You might unknowingly be asking a FAQ. See the end of this section:
http://www.postgresql.org/docs/faqs.FAQ.html#4.6
Depending on the version of postgres you're running and the data types of the symbol and source columns, you might need to quote (or cast) your constant data for symbol and source, e.g.:
SELECT * FROM article WHERE symbol='12646' AND source = '19' ORDER BY time DESC LIMIT 1000;
-tfo
-- Thomas F. O'Connell Co-Founder, Information Architect Sitening, LLC http://www.sitening.com/ 110 30th Avenue North, Suite 6 Nashville, TN 37203-6320 615-260-0005
On Mar 18, 2005, at 3:56 AM, Alex Stapleton wrote:
Woops sorry we have indexes on (symbol, source, time) and there is no date
column :/
SELECT * FROM article WHERE symbol=12646 AND source = 19 ORDER BY time DESC
LIMIT 1000;
QUERY PLAN
----------------------------------------------------------------------- -----
----------------------------------------
Limit (cost=321163.29..321165.79 rows=1000 width=466) (actual
time=33243.924..33246.021 rows=1000 loops=1)
-> Sort (cost=321163.29..321402.31 rows=95609 width=466) (actual
time=33243.917..33244.626 rows=1000 loops=1)
Sort Key: "time"
-> Seq Scan on article (cost=0.00..301724.00 rows=95609
width=466) (actual time=0.022..32979.685 rows=42959 loops=1)
Filter: ((symbol = 12646) AND (source = 19))
Total runtime: 33258.706 ms
(6 rows)
explain analyze SELECT * FROM article WHERE symbol=12646 AND source = 19;
QUERY PLAN
----------------------------------------------------------------------- -----
----------------------------------------
Seq Scan on article (cost=0.00..301724.00 rows=95609 width=466) (actual
time=0.021..33275.433 rows=42959 loops=1)
Filter: ((symbol = 12646) AND (source = 19))
Total runtime: 33320.920 ms
(3 rows)
We can't use CLUSTER because we need the DB up all the time.
The analyze suggests that it's the I/O taking most of the time to me.
-----Original Message----- From: Alban Hertroys [mailto:[EMAIL PROTECTED] Sent: 18 March 2005 09:48 To: Alex Stapleton Cc: pgsql-general@postgresql.org Subject: Re: [GENERAL] Select Last n Rows Matching an Index Condition (and caches)
Alex Stapleton wrote:SELECT * FROM article WHERE symbol=12646 AND source = 19 ORDER BY timeDESCLIMIT 1000;
To get the latest 1000 rows for that symbol and source.
My (not yet implemented) solution to this problem is to add a SEQUENCE andwith
index it so that by adding a WHERE id > [max_id]-1000 and ordering by time
DESC will reduce the I/O quite a lot. Am I right here? It would be nice if
there was a way to get PostgreSQL to try and precache the tables pages as
well, is there anyway I could achieve something like that? I have toyedcreating a ramdisk to store a lookup table of sorts on (we only care aboutafew columns initially) to speed this up a bit but its a right pain in the
arse to do by the looks of things.
First question that always gets asked here: What's the output of explain
analyse? Without that, people here can't see where the slowdown is.
I expect though, that the problem is the ordering by time. I imagine
that you could create an index on time, maybe truncated to months or
something similar (You can create indices based on functions). That
index alone should speed up the ordering already.
It could also be used to cluster the table, which should speed up things
some more, I suppose.
-- Alban Hertroys MAG Productions
T: +31(0)53 4346874 F: +31(0)53 4346876 E: [EMAIL PROTECTED] W: http://www.magproductions.nl
---------------------------(end of broadcast)---------------------------
TIP 3: if posting/reading through Usenet, please send an appropriate
subscribe-nomail command to [EMAIL PROTECTED] so that your
message can get through to the mailing list cleanly
---------------------------(end of broadcast)--------------------------- TIP 1: subscribe and unsubscribe commands go to [EMAIL PROTECTED]