Re: [HACKERS] Push down more UPDATEs/DELETEs in postgres_fdw

On 2017/02/13 18:24, Rushabh Lathia wrote:

I started reviewing the patch again. Patch applied cleanly on latest source
as well as regression pass through with the patch. I also performed
few manual testing and haven't found any regression. Patch look
much cleaner the earlier version, and don't have any major concern as
such.

Thanks for the review!

Here are few comments:

1)

@@ -211,6 +211,12 @@ typedef struct PgFdwDirectModifyState
     PGresult   *result;            /* result for query */
     int            num_tuples;        /* # of result tuples */
     int            next_tuple;        /* index of next one to return */
+    Relation    resultRel;        /* relcache entry for the target table */

Why we need resultRel? Can't we directly use dmstate->rel ?

The reason why we need that is because in get_returning_data, we pass dmstate->rel to make_tuple_from_result_row, which requires that dmstate->rel be NULL when the scan tuple is described by fdw_scan_tlist.  So in that case we set dmstate->rel to NULL and have dmstate->resultRel that is the relcache entry for the target relation in postgresBeginDirectModify.