Discussion:
[linux-lvm] Shared VG, Separate LVs
Indivar Nair
7 years ago
Permalink
Hi...,

With CLVM / HA-LVM on a 2 node cluster -

Is it possible to have a shared VG but separate LVs, with each LV
exclusively activated on different nodes in a 2 node cluster.
In case of a failure, the LV of the failed node will be activated on the
other node.

Regards,


Indivar Nair
Eric Ren
7 years ago
Permalink
Hi,
Post by Indivar Nair
With CLVM / HA-LVM on a 2 node cluster -
Is it possible to have a shared VG but separate LVs, with each LV
exclusively activated on different nodes in a 2 node cluster.
In case of a failure, the LV of the failed node will be activated on
the other node.
I think clvm can do what you want if you perform LVM commands by hand.
But, with HA cluster manager (pacemaker) you cannot
do it with the current resource agents (clvm + LVM) [1] [2], because
they do failover at a VG basis.

We are currently working on new resource agents [3] for lvmlockd [4].
The new agents can do activation on LV basis, but I won't
recommend doing that if there's no strong reason. It makes things much
more complicated.

[1]
https://github.com/ClusterLabs/resource-agents/blob/master/heartbeat/clvm
[2] https://github.com/ClusterLabs/resource-agents/blob/master/heartbeat/LVM
[3] https://www.redhat.com/archives/linux-lvm/2017-January/msg00025.html
[4] https://github.com/ClusterLabs/resource-agents/pull/1040

Eric
Eric Ren
7 years ago
Permalink
Hi,
...
OK! It sounds reasonable.
In my case, we will have the same PVs and VG activated on both hosts,
but LV1 on Host01 and LV2 on Host02. I paln to use ext4 or XFS
filesystems.
Is there some possibility that it would work?
As said in the last mail, the new resource agent [4] will probably work
for you, but I didn't test this case yet. It's easy to have a try - the
RA is just shell
script, you can just copy LVM-activate to
/usr/lib/ocf/resource.d/heartbeat/ (assume you've installed
resource-agents package), and then configure
"clvm + LVM-activate" for pacemaker [5]. Please report back if it
doesn't work for you.

The LVM-activate RA is WIP. We are thinking if we should merge it into
the old LVM RA. So it may changes at any time.

[5]
https://www.suse.com/documentation/sle-ha-12/book_sleha/data/sec_ha_clvm_config.html
...
Eric
Indivar Nair
7 years ago
Permalink
Hi Eric, All,

Thanks for the input. I have got it working.

Here is what I did -
-------------------------------------------------------------------------------------------------------------------------------------------------------
Cluster Setup:
2 Nodes with CentOS 7.x: clstr01-nd01, clstr01-nd02
Common storage array between both nodes (8 shared volumes, presented as
/dev/mapper/mpatha to /dev/mapper/mpathh)
2 Port NICs, bonded (bond0) in each node

Resource group grp_xxx (nd01 preferred) -
Mount Point: /clstr01-xxx
Cluster IP: 172.16.0.101/24

Resource group grp_yyy (nd02 preferred) -
Mount Point: /clstr01-yyy
Cluster IP: 172.16.0.102/24


On both nodes:
--------------
Edit /etc/lvm/lvm.conf, and configure 'filter' and 'global_filter'
parameters to scan only the required (local and shared) devices.

Then run -
# /sbin/lvmconf --enable-cluster
Rebuild initramfs -
# mv /boot/initramfs-$(uname -r).img /boot/initramfs-$(uname -r).img-orig
# dracut -H -f /boot/initramfs-$(uname -r).img $(uname -r)

Reboot both nodes.
--------------



After rebooting both nodes, run the following commands on any one node:
--------------
# pcs cluster start --all
# pcs resource create dlm ocf:pacemaker:controld op monitor interval=30s
on-fail=fence clone interleave=true ordered=true
# pcs resource create clvmd ocf:heartbeat:clvm op monitor interval=30s
on-fail=fence clone interleave=true ordered=true
# pcs constraint order start dlm-clone then clvmd-clone
# pcs constraint colocation add clvmd-clone with dlm-clone


