Message-ID: <54B83F70.7080004@agliodbs.com>
Date: Fri, 16 Jan 2015 11:30:08 +1300
From: Josh Berkus <josh@agliodbs.com>
Organization: PostgreSQL Experts Inc.
User-Agent: Mozilla/5.0 (X11; Linux x86_64;
 rv:31.0) Gecko/20100101 Thunderbird/31.3.0
MIME-Version: 1.0
To: postgres performance list <pgsql-performance@postgresql.org>
Subject: Strange choice of general index over partial index
Content-Type: text/plain; charset=utf-8
Content-Transfer-Encoding: 7bit
Precedence: bulk
Sender: pgsql-performance-owner@postgresql.org

This is an obfuscation and mock up, but:

table files (
	id serial pk,
	filename text not null,
	state varchar(20) not null
	... 18 more columns
)

index file_state on (state)
	(35GB in size)
index file_in_flight_state (state) where state in (
'waiting','assigning', 'processing' )
	(600MB in size)
... 10 more indexes

More important facts:
* state = 'done' 95% of the time.  thereform the partial index
represents only 5% of the table
* all indexes and the table are very bloated
* server has 128GB RAM
* Version 9.2.

Given this setup, I would expect the planner to *always* choose
file_in_flight_state over file_state for this query:

SELECT id, filename FROM files WHERE state = 'waiting';

... and yet it keeps selecting file_state based on extremely small
changes to the stats.   This is important because the same query, using
file_state, is 20X to 50X slower, because that index frequently gets
pushed out of memory.

What am I missing?  Or is this potentially a planner bug for costing?

-- 
Josh Berkus
PostgreSQL Experts Inc.
http://pgexperts.com


-- 
Sent via pgsql-performance mailing list (pgsql-performance@postgresql.org)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-performance