From: Tom Lane <tgl@sss.pgh.pa.us>
To: Matthew T. O'Connor <matthew@zeut.net>
Cc: pgsql-hackers@postgreSQL.org, Otto Blomqvist <o.blomqvist@secomintl.com>
Cc: pgsql-performance@postgreSQL.org
Subject: lazy_update_relstats considered harmful (was Re: [PERFORM] pg_autovacuum not having enough suction ?)
Date: Fri, 25 Mar 2005 15:22:49 -0500
Message-ID: <12121.1111782169@sss.pgh.pa.us> (raw)
In-Reply-To: <11991.1111780527@sss.pgh.pa.us>
References: <d1v0a7$iau$1@news.hub.org>
<42430FA0.9080204@zeut.net>
<002b01c530c5$ad5a1620$e602a8c0@blomqvist>
<42435408.30303@zeut.net>
<00f201c53168$96903480$e602a8c0@blomqvist>
<42446A66.6010504@zeut.net>
<11991.1111780527@sss.pgh.pa.us>
I wrote:
> One thing that is possibly relevant here is that in 8.0 a plain VACUUM
> doesn't set reltuples to the exactly correct number, but to an
> interpolated value that reflects our estimate of the "steady state"
> average between vacuums. I wonder if that code is wrong, or if it's
> operating as designed but is confusing autovac.
Now that I think it over, I'm thinking that I must have been suffering
severe brain fade the day I wrote lazy_update_relstats() (see
vacuumlazy.c). The numbers that that routine is averaging are the pre-
and post-vacuum physical tuple counts. But the difference between them
consists of known-dead tuples, and we shouldn't be factoring dead tuples
into reltuples. The planner has always considered reltuples to count
only live tuples, and I think this is correct on two grounds:
1. The numbers of tuples estimated to be returned by scans certainly
shouldn't count dead ones.
2. Dead tuples don't have that much influence on scan costs either, at
least not once they are marked as known-dead. Certainly they shouldn't
be charged at full freight.
It's possible that there'd be some value in adding a column to pg_class
to record dead tuple count, but given what we have now, the calculation
in lazy_update_relstats is totally wrong.
The idea I was trying to capture is that the tuple density is at a
minimum right after VACUUM, and will increase as free space is filled
in until the next VACUUM, so that recording the exact tuple count
underestimates the number of tuples that will be seen on-the-average.
But I'm not sure that idea really holds water. The only way that a
table can be at "steady state" over a long period is if the number of
live tuples remains roughly constant (ie, inserts balance deletes).
What actually increases and decreases over a VACUUM cycle is the density
of *dead* tuples ... but per the above arguments this isn't something
we should adjust reltuples for.
So I'm thinking lazy_update_relstats should be ripped out and we should
go back to recording just the actual stats.
Sound reasonable? Or was I right the first time and suffering brain
fade today?
regards, tom lane
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Reply to all the recipients using the --to and --cc options:
reply via email
To: pgsql-performance@postgresql.org
Cc: tgl@sss.pgh.pa.us, matthew@zeut.net, o.blomqvist@secomintl.com
Subject: Re: lazy_update_relstats considered harmful (was Re: [PERFORM] pg_autovacuum not having enough suction ?)
In-Reply-To: <12121.1111782169@sss.pgh.pa.us>
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
This inbox is served by DDX for PostgreSQL; see mirroring instructions
for how to clone and mirror all data and code used for this inbox