[Pgpool-hackers] removing backend_socket_dir parameter ?

[Guillaume Lelarge]

Hi,

Le 04/01/2011 14:27, Jehan-Guillaume (ioguix) de Rorthais a écrit :
> [...]
> pgPool is currently using the parameter "backend_socket_dir" to define where it
> can find the backend unix socket files.
> 
> However, from the libpq world, we just use one parameter for both unix or inet
> socket: host. If this parameter starts with '/', then this is a unix socket. All
> other values are inet connections. See:
> 
> http://www.postgresql.org/docs/9.0/static/libpq-connect.html#LIBPQ-CONNECT-HOST
> 
> Back in pgPool world, the equivalent parameter is "backend_hostnameN". So, when
> I was setting up a dev environment yesterday, I naturally set this parameter  to
> my unix path (I'm using Debian) and end up with a hostname resolution error.
> 
> So you'll find in attachment a patch to remove the "backend_socket_dir"
> parameter and use the libpq policy. Moreover, on empty "backend_hostnameN"
> value, the patch fall back to DEFAULT_SOCKET_DIR.
> 

This patch is really interesting. This is something I had on my todo
list for quite some time.

> I realize it will break compatibility with older configuration file. But in a
> first step, I prefer something clean and start discussing this issue with you if
> you really mind this.
> 

On the configuration file, compatibility of configuration file is never
garantied. So I don't think this is such an issue. We need to make it
clear that the parameter has a new mail.

If we really want to maintain the compatibility on this issue, we could
still support the old parameter during two or three major releases. I
guess we need to put a warning in the log to say "hey guy, you're still
using that old obsolete configuration parameter, you should better
change with this one".

> After discussing it, if you agree to commit it at some point, I'll patch the
> documentation about this behaviour as well.
> 
> Thank you,

You're welcome :)


-- 
Guillaume
 http://www.postgresql.fr
 http://dalibo.com