Message-ID: <611ad3510712170546g11e4003cxb54f400ba7a6c9d8@mail.gmail.com>
Date: Mon, 17 Dec 2007 07:46:21 -0600
From: "Jeremy Schneider" <jeremy.schneider@ardentperf.com>
To: "Alex Gorbachev" <ag@oracloid.com>
Subject: Re: superblock backups, ASM vs OCFS2
Cc: Oracle-L <oracle-l@freelists.org>
In-Reply-To: <c2213f680712160653t50926d67wfe637be46538c8a1@mail.gmail.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=ISO-8859-1
Content-Disposition: inline
References: <611ad3510712130850j4575aea3k4d43481b27699908@mail.gmail.com>
	 <c2213f680712160653t50926d67wfe637be46538c8a1@mail.gmail.com>
Sender: oracle-l-bounce@freelists.org
Errors-to: oracle-l-bounce@freelists.org
Precedence: normal
Reply-to: jeremy.schneider@ardentperf.com

I totally agree about always using partitions in Linux.

However in this particular case the ASMLIB labels were created on top
of entire disks - and I don't think that ASMLIB never issued any
warnings about not using partitions.  (I don't remember ever seeing
any warnings - is this a new feature in 11g?)  FYI, this was a rushed
setup of a test environment for some basic load testing...  I always
create partitions for production setups and somehow this particular
case was the exception.  (I don't remember; they may have had ASM
setup for this machine before I arrived onsite.  The entire setup and
load testing of five different platforms was done in 4 days, according
to their requirements.)

An SA apparently went in later and created partition tables and LVM
labels on the partitions.  I'm pretty sure that partition tables are
written at the *end* of the first block - which is why the ASMLIB
labels were fine (at the top of the block).  And the LVM labels were
in the partition - corrupting the ASM disk headers but leaving the
ASMLIB labels untouched again.

It was actually a pretty tricky situation to figure out at first -
they called back several months after I'd come onsite to do the
testing and wanted help figuring out why they couldn't mount the ASM
disks.  And I assumed that I'd made the partitions at first and
couldn't figure out why the ASM disk headers didn't point to the
partitions.  Took a little detective work.  :)

Jeremy


On 12/16/07, Alex Gorbachev <ag@oracloid.com> wrote:
> Jeremy,
>
> Oracle normally skips the first block of datafiles on raw devices to
> avoid intervention with LVM headers. I wish they do that for ASM but
> in ASM that space if used by ASM itself for similar purpose as in LVM.
>
> On the other hand, ASM disks should be created on top of partitions
> (in Linux terminology) but not the whole disks themselves. In fact,
> ASMLib (if you happen to use it) requires that - it won't mark ASM
> disk unless it's a partition. This would save your disk in case when
> the header of the LUN/physical disk is corrupted. Though, partition
> table may be hosed but that's recoverable and you can easilly back it
> up with "dd" - it's static.
>
> Cheers,
> Alex
>

-- 
Jeremy Schneider
Chicago, IL
http://www.ardentperf.com/category/technical
--
http://www.freelists.org/webpage/oracle-l