X-Original-To: pgsql-performance-postgresql.org@localhost.postgresql.org Received: from localhost (av.hub.org [200.46.204.144]) by svr1.postgresql.org (Postfix) with ESMTP id 41961D7318 for ; Thu, 22 Sep 2005 20:07:52 -0300 (ADT) Received: from svr1.postgresql.org ([200.46.204.71]) by localhost (av.hub.org [200.46.204.144]) (amavisd-new, port 10024) with ESMTP id 45703-05 for ; Thu, 22 Sep 2005 23:07:37 +0000 (GMT) Received: from BAYC1-PASMTP04.bayc1.hotmail.com (bayc1-pasmtp04.bayc1.hotmail.com [65.54.191.164]) by svr1.postgresql.org (Postfix) with ESMTP id 4433BD7204 for ; Thu, 22 Sep 2005 20:07:38 -0300 (ADT) Message-ID: X-Originating-IP: [70.50.100.93] X-Originating-Email: [pelletier_32@sympatico.ca] Received: from JP ([70.50.100.93]) by BAYC1-PASMTP04.bayc1.hotmail.com over TLS secured channel with Microsoft SMTPSVC(6.0.3790.1830); Thu, 22 Sep 2005 16:07:09 -0700 Message-ID: <021501c5bfca$6fe82130$6501a8c0@JP> From: "Jean-Pierre Pelletier" To: "Michael Fuhr" , "Josh Berkus" Cc: , "John Arbash Meinel" References: <433326A6.8010708@arbash-meinel.com> <200509221519.05555.josh@agliodbs.com> <20050922225432.GA25103@winnie.fuhr.org> Subject: Re: Queries 15 times slower on 8.1 beta 2 than on 8.0 Date: Thu, 22 Sep 2005 19:07:41 -0400 MIME-Version: 1.0 Content-Type: text/plain; format=flowed; charset="iso-8859-1"; reply-type=original Content-Transfer-Encoding: 7bit X-Priority: 3 X-MSMail-Priority: Normal X-Mailer: Microsoft Outlook Express 6.00.2900.2670 X-MimeOLE: Produced By Microsoft MimeOLE V6.00.2900.2670 X-OriginalArrivalTime: 22 Sep 2005 23:07:09.0288 (UTC) FILETIME=[5BB5FE80:01C5BFCA] X-Virus-Scanned: by amavisd-new at hub.org X-Spam-Status: No, hits=0.519 required=5 tests=[AWL=0.003, FROM_ENDS_IN_NUMS=0.516, MSGID_FROM_MTA_HEADER=0] X-Spam-Level: X-Archive-Number: 200509/260 X-Sequence-Number: 14568 I don't know if it makes a difference but in my tables, content.supplierid and content.priceid were nullable. ----- Original Message ----- From: "Michael Fuhr" To: "Josh Berkus" Cc: ; "Jean-Pierre Pelletier" ; "John Arbash Meinel" Sent: Thursday, September 22, 2005 6:54 PM Subject: Re: [PERFORM] Queries 15 times slower on 8.1 beta 2 than on 8.0 > On Thu, Sep 22, 2005 at 03:19:05PM -0700, Josh Berkus wrote: >> > " -> Seq Scan on price p (cost=0.00..11317.75 rows=581475 width=4) >> > (actual time=0.004..1143.720 rows=581475 loops=1)" >> >> Well, this is your pain point. Can we see the index scan plan on 8.1? >> Given that it's *expecting* only one row, I can't understand why it's >> using a seq scan ... > > I've created a simplified, self-contained test case for this: > > CREATE TABLE price ( > priceid integer PRIMARY KEY > ); > > CREATE TABLE supplier ( > supplierid integer PRIMARY KEY > ); > > CREATE TABLE content ( > contentid integer PRIMARY KEY, > supplierid integer NOT NULL REFERENCES supplier, > priceid integer NOT NULL REFERENCES price > ); > > INSERT INTO price (priceid) SELECT * FROM generate_series(1, 50000); > INSERT INTO supplier (supplierid) SELECT * FROM generate_series(1, 10000); > INSERT INTO content (contentid, supplierid, priceid) VALUES (1, 1, 50000); > > ANALYZE price; > ANALYZE supplier; > ANALYZE content; > > EXPLAIN ANALYZE > SELECT 0 > FROM content c > LEFT OUTER JOIN supplier s ON c.supplierid = s.supplierid > LEFT OUTER JOIN price p ON c.priceid = p.priceid; > > Here's the EXPLAIN ANALYZE from 8.0.3: > > Nested Loop Left Join (cost=0.00..7.06 rows=1 width=0) (actual > time=0.180..0.232 rows=1 loops=1) > -> Nested Loop Left Join (cost=0.00..4.04 rows=1 width=4) (actual > time=0.105..0.133 rows=1 loops=1) > -> Seq Scan on content c (cost=0.00..1.01 rows=1 width=8) > (actual time=0.021..0.029 rows=1 loops=1) > -> Index Scan using supplier_pkey on supplier s (cost=0.00..3.01 > rows=1 width=4) (actual time=0.052..0.059 rows=1 loops=1) > Index Cond: ("outer".supplierid = s.supplierid) > -> Index Scan using price_pkey on price p (cost=0.00..3.01 rows=1 > width=4) (actual time=0.046..0.055 rows=1 loops=1) > Index Cond: ("outer".priceid = p.priceid) > Total runtime: 0.582 ms > > Here it is from 8.1beta2: > > Merge Right Join (cost=4.05..1054.06 rows=1 width=0) (actual > time=676.863..676.895 rows=1 loops=1) > Merge Cond: ("outer".priceid = "inner".priceid) > -> Index Scan using price_pkey on price p (cost=0.00..925.00 > rows=50000 width=4) (actual time=0.035..383.345 rows=50000 loops=1) > -> Sort (cost=4.05..4.05 rows=1 width=4) (actual time=0.152..0.159 > rows=1 loops=1) > Sort Key: c.priceid > -> Nested Loop Left Join (cost=0.00..4.04 rows=1 width=4) > (actual time=0.082..0.111 rows=1 loops=1) > -> Seq Scan on content c (cost=0.00..1.01 rows=1 width=8) > (actual time=0.016..0.024 rows=1 loops=1) > -> Index Scan using supplier_pkey on supplier s > (cost=0.00..3.01 rows=1 width=4) (actual time=0.039..0.047 rows=1 loops=1) > Index Cond: ("outer".supplierid = s.supplierid) > Total runtime: 677.563 ms > > If we change content's priceid then we get the same plan but faster > results: > > UPDATE content SET priceid = 1; > > Merge Right Join (cost=4.05..1054.06 rows=1 width=0) (actual > time=0.268..0.303 rows=1 loops=1) > Merge Cond: ("outer".priceid = "inner".priceid) > -> Index Scan using price_pkey on price p (cost=0.00..925.00 > rows=50000 width=4) (actual time=0.049..0.061 rows=2 loops=1) > -> Sort (cost=4.05..4.05 rows=1 width=4) (actual time=0.187..0.192 > rows=1 loops=1) > Sort Key: c.priceid > -> Nested Loop Left Join (cost=0.00..4.04 rows=1 width=4) > (actual time=0.099..0.128 rows=1 loops=1) > -> Seq Scan on content c (cost=0.00..1.01 rows=1 width=8) > (actual time=0.025..0.033 rows=1 loops=1) > -> Index Scan using supplier_pkey on supplier s > (cost=0.00..3.01 rows=1 width=4) (actual time=0.046..0.053 rows=1 loops=1) > Index Cond: ("outer".supplierid = s.supplierid) > Total runtime: 0.703 ms > > -- > Michael Fuhr