Hi all,
Thank you for the updated v2 patch. I reproduced the issue using the
provided test case. With the unpatched code, the planner chose a local
Hash Semi Join, followed by Sort and Limit, instead of pushing down
the semi-join. After applying the patch, the planner selected a single
Foreign Scan, with the semi-join, filters, ORDER BY, LIMIT, and OFFSET
pushed down to the remote server as intended. I also ran the
postgres_fdw regression tests, including postgres_fdw and
query_cancel, along with the eval_plan_qual isolation test. All three
tests passed. The change looks good to me.
Thank you for addressing this issue.
Regards,
Solai
