[Linux-cluster] Storage Cluster Newbie Questions - any helpwith answers greatly appreciated!

Andreas Pfaffeneder apfaffeneder at pfaffeneder.org
Tue Mar 16 06:28:24 UTC 2010


Hi Michael,

Am 16.03.2010 00:23, schrieb Michael @ Professional Edge LLC:
> Hello again Andreas,
>
> Interesting mix... assuming I follow your logic... you are suggesting
> I mix the - "single non-clustered LVM mirror" with the "HA-LVM tag
> based fail-over".

yes. The other way is to use clustered-lvm which has the advantage of
keeping its mirrors-state even when moving the volume(s) between nodes.
In the setups I take care of, GFS(2) and concurrent access from both
nodes is not a requirement and I like to be able to take a LV online
without the cluster-framework.

>
> HA-LVM is where I started on this concept... so I have already setup
> the basic /etc/lvm/lvm.conf - "volume_list = [ "VolGroup00",
> "@`hostname`" ] type stuff for root mount only.
>
> When I looked at the LVM mirroring in general the concept of the
> dedicated disk for mirror logging and/or use of memory made me
> pause... but... you are saying do something like on node#1 - take
> "shelf1: disk1" and LVM mirror it to "shelf2: disk1" - and put the
> mirror logging in memory?  
true. lvcreate -m1 --corelog etc.

> Therefore in a fail-over condition - I obviously loose the in memory
> mirror logging - but node#2 would just see the LVM as in an
> inconsistent state - and resync the mirror?

Because of the lack of a logging-lun, LVM would always assume, that your
devices are not in sync, so yes, it would sync everytime the LV is
activated/moved between cluster-systems. If your LUNs are too large or
you do not want the overhead, clvm is the way to go.

Andreasw
>
> -Michael
>
>
> Andreas Pfaffeneder wrote, On 3/15/2010 7:52 AM:
>> Hi Michael,
>>
>> a way to prevent both system from thinking that they're responsible
>> for the
>> FC-devices is to use LVM for building a host-based-mirror and 
>> LVM-filter
>> and -tags:
>>
>> - Set up your RHEL-Cluster
>> - Modify /etc/lvm/lvm.conf: volume_list =
>> ["local-vg-if-used-to-store-/","@local_hostname-as-in-cluster.conf"]
>> (all
>> systems of the cluster)
>> - rebuild initrd, reboot
>> -->  LVM only accesses local LVs and LVs with the local hostname in the
>> lvm-tag
>> - create an mirrored lv with lvcreate --addtag or lvchange --addtag
>> so the
>> lv will be active on one system
>>
>> RH-Cluster supports floating LVs on its own, just add the LV + FS as
>> resource.
>>
>> If you're not using a LUN for mirror-logging, your're moving the LVs
>> from
>> one system to the other at the cost of one full rebuild of the host
>> based
>> mirror.
>>
>> Regards
>> Andreas
>>




More information about the Linux-cluster mailing list