[linux-lvm] LVM1 on top of DRBD
Joachim Banzhaf
jbanzhaf at ngi.de
Wed Aug 20 06:55:02 UTC 2003
Am Mittwoch, 20. August 2003 12:46 schrieb Heinz J . Mauelshagen:
> Joachim,
First, thanks for your quick response, Heinz.
> LVM1 isn't cluster aware.
I knew that - as I stated, I read the FAQ :-).
> DRBD devices are just some other "local" ones for it.
OK, but as i stated, I used them that way. LVM1 just seems to use the lower
level devices drbd works on instead. That's what's bothering me.
To make clear what I mean, here some details:
I have /dev/nb1 as resource lvm1 on top of /dev/sdb1 in /etc/drbd.conf
I have /dev/nb2 as resource lvm2 on top of /dev/sdb2 in /etc/drbd.conf
On both nodes I did:
# /etc/init.d/drbd start
On one node I did (as far as I can remember):
# /etc/ha.d/resource.d/datadisk lvm1 start
# /etc/ha.d/resource.d/datadisk lvm2 start
# pvcreate /dev/nb1
# pvcreate /dev/nb2
# vgcreate lvm-drbd /dev/nb1 /dev/nb2
# lvcreate -s 2 -n lv1-s2 lvm_drbd
# vgchange -a n lvm-drbd
On the other node I did (while the vg was inactive)
# rm -r /etc/lvm*
# vgscan
# vgcfgbackup
Now, look what vgdisplay says on PV names (on both nodes):
# vgdisplay -vD lvm-drbd
--- Volume group ---
VG Name lvm-drbd
<skip>
--- Logical volume ---
LV Name /dev/lvm-drbd/lv1-s2
VG Name lvm-drbd
<skip>
--- Physical volumes ---
PV Name (#) /dev/sdb1 (1)
PV Status available / allocatable
Total PE / Free PE 30 / 15
PV Name (#) /dev/sdb2 (2)
PV Status available / allocatable
Total PE / Free PE 30 / 15
It just seems, the partitions are found before their associated network block
devices and used despite of pvcreate commandline options.
As I stated, to follow this hypothesis I used a patch to make lvm tools ignore
the partitions /dev/sdbx.
The vgdisplay output then shows the nbx devices (even without vgscan'ning or
something). But they are not used by the kernel anyway as I see no activity
on /proc/drbd :-(
> No replication is done by LVM1.
Yes, that's why I use drbd :-)
> The whole replication process is up to DRBD, that's why you
> need to start it using a DRBD command.
>
> You can use hartbeat to recognize if a node dies and scripts to start LVM
> on the living node to access logical volumes there.
I have a fully working set of linux, lvm1, heartbeat and drbd with services on
top. I know how they work (from a user/admin perspective - and a bit more).
And they work fine for me (including failovers of course) as long as I use
drbd resources on top of lv's.
But I want to use drbd resources as pv's.
That seems not to work, because LVM1 persists in using the partition and not
the drbd network block device although I configured it that way (as stated my
original post).
Again, to clear the ground: Should it be possible at all to use pv's on top of
drbd devices or am I forced to use drbd devices on top of lv's and loose
things like online resize feature?
Have you - or anybody reading the list - ever sucessfully used that scenario?
> Regards,
> Heinz -- The LVM Guy --
>
> On Wed, Aug 20, 2003 at 11:41:20AM +0200, Joachim Banzhaf wrote:
> > Hi list,
> >
> > I'd like to use LVM1 on top of DRBD 0.6.6 and not the other way round,
> > because I then can resize my partitions online (add a new drbd device,
> > make it a pv, add it to a vg, be happy).
> >
> > It just seems not to be that simple.
> >
> > The HOWTO warns about LVM not being cluster aware and describes how to do
> > it anyway on shared storage. I (think I) have these requirements met by
> > using heartbeat to have vg's active at one node only and running vgscan
> > on the secondary after changes on the primary.
> >
> > I created the pv's using the /dev/nbx devices (x=1,2).
> > I created the vg lvm-drbd with lv lv1-s2 using these pv's.
> >
> > The corresponding haresources file is this:
> > job-vm1 192.168.1.101 datadisk::lvm1 datadisk::lvm2 LVM::lvm-drbd \
> > Filesystem::/dev/lvm-drbd/lv1-s2::/ha/export::reiserfs
> >
> > What I get is lvm using the lower level device (sdbx in my case).
> > vgdisplay -vD shows the pv's using /dev/sdbx.
> > Nothing gets replicated to the other node (not surprising then).
> > I even patched lvm_check_dev.c and rebuilt the lvm tools to ignore
> > /dev/sdbx devices, removed all lvm infos from /etc and did a new vgscan
> > and vgcfgbackup. vgdisplay -vD now shows the /dev/nbx devices for the
> > pv's. But still no replication without drbdsetup /dev/nbx replicate :-(.
> >
> > I use suse 8.2 with suse 2.4.20-4GB kernel (you'ed a few weeks ago)
> > for the lvm_check_dev.c-patch I installed the lvm source rpm, modified
> > the spec file not to use the mp patches (because a file was missing),
> > added my patchfile and did a rpm -bi /usr/src/packages/SPECS/lvm.spec.
> >
> > Btw. the build process did not stop on compile errors in lvm_check_dev.c.
> > Is that intended? It is confusing...
> >
> > What am I doing wrong?
> > Is it possible doing LVM on top of DRBD at all?
> >
> > I'd happily provide more details if it helps.
> >
> > Ciao,
> >
> > Joachim Banzhaf
> >
> >
> > _______________________________________________
> > linux-lvm mailing list
> > linux-lvm at sistina.com
> > http://lists.sistina.com/mailman/listinfo/linux-lvm
> > read the LVM HOW-TO at http://tldp.org/HOWTO/LVM-HOWTO/
>
> *** Software bugs are stupid.
> Nevertheless it needs not so stupid people to solve them ***
>
> =-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=
>-=-
>
> Heinz Mauelshagen Sistina Software Inc.
> Senior Consultant/Developer Am Sonnenhang 11
> 56242 Marienrachdorf
> Germany
> Mauelshagen at Sistina.com +49 2626 141200
> FAX 924446
> =-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=
>-=-
>
> _______________________________________________
> linux-lvm mailing list
> linux-lvm at sistina.com
> http://lists.sistina.com/mailman/listinfo/linux-lvm
> read the LVM HOW-TO at http://tldp.org/HOWTO/LVM-HOWTO/
More information about the linux-lvm
mailing list