Re: UPDATE and Indexes and Performance - Mailing list pgsql-general

From Joshua Tolley
Subject Re: UPDATE and Indexes and Performance
Date
Msg-id e7e0a2570810150957u492ad26t3a586d84735df36e@mail.gmail.com
Whole thread Raw
In response to UPDATE and Indexes and Performance  (Bill Thoen <bthoen@gisnet.com>)
List pgsql-general
On Wed, Oct 15, 2008 at 10:42 AM, Bill Thoen <bthoen@gisnet.com> wrote:
> Does PG (8.1) ever use existing indexes when executing an UPDATE?
>
> I've got some tables with millions of records and whenever I update a column
> that involves most or all the records the EXPLAIN command seems to indicate
> that it isn't using the pre-existing indexes. This result in a slow update,
> which is further slowed by the presence of indexes. So when doing a large
> update should I just drop the indexes first, or is there some good reason to
> keep them?
>
> Thanks,
> - Bill Thoen

You have to hit the disk twice if you use an index -- once to read the
index and once to read the table itself. If the planner guesses that
an operation involves most of the records in a table, it will avoid
indexes and just seqscan, because it's faster.

- Josh / eggyknap

pgsql-general by date:

Previous
From: Bill Thoen
Date:
Subject: UPDATE and Indexes and Performance
Next
From: "Andrus"
Date:
Subject: Re: Opptimizing projections containing unused columns