MIME-Version: 1.0
References: <20230317230930.nhsgk3qfk7f4axls@awork3.anarazel.de>
 <20230318093357.g4e3eexqg6dxmuel@alvherre.pgsql> <ZBswZOh68Yh7t9df@telsasoft.com>
 <20230322170048.ffbnfr5o2jrjgb2f@awork3.anarazel.de> <ZDWLvS1GghESoX/d@telsasoft.com>
 <20230411182010.zsldtjrxe5m2bqzp@awork3.anarazel.de> <CAA4eK1+s7hyed2ZSwTC42N6VqCV73CGtzBRfcf1qgjCSJbO3Gw@mail.gmail.com>
 <bf42784f-4d57-0a3d-1a06-ffac1a09318c@postgresql.org> <490f1ebd-03fc-d996-cc5e-7991569a9e45@postgresql.org>
 <f8b69c30afd245475c70d2c629f0eb0860e997b6.camel@cybertec.at>
 <CA+TgmoaZXaY_xJakDxZMeUUG+CuuuDmL+Yg8-LjPL5kiT-2Zjg@mail.gmail.com> <2755ebe5-5ab6-7509-cf47-c606be5583cd@postgresql.org>
In-Reply-To: <2755ebe5-5ab6-7509-cf47-c606be5583cd@postgresql.org>
From: Greg Stark <stark@mit.edu>
Date: Fri, 14 Apr 2023 11:25:34 -0400
Message-ID: <CAM-w4HOiHaNgrBZGWGhOeset21jdFN5grdqZFLXXos6uCNnSHQ@mail.gmail.com>
Subject: Re: Should we remove vacuum_defer_cleanup_age?
To: "Jonathan S. Katz" <jkatz@postgresql.org>
Cc: Robert Haas <robertmhaas@gmail.com>, Laurenz Albe <laurenz.albe@cybertec.at>, 
	Amit Kapila <amit.kapila16@gmail.com>, Andres Freund <andres@anarazel.de>, 
	Justin Pryzby <pryzby@telsasoft.com>, Alvaro Herrera <alvherre@alvh.no-ip.org>, 
	pgsql-hackers@postgresql.org
Content-Type: text/plain; charset="UTF-8"
Archived-At: <https://www.postgresql.org/message-id/CAM-w4HOiHaNgrBZGWGhOeset21jdFN5grdqZFLXXos6uCNnSHQ%40mail.gmail.com>
Precedence: bulk

On Fri, 14 Apr 2023 at 09:47, Jonathan S. Katz <jkatz@postgresql.org> wrote:
>
> Let me restate [1] in a different way.
>
> Using a large enough dataset, I did qualitatively look at overall usage
> of both "vacuum_defer_cleanup_age" and compared to
> "hot_standby_feedback", given you can use both to accomplish similar
> outcomes.

I assume people would use hot_standby_feedback if they have streaming
replication. The main use cases for vacuum_defer_cleanup_age would be
if you're replaying WAL files. That may sound archaic but there are
plenty of circumstances where your standby may not have network access
to your primary at all or not want to be replaying continuously.

I wonder whether your dataset is self-selecting sites that have
streaming replication. That does seem like the more common usage
pattern.

Systems using wal files are more likely to be things like data
warehouses, offline analytics systems, etc. They may not even be well
known in the same organization that runs the online operations -- in
my experience they're often run by marketing or sales organizations or
in some cases infosec teams and consume data from lots of sources. The
main reason to use wal archive replay is often to provide the
isolation so that the operations team don't need to worry about the
impact on production and that makes it easy to forget these even
exist.

-- 
greg