Re: postgres_fdw join pushdown (was Re: Custom/Foreign-Join-APIs)

On 2016/02/04 21:42, Ashutosh Bapat wrote:

>     * Is it safe to replace outerjoinpath with its fdw_outerpath the
>     following way?  I think that if the join relation represented by
>     outerjoinpath has local conditions that can't be executed remotely,
>     we have to keep outerjoinpath in the path tree; we will otherwise
>     fail to execute the local conditions.  No?
>
>     +                       /*
>     +                        * If either inner or outer path is a
>     ForeignPath corresponding to
>     +                        * a pushed down join, replace it with the
>     fdw_outerpath, so that we
>     +                        * maintain path for EPQ checks built
>     entirely of local join
>     +                        * strategies.
>     +                        */
>     +                       if (IsA(joinpath->outerjoinpath, ForeignPath))
>     +                       {
>     +                               ForeignPath *foreign_path;
>     +                               foreign_path = (ForeignPath
>     *)joinpath->outerjoinpath;
>     +                               if
>     (foreign_path->path.parent->reloptkind == RELOPT_JOINREL)
>     +                                       joinpath->outerjoinpath =
>     foreign_path->fdw_outerpath;
>     +                       }

> all the conditions (local and remote) should be part of fdw_outerpath as
> well, since that's the alternate local path, which should produce (when
> converted to the plan) the same result as the foreign path.
> fdw_outerpath should be a local path set when paths for
> outerjoinpath->parent was being created. Am I missing something?

I assumed by mistake that only the remote conditions were evaluated in a
plan created from each fdw_outerpath.  Sorry for that.  I think that is
a good idea!

Btw, IIUC, I think the patch fails to adjust the targetlist of the top
plan created that way, to output the fdw_scan_tlist, as discussed in [1]
(ie, I think the attached patch is needed, which is created on top of
your patch pg_fdw_join_v8.patch).

Best regards,
Etsuro Fujita

[1]
http://www.postgresql.org/message-id/CA+TgmobA4MSKgquicgt5CkbpQJ-TmpqEfHt_wy49ndwa91Wkpw@mail.gmail.com