Re: [Gluster-devel] AFR Replication

[Daniel Maher]

On Fri, 18 Apr 2008 10:34:18 -0400 "Christopher Hawkins"
<address@hidden> wrote:

> The big issue right now seems to be the inability of Unify to
> gracefully handle the failure of one of its subvolume servers. So in
> its current state, it seems that AFR works for mirrors, Unify works
> for aggregation, and when you put them together you get both but it
> becomes inoperable if a Unify subvolume fails.

Of course, if you let the servers handle the AFR, then there's no need
for Unify, as the AFR'd volumes can be exported directly to the
clients.  If combined with RRDNS, this creates a fault-tolerant
mirrored cluster, as the clients will just keep trying until they get
a working server, and the servers will replicate among themselves
until the missing node(s) return(s).


-- 
Daniel Maher <dma AT witbe.net>