----- "Elbert Lai" <elbert.lai at eng.admob.com> wrote: > It seems to me that this must be inherent in the concept of a split > namenode, namely that the potential for splitbrain exists. Is this no > > longer the case with gluster v2? Or if one host becomes unreachable > for long enough, then this type of scenario is unavoidable? A split-brain situation is inherent to any sort of distributed system. If you have two servers A & B, and for a time A is up, B is down, and later B comes up but A goes down, then the servers will obviously have different data and this will be detected as a split-brain. > It looks like to upgrade, there is a post-install script that must be > run on the exported volumes to clean up extended attributes. I haven't > spent much time look at any upgrade docs yet (which I fully intend to > do), but are there any big gotchas? The gluster servers house our > production data storage, about 1.7TB on the average. You just need to stop GlusterFS 1.3, run the script backend-xattr-sanitize.sh (found in the extras/ directory of the source tarball) on all your backend directories, and start GlusterFS 2.0. Vikas -- Engineer - Z Research http://gluster.com/
