On 10/25/2012 12:01 PM, Nux! wrote:> I'm planning to deploy a largish Glusterfs replicated-distributed
> setup. I'm talking large servers with 6-12 hotswap HDD bays. My idea is
> to use each HDD as a brick as to maximise I/O. The replica number at
> this stage would be 3 probably.
> The one issue that's bothering me is - since one server will hold many
> HDDs, how can I instruct Gluster not to keep all replicas in bricks
> belonging to the same server?
GlusterFS is sensitive to the order in which you specify the bricks for a
replicated setup. For example:
gluster volume create x replica 2 abc def ghi jkl
In this case, abc and def will be combined into a replica set. Ditto for ghi
and jkl. Finally, those two replica sets will be combined into a single
volume. It's up to you, when you construct the "create" line, to
make sure
that the adjacently-listed members of a replica set are on different servers.
There is actually code in there to check for the bad cases and warn you, but we
won't try to fix it for you.