Horton

Re: postgres_fdw: batch inserts vs. before row triggers

Etsuro Fujita <etsuro.fujita@gmail.com>

View thread View on postgr.es
From: Etsuro Fujita <etsuro.fujita@gmail.com>
To: Tom Lane <tgl@sss.pgh.pa.us>
Cc: Tomas Vondra <tomas.vondra@enterprisedb.com>, PostgreSQL Hackers <pgsql-hackers@lists.postgresql.org>
Date: 2022-12-02T07:54:34Z
Lists: pgsql-hackers

Attachments

On Sun, Nov 27, 2022 at 12:11 AM Tom Lane <tgl@sss.pgh.pa.us> wrote:
> Etsuro Fujita <etsuro.fujita@gmail.com> writes:
> > On Sat, Nov 26, 2022 at 1:57 AM Tom Lane <tgl@sss.pgh.pa.us> wrote:
> >> Couldn't we add the field to ModifyTableState, instead?
>
> > We could probably do so, but I thought having a global list would be
> > more efficient to handle pending buffered inserts than that.
>
> OK, as long as there's a reason for doing it that way, it's OK
> by me.  I don't think that adding a field at the end of EState
> is an ABI problem.
>
> We have to do something else than add to ResultRelInfo, though.

OK, I removed from ResultRelInfo a field that I added in the commit to
save the owning ModifyTableState if insert-pending, and added to
EState another List member to save such ModifyTableStates, instead.  I
am planning to apply this to not only back branches but HEAD, to make
back-patching easy, if there are no objections.

Best regards,
Etsuro Fujita

Commits

  1. Remove new structure member from ResultRelInfo.

  2. Fix handling of pending inserts in nodeModifyTable.c.

  3. postgres_fdw: Disable batch insertion when there are WCO constraints.

  4. postgres_fdw: Disable batch insert when BEFORE ROW INSERT triggers exist.

  5. Implement support for bulk inserts in postgres_fdw