[Gluster-devel] Re: AFR + unify + namespace question

[Brandon Lamb]

On Sat, May 3, 2008 at 2:56 PM, Brandon Lamb <brandonlamb at gmail.com> wrote:
>
> On Sat, May 3, 2008 at 2:50 PM, Brandon Lamb <brandonlamb at gmail.com> wrote:
> > I couldnt find anything related to this with a quick search.
> >
> > Lets take a case where we have 3 data servers, so we do AFR using
> > unify. For unify we specify a namespace volume that exists on lets say
> > server1.
> >
> > Ok so what happens when server1 goes down, now there is no namespace.
> > What are the implications and is there a common solution to avoid
> > this?
>
> http://gluster.org/docs/index.php/Aggregating_Three_Storage_Servers_with_Unify
>
> Ok so I started to attempt to create a 3 server 2 client setup. I
> thought I would use 3 servers and i would only want 2 copies of files.
> I thought to do this you needed afr + unify.
>
> However, AFR uses the number of servers as the number of copies, so
> even using 3 servers I dont see a reason to use unify? Wont it just
> send every write to all 3 servers anyway? So what is unify doing in
> the example?
>
> Is that just a bad example that I am trying to follow?

http://gluster.org/docs/index.php/Understanding_Unify_Translator

I just finished reading the very bottom of this page. So what I got
from that page is that if the server which holds the namespace goes
down, something bad will happen?

So what is the "recommended" solution for this? Have a seperate
machine that does nothing export only a name space directory? At least
until 1.4 and having the addition of distributed namespace?