Gluster users - Feb 2017 - chances of split brain with a distributed replicated volume where replica is 3

All:

I read this in the gluster docs. Note I am not using arbiter -- I am
setting up volumes with full 3 replicas. In this case, is this split brain
scenario theoretical or has this actually occurred? If so, what are the
chances that this could happen? In other words, aside from doing regular
snapshots, is this type of split brain scenario something I should be
planning in an unlikely disaster recovery scenario *or *part of daily
maintenance? Since the docs say its a corner case, I am inferring that this
is pretty unlikely.

"There is a corner case even with replica 3 volumes where the file can end
up in a split-brain. AFR usually takes range locks for the {offset, length}
of the write. If 3 writes happen on the same file at non-overlapping
{offset, length} and each write fails on (only) one different brick, then
we have AFR xattrs of the file blaming each other."

https://gluster.readthedocs.io/en/latest/Administrator%20Guide/arbiter-volumes-and-quorum/#replica-2-and-replica-3-volumes

Thanks,
Joe
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.gluster.org/pipermail/gluster-users/attachments/20170209/96c083cb/attachment.html>

On 02/09/2017 06:39 PM, Joseph Lorenzini wrote:
> All:
>
> I read this in the gluster docs. Note I am not using arbiter -- I am 
> setting up volumes with full 3 replicas. In this case, is this split 
> brain scenario theoretical or has this actually occurred? If so, what 
> are the chances that this could happen? In other words, aside from 
> doing regular snapshots, is this type of split brain scenario 
> something I should be planning in an unlikely disaster recovery 
> scenario *or *part of daily maintenance? Since the docs say its a 
> corner case, I am inferring that this is pretty unlikely.
>
> "There is a corner case even with replica 3 volumes where the file can
> end up in a split-brain. AFR usually takes range locks for the 
> {offset, length} of the write. If 3 writes happen on the same file at 
> non-overlapping {offset, length} and each write fails on (only) one 
> different brick, then we have AFR xattrs of the file blaming each
other."
>
>
https://gluster.readthedocs.io/en/latest/Administrator%20Guide/arbiter-volumes-and-quorum/#replica-2-and-replica-3-volumes
I don't think you should be bothered about it too much as it is  rare 
for split-brains to happen in replica 3 (folks who have had experience 
otherwise, please correct me:)). In the worst case, there are commands 
to resolve split-brains as described in the link you shared above.

Thanks,
Ravi
>
> Thanks,
> Joe
>
>
> _______________________________________________
> Gluster-users mailing list
> Gluster-users at gluster.org
> http://lists.gluster.org/mailman/listinfo/gluster-users
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.gluster.org/pipermail/gluster-users/attachments/20170209/cee5ad3a/attachment.html>