[Gluster-users] Avoid Split-brain and other stuff

Brian Candler B.Candler at pobox.com
Thu Nov 15 17:20:14 UTC 2012


On Thu, Nov 15, 2012 at 03:12:09PM +0100, Martin Emrich wrote:
> Hi!
> 
> > Apologies if this is a dumb comment, but isn't master-slave, by definition, only in one direction? If you could write on all nodes, it would be multi-master, yes?
>  
> I would say it depends on the definition. Master-slave is surely asymmetrical, as the replication can only be triggered by the master. But the data flow could be bidirectional anyways...

It's unidirectional: writes go to master, writes replicate from master to
slave.  I believe it's basically rsync on steroids (with some help from
xattrs to track changed files)

Any multi-master replication suffers from exactly the same split-brain
scenarios as you described earlier.

In glusterfs, geo-replication is what you should use for WAN-separated
sites.  Replicated volumes are for LAN scenarios where partitioning (and
hence split-brain) should not be expected to occur.

If this doesn't meet your requirements then glusterfs is not the right tool
for you.

Regards,

Brian.



More information about the Gluster-users mailing list