[Gluster-users] Docs on setting up translators

[Jeff Darcy]

On 08/01/2013 11:39 AM, Joel Young wrote:
> Sorry, no. This doesn't help.  It shows text entries for some configuration
> file and doesn't name which configuration file these text entries go in.  It
> doesn't say if setting these options with "gluster volume set" is
> persistent.  There is so much context missing that it is unusable for a new
> admin.

Options set via the Gluster CLI are both global (for the volume) and
persistent.  You shouldn't need to know which actual file(s) the changes go
into, because the fetching and use of those files is automatic as part of
volume startup (on the server side) or mounting (on the client side).  If you
really do want to know, you can look at /var/lib/glusterd/vols/$myvol/*.vol on
any server but you should never change anything you find there (it will be
overwritten).

> For example it says "The IO-Cache translator is useful on both the client
> and server sides of a connection."  Great!  How?  How do I specify that I
> want it on the client side? How do I specify which client?  How do I check
> to see if the change has "taken"?

Yes, that particular piece of documentation seems rather useless.  While
translators *can* be loaded and even be beneficial on either side, the CLI
tends to treat them as server-only or client-only.  The reason is simply that
if we supported every possible ordering and placement of translators we'd never
be able to finish testing.  In this case, io-cache is loaded only on the client
side and the doc should say so.

> Is the change persistent across restarts?  What configuration files are
> modified (hopefully in /etc) so that I can make sure they are apprpriately
> backed up?  It even gives a sample piece of configuration file but doesn't
> say where it is supposed to go if I want it for the server or for the
> client.  Do I need to do this on each server, one one server node and it is
> propogated?  On one client?  On all clients?
>
> Sorry if I am obtuse.  I really feel like there is some statement of "
> administrative philosophy that is missing.

The philosophy is that other distributed filesystems (including GlusterFS 2.x)
have done a more than adequate job demonstrating that configuration of such a
system by manual hacking of config files on multiple machines doesn't scale and
will drive users crazy.  Therefore, we assume responsibility for distribution,
durability, and consistency of configuration info.  If you use the CLI, the
"right" config files will be updated automatically, and since there are
multiple machines holding copies of that information each can effectively be a
backup for the others.  I know nobody ever believes that "you don't need to 
worry about that" but we do what we can to make it as true as possible.