Re: function optimization ???

From: Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us>
To: Brent Verner <brent(at)rcfile(dot)org>
Cc: pgsql-hackers(at)postgresql(dot)org
Subject: Re: function optimization ???
Date: 2001-01-24 17:14:20
Message-ID: 13152.980356460@sss.pgh.pa.us
Views: Raw Message | Whole Thread | Download mbox | Resend email
Thread:
Lists: pgsql-hackers

Brent Verner <brent(at)rcfile(dot)org> writes:
> calling it as:
> SELECT p.*, p.book_info.title FROM pricing p WHERE vista_ans='POD';
> background and observation:
> the pricing table is fairly large, but only a small number meet
> "WHERE vista_ans='POD'". I can select all where vista_ans='POD'
> very quickly (.2 sec), but adding in the get_book(pricing) call
> slows this down to about 20sec. I can, with an external sql query,
> select all of the desired records in about 1 sec, so it appears
> to me that the function is being called regardless of whether
> or not the WHERE clause is being satisfied.

This conclusion is absolutely false: the SELECT target list is NOT
evaluated except at rows where the WHERE condition is satisfied.

I suspect the real problem is that the select inside the function
is not being done as efficiently as you'd like. How big is
catalog_general, and would a sequential scan over it inside the
function account for the performance discrepancy?

IIRC, 7.0.* is not very bright about using indexscans in situations
where the righthand side of the WHERE clause is anything more complex
than a literal constant or simple parameter reference ($n). The
fieldselect you have here would be enough to defeat the indexscan
recognizer. This is fixed in 7.1, however. For now, you could
declare book_info as taking a simple datum and invoke it as
p.vista_isbn.book_info.title

BTW, star_isbn and vista_isbn are the same datatype, I trust, else
that might cause failure to use an indexscan too.

regards, tom lane

In response to

Responses

Browse pgsql-hackers by date

  From Date Subject
Next Message Peter Eisentraut 2001-01-24 18:03:32 Re: LEFT JOIN in pg_dumpall is a bug
Previous Message Bruce Momjian 2001-01-24 17:01:17 Re: Open 7.1 items