Content-Type: text/plain;
	charset=utf-8
Mime-Version: 1.0 (Mac OS X Mail 13.4 \(3608.80.23.2.2\))
Subject: Re: Make autovacuum sort tables in descending order of xid_age
From: Mark Dilger <mark.dilger@enterprisedb.com>
In-Reply-To: <2e45935a-d492-2b19-a2c4-991e1efe0895@pgmasters.net>
Date: Mon, 30 Mar 2020 09:20:15 -0700
Cc: Robert Haas <robertmhaas@gmail.com>,
 Mark Dilger <hornschnorter@gmail.com>,
 David Fetter <david@fetter.org>,
 PostgreSQL Development <pgsql-hackers@postgresql.org>,
 Christophe Pettus <xof@thebuild.com>
Content-Transfer-Encoding: quoted-printable
Message-Id: <019573E6-B914-4FD4-A46B-AD727CF500AF@enterprisedb.com>
References: <20191129180138.GD8731@fetter.org>
 <20191129222152.GE8731@fetter.org>
 <2488acea-328b-1328-d437-8c156bc8f7a4@gmail.com>
 <20191130222314.GF8731@fetter.org>
 <877ac074-e95d-bdee-09e7-c51f0ee0fde2@gmail.com>
 <20191212192631.GA29201@fetter.org>
 <CA+TgmoZPsiERnWf2eFSk7F2AkmOymhQKGyVFsB5t4BPim7p5Og@mail.gmail.com>
 <20200111175335.GB32763@fetter.org>
 <2e45935a-d492-2b19-a2c4-991e1efe0895@pgmasters.net>
To: David Steele <david@pgmasters.net>
Precedence: bulk


> On Mar 30, 2020, at 7:09 AM, David Steele <david@pgmasters.net> wrote:
>=20
> On 1/11/20 12:53 PM, David Fetter wrote:
>> I agree that it's a complex situation, and that many different
>> approaches will eventually need to be brought to bear.
>> What concerns me about introducing a big lump of complexity here is
>> disentangling the effects of each part and of their interaction =
terms.
>> We're not, to put it mildly, set up to do ANOVA
>> (https://en.wikipedia.org/wiki/Analysis_of_variance ) , ANCOVA (
>> https://en.wikipedia.org/wiki/Analysis_of_covariance ), etc. on
>> changes.
>> Given the above, I'd like to make the case for changing just this one
>> thing at first and seeing whether the difference it makes is =
generally
>> positive.
>=20
> Mark, Robert, thoughts on this?

I have not been working on this issue lately, but as I recall, my =
concern was that changing the behavior of autovacuum could introduce =
regressions for some users, so we should be careful to get it right =
before we rush to release anything.  It didn't seem like the proposed =
changes took enough into account.  But that's clearly a judgement call, =
having to do with how cautious any particular person thinks we should =
be.  I don't feel strongly enough to stand in the way if the general =
concensus is that this is a good enough implementation.

=E2=80=94
Mark Dilger
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company