From: Rod Taylor <rbt@rbt.ca>
To: Scott Cain <cain@cshl.org>
Cc: Postgresql Performance <pgsql-performance@postgresql.org>
Subject: Re: force the use of a particular index
Date: 11 Jul 2003 13:38:16 +0000
Message-ID: <1057930695.46100.172.camel@jester> (raw)
In-Reply-To: <1057929459.5766.7.camel@localhost.localdomain>
References: <1057850280.1451.22.camel@localhost.localdomain>
<1057920672.46100.149.camel@jester>
<1057929459.5766.7.camel@localhost.localdomain>
On Fri, 2003-07-11 at 13:17, Scott Cain wrote:
> The problem (at least as it appears to me) is not that it is performing
> a table scan instead of an index scan, it is that it is using the wrong
> index. Here is the output from EXPLAIN ANALYZE:
>
> QUERY PLAN
> ------------------------------------------------------------------------------------------------------------------------------------------------------------
> Unique (cost=494008.47..494037.59 rows=166 width=54) (actual time=114660.37..114660.38 rows=1 loops=1)
> -> Sort (cost=494008.47..494012.63 rows=1664 width=54) (actual time=114660.37..114660.37 rows=1 loops=1)
> Sort Key: f.name, fl.fmin, fl.fmax, fl.strand, f.type_id, f.feature_id
> -> Nested Loop (cost=0.00..493919.44 rows=1664 width=54) (actual time=2596.13..114632.90 rows=1 loops=1)
> -> Index Scan using feature_pkey on feature f (cost=0.00..134601.43 rows=52231 width=40) (actual time=105.74..56048.87 rows=13825 loops=1)
> Filter: (type_id = 219)
> -> Index Scan using featureloc_idx1 on featureloc fl (cost=0.00..6.87 rows=1 width=14) (actual time=4.23..4.23 rows=0 loops=13825)
> Index Cond: ("outer".feature_id = fl.feature_id)
> Filter: ((srcfeature_id = 6) AND (fmin <= 2585581) AND (fmax >= 2565581))
> Total runtime: 114660.91 msec
> it is using on featureloc (featureloc_idx1) is on the foreign key
> feature_id. It should instead be using another index, featureloc_idx3,
> which is built on (srcfeature_id, fmin, fmax).
Nope.. The optimizer is right in the decision to use featureloc_idx1.
You will notice it is expecting to retrieve a single row from this
index, but the featureloc_idx3 is bound to be larger (due to indexing
more data), thus take more disk reads for the exact same information (or
in this case, lack thereof).
What is taking a long time is the scan on feature_pkey. It looks like it
is throwing away a ton of rows that are not type_id = 219. Either that,
or you do a pile of deletes and haven't run REINDEX recently.
Create an index consisting of (feature_id, type_id). This will probably
make a significant different in execution time.
Attachments:
signature.asc (application/pgp-signature, 186B) download
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 from your mail client; the link fills in
the headers below: reply via email
Some clients, webmail especially, drop In-Reply-To and References from
a link, and the reply then starts a new thread: use the mbox below.
To: pgsql-performance@postgresql.org
Cc: rbt@rbt.ca, cain@cshl.org
Subject: Re: force the use of a particular index
In-Reply-To: <1057930695.46100.172.camel@jester>
References: <1057850280.1451.22.camel@localhost.localdomain> <1057920672.46100.149.camel@jester> <1057929459.5766.7.camel@localhost.localdomain> <1057930695.46100.172.camel@jester>
* 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