Date: Wed, 2 Aug 2017 02:01:15 +0000
From: Noah Misch <noah@leadboat.com>
To: Tomas Vondra <tomas.vondra@2ndquadrant.com>
Cc: Tom Lane <tgl@sss.pgh.pa.us>, pgsql-hackers@postgresql.org
Subject: Re: VACUUM and ANALYZE disagreeing on what reltuples means
Message-ID: <20170802020115.GE2611764@rfd.leadboat.com>
References: <16db4468-edfa-830a-f921-39a50498e77e@2ndquadrant.com>
 <19990.1500936945@sss.pgh.pa.us>
 <3d250597-384e-c5f4-7533-b513e366888f@2ndquadrant.com>
 <7087.1500995043@sss.pgh.pa.us>
 <9b594258-57b1-f2cf-f25c-d2caa0a1bc8d@2ndquadrant.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <9b594258-57b1-f2cf-f25c-d2caa0a1bc8d@2ndquadrant.com>
User-Agent: Mutt/1.5.24 (2015-08-30)
Precedence: bulk
Sender: pgsql-hackers-owner@postgresql.org

On Tue, Jul 25, 2017 at 07:02:28PM +0200, Tomas Vondra wrote:
> On 7/25/17 5:04 PM, Tom Lane wrote:
> >Tomas Vondra <tomas.vondra@2ndquadrant.com> writes:
> >>Attached is a patch that (I think) does just that. The disagreement
> >>was caused by VACUUM treating recently dead tuples as live, while
> >>ANALYZE treats both of those as dead.
> >
> >>At first I was worried that this will negatively affect plans in
> >>the long-running transaction, as it will get underestimates (due
> >>to reltuples not including rows it can see). But that's a problem
> >>we already have anyway, you just need to run ANALYZE in the other
> >>session.
> >
> >This definitely will have some impact on plans, at least in cases
> >where there's a significant number of unvacuumable dead tuples. So I
> >think it's a bit late for v10, and I wouldn't want to back-patch at
> >all. Please add to the next commitfest.
> >
> 
> I dare to disagree here, for two reasons.
> 
> Firstly, the impact *is* already there, it only takes running ANALYZE. Or
> VACUUM ANALYZE. In both those cases we already end up with
> reltuples=n_live_tup.
> 
> Secondly, I personally strongly prefer stable predictable behavior over
> intermittent oscillations between two values. That's a major PITA on
> production, both to investigate and fix.

> FWIW I personally see this as a fairly annoying bug, and would vote to
> backpatch it, although I understand people might object.

I tend to agree.  If you have a setup that somehow never uses ANALYZE or never
uses VACUUM on a particular table, you might prefer today's (accidental)
behavior.  However, the discrepancy arises only on a table that gets dead
tuples, and a table that gets dead tuples will see both VACUUM and ANALYZE
soon enough.  It does seem like quite a stretch to imagine someone wanting
plans to depend on which of VACUUM or ANALYZE ran most recently.


-- 
Sent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-hackers