Re: [ROOT] question as to meaning of R__unzip

From: Rene Brun (Rene.Brun@cern.ch)
Date: Wed Jun 26 2002 - 07:54:11 MEST


Hi David,

We have no time-out logic when reading in ROOT itself. Time-out
should be in NFS. One could consider adding a "retry" call
in case the TFile read function returns 0 bytes.
It would be nice if you could try this small addition and test it
with your configuration. Let me know.

Rene Brun

On Tue, 25 Jun 2002, David R. Relyea wrote:

> Before I upgrade ROOT here (since I have to get someone here at SLAC to do
> that, and they're difficult to find), may I ask a little more?
> 
> We have determined that the crashes are caused by the poor performance of
> an NFS server at SLAC - when it pegs (it has a very slow network card), we
> see this problem.  On other servers, I get no such errors, since they can
> handle the traffic.  It's as if there's a timeout on a read function
> somwhere (returning zero bytes, causing the R__unzip error).  Can ROOT
> time out on reads?  (we're trying to isolate the source of the timeout)
> And if so, what's the value of the timeout?
> 
> David Relyea
> 
> 
> On Sun, 23 Jun 2002, Rene Brun wrote:
> 
> Hi David,
> 
> Could you try with version 3.03/06?
> We fixed a problem in February in some special cases of branch
> buffers when the compressed buffer is bigger (yes) than the
> uncompressed buffer.
> 
> Rene Brun
> 
> On Sat, 22 Jun 2002, David R. Relyea wrote:
> 
> > I have a large subroutine which opens three trees and loops through all
> > three (all the same length) multiple times.  I use SetBranchAddress and
> > GetEntry only (no other tree subroutines called).  About 50% of the time,
> > this subroutine works.  The other 50% of the time, it seg faults after
> > having given me this kind of error (thousands of times):
> >
> > R__unzip: discrepancy with initial size
> > Error in <TBasket::ReadBasketBuffers>: fObjlen = 31936, nout = 0
> > Error in <TBranch::GetBasket>: File problem at address:50142285, basket
> > seekkey=50142285, branch:cep00
> >
> > The error is non-reproducible - it's a random occurrence, as far as I can
> > tell.  Where does the R__unzip come from?  Looking through roottalk, it
> > seems as if it should only occur on odd zero-length files (trees?), and
> > I'm positive I have none (this routine works 50% of the time on the exact
> > same files...).  Also, the branch variable shown above is not the only
> > variable it crashes on - it seems to pick a variable at random from one of
> > the three trees each time it crashes.  My only guess is that there's some
> > odd memory leak, but since I only have two new statements in the entire
> > routine (one per file opened - two of the trees are in one file, another
> > is in the other file), I'm absolutely sure I delete things properly.  Any
> > information is appreciated.
> >
> > David Relyea
> >
> 
> 
> 



This archive was generated by hypermail 2b29 : Sat Jan 04 2003 - 23:50:58 MET