[Gluster-devel] Unify translator with AFR ? How to implement NS in AFR without Unify ?

[Daniel Maher]

On Wed, 9 Apr 2008 17:22:35 +0530 "Krishna Srinivas"
<krishna at zresearch.com> wrote:

> you don't need unify/namespace for your setup here, you can
> directly export gfs-ds-afr from the server which can be mounted
> on the client. Can you paste your client spec? if you are connecting

Hello again :)

As you note above, using "unify/namespace" is not necessary, nor
recommended for a two-node HA cluster using AFR.  During failover
tests, using the Unify translator caused the mountpoint to become
completely inaccessible when one of the nodes failed - removing said
translator solved the problem.

You also noted (in the same thread) that the following wiki page is
wrong :
http://www.gluster.org/docs/index.php/GlusterFS_1.3_High_Availability_Storage_with_GlusterFS
As it indicates the usage of Unify in this context.

I then noted that the AFR page *itself* suggests using Unify :
http://www.gluster.org/docs/index.php/Understanding_AFR_Translator
In this case, it is done in order to link the namespace to the remote
volumes.

Are we or are we not supposed to use the Unify translator in
conjunction with AFR ?

If we are, why does it cause the mountpoint to fail if one of the
AFR'd nodes fails, and how can this be solved ?

If we are not, what is the recommended way to implement a namespace
with AFR ?

For reference purposes, the relavent threads are :
http://lists.nongnu.org/archive/html/gluster-devel/2008-04/msg00074.html
http://lists.nongnu.org/archive/html/gluster-devel/2008-04/msg00078.html


-- 
Daniel Maher <dma AT witbe.net>