Skip site navigation (1) Skip section navigation (2)

Re: vacuum, analyze and reindex

From: Christopher Browne <cbbrowne(at)acm(dot)org>
To: pgsql-performance(at)postgresql(dot)org
Subject: Re: vacuum, analyze and reindex
Date: 2006-02-28 13:34:01
Message-ID: (view raw, whole thread or download thread mbox)
Lists: pgsql-performance
Centuries ago, Nostradamus foresaw when jsomoza(at)pandasoftware(dot)es (Javier Somoza) would write:
>             Hi all
>             I've a question about vacuuming, ...
>             Vacuum: cleans out obsolete and deleted registers...
>             Analyze:  update statistics for the planner
>             Reindex:  rebuild indexes
>             I think the correct order to perform the database maintenance for performance is:
>             1 - Vacuum
>             2 - Reindex
>             3 - Analyze
>             So the planner is updated with the updated indexes.

There is a misunderstanding there.  The planner isn't aware of
"updates" to the indexes; it is aware of how they are defined.

ANALYZE doesn't calculate index-specific things; it calculates
statistical distributions for the contents of each column.  As soon as
it runs, the planner will be better able to choose from the available
indexes.  If you add another index, the planner will, without another
ANALYZE, be able to choose that index, if it is useful to do so, based
on the existing statistical distributions.

>             The autovacuum daemon does vacuum and analyze. Not reindex.
>             So, no way to perform it in that order.
>             What do you think?
>             How often do you reindex your tables?

If tables are being vacuumed frequently enough, the answer to that can
be "almost never."

Back in the days of 7.2, there were conditions where indexes could
bloat mercilessly, so that heavily updated tables needed a reindex
every so often.  But if you're running a reasonably modern version,
that shouldn't be necessary.
wm(X,Y):-write(X),write('@'),write(Y). wm('cbbrowne','').
Rules of the Evil Overlord #219. "I will be selective in the hiring of
assassins.   Anyone who  attempts to  strike down  the hero  the first
instant his back is turned will not even be considered for the job."

In response to

pgsql-performance by date

Next:From: RonDate: 2006-02-28 13:40:56
Subject: Re: Different disks for xlogs and data
Previous:From: Javier SomozaDate: 2006-02-28 12:11:05
Subject: vacuum, analyze and reindex

Privacy Policy | About PostgreSQL
Copyright © 1996-2017 The PostgreSQL Global Development Group