[Gluster-devel] Client side afr versus server side, doing a self-heal

[Amar S. Tumballi]

On Thu, May 1, 2008 at 12:43 AM, Brandon Lamb <brandonlamb at gmail.com> wrote:

> Basically yea. So with server side afr server1 would send a copy to
> server2.
>
> With client side, client1 copies from server1 to itself, then copies
> from itself to server2, correct?
>
>
You have made a very good point. Generally when we setup a storage, or
recommend it, we consider a fresh setup. We recommend client side afr (or
any other clustering translators) because it got its own benefits like, open
fds will remain intact even if the subvolume goes down.

In the scenario like you described, I would say, server side afr helps a
lot. But if I am setting up GlusterFS, I will rather use rsync or scp to
copy the data to server2 from server1 directly, and then start afr on client
node.  I know GlusterFS should work fine for healing even that, but
considering opening each file to heal it through GlusterFS, its the same or
less effort to start with symmetric export points.

Krishna,
 How does it handle two files without any attributes but have same data?

Regards,
Amar


-- 
Amar Tumballi
Gluster/GlusterFS Hacker
[bulde on #gluster/irc.gnu.org]
http://www.zresearch.com - Commoditizing Super Storage!