On 11/26/2016 04:54 PM, Alexandr Porunov wrote:
Hello,
I am using GlusterFS 3.8.5. I wonder to know when to use Arbiter
volumes and quorum options in GlusterFS and what difference between them?
I tried to achive a split-brain in a replicated voulume 3 but
couldn't. Does GlusterFS manage split-brain situations automaticaly?
I tried following:
node1, node2, node3 is the nodes with a replicated volume.
I have created a network crash between node1 and node3 but node1 and
node3 still can speak with node2. When I write a file to node1 it
automatically creates on node2 and node3. When I write a file to node3
it automaticaly creates on node2 and node1.
I tried to crash node1 then write to node3 and then crash a node3 and
recover node1. Still it works just fine. I tried to crash two nodes
and then write to the healthy node and it doesn't work. After I have
run one more node it started work.
Is it possible to achive a split-brain in a GlusterFS cluster 3.8.5?
If yes then how and what options we have to set to prevent
split-brain? If no then why do we need an arbiter and quorum options?
Split-brains are possible in replica-2 volumes. They don't occur in
replica-3 volumes or arbiter-volumes. Client-quorum is enabled and set
to 'auto' when you create replica-3 or arbiter vols. You're not supposed
to change it. Arbiter vols can be used if you want to avoid split-brains
but don't want to invest in a 3rd data brick for replica-3. Of course,
the availability of arbiter is less compared to replica-3 because we
don't store the actual file content in the 3rd (arbiter) brick.
You can find more information @
http://gluster.readthedocs.io/en/latest/Administrator%20Guide/arbiter-volumes-and-quorum/
and
http://events.linuxfoundation.org/sites/events/files/slides/glusterfs-arbiter-VAULT-2016.pdf
Regards,
Ravi
Sincerely,
Alexandr
_______________________________________________
Gluster-users mailing list
[email protected]
http://www.gluster.org/mailman/listinfo/gluster-users
_______________________________________________
Gluster-users mailing list
[email protected]
http://www.gluster.org/mailman/listinfo/gluster-users
