Re: Sharing disks between two kvm guests

[Anthony Liguori <anthony@codemonkey.ws>]

Laurent Vivier wrote:
> Le vendredi 20 juin 2008 à 09:07 -0500, Javier Guerra a écrit :
>   
>> On Fri, Jun 20, 2008 at 7:23 AM, carlopmart <carlopmart@gmail.com> wrote:
>>     
>>> Felix Leimbach wrote:
>>>       
>>>>>  This is my first post to this list. I have already installed kvm-70
>>>>> under rhel5.2. My intention is to share on disk image betwwen two rhel5.2
>>>>> kvm guests. Is it possible to accomplish this in kvm like xen or vmware
>>>>> does?? How can I do?? I didn't find any reference abou this on kvm
>>>>> documentation ...
>>>>>           
>> i tried this looong ago and didn't really work because there was some
>> userspace cache on each QEMU instance.  but the -drive option has a
>> 'cache=off' setting that should be enough.
>>
>> in theory (i haven't tested, but Avi 'blessed' it):
>> - create a new image with qemu-img
>> - add it to the command line using -drive file=xxx,cache=off on both
>> KVM instances
>> - use a cluster filesystem!
>>     
>
> RFC:
>
> Well, well, perhaps it is delusions of a sick mind but since the
> introduction of qemu-nbd I think we can develop easily something to
> share a disk between several virtual hosts:
>
> I- in a first step, we can modify qemu-nbd to accept several connections
> for one disk image, for instance:
>
> # qemu-nbd my-disk.qcow2
> # nbd-client localhost 1024 /dev/nbd0
> # nbd-client localhost 1024 /dev/nbd1
>
> and start two virtual hosts:
>
> "qemu -hda v1.img -hdb /dev/nbd0" and "qemu -hda v2.img -hdb /dev/nbd1"
>
> Of course the filesystem must know how to share the access to the disk
> with others (-> "cluster filesystem")
>
> II- in a second step, we can include directly the nbd protocol in qemu
> (block-nbd.c, "-drive file=nbd:localhost:1024") to connect to the
> server. We can also add some commands to the protocol to manage lock,
> HA, "what else ?" (Hi George).
>   

http://hg.codemonkey.ws/qemu-pq/file/25ca451f2040/block-nbd.diff

Regards,

Anthony Liguori

> Any comments ?
>
> Cheers,
> Laurent
>