[Ocfs2-users] Can't delete LV snapshot after mounting

[Armin Wied]

Dear Sunil

Thank's a lot for your reply.

I think I maybe wasn't clear enough on one point and you got me wrong.
I don't have LVM on top of DRBD but the other way round:
OCFS2 -> DRBD -> LVM -> MD-RAID -> physical disk
Same on both machines, so do you still think, that's queasy?

I have used iSCSI already for some other setup, but wanted to use just 2
single PCs for this setup to have a cluster system with no single point of
failure, so I don't see any advantage of using iSCSI in this scenario.

You are right with your guess reagrding the o2hb thread. There is one after
the system is rebootet and a second one is started, as soon as I mount the
snapshot. After unmounting, both are still there, so that appears to be the
reason, why I can't remove the snapshot.

Is there any (safe) way get rid of it after unmounting?

Thank's
Armin



> -----Original Message-----
> From: Sunil Mushran [mailto:sunil.mushran at oracle.com]
> Sent: Thursday, March 18, 2010 6:28 PM
> To: Armin Wied
> Cc: ocfs2-users at oss.oracle.com
> Subject: Re: [Ocfs2-users] Can't delete LV snapshot after mounting
>
>
> I am queasy recommending such a setup to anyone. It is one thing to handle
> a workload. The problem is about handling user/admin errors. You are
> essentially
> running a local volume manager that is unaware of the other node. Any
> reconfig
> that is not coordinated will lead to corruption. Below that you have
> drbd which
> a fine block device replication solution. But I would personally choose
> iscsi
> which is a excellent lowcost shared device. Also, does not limit you to
> 2 nodes.
> The iscsi target in sles is known to be good. Why not just use that and
> use drbd
> to replicate the device (like emc srdf) for still higher availability.
>
> Having said that, the 2 you are seeing is not because of the number of
> nodes but
> because of the hb thread. umount is supposed to stop that hb thread.
> Maybe that is
> not happening.
>
> # ps aux | grep o2hb
> You should see one when the volume is mounted and not when umounted.
>
> Sunil
>
> Armin Wied wrote:
> > Hello group!
> >
> > I'm pretty new to ocfs2 and clustered file systems in general.
> > I was able to set up a 2 node cluster (CentOS 5.4) with ocfs2
> 1.4.4 on DRBD
> > on top of a LVM volume.
> >
> > Everything works like a charm and is rock solid, even under heavy load
> > conditions, so I'm really happy with it.
> >
> > However, there remains one little problem: I'd like to do backups with
> > snapshots. Creating the snapshot volume, mounting, copying and
> dismounting
> > works like expected. But I can't delete the snapshop volume after it was
> > mounted once.
> >
> > What I do is:
> >
> > lvcreate -L5G -s -n lv00snap /dev/vg00/lv00
> > tunefs.ocfs2 -y --cloned-volume /dev/vg00/lv00snap
> > mount -t ocfs2 /dev/vg00/lv00snap /mnt/backup
> >
> > (copy stuff)
> >
> > umount /mnt/backup
> > lvremove -f /dev/vg00/lv00snap
> >
> > lvremove fails, saying, that the volume is open. Checking with
> lvdisplay it
> > tells me "# open" is 1.
> > And that's the funny thing: After creating the snapshot volume,
> # open is 0,
> > what's not a surprise. After mounting the volume, # open is 2 -
> which is the
> > same for the other ocfs2 volume and makes sense to me, as there
> are 2 nodes.
> > But after unmounting the snapshot volume, the number decreases
> to 1, not to
> > 0 so LVM consideres the volume still open.
> >
> > I also tried mounting read only and/or adding "--fs-features=local" to
> > tunefs.ocfs2 without success. In the moment I have to reboot
> the node to be
> > able to remove the snapshot.
> >
> > So what am I doing wrong?
> >
> > Thank's a lot for any hint!
> >
> > Armin
>