Re: pgbench: using prepared BEGIN statement in a pipeline could cause an error

On Fri, 25 Mar 2022 16:19:54 -0400
Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> wrote:

> Fabien COELHO <coelho(at)cri(dot)ensmp(dot)fr> writes:
> >> [...] One way to avoid these errors is to send Parse messages before
> >> pipeline mode starts. I attached a patch to fix to prepare commands at
> >> starting of a script instead of at the first execution of the command.
>
> > ISTM that moving prepare out of command execution is a good idea, so I'm
> > in favor of this approach: the code is simpler and cleaner.
> > ISTM that a minor impact is that the preparation is not counted in the
> > command performance statistics. I do not think that it is a problem, even
> > if it would change detailed results under -C -r -M prepared.
>
> I am not convinced this is a great idea. The current behavior is that
> a statement is not prepared until it's about to be executed, and I think
> we chose that deliberately to avoid semantic differences between prepared
> and not-prepared mode. For example, if a script looks like
>
> CREATE FUNCTION foo(...) ...;
> SELECT foo(...);
> DROP FUNCTION foo;
>
> trying to prepare the SELECT in advance would lead to failure.
>
> We could perhaps get away with preparing the commands within a pipeline
> just before we start to execute the pipeline, but it looks to me like
> this patch tries to prepare the entire script in advance.
>
Well, the semantic differences is already in the current behavior.
Currently, pgbench fails to execute the above script in prepared mode
because it prepares the entire script in advance just before the first
command execution. This patch does not change the semantic.

> BTW, the cfbot says the patch is failing to apply anyway ...
> I think it was sideswiped by 4a39f87ac.

I attached the rebased patch.

Regards,
Yugo Nagata

--
Yugo NAGATA <nagata(at)sraoss(dot)co(dot)jp>