 9/9/10 10:34 AM, OKUMURA, Akira wrote:
>>> Hello all,
>>>
>>> Here is a minimum example. You don't need to download http://slac.stanford.edu/~oxon/tmp.tgz
>>>
>>> ==============================
>>> #include<TFile.h>
>>> #include<TCanvas.h>
>>> #include<TH2D.h>
>>> #include<iostream>
>>>
>>> int main(int argc, char** argv)
>>> {
>>> TCanvas can("can", "can");
>>> TH2D h("h", "h", 100, -3, 3, 100, -3, 3);
>>> h.FillRandom("gaus", 100000);
>>> h.Draw("col");
>>> can.SaveAs("tmp.root");
>>>
>>> const Int_t kN = 10000;
>>> for(Int_t i = 0; i< kN; i++){
>>> TFile f("tmp.root");
>>> std::cout<< "Reading"<< f.GetName()<< "...\n";
>>> TCanvas* c = (TCanvas*)f.Get("can");
>>> f.Close();
>>> } // i
>>>
>>> return 0;
>>> }
>>> ==============================
>>>
>>> I would like to know how to delete the allocated memory to TCanvas* c. C++'s delete does not work.
>>>
>>> Regards,
>>>
>>> --
>>> OKUMURA, Akira oxon@{astro.isas.jaxa.jp,stanford.edu}
>>> Institute of Space and Astronautical Science (ISAS/JAXA)
>>> Now at KIPAC/SLAC/Stanford
>>> Varian Physics #306
>>> 382 Via Pueblo Mall, MC 406 Stanford, CA94305
>>> TEL 650-736-0971/FAX 650-724-5065
>>> Skype : okumura.akira
>>>
>>> On 2010/09/09, at 8:20, OKUMURA, Akira wrote:
>>>
>>>> Hello John and all,
>>>>
>>>> I uploaded the log file here.
>>>> http://slac.stanford.edu/~oxon/valgrind.log
>>>>
>>>> Regards,
>>>>
>>>> --
>>>> OKUMURA, Akira oxon@{astro.isas.jaxa.jp,stanford.edu}
>>>> Institute of Space and Astronautical Science (ISAS/JAXA)
>>>> Now at KIPAC/SLAC/Stanford
>>>> Varian Physics #306
>>>> 382 Via Pueblo Mall, MC 406 Stanford, CA94305
>>>> TEL 650-736-0971/FAX 650-724-5065
>>>> Skype : okumura.akira
>>>>
>>>> On 2010/09/09, at 2:06, John Idarraga wrote:
>>>>
>>>>> Hello Akira
>>>>>
>>>>> Can you&> that to a log file and send it ?
>>>>>
>>>>> John
>>>>>
>>>>> OKUMURA, Akira wrote:
>>>>>> Hello Claudi and all,
>>>>>>
>>>>>> Thank you for your advice. I attach a part of the valgrind output. Even though this is a first time to use valgrind for me, the output seems that there is memory leak.
>>>>>>
>>>>>> ==17737== 5,189,819 (27,328 direct, 5,162,491 indirect) bytes in 61 blocks are definitely lost in loss record 45,809 of 45,811
>>>>>> ==17737== at 0x1000146CE: operator new(unsigned long) (vg_replace_malloc.c:261)
>>>>>> ==17737== by 0x1000386C9: TStorage::ObjectAlloc(unsigned long) (in /usr/local/root_svn/lib/libCore.so)
>>>>>> ==17737== by 0x100233BB4: TObject::operator new(unsigned long) (in /usr/local/root_svn/lib/libCore.so)
>>>>>> ==17737== by 0x102DE6955: ROOT::new_TPaletteAxis(void*) (in /usr/local/root_svn/lib/libHistPainter.so)
>>>>>> ==17737== by 0x10010AAA5: TClass::New(TClass::ENewType) const (in /usr/local/root_svn/lib/libCore.so)
>>>>>> ==17737== by 0x100FEB2BF: TBufferFile::ReadObjectAny(TClass const*) (in /usr/local/root_svn/lib/libRIO.so)
>>>>>> ==17737== by 0x1000EF390: TBuffer& operator>><TObject>(TBuffer&, TObject*&) (in /usr/local/root_svn/lib/libCore.so)
>>>>>> ==17737== by 0x1000F2C22: TList::Streamer(TBuffer&) (in /usr/local/root_svn/lib/libCore.so)
>>>>>> ==17737== by 0x100FE8A18: TBufferFile::ReadFastArray(void**, TClass const*, int, bool, TMemberStreamer*, TClass const*) (in /usr/local/root_svn/lib/libRIO.so)
>>>>>> ==17737== by 0x1010A1864: int TStreamerInfo::ReadBuffer<char**>(TBuffer&, char** const&, int, int, int, int) (in /usr/local/root_svn/lib/libRIO.so)
>>>>>> ==17737== by 0x100FE9C15: TBufferFile::ReadClassBuffer(TClass const*, void*, int, unsigned int, unsigned int, TClass const*) (in /usr/local/root_svn/lib/libRIO.so)
>>>>>> ==17737== by 0x101363C23: TH1::Streamer(TBuffer&) (in /usr/local/root_svn/lib/libHist.so)
>>>>>> ==17737== ==17737== 69,739,543 (823,176 direct, 68,916,367 indirect) bytes in 999 blocks are definitely lost in loss record 45,810 of 45,811
