[DRBD-user] DRBD and KVM for a HA-Cluster ?

Digimer linux at alteeve.com
Thu Jan 6 17:04:08 CET 2011

Note: "permalinks" may not be as permanent as we would like,
direct links of old sources may well be a few messages off.


On 01/06/2011 10:29 AM, Lentes, Bernd wrote:
> Digimer wrote:
>>
>>   Using DRBD resources as LVM PVs is perfectly fine. Keep in 
>> mind though that I don't think DRBD resources can be resized 
>> themselves, though I could be wrong. What I've done in the 
>> past is to simply setup a second DRBD resource, set it as a 
>> PV and then add that PV to the common VG.
>>
>>   If you're building a 2-node cluster and you are running 
>> Primary/Primary, then yes, I think Clustered LVM is best (perhaps not
>> *strictly* required). Given that each VM is on a dedicated 
>> LV, in theory you should not need clustered locking, though I 
>> think having CLVM is a good idea. It gives you the option of 
>> creating a GFS2 partition with the VMs XML config files, 
>> shared ISOs, etc. I always do this, so I should say that I've 
>> not looked at the feasibility of *not* using CLVM/GFS2.
>>
>> Cheers
>>
> 
> I don't want to setup a DRBD as a PV. My idea is to set up a DRBD on top of a LV, and to format or not format this LV, depending on the decision to install a KVM into a file or a plain device.
> Is this setup ok ?

That would end up being a stacked LVM:

<sda1><sdb1>
 |
<md0>
 |
<pv>
 |
<vg>
 |
<lv(s)>
 |
<drbd>
 |
<pv>
 |
<vg>
 |
<lv(s)>
 |
<fs-or-lv/per VM>

Is this what you had in mind? Or are you thinking of just creating raw,
standard partitions on the DRBD resource?

If it is this, I was doing something similar some time ago, and it
worked. However, after discussing this setup with Chrissie from RH, she
recommended against it when using CLVM (which I was for a GFS2 partition
for shared storage).

The reasoning was that it could cause problems as the LVM would start
before the cluster, so the clustered LVM members would not be available.
Unavoidable as I need LVM at node boot-time, long before the cluster
could start. In the end, I decided to forgo LVM on each node and instead
only use LVM on the DRBD.

If you are not planning to use LVM on the DRBD resource, but instead use
raw partitions, then you won't use clvm. As for how well it works, I
can't say as I've not tried that. I can't *think* of any reason why it
wouldn't work though.

-- 
Digimer
E-Mail: digimer at alteeve.com
AN!Whitepapers: http://alteeve.com
Node Assassin:  http://nodeassassin.org



More information about the drbd-user mailing list