[ovirt-users] why replica 3

[Jorick Astrego]

On 09/09/2015 10:14 AM, Richard Neuboeck wrote:
> On 04.09.15 10:02, Simone Tiraboschi wrote:
>>     Is there a reason why it has to be exactly replica 3?
>>
>>
>> To have a valid quorum having the system being able to decide witch is
>> the right and safe copy avoiding an issue called split brain.
>> Under certain circumstances/issues (network issue, hosts down or
>> whatever could happen) the data on different replica could diverge: if
>> you have two and just two different hosts that claim each other
> that its
>> copy is the right one there is no way to automatically take the right
>> decision. Having three hosts and setting the quorum according to that
>> solves/mitigates the issue.
>
> Thanks for the explanation. I do understand the problem but since
> I'm somewhat limited in my hardware options is there a way to
> override this requirement? Meaning if I change the checks for
> replica 3 in the installation scripts does something else fail on
> the way?
>
> In my case coherence checks would come from outside the storage and
> vm host setup and fencing would be applied appropriately.
>
> I would very much appreciate it if the particulars of the storage
> setup could be either selected from a list of possibilities or be
> ignored and just a warning be issued that this setup is not recommended.
>
> Thanks!
> Richard
>
>
As a side question, in Glusterfs 3.7 there is an "AFR arbiter volume"
(http://gluster.readthedocs.org/en/release-3.7.0-1/Features/afr-arbiter-volumes/)
that will only contain the metadata.

Will ovirt support this in 3.6?

Kind regards,

Jorick Astrego




Met vriendelijke groet, With kind regards,

Jorick Astrego

Netbulae Virtualization Experts 

----------------

	Tel: 053 20 30 270 	info at netbulae.eu 	Staalsteden 4-3A 	KvK 08198180
 	Fax: 053 20 30 271 	www.netbulae.eu 	7547 TA Enschede 	BTW NL821234584B01

----------------

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.ovirt.org/pipermail/users/attachments/20150909/354c251b/attachment.html>