# pvcreate /dev/mapper/mpath{a,b,c,d,e,f,g,h}
# vgcreate -Ay -cy clstr_vg01 /dev/mapper/mpath{a,b,c,d,e,f,g,h}
# lvcreate -L 100T -n lv01 clstr_vg01
# mkfs.xfs /dev/clstr_vg01/lv01
# lvcreate -L 100T -n lv02 clstr_vg01
# mkfs.xfs /dev/clstr_vg01/lv02


# pcs resource create xxx_mount ocf:heartbeat:Filesystem
device=/dev/clstr_vg01/lv01 directory=/clstr01-xxx fstype=xfs --group
xxx_grp --disabled

# pcs resource create xxx_ip_01 ocf:heartbeat:IPaddr2 ip=172.16.0.101
cidr_netmask=24 nic=bond0:0 op monitor interval=30s --group xxx_grp
--disabled

# pcs constraint location xxx_grp prefers clstr01-nd01=50
# pcs constraint order start clvmd-clone then xxx_grp

# pcs resource enable xxx_mount
# pcs resource enable xxx_ip_01


# pcs resource create yyy_mount ocf:heartbeat:Filesystem
device=/dev/clstr_vg01/lv02 directory=/clstr01-yyy fstype=xfs --group
yyy_grp --disabled

# pcs resource create yyy_ip_01 ocf:heartbeat:IPaddr2 ip=172.16.0.102
cidr_netmask=24 nic=bond0:1 op monitor interval=30s --group yyy_grp
--disabled

# pcs constraint location yyy_grp prefers clstr01-nd02=50
# pcs constraint order start clvmd-clone then yyy_grp

# pcs resource enable yyy_mount
# pcs resource enable yyy_ip_01
--------------


# pcs resource show
--------------
-------------------------------------------------------------------------------------------------------------------------------------------------------


Regards,


Indivar Nair
...
Eric Ren
7 years ago
Permalink
Had a look at your setup, I have one question:

Did you check if your active-passive model HA stack can always work
correctly and stably by
putting one node into offline state?

I noticed you didn't configure LVM resource agent to manage your VG's
(de)activation task,
not sure if it can always work as expect, so have more exceptional
checking :)

Eric
...
Indivar Nair
7 years ago
Permalink
Hi Eric,

Answering your queries -


*"Did you check if your active-passive model HA stack can always work
correctly and stably byputting one node into offline state?"*

Yes, it works perfectly while failing over and failing back.



*"I noticed you didn't configure LVM resource agent to manage your VG's
(de)activation task,not sure if it can always work as expect, so have more
exceptional checking :)"*

Strangely the Pacemaker active-passive configuration example
shows VG controlled by Pacemaker, while the active-active one does not. I
have taken the active-active configuration for Pacemaker and created 2 LVs,
then instead of formatting it using the GFS2 clustered filesystem, I used
normal XFS and made sure that it is mounted only on one node at a time.
(lv01 on node 2, lv02 on node2)


https://access.redhat.com/documentation/en-us/red_hat_enterprise_linux/7/html/global_file_system_2/ch-clustsetup-gfs2

I can see the clustered VG and LVs as soon ocf:heartbeat:clvm
is started.

Is there anything I am missing here?

Regards,


Indivar Nair
...
Eric Ren
7 years ago
Permalink
Hi,
...
Good. "clvm" will activate all VGs by default. If you have more than one
VG in your cluster,  you may want to
activate/deactivate one VG for each group of "vg" and "xfs", then you
may need to look at LVM for each VG:

https://github.com/ClusterLabs/resource-agents/blob/master/heartbeat/LVM

Eric
Indivar Nair
7 years ago
Permalink
Sure. Will keep that in mind.
Thanks a lot, Eric.

Regards,


Indivar Nair
...
Loading...