Re: timestamp with time zone - Mailing list pgsql-performance

From Tom Lane
Subject Re: timestamp with time zone
Date
Msg-id 7688.1328854792@sss.pgh.pa.us
Whole thread Raw
In response to Re: timestamp with time zone  (Alessandro Gagliardi <alessandro@path.com>)
Responses Re: timestamp with time zone  (Alessandro Gagliardi <alessandro@path.com>)
List pgsql-performance
Alessandro Gagliardi <alessandro@path.com> writes:
> Still slow as mud: http://explain.depesz.com/s/Zfn
> Now I've got indices on created, timezone, created at time zone timezone,
> and (created at time zone timezone)::date. Clearly the problem isn't a lack
> of indices!...except, wait, it's not actually using blocks_created_date_idx
> (or blocks_created_at_timezone_idx). How do I make that happen?

Did you ANALYZE the table after creating those indexes?  Generally you
need an ANALYZE so that the planner will have some stats about an
expression index.

It might still think that the other index is a better option.  In that
case you can experiment to see if it's right or not; the general idea
is

    begin;
    drop index index_that_planner_prefers;
    explain analyze your_query;
    rollback;    -- revert the index drop

If that EXPLAIN isn't actually any better than what you had, then the
planner was right.  If it is better, let's see 'em both.

            regards, tom lane

pgsql-performance by date:

Previous
From: Scott Marlowe
Date:
Subject: Re: random_page_cost = 2.0 on Heroku Postgres
Next
From: David Yeu
Date:
Subject: Performance on large, append-only tables