[Gluster-devel] volume stanzas for non-existent directories in server volume specification

[John Marshall]

Brandon Lamb wrote:
> Basically when glusterfs parses the spec file there would have to be a
> way for it to know that a volume specification was for *itself* and if
> not (ie it was defining a volume for another server) it would ignore
> the volume.
>   
Right. That's why I asked about an "option hosts".
> Pros and cons to this I guess, a pro is definately a single config to
> edit and copy for all servers. Howerver, I believe this would lead to
> a possibly huge and confusing (hence more room for error)
> configuration file as you would have ALL definitions for all servers
> in the file rather than *only* what it needs.
>   

For me, this actually brings up the question of the usefulness of having
two spec files: server and client. It may simplify some things, but, from
what I can tell, the server part is usually quite small. Wrt to the "what
host do I contact", this information is actually in the client spec so an
integration of the two could be useful (at least in the way I am seeing
things).

Amar Tumballi wrote:
    if you want to do that, you can use the server spec file you sent in the
    first mail , on all the servers. It will create those directories, 
but the
    volume is not used at all (if not linked by the client protocol). Its
    possible with current code base itself.

That is helpful. Thanks.

Thanks,
John