the lvm shared storage question of the week

Hello there,

Is there any hint or any links anyone could share for the use of lvm
on an AOE storage (shared block device over the network). I wanted to know how to create lvm group on a device that is shared among several host.

so i use debian 4 servers that acces a AOE target drive. The servers all mount the etherdrive and i want to use lvm to partition this 4Terabyte block device in smaller chunck. what should be the best way to do that ?

While googling about this i found references to clvm but no documentation i could use on how to set it up. Is this the way to go ?

regards,
Jean.

Our testing group has machines that use AOE. It works well with CLVM. However, they use their machines as a cluster. It seems to me that you want to carve up your storage, but you only want one machine to use each piece.

LVM manual briefly mentions CLVM (should apply equally well to debian):
http://www.redhat.com/docs/manuals/enterprise/RHEL-5-manual/Cluster_Logical_Volume_Manager/index.html

You will want to come up to speed on the cluster infrastructure (at least the setup) if you want to use CLVM....

brassow

Jonathan Brassow a �crit :

yes each server will mount a part of the filesystem and not the same part on each. This mean a single lvm logical volume will be mounted only on one host.

yes it just speak about it but there is no doc at all, nothing tells us if it listen on a tcp port or just communicate by using data on the disk or anything about the setup. Also it speak about special configuration in lvm.conf but nowhere i see WHAT should be configured :)

i guess with:

locking_type What type of locking to use. 1 is the default, which use flocks on files in locking_dir (see below) to avoid conflicting LVM2 commands running concurrently on a single machine. 0 disables locking and risks corrupting your metadata. If set to 2, the tools will load the external locking_library (see below). If the tools were configured --with-cluster=internal (the default) then 3 means to use built-in cluster-wide locking. All changes to logical volumes and their states are communicated using locks.

that it should use

locking_type=2

but is this enough ?

sorry i am not native english speaker, what do you mean by that ?

regards,
Jean.

On Wed, 2008-08-06 at 08:38 +0200, Jean Spirat wrote:

Hi,

CLVM needs a distributed lock manager to work (which communicates over the network, not via the disk). This is usually provided with the help of the RedHat Cluster Suite.

by default, locking type 2 or 3 requires clvmd, which as I say will require a cluster setup of some kind.

You *can* work without a cluster manager but I'm not sure how advised it is. So you leave your locking type as 1 and use "lvchange --refresh /dev/vg/*" on all nodes to reload metadata from disk whenever you make a change. Can be a bit risky.

John.

http://johnleach.co.uk

locking on the disk seems easier to setup than network port firewall rules and so on but if it was not done i assume there was a good reason for it that my poor brain cannot see :) if each node is gived a unique key then you could use on disk locking with a special lock metadata and a timestamps for reloading configuration for the other node. Of course it seems too easy to have not been thinked about so there must be a major flaw in my thinking :)

strange that this use case has not handled in an safer/easier way :)

if each lvm partition is mounted on only one box where does the risk lies in your opinion ?

regards,
Jean.