Thread: REPEATED INSERT INTO ...
I've noticed subsequent executions of the same insert command are slow. I've searched the list archives for this matter and found several entries related, including suggestions how to speed up. The standard answer from the core team is, use COPY. Sorry, but this is from an application point of view not really an option if you're dealing with program variables. We used to have an INGRES installation around, and since I know Postgres is based on it, may be this old INGRES feature is worth to consider: Taken from the Reference Manual [REPEATED] INSERT INTO ... The key word REPEATED directs INGRES to encode the INSERT and save its execution plan when it is first executed. This encoding can account for significant performance improvements on subsequent executions of the same INSERT. What do you others think of it? Regards, Christoph
Christoph Haller <ch@rodos.fzk.de> writes:
> Taken from the Reference Manual
> [REPEATED] INSERT INTO ...
> The key word REPEATED directs INGRES to encode the INSERT and save its
> execution plan when it is first executed. This encoding can account for
> significant performance improvements on subsequent executions of the
> same INSERT.
> What do you others think of it?
You can do that today with PREPARE/EXECUTE; there's no need to invent
specialized syntax for it.
regards, tom lane
On Mon, 2003-02-24 at 07:22, Christoph Haller wrote: > I've noticed subsequent executions of the same insert command are slow. > I've searched the list archives for this matter and found several > entries > related, including suggestions how to speed up. > The standard answer from the core team is, use COPY. > Sorry, but this is from an application point of view not really an > option > if you're dealing with program variables. What do you mean by "program variables"? If you're just referring to variables in the programming language the DB client is written in, I see no reason why you couldn't use COPY to input those. > The key word REPEATED directs INGRES to encode the INSERT and save its > execution plan when it is first executed. This encoding can account for > significant performance improvements on subsequent executions of the > same INSERT. As Tom mentioned, we already support PREPARE/EXECUTE for this functionality -- but I'd suspect that for most INSERT queries, the parsing/rewriting/planning stages are relatively inexpensive, and the bulk of the runtime is spent in the executor (and writing WAL records, etc). So my guess would be that you won't see that much of a performance improvement from this... Cheers, Neil -- Neil Conway <neilc@samurai.com> || PGP Key ID: DB3C29FC
> > > The key word REPEATED directs INGRES to encode the INSERT and save its > > execution plan when it is first executed. This encoding can account for > > significant performance improvements on subsequent executions of the > > same INSERT. > > > What do you others think of it? > > You can do that today with PREPARE/EXECUTE; there's no need to invent > specialized syntax for it. > Excellent. As often, a closer look into the documentation would have saved Tom's time. Nevertheless, I think the PREPARE/EXECUTE functionality could definitely use some more promotion, especially under Performance Tips. Thanks for your quick reply. Regards, Christoph