Hi Derek,
Yes, you point out the issue that xrootd does not interlock meta-data
commands with active open requests. That means that, indeed, you can erase
files that are currently being written or read. One could argue that this
is consistent with Unix (well, in certain unix systems anyway). Keeping
with consistency, erasing a Unix file does nothing about current I/O to
the erased file. The data is wriiten into the file but once the file is
closed, all of the data is thrown away. So, the question is whether the
meta-data command should be interlocked with usage. That is, once a file
is opened it's attributes cannot be modified (i.e., name, mode, existence,
etc).
As for eliminating the file from the MSS, that is the default operation.
Do you need an xrootd interface to only erase the file from the cache?
Andy
On Thu, 28 Apr 2005, Derek Feichtinger wrote:
> Hi,
>
> I have a r/w + MSS setup.
>
> My little test from the ROOT shell:
> root [0] File *f
> =TFile::Open("root://redirector.cern.ch//tmp/xrootd/new.root","create")
> root [1] TTree *t = new TTree("t","title")
> root [2] t->Write()
>
> Now I use another shell to execute this perl client command:
> @ans = XrdClientAdmin::XrdRm("/tmp/xrootd/new11.root");
>
> This succeeds and I can see, that the file was indeed erased from the
> respective leaf node (and the cache and the MSS!), even though the other
> client should have a lock on it!!!
>
> Moreover, in my ROOT session I go happily on writing to the file:
> root [3] TTree *t2 = new TTree("t2","title")
> root [4] t2->Write()
>
> (I can see on the leaf node log that it correctly acknowledges the bytes sent)
>
> root [5] f->Close()
>
> (here, too, I receive no errors upon closing and the leaf node log also shows
> no peculiar entries.).
>
> I try to open my file again, and I fail
> root [0] File *f
> =TFile::Open("root://redirector.cern.ch//tmp/xrootd/new.root","read")
>
> (I can see that it has not been recreated!). Where have my bytes gone?
>
>
> Background:
> I'm trying to investigate how an external MPS (Migration, Purge, Staging)
> script could interact with the xrootd system.
>
> Cheers,
> Derek
>
>
> --
> Dr. Derek Feichtinger Tel: +41 22 767 10 07
> LCG/ARDA Group email: [log in to unmask]
> CERN http://people.web.psi.ch/feichtinger
> CH-1211 Genève 23
>
>
>
