| From: | Pierre Forstmann <pierre(dot)forstmann(at)gmail(dot)com> |
|---|---|
| To: | Jeevan Chalke <jeevan(dot)chalke(at)enterprisedb(dot)com> |
| Cc: | pgsql-hackers(at)lists(dot)postgresql(dot)org |
| Subject: | Re: PL/pgSQL: allow parallel query for assignment statements |
| Date: | 2026-11-07 14:10:45 |
| Message-ID: | b51009fe-8ca0-418d-a14b-46cd82e1cb1d@gmail.com |
| Views: | Whole Thread | Raw Message | Download mbox | Resend email |
| Thread: | |
| Lists: | pgsql-hackers |
Hello,
This patch looks good to me.
I have applied it on latest 20-devel master branch and I have run make
installcheck-world successfully.
I have set status to Ready For Committed.
Regards,
PF.
Le 26/10/2026 à 09:58, Jeevan Chalke a écrit :
> Hi,
>
> Commit 61c2e1a95f ("Improve access to parallel query from procedural
> languages") made exec_eval_expr() prepare expressions with
> CURSOR_OPT_PARALLEL_OK, so that queries in IF, RETURN, RAISE, etc. can use
> parallel plans. It missed exec_assign_expr(), which creates the plan
> itself,
> with cursorOptions 0, before calling exec_eval_expr(). As the plan already
> exists by then, the PARALLEL_OK prepare in exec_eval_expr() never runs.
>
> As a result, a query used in an assignment never gets a parallel plan:
> x := sum(a) from tab;
> while the same query used in, say, an IF condition can. The same
> applies to
> DECLARE default expressions, which also go through exec_assign_expr().
>
> The pre-planning in exec_assign_expr() is redundant, since
> exec_eval_expr()
> creates the plan if needed, so the attached patch just removes it.
>
> The patch also adds regression tests to select_parallel.sql, one for an
> assignment statement and one for a DECLARE default. Each provokes an
> error
> in a worker with debug_parallel_query = 1 and checks for the "parallel
> worker" context line. Both tests fail without the fix.
>
> Thoughts?
>
> Thanks
>
> --
> *Jeevan Chalke*
> /Senior Principal Engineer, Engineering Manager//
> Product Development/
> enterprisedb.com <https://www.enterprisedb.com>
| From | Date | Subject | |
|---|---|---|---|
| Next Message | Ashutosh Bapat | 2026-11-07 14:31:07 | Re: Call CalculateShmemSize() only once |
| Previous Message | Jeevan Chalke | 2026-11-07 13:53:00 | Re: postgres_fdw: virtual generated columns vs. IMPORT FOREIGN SCHEMA |