Re: FDW-based dblink (WIP)

On Wed, 2009-08-19 at 17:07 +0900, Itagaki Takahiro wrote:
> Here is a WIP patch for a foreign data wrapper based dblink.
>
> It integrates dblink module into core and adds a new functionality,
> automatic transaction management. The new interface of dblink is
> exported by include/foreign/dblink.h. We can easily write a connector
> module for another database because we can reuse transaction and
> resource management parts in core.

This patch is listed in the commitfest, but I think the consensus was
that it needed some rework. I think the idea is that we will have
support for syntax like

> Syntax to create FDW with connector is below:
> CREATE FOREIGN DATA WRAPPER postgresql
> VALIDATOR postgresql_fdw_validator
> CONNECTOR postgresql_fdw_connector
> OPTIONS (...);

in core, but the actual implementation of postgresql_fdw_connector would
be a loadable module.

Personally, I'm undecided whether the single-function connector
implementation is the best. The other approach would be to use a
multiple-function interface based directly on the functions currently
provided by dblink.

More generally, what does this really buy us? It doesn't advance the
SQL/MED implementation, because you are not adding, say, some kind of
CREATE FOREIGN TABLE support. You are just changing the dblink
implementation to go through the FDW. I would argue that it should be
the other way around: The FDW should go through dblink.