[DRBD-user] Partitioning /dev/drbd1

Dennis Jacobfeuerborn dennisml at conversis.de
Wed Mar 9 13:01:35 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 03/09/2011 09:18 AM, Felix Frank wrote:
>>> while partitioning a partition is possible and rather straight-forward,
>>> it sure isn't standard practice.
>>
>> I wasn't actually suggesting to create "partitions in partitions" I just
>> wasn't aware that the drbd device nodes are partitions and not basic
>> block devices (like i.e. the /dev/xvdN nodes that xen adds). That
>> certainly explains the behavior I'm seeing.
>
> You're right, technically DRBDs are not harddrive partitions (I think).
> This had me wondering: What is your backing device? I was assuming you
> created a DRBD on top of a physical partition.

All of this is happening in a xen VM an I use a dedicated virtio disk as 
backing device i.e. in this particular case it is /dev/xvdc for drbd 
resource r1 (/dev/drbd1).
/dev/xvda is the system disk and /dev/xvdb is the backing device for drbd 
resource r0 (/dev/drbd0).
r0 is already setup as redundant nfs share using pacemaker which works 
nicely. What I am now trying to do is to also make iSCSI LUNs highly 
available next to the nfs share.

>>> Also, please make sure that replication actually works when you mount
>>> your "inner partitions" this way. I think it should, but I might be
>>> wrong.
>>
>> I'm sure one could make this work but I don't intend to try it. I've
>> seen people use logical LVM volumes as physical volumes for new volume
>> groups which *seemed* to work though I didn't know for sure because I
>> was too busy trying to get out of there. Always fun to see these kinds
>> of M.C.Escher-esque topologies....as long as I don't have to maintain
>> them :)
>
> Within the DRBD paradigm, this is actually a common and documented
> approach: http://www.drbd.org/users-guide/s-nested-lvm.html

Yes, though I'm keen to avoid going too deep with kind of nesting because 
of my experiences. In the case I mentioned the guy managed to create loops 
where a volume nested inside a logical volume was added as a physical 
volume for the volume group of it's parent. The results weren't pretty.

So eventually I will probably end up using LVM maybe even in a nested 
fashion but for now I try to keep things simple for the sake of making 
experimenting/debugging easier.

> What I'm saying is: You *are* mounting nested partitions using kpartx.
> I'd be interested in whether this is supported and works wrt. replication.

So far I've used kpartx mostly to mount partitions inside the lvm backing 
device for a virtual machine on the host. If I were to use this in this 
case that would probably require a modification of the pacemaker ocf 
scripts so that kpartx gets called automatically when the state of the drbd 
device changes. That sounds rather fiddly and non-robust so I will just 
skip this experimental step and go straight to the lvm on top of drbd step.

Regards,
   Dennis



More information about the drbd-user mailing list