Re: Join push-down support for foreign tables

On Thu, Sep  4, 2014 at 08:41:43PM +0530, Atri Sharma wrote:
>
>
> On Thursday, September 4, 2014, Bruce Momjian <bruce@momjian.us> wrote:
>
>     On Thu, Sep  4, 2014 at 08:37:08AM -0400, Robert Haas wrote:
>     > The main problem I see here is that accurate costing may require a
>     > round-trip to the remote server.  If there is only one path that is
>     > probably OK; the cost of asking the question will usually be more than
>     > paid for by hearing that the pushed-down join clobbers the other
>     > possible methods of executing the query.  But if there are many paths,
>     > for example because there are multiple sets of useful pathkeys, it
>     > might start to get a bit expensive.
>     >
>     > Probably both the initial cost and final cost calculations should be
>     > delegated to the FDW, but maybe within postgres_fdw, the initial cost
>     > should do only the work that can be done without contacting the remote
>     > server; then, let the final cost step do that if appropriate.  But I'm
>     > not entirely sure what is best here.
>
>     I am thinking eventually we will need to cache the foreign server
>     statistics on the local server.
>
>
>
>
> Wouldn't that lead to issues where the statistics get outdated and we have to
> anyways query the foreign server before planning any joins? Or are you thinking
> of dropping the foreign table statistics once the foreign join is complete?

I am thinking we would eventually have to cache the statistics, then get
some kind of invalidation message from the foreign server.  I am also
thinking that cache would have to be global across all backends, I guess
similar to our invalidation cache.