Error in TTree::Fill

[vfriese]

With cbmroot, we recently encountered the error message

Error in <TTree::Fill>: Failed filling branch:cbmsim.MvdDigi, nbytes=-1845880316, entry=1
 This error is symptomatic of a Tree created as a memory-resident Tree
 Instead of doing:
    TTree *T = new TTree(...)
    TFile *f = new TFile(...)
 you should do:
    TFile *f = new TFile(...)
    TTree *T = new TTree(...)

The message occurs when the number of objects (or the associated memory) to be filled into the tree branch exceeds some threshold. The run proceeds to the end, and the output ROOT file appears intact and can be opened without error. However, some data are not present there (strangely enough, in this case, the MVD data, on which the error complains, are there, but the STS data are not):

root [0] 
Processing checkdata.C...
Entries 1
MVD vector 0x600001d4fc80 MVD digis 43145913
STS vector 0x600001d4efa0 STS digis 0

More strangely, when I look at the file with TBrowser, I can see the all data, including the STS (see attached plot).

Question: Has this kind of error been encountered before by somebody? Is a cure within fairroot available? I know that there are some ROOT methods concerning the tree I/O, like TTree::SetCacheSize(), but since we use the ROOT I/O scheme through FairRootManager, we have no direct way to try this out.

For more details, see https://redmine.cbm.gsi.de/issues/3785 (needs CBM Redmine Account).

[fuhlig1]

Hi @vfriese,

after some investigation I think the underlying issue is that the MvdDigi branch hits the limit of 1GB which is a hard limit in ROOT due to the usage of 32bit Integers. I the size of the written bytes exactly indicates the problem. The number should be positive but in the example it is negative

nbytes=-1845880316

If I am not wrong doing some simple calculation the size of the MvdDigi data is more than 3GB

43145913 * 80byte =~ 3.2 GB

The problem was also seen by the ALICE experiment.

Lets further discuss the issue in the Redmine issue.

[vfriese]

Thanks. I tried to look into TTree::Fill() of ROOT 6.30 and confirm that for the byte count, Int_t is used (signed 32-bit integer). This sets the allowed range to 2 GiB (2^31-1). In our example, we surpass this limit.

The error message by TTree::Fill() is not caught, so the run continues, which I find dangerous. For instance, we are currently trying to find out why in the output file, the corresponding branch (MvdDigi) is properly filled, while the next ones are not.

I suggest that the return value of TTree::Fill() in case of this exception (the method returns then -1) be caught in FairRootFileSink::Fill(), preferably with a Fatal error.