Note: "permalinks" may not be as permanent as we would like,
direct links of old sources may well be a few messages off.
On Fri, Oct 26, 2012 at 12:55:44AM +1100, Adam Goryachev wrote:
> In your case, it sounds like you used LVM (block devices) instead of
> /dev/sda1, and that you used DRBD instead of mdadm, and you used kvm
> instead of mount... but at the end of the day, you told kvm to slip
> under DRBD, which is allowed, but it hurts.....
It's always a struggle to draw the right line between stuff where deep
technical knowledge should be aspired to and stuff where it's just fine to
take an "and then the magic happens" approach. Those with knowledge both
general and deep are few. At some point, we have to stop studying the stuff
and implement, choose our deep specialties and trust other stuff to make
sense on the surface.
I like recipes at even more than ASCII art! Yours makes the task of fixing
my mess look more straight forward than I'd thought. One question;
> I would suggest that at the least, you will need to
> 1) Stop one VM (at a time)
> 2) Start the VM on top of DRBD instead of LVM
In SomeVM.xml where I currently have:
<disk type='block' device='disk'>
<driver name='qemu' type='raw'/>
<source dev='/dev/VolumeGroup/SomeVolume'/>
<target dev='hda' bus='ide'/>
<address type='drive' controller='0' bus='0' unit='0'/>
</disk>
I should be able to stop the domain, change the middle line using "virsh
edit" to:
<source dev='/dev/drbdN'/>
restart the domain, do a full sync, and I'm good? I don't need to create an
emptly volume, put DRBD on it, and then clone an existing VM into that
space? I can take an existing raw-VM-on-LVM volume, add DRBD, stop the VM,
edit the XML to mount it through DRBD, and internal metadata should work,
despite room not being initially set aside for it?
If that works, then LCMC could even accomplish this in a future release.
> 3) Instruct DRBD to throw away the secondary
> 4) do a full sync from the primary to the secondary
>
> Repeat for each VM
> Internal or External metadata is irrelevant, as long as the VM's are
> reading/writing to DRBD and not LVM.
Thanks,
Whit