From: Mattias Kregert <mattias@kregert.se>
To: pgsql-performance@postgresql.org
Subject: Strange plan in pg 8.1.0
Date: Mon, 30 Oct 2006 13:05:07 +0200
Message-ID: <200610301205.08087.mattias@kregert.se> (raw)
Look at this insane plan:
lucas=# explain analyse select huvudklass,sum(summa) from kor_tjanster left outer join prislist on prislista=listid and tjanst=tjanstid where kor_id in (select id from kor where lista=10484) group by 1;
QUERY PLAN
--------------------------------------------------------------------------------------------------------------------------------------------
HashAggregate (cost=212892.07..212892.10 rows=2 width=23) (actual time=4056.165..4056.167 rows=2 loops=1)
-> Hash IN Join (cost=102.84..212889.04 rows=607 width=23) (actual time=4032.931..4056.017 rows=31 loops=1)
Hash Cond: ("outer".kor_id = "inner".id)
-> Hash Left Join (cost=59.66..206763.11 rows=1215336 width=27) (actual time=4.959..3228.550 rows=1216434 loops=1)
Hash Cond: (("outer".prislista = ("inner".listid)::text) AND ("outer".tjanst = ("inner".tjanstid)::text))
-> Seq Scan on kor_tjanster (cost=0.00..23802.36 rows=1215336 width=26) (actual time=0.032..1257.241 rows=1216434 loops=1)
-> Hash (cost=51.77..51.77 rows=1577 width=29) (actual time=4.898..4.898 rows=1577 loops=1)
-> Seq Scan on prislist (cost=0.00..51.77 rows=1577 width=29) (actual time=0.034..2.445 rows=1577 loops=1)
-> Hash (cost=41.79..41.79 rows=557 width=4) (actual time=0.185..0.185 rows=29 loops=1)
-> Index Scan using kor_lista on kor (cost=0.00..41.79 rows=557 width=4) (actual time=0.070..0.150 rows=29 loops=1)
Index Cond: (lista = 10484)
Total runtime: 4056.333 ms
I have an index on kor_tjanster(kor_id), an index on prislist(prislist_id), did ANALYZE and all that stuff... but those indexes are not used.
Why does it come up with this strange plan? It does a seqscan of 1.2 million rows and then a join!? Using the index would be much faster...
I expected something like this:
1. Index Scan using kor_lista on kor (use lista_id 10484 to get a list of kor_id's - 29 rows (expected 557 rows))
2. Index Scan using kor_id on kor_tjanster (use the kor_id's to get a list of kor_tjanster - 31 rows)
3. Index Scan using prislist_listid on prislist (use the 31 kor_tjanster rows to find the corresponding 'huvudklass' for each row)
29+31+31=91 index lookups... which is MUCH faster than seq-scanning millions of rows...
I need to speed up this query. How can i make it use the correct index? Any hints?
I have pg 8.1.0, default settings.
/* m */
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Reply to all the recipients using the --to and --cc options:
reply via email
To: pgsql-performance@postgresql.org
Cc: mattias@kregert.se
Subject: Re: Strange plan in pg 8.1.0
In-Reply-To: <200610301205.08087.mattias@kregert.se>
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
This inbox is served by DDX for PostgreSQL; see mirroring instructions
for how to clone and mirror all data and code used for this inbox