[ovirt-users] Ha cluster 3 nodes but 1 slow

[Ravishankar N]

On 10/18/2015 07:27 PM, Nicolas LIENARD wrote:
> Hey Nil
>
> What about 
> https://gluster.readthedocs.org/en/release-3.7.0/Features/afr-arbiter-volumes/ 
> ?
>
> Regards
> Nico
>
>
> Le 18 octobre 2015 15:12:23 GMT+02:00, Nir Soffer <nsoffer at redhat.com> 
> a écrit :
>
>     On Sat, Oct 17, 2015 at 12:45 PM, Nicolas LIENARD <gluster at distran.org> wrote:
>
>         Hi Currently, i ve 3 nodes, 2 in same DC and a third in
>         another DC. They are all bridged together through a vpn. I
>         know a cluster is at least 3 nodes to satisfy the quorum. 
>


Just adding a 3rd node (without actually using it for 3 way replication) 
might not help in preventing split-brains. gluster has client-quorum and 
server-quorum. Have a look at 
http://comments.gmane.org/gmane.comp.file-systems.gluster.user/22609 for 
some information.

If you are indeed using it as a replica-3, then it is better to have all 
3 nodes in the same DC. gluster clients sends every write() to all 
bricks of the replica (and waits for their responses too), so if one of 
them is in another DC, it might slow writes due to network latency.
>
>         My question is to know if i can have my VM balancing on the 2
>         fast nodes with HA and glusterfs replica 2. 
>


replica 2 definitely provides HA but if you have more chances of files 
ending in split-brain if you have frequent network disconnects , which 
is why replica 3 with client-quorum set to 'auto' is better for 
preventing split-brains.
arbiter-volumes are a kind of a sweet-spot between replica-2 and 
replica-3 that can prevent split-brains. The link that Nir shared 
describes it and how to create one etc.

Regards,
Ravi
>
>     gluster replica 2 is not supported.
>
>     Nir
>
>
>
> _______________________________________________
> Users mailing list
> Users at ovirt.org
> http://lists.ovirt.org/mailman/listinfo/users

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.ovirt.org/pipermail/users/attachments/20151019/af3a47e7/attachment-0001.html>