Re: check_postgres reports most of my tables are unanalyzed for 8 weeks; why isn't autovacuum analyzing them? - Mailing list pgsql-general

From Merlin Moncure
Subject Re: check_postgres reports most of my tables are unanalyzed for 8 weeks; why isn't autovacuum analyzing them?
Date
Msg-id AANLkTinI0jKv8GwZuqy6SWDqsT5_QAfE0OJ4uEkeQMAM@mail.gmail.com
Whole thread Raw
In response to check_postgres reports most of my tables are unanalyzed for 8 weeks; why isn't autovacuum analyzing them?  (Aleksey Tsalolikhin <atsaloli.tech@gmail.com>)
List pgsql-general
On Tue, Jul 20, 2010 at 4:31 PM, Aleksey Tsalolikhin
<atsaloli.tech@gmail.com> wrote:
> Hi.  I've just discovered the check_postgres utility and am running
> all the tests against my database.
>
> The "last_analyze" test comes out critical - many tables unanalyzed
> for 8 weeks.

have those tables been modified at all?  this is only an issue if the
database's understanding of the table is different from reality AFIAK.

> P.S. I checked the PostgreSQL logs, and in the last week, I just have
> three messages with "vacuum" in them:
> (I am running at default log levels)
>
> 2010-07-20 02:05:05 PDT ERROR:  canceling autovacuum task
> 2010-07-20 02:05:05 PDT CONTEXT:  automatic vacuum of table
> "mydb.pg_catalog.pg_listener"
>
> 2010-07-20 07:27:14 PDT ERROR:  canceling autovacuum task
> 2010-07-20 07:27:14 PDT CONTEXT:  automatic vacuum of table
> "mydb._slony_cluster.sl_event"
>
> 2010-07-20 07:27:53 PDT ERROR:  canceling autovacuum task
> 2010-07-20 07:27:53 PDT CONTEXT:  automatic vacuum of table
> "mydb._slony_cluster.sl_log_2"

this is normal.  note that slony abuses the listener notify system,
and autovac punts to avoid messing with user sessions.  9.0 has a much
cleaner async notify mechanism so won't have this issue.

merlin

pgsql-general by date:

Previous
From: Aleksey Tsalolikhin
Date:
Subject: Re: check_postgres reports most of my tables are unanalyzed for 8 weeks; why isn't autovacuum analyzing them?
Next
From: Jeff Hamann
Date:
Subject: printing vector column seems to print the rest of the row too