XFS Journal on md device

XFS Journal on md device

[Chris Seufert]

Hey,

I currently have a 2.1Tb xfs partition, sitting on a hardware SATA raid card.

However i also have 2 hdd's (for OS etc) in software raid1, with a md
device for the xfs log file. Its a 100mb RAID1 (under /dev/md4), now
when i halt/reboot the box, even after the xfs partition is unmounted
(as part of the shutdown sequence as normal on debian etch) the
/dev/md4 device cant be cleanly stopped.

Is having the log on a redundant partition a good idea or is it better
to leave it as an internal log or is there another way round this
problem.

-Chris

Re: XFS Journal on md device

[Chris Seufert]

unable to stop /dev/md3, right before it does the powering off
/dev/sda /dev/sdb etc...

I believe its from the init scripts for md. Its only there for a sec,
so im not completely sure of the message.

However on reboot xfs does a journal rebuild/repair. and the md does a
re-sync of the md device.

On 9/6/06, linux-xfs@oss.sgi.com <linux-xfs@oss.sgi.com> wrote:
> Chris Seufert wrote:
> > Hey,
> >
> > I currently have a 2.1Tb xfs partition, sitting on a hardware SATA raid
> > card.
> >
> > However i also have 2 hdd's (for OS etc) in software raid1, with a md
> > device for the xfs log file. Its a 100mb RAID1 (under /dev/md4), now
> > when i halt/reboot the box, even after the xfs partition is unmounted
> > (as part of the shutdown sequence as normal on debian etch) the
> > /dev/md4 device cant be cleanly stopped.
>
> What sorts of related error messages do you get on shutdown?  (both from the
> initscripts & the kernel?)
>
> -Eric
>

Re: XFS Journal on md device

[Chris Seufert]

Sorry i meant to say md4, not that it really matters. if I unmount the
xfs volume before i halt it seems fine (althou this is not that easy
with nfs+samba shares on it)

On 9/6/06, Chris Seufert <seufert@gmail.com> wrote:
> unable to stop /dev/md3, right before it does the powering off
> /dev/sda /dev/sdb etc...
>
> I believe its from the init scripts for md. Its only there for a sec,
> so im not completely sure of the message.
>
> However on reboot xfs does a journal rebuild/repair. and the md does a
> re-sync of the md device.
>
> On 9/6/06, linux-xfs@oss.sgi.com <linux-xfs@oss.sgi.com> wrote:
> > Chris Seufert wrote:
> > > Hey,
> > >
> > > I currently have a 2.1Tb xfs partition, sitting on a hardware SATA raid
> > > card.
> > >
> > > However i also have 2 hdd's (for OS etc) in software raid1, with a md
> > > device for the xfs log file. Its a 100mb RAID1 (under /dev/md4), now
> > > when i halt/reboot the box, even after the xfs partition is unmounted
> > > (as part of the shutdown sequence as normal on debian etch) the
> > > /dev/md4 device cant be cleanly stopped.
> >
> > What sorts of related error messages do you get on shutdown?  (both from the
> > initscripts & the kernel?)
> >
> > -Eric
> >
>

Re: XFS Journal on md device

[martin f krafft]

[-- Attachment #1: Type: text/plain, Size: 738 bytes --]

also sprach Chris Seufert <seufert@gmail.com> [2006.09.06.0353 +0200]:
> However on reboot xfs does a journal rebuild/repair. and the md
> does a re-sync of the md device.

Which distro?

I am the Debian maintainer for mdadm and have run into the problem
that the array used for / cannot be stopped until after / is
unmounted, at which point nothing stops the array for there is no
shutdownramfs.

However, we (Debian) remount / read-only and I never see
a filesystem check on reboot.

-- 
martin;              (greetings from the heart of the sun.)
  \____ echo mailto: !#^."<*>"|tr "<*> mailto:" net@madduck
 
spamtraps: madduck.bogus@madduck.net
 
the micro$oft hoover: finally, a product that's supposed to suck!

[-- Attachment #2: Digital signature (GPG/PGP) --]
[-- Type: application/pgp-signature, Size: 189 bytes --]

Re: XFS Journal on md device

[Chris Seufert]

I'm running Debian Testing (is that Etch), updated about a week ago,
on AMD64, with Kernel 2.6.18-rc5-mm1 + md hotfix(1) with md built-in,
no initrd, so raid autodetect works.

My root (/) volume is ext3 running on /dev/md0 (RAID 1), the problem
is with my /data volume thats xfs, running on /dev/sda, with log on
/dev/md4.

1: The patch is required becase mm1 killed the KConfig for md devices.

On 9/6/06, martin f krafft <madduck@madduck.net> wrote:
> also sprach Chris Seufert <seufert@gmail.com> [2006.09.06.0353 +0200]:
> > However on reboot xfs does a journal rebuild/repair. and the md
> > does a re-sync of the md device.
>
> Which distro?
>
> I am the Debian maintainer for mdadm and have run into the problem
> that the array used for / cannot be stopped until after / is
> unmounted, at which point nothing stops the array for there is no
> shutdownramfs.
>
> However, we (Debian) remount / read-only and I never see
> a filesystem check on reboot.
>
> --
> martin;              (greetings from the heart of the sun.)
>   \____ echo mailto: !#^."<*>"|tr "<*> mailto:" net@madduck
>
> spamtraps: madduck.bogus@madduck.net
>
> the micro$oft hoover: finally, a product that's supposed to suck!
>
>
> -----BEGIN PGP SIGNATURE-----
> Version: GnuPG v1.4.5 (GNU/Linux)
>
> iD8DBQFE/kMrIgvIgzMMSnURAoSgAJ4mQ8a1RH6sYd7VRn4yZsRNKxbeSACdEGFv
> HeWvLK1N+R1nvxMfeqlDZk8=
> =LyfV
> -----END PGP SIGNATURE-----
>
>
>

Re: XFS Journal on md device

[martin f krafft]

[-- Attachment #1: Type: text/plain, Size: 1115 bytes --]

also sprach Chris Seufert <seufert@gmail.com> [2006.09.06.0553 +0200]:
> My root (/) volume is ext3 running on /dev/md0 (RAID 1), the
> problem is with my /data volume thats xfs, running on /dev/sda,
> with log on /dev/md4.

Can you tell when /data gets umounted during the shutdown sequence?
Correct me if I'm wrong, but once that happened, /dev/md4 should
become free as far as XFS is concerned.

However, since mdadm or the kernel fails to stop the device during
shutdown, I am guessing that the partition is simply not being
umounted.

Does it have an entry in /etc/fstab?

Try changing the 

  #! /bin/sh

in line 1 of /etc/rc6.d/S40umountfs to

  #! /bin/sh -x
  exec 2> /root/umountfs.out.2

then reboot and paste that file somewhere (http://rafb.net/paste).

-- 
martin;              (greetings from the heart of the sun.)
  \____ echo mailto: !#^."<*>"|tr "<*> mailto:" net@madduck
 
spamtraps: madduck.bogus@madduck.net
 
"you don't sew with a fork, so I see no reason
 to eat with knitting needles."
                               -- miss piggy, on eating chinese food

[-- Attachment #2: Digital signature (GPG/PGP) --]
[-- Type: application/pgp-signature, Size: 189 bytes --]