Doesn't *anyone* (else) use an atacontrol-built software RAID1 (mirror)?

Hi,

I've posted this sort of question to freebsd-questions and
freebsd-stable, back on the 22nd, when I first noticed a
problem.  Much to my surprise, I've not received any answer
at
all, which makes me wonder if, perhaps, I'm the only one
actually using this sort of configuration.  If that's the
case,
then I'll happily drop it like a hot rock and try again with
a
GEOM mirror, or something like that.

Here's the story:

I just (a week ago) tried to do an in-place upgrade from
RELENG_5 to RELENG_6 on a system that was running an ataraid
(built with atacontrol) mirror on a pair of SATA disks.  The
root file system (256M) mounted fine, but fsck -p fails
(unable
to find a superblock, from memory) under the new kernel on
my
/usr partition, which is about 73G (most of the 80G of the
disks).  Falling back to the RELENG_5 /boot/kernel and fsck
finds no problem with that filesystem.

Is there a known behaviour change in ataraid behaviour
between
the two versions?  The hardware is Intel P4, ICH6 SATA150,
and a
pair of Seagate 80G SATA drives.  I can dig up more details
if
necessary.  Full dmesg and config file details are included
in
the message to freebsd-questions on the 22nd, entitled
"ataraid/fsck glitch on upgrade from 5.5 to
6-stable?"

One odd-looking thing that I noticed (in RELENG_5, of
course) is that fdisk ar0 says that cylinders=9729 heads=255
sectors/track=63, but fdisk ad4 (the first of the
"real" SATA
disks) says cylinders=155061, heads=16, sectors/track=63.
Should these really be so different?  I haven't tested it,
but
wouldn't this geometry difference make recovery with one
dead
drive kind of awkward?

Is backup and start again from label/newfs (perhaps using
gmirror instead of atacontrol?) my only option?

Cheers,

-- 
Andrew
_______________________________________________
freebsd-fsfreebsd.org mailing list

http://lists.freebsd.org/mailman/listinfo/freebsd-fs
To unsubscribe, send any mail to
"freebsd-fs-unsubscribefreebsd.org"

On Fri, Sep 29, 2006 at 11:53:52AM +1000, Andrew Reilly
wrote:
> Hi,

Hi,

I normally would reinstall a machine from scratch rather
than having a mission
to upgrade from one major release to another major release.

Also the software pseudo raid behaves quite well from my
usage of it in
production environments.

Regards
--jm
> 
> I've posted this sort of question to freebsd-questions
and
> freebsd-stable, back on the 22nd, when I first noticed
a
> problem.  Much to my surprise, I've not received any
answer at
> all, which makes me wonder if, perhaps, I'm the only
one
> actually using this sort of configuration.  If that's
the case,
> then I'll happily drop it like a hot rock and try again
with a
> GEOM mirror, or something like that.
> 
> Here's the story:
> 
> I just (a week ago) tried to do an in-place upgrade
from
> RELENG_5 to RELENG_6 on a system that was running an
ataraid
> (built with atacontrol) mirror on a pair of SATA disks.
 The
> root file system (256M) mounted fine, but fsck -p fails
(unable
> to find a superblock, from memory) under the new kernel
on my
> /usr partition, which is about 73G (most of the 80G of
the
> disks).  Falling back to the RELENG_5 /boot/kernel and
fsck
> finds no problem with that filesystem.
> 
> Is there a known behaviour change in ataraid behaviour
between
> the two versions?  The hardware is Intel P4, ICH6
SATA150, and a
> pair of Seagate 80G SATA drives.  I can dig up more
details if
> necessary.  Full dmesg and config file details are
included in
> the message to freebsd-questions on the 22nd, entitled
> "ataraid/fsck glitch on upgrade from 5.5 to
6-stable?"
> 
> One odd-looking thing that I noticed (in RELENG_5, of
> course) is that fdisk ar0 says that cylinders=9729
heads=255
> sectors/track=63, but fdisk ad4 (the first of the
"real" SATA
> disks) says cylinders=155061, heads=16,
sectors/track=63.
> Should these really be so different?  I haven't tested
it, but
> wouldn't this geometry difference make recovery with
one dead
> drive kind of awkward?
> 
> Is backup and start again from label/newfs (perhaps
using
> gmirror instead of atacontrol?) my only option?
> 
> Cheers,
> 
> -- 
> Andrew
> _______________________________________________
> freebsd-fsfreebsd.org mailing list
> 
http://lists.freebsd.org/mailman/listinfo/freebsd-fs
> To unsubscribe, send any mail to
"freebsd-fs-unsubscribefreebsd.org"
_______________________________________________
freebsd-fsfreebsd.org mailing list

http://lists.freebsd.org/mailman/listinfo/freebsd-fs
To unsubscribe, send any mail to
"freebsd-fs-unsubscribefreebsd.org"

On Fri, Sep 29, 2006 at 04:37:49PM +0200, Jacques Marneweck
wrote:
> On Fri, Sep 29, 2006 at 11:53:52AM +1000, Andrew Reilly
wrote:
> I normally would reinstall a machine from scratch
rather than having a mission
> to upgrade from one major release to another major
release.

That's what I think that I'll have to do.  It's a pity, as
I've
had good success doing the 5->6 transition in place on
other
machines.  That's a much smaller step than the gargantuan
4->5
upgrade, which definitely needed a from-scratch switch (to
get
UFS2 on the disks).

I suppose that I should report the raid breakage in a pr, so
that
it becomes a known problem.

> Also the software pseudo raid behaves quite well from
my usage of it in
> production environments.

Yes, it's certainly been working well for me on 5-stable, so
far.

Cheers,

-- 
Andrew
_______________________________________________
freebsd-fsfreebsd.org mailing list

http://lists.freebsd.org/mailman/listinfo/freebsd-fs
To unsubscribe, send any mail to
"freebsd-fs-unsubscribefreebsd.org"