Reply-To: <gnanam@zoniac.com>
From: "Gnanakumar" <gnanam@zoniac.com>
To: "'Alvaro Herrera'" <alvherre@commandprompt.com>
Cc: <pgsql-admin@postgresql.org>
References: <00c001cad728$22f92f20$68eb8d60$@com>
	<20100408145328.GB4126@alvh.no-ip.org>
In-Reply-To: <20100408145328.GB4126@alvh.no-ip.org>
Subject: Re: Autovacuum daemon functionality questions
Date: Fri, 9 Apr 2010 13:07:20 +0530
Organization: Zoniac Software Private Limited
Message-ID: <001701cad7b7$7f1e2730$7d5a7590$@com>
MIME-Version: 1.0
Content-Type: text/plain;
	charset="us-ascii"
Content-Transfer-Encoding: 7bit
Thread-Index: AcrXK0JWOVizLxLlTRWvcf6wlI8oRQAiiJrw
Content-Language: en-us

>> 1. Why need to have 2 parameters (base threshold and scale factor) to
define
>> the threshold value, when either one of the parameter is more than enough
to
>> define the threshold value.  Can you explain the significance of having
both
>> parameters.  What is the real-time advantage of this?

>real-time advantage?  They are just the two factors in a linear
>equation.

>> 2. Documentation says ".... If the number of obsolete tuples since the
last
>> VACUUM exceeds the "vacuum threshold", the table is vacuumed ...".  
>> I also know about this table "pg_stat_user_tables" which has columns
>> n_tup_ins, n_tup_upd, n_tup_del, last_autovaccum and last_autoanalyze.
>> Since INSERT, UPDATE and DELETE count gets incremented everytime and do
not
>> reset after running autovacuum/autoanalyze, how does autovacuum
identifies
>> obsolete tuples since last VACUUM from this entries.

>There are two separate counters for live and dead tuples, IIRC (though
>they may not be exposed in the pg_stat views)

I've a stop/start of PostgreSQL service on a daily basis.  Since these 2
counters are not stored/saved in tables and not available in pg_stat views
also, will these values be persisted/retained even after stop/start/restart?

>There are two separate counters for live and dead tuples
As per documentation, live tuples can always be obtained from
pg_class.reltuples.  So, the question/answer here again would be to have
only counter on dead tuples since last VACUUM.  Is my understanding right?
Is there some other/alternative way where I can check obsolete tuples at any
time since last VACUUM?

>> 3. Is there a way to see autovacuum daemon log entries? 

>Not in 8.2.