[Info-vax] OT: ill-behaved filesystems and databases on Linux? Thank you but so what?

[John Wallace]

On Jul 12, 2:20 am, onedbguru <onedbg... at yahoo.com> wrote:
> On Jul 11, 6:33 pm, John Wallace <johnwalla... at yahoo.co.uk> wrote:
>
>
>
> > On Jul 11, 8:44 pm, "Richard B. Gilbert" <rgilber... at comcast.net>
> > wrote:
>
> > > On 7/11/2011 2:51 PM, Phillip Helbig---undress to reply wrote:
>
> > > > Obviously, if I see the error count increasing quickly on a physical
> > > > disk, I will replace it.  My hope is that until I do so, HBVS will keep
> > > > my data safe.  (For really important shadow sets, I have 3 members; for
> > > > others, 2.)  But what about for SLOWLY increasing error counts?  And
> > > > what about errors on the shadow set itself, rather than on the members?
>
> > > For SLOWLY increasing error counts, I'd have a look at the error log to
> > > see what's going on.  If you are getting errors, your hardware is trying
> > > to tell you something.  Sometimes, it's telling you REPLACE ME ASAP!
>
> > > You had better be paying attention!
>
> > > If you can't interpret the errorlog, your Field Service Engineer can.
> > > Yes, he, or she, costs money.  If your system is business critical,
> > > spend the money!
>
> > > > Obviously, physically bad sections of a disk can cause errors, but what
> > > > are other causes of error on physical disks and on shadow sets?
>
> > > Users!  I recall a user many years ago who took a two week vacation and
> > > left a program running the filled a disk directory with several
> > > thousands of small files!  All those files were cataloged in the SAME
> > > DIRECTORY.
>
> > > Take my word for it, a directory with fifteen thousand files does NOT
> > > WORK QUICKLY or well!
> > > <snip>
>
> > A directory with too many files may well lead to poor performance, but
> > under what circumstances do you think it will lead to an increase in
> > the device error count on a physical disk?
>
> And a directory on Linux with 100K files will crash a running Oracle
> db (audit logs or trace files).

Maybe a particular Linux version with a particular Linux filesystem
will crash a particular database product, which if correct is pretty
poor in anybody's book, but:
1) a database is not a hardware device-level error logger
2) Linux is not VMS
3) The conversation was about hardware device-level errors on VMS
devices not ill-behaved fillesystems and databases on a totally
different OS which last time I checked didn't even have any concept of
a device-level error logging and analysis infrastructure