>>>>>> ==17737== at 0x1000146CE: operator new(unsigned long) (vg_replace_malloc.c:261)
>>>>>> ==17737== by 0x1000386C9: TStorage::ObjectAlloc(unsigned long) (in /usr/local/root_svn/lib/libCore.so)
>>>>>> ==17737== by 0x100233BB4: TObject::operator new(unsigned long) (in /usr/local/root_svn/lib/libCore.so)
>>>>>> ==17737== by 0x101A9EA7E: ROOT::new_TCanvas(void*) (in /usr/local/root_svn/lib/libGpad.so)
>>>>>> ==17737== by 0x10010AAA5: TClass::New(TClass::ENewType) const (in /usr/local/root_svn/lib/libCore.so)
>>>>>> ==17737== by 0x101013FBB: TKey::ReadObj() (in /usr/local/root_svn/lib/libRIO.so)
>>>>>> ==17737== by 0x100FF5AEE: TDirectoryFile::Get(char const*) (in /usr/local/root_svn/lib/libRIO.so)
>>>>>> ==17737== by 0x100000EFC: main (test.C:19)
>>>>>> ==17737== ==17737== LEAK SUMMARY:
>>>>>> ==17737== definitely lost: 859,176 bytes in 1,222 blocks
>>>>>> ==17737== indirectly lost: 85,444,368 bytes in 14,771 blocks
>>>>>> ==17737== possibly lost: 387,824 bytes in 1,534 blocks
>>>>>> ==17737== still reachable: 9,549,367 bytes in 84,421 blocks
>>>>>> ==17737== suppressed: 852,204 bytes in 23,733 blocks
>>>>>> ==17737== Reachable blocks (those to which a pointer was found) are not shown.
>>>>>> ==17737== To see them, rerun with: --leak-check=full --show-reachable=yes
>>>>>> ==17737== ==17737== For counts of detected and suppressed errors, rerun with: -v
>>>>>> ==17737== Use --track-origins=yes to see where uninitialised values come from
>>>>>> ==17737== ERROR SUMMARY: 705 errors from 705 contexts (suppressed: 1692 from 1692)
>>>>>>
>>>>>> Regards,
>>>>>>
>>>>>> --
>>>>>> OKUMURA, Akira oxon@{astro.isas.jaxa.jp,stanford.edu}
>>>>>> Institute of Space and Astronautical Science (ISAS/JAXA)
>>>>>> Now at KIPAC/SLAC/Stanford
>>>>>> Varian Physics #306
>>>>>> 382 Via Pueblo Mall, MC 406 Stanford, CA94305
>>>>>> TEL 650-736-0971/FAX 650-724-5065
>>>>>> Skype : okumura.akira
>>>>>>
>>>>>>
>>>>>> On 2010/09/08, at 14:31, Claudi Astres wrote:
>>>>>>
>>>>>>
>>>>>>> Hi Akira,
>>>>>>> You could use "valgrind" utility to check whether the problem is really a memory leak. Maybe you already knew it.
>>>>>>> Valgrind has become crucial for me: http://valgrind.org/
>>>>>>>
>>>>>>> Regards.
>>>>>>> Claudi
>>>>>>>
>>>>>>>
>>>>>>> El 08/09/2010, a las 22:14, OKUMURA, Akira escribió:
>>>>>>>
>>>>>>>
>>>>>>>> Hello ROOTers,
>>>>>>>>
>>>>>>>> This question may be a very basic one but I could not find similar questions on the web.
>>>>>>>>
>>>>>>>> http://slac.stanford.edu/~oxon/tmp.tgz
>>>>>>>>
>>>>>>>> This tgz file includes a .root file and a very simple script which opens TFile 10000 times. When I execute the script,
>>>>>>>>
>>>>>>>> root [0] .x test.C
>>>>>>>>
>>>>>>>> the memory usage of the process becomes more than 1 GB while all TFile instances are closed.
>>>>>>>>
>>>>>>>> The .root file was created via TPad::SaveAs, as follows.
>>>>>>>>
>>>>>>>> root [0] TCanvas can("can", "can")
>>>>>>>> root [1] TH2D* h = new TH2D("h", "h", 100, -3, 3, 100, -3, 3)
>>>>>>>> root [2] h->FillRandom("gaus", 1000000)
>>>>>>>> root [3] h->Draw()
>>>>>>>> root [4] can.SaveAs("aho.root")
>>>>>>>>
>>>>>>>> How can I suppress this large memory consumption? I am not sure if this is an intended behavior or memory leak.
>>>>>>>>
>>>>>>>> I am using svn trunk on Snow Leopard and SL 5.
>>>>>>>>
>>>>>>>> Regards,
>>>>>>>>
>>>>>>>> --
>>>>>>>> OKUMURA, Akira oxon@{astro.isas.jaxa.jp,stanford.edu}
>>>>>>>> Institute of Space and Astronautical Science (ISAS/JAXA)
>>>>>>>> Now at KIPAC/SLAC/Stanford
>>>>>>>> Varian Physics #306
>>>>>>>> 382 Via Pueblo Mall, MC 406 Stanford, CA94305
>>>>>>>> TEL 650-736-0971/FAX 650-724-5065
>>>>>>>> Skype : okumura.akira
>>>>>>>>
>>>>>>>>
Received on Thu Sep 09 2010 - 20:07:16 CEST