> -----Original Message-----
> From: LHC Computer Grid - Rollout
> [mailto:[log in to unmask]] On Behalf Of Burke, S (Stephen)
> Sent: dinsdag 18 januari 2005 11:00
> To: [log in to unmask]
> Subject: Re: [LCG-ROLLOUT] [ATLAS-LCG] Disk failure at Prague
>
> > [mailto:[log in to unmask]] On Behalf Of Jules
> Wolfrat said:
> > That's true, but you must make a distinction between the
> database that
> > RLS is and the physical file itself. The file itself can be
> destroyed
> > while the RLS itself still thinks that the file is present. What you
> > need is some mechanism to check the integrity of the RLS,
> > e.g. run every night a consistency check on the RLS!
>
> I certainly agree that we should be checking consistency, but
> you can't
> recreate the catalogues just from what's on disk, you don't
> know the guid or
> LFN, or indeed which files are replicas of each other without doing
> something fairly elaborate with checksums. If the catalogues
> are corrupted
> and the experiment doesn't keep its own record the files are
> effectively
> lost even if they still exist.
>
> Stephen
>
There are two situations.
Firstly where the physical file is lost (due to whatever unwanted
reason). This can be detected by a consistency check on the RLS and a
restore of the physical file can be done if it's still supposed to be
there following the RLS (and of course it must be investigated why the
file got lost).
Secondly if the RLS itself is corrupted (a physical file can't be found
in the RLS), the situation you describe, that's indeed not so easy to
handle. You can do a restore of the whole RLS, but that's tricky because
all changes between the time of backup and the time of restore are lost
(ideally you need a rollback log too). And you must be sure that you
want to do this. A consistency check can help here too at least in
detecting anomolies.
On the other hand the RLS is supposed to be very secure and protected
against unwanted actions from users. The unwanted loss of physical files
is more likely too happen.
Jules
|