Tracking table modifications / table stats - Mailing list pgsql-general

From Derrick Rice
Subject Tracking table modifications / table stats
Date
Msg-id AANLkTikKofBG4PUQYU=Y6+-GhGoicvUwKjhpCMXdjWAw@mail.gmail.com
Whole thread Raw
Responses Re: Tracking table modifications / table stats  (Merlin Moncure <mmoncure@gmail.com>)
Re: Tracking table modifications / table stats  (Andy Colson <andy@squeakycode.net>)
List pgsql-general
Hey folks,

I was looking through the contrib modules with 8.4 and hoping to find something that satisfies my itch.  http://www.postgresql.org/docs/8.4/static/pgstatstatements.html comes the closest.

I'm inheriting a database which has mostly unknown usage patterns, and would like to figure them out so that I can allocate tablespaces and set autovacuum settings appropriately.  To do this, it seems I need to know (at least) the number of rows read, rows updated, rows deleted, and rows inserted for each table (over time, or until reset).

I suppose things like disk usage and CPU usage would be interesting as well, but I'm somewhat less concerned with those.  For one, CPU usage can't be tied to a table as easily and is more about query optimization than PostgreSQL configuration (excluding cost coefficients and memory size settings).  For the other, disk usage can be mostly inferred from the row size and and number of operations per table (this does exclude seq. scans and heavy heavy index use, though).  I realize those statements are fuzzy and short-sighted, but I'm trying to get "good enough" information, not optimize a space shuttle.

There's no way I'm the first person to feel the need for this.  Is there a doc or wiki which gives some recommendations?  I'd like to avoid parsing logs or installing triggers.  I'd also like to avoid heavy statement-level tracking like the above mentioned contrib does (sounds expensive, and I'm not sure the users have parameterized SQL).

Thanks,

Derrick

pgsql-general by date:

Previous
From:
Date:
Subject: Why count(*) doest use index?
Next
From: Adrian Klaver
Date:
Subject: Re: Why count(*) doest use index?