Date: Mon, 30 Aug 2010 18:45:26 +0200 (CEST)
From: Gerhard Wiesinger <lists@wiesinger.com>
To: Tom Lane <tgl@sss.pgh.pa.us>
cc: Pavel Stehule <pavel.stehule@gmail.com>,
	Andreas Kretschmer <akretschmer@spamfence.net>,
	pgsql-performance@postgresql.org
Subject: Re: Major performance problem after upgrade from 8.3 to
 8.4
In-Reply-To: <11111.1283185373@sss.pgh.pa.us>
Message-ID: <alpine.LFD.2.01.1008301838480.4133@bbs.intern>
References: <alpine.LFD.2.01.1008300802300.4121@bbs.intern>
	<20100830072230.GA28373@tux>
	<AANLkTi=Te9ToTG-bqVy+fHQ0wM1YWv0nhBqqYiTk2ePg@mail.gmail.com>
	<alpine.LFD.2.01.1008300956320.19598@bbs.intern>
	<AANLkTik14X_MYoZA5CHNGC8M7Rc0zXrRhoGw6FrfUPXm@mail.gmail.com>
	<alpine.LFD.2.01.1008301808330.31277@bbs.intern>
	<11111.1283185373@sss.pgh.pa.us>
User-Agent: Alpine 2.01 (LFD 1266 2009-07-14)
MIME-Version: 1.0
Content-Type: TEXT/PLAIN; charset=ISO-8859-15; format=flowed
Content-Transfer-Encoding: 8BIT

On Mon, 30 Aug 2010, Tom Lane wrote:

> Gerhard Wiesinger <lists@wiesinger.com> writes:
>> I know the drawbacks of an EAV design but I don't want to discuss that. I
>> want to discuss the major performance decrease of PostgreSQL 8.3
>> (performance was ok) to PostgreSQL 8.4 (performance is NOT ok).
>
>> Any further ideas how I can track this down?
>> Can someone explain the difference in query plan from an optimizer point
>> of view?
>
> Since you haven't shown us the 8.3 plan, it's kind of hard to speculate ;-)
>
> One thing that jumped out at me was that 8.4 appears to be expecting
> multiple matches in each of the left-joined tables, which is why the
> total rowcount estimate balloons so fast.  I rather imagine that you are
> expecting at most one match in reality, else the query isn't going to
> behave nicely.  Is this correct?  Are you *sure* you analyzed all these
> tables?  And if that is how the data looks, where is the actual
> performance problem?  A bad rowcount estimate isn't in itself going
> to kill you.
>
> FWIW, in a similar albeit toy example, I don't see any difference
> between the 8.3 and 8.4 plans or cost estimates.

Yes, I'm expecting only one match in reality and I thing PostgreSQL should 
also know that from table definition and constraints. Long answer below.

Query doesn't "end" in PostgreSQL.