Date: Fri, 23 Apr 2021 19:53:03 -0700
From: Andres Freund <andres@anarazel.de>
To: Peter Geoghegan <pg@bowt.ie>
Cc: PostgreSQL Hackers <pgsql-hackers@postgresql.org>
Subject: Re: Testing autovacuum wraparound (including failsafe)
Message-ID: <20210424025303.wtehilaqjwhk2gb3@alap3.anarazel.de>
References: <20210423204306.5osfpkt2ggaedyvy@alap3.anarazel.de>
 <CAH2-Wz=f2dfc-0BcM5X5ttXTeRNPg+JtxEMCuSZR82a7XHfRkg@mail.gmail.com>
 <20210424002921.pb3t7h6frupdqnkp@alap3.anarazel.de>
 <CAH2-WzkDM7nC+j2b3kB-2w4YkjXmrT+i0ppEcCNwTkpRehKLyQ@mail.gmail.com>
 <20210424023314.iem2gxm7cuaj66au@alap3.anarazel.de>
 <CAH2-Wz=wsX2_53-pjrNNiu5481QKdQX3oOj+APHZgtMB6Y16Sw@mail.gmail.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: 
 <CAH2-Wz=wsX2_53-pjrNNiu5481QKdQX3oOj+APHZgtMB6Y16Sw@mail.gmail.com>
Archived-At: 
 <https://www.postgresql.org/message-id/20210424025303.wtehilaqjwhk2gb3%40alap3.anarazel.de>
Precedence: bulk

Hi,

On 2021-04-23 19:42:30 -0700, Peter Geoghegan wrote:
> On Fri, Apr 23, 2021 at 7:33 PM Andres Freund <andres@anarazel.de> wrote:
> > Check it every so often, independent of whether there are indexes or
> > dead tuples? Or just check it at the boundaries.
>
> I think that the former suggestion might be better -- I actually
> thought about doing it that way myself.

Cool.


> The latter suggestion sounds like you're suggesting that we just check
> it at the beginning and the end in all cases (we do the beginning in
> all cases already, but now we'd also do the end outside of the loop in
> all cases). Is that right?

Yes.


> If that is what you meant, then you should note that there'd hardly be
> any check in the one-pass case with that scheme (apart from the
> initial check that we do already). The only work we'd be skipping at
> the end (in the event of that check triggering the failsafe) would be
> heap truncation, which (as you've pointed out yourself) doesn't seem
> particularly likely to matter.

I mainly suggested it because to me the current seems hard to
understand. I do think it'd be better to check more often. But checking
depending on the amount of dead tuples at the right time doesn't strike
me as a good idea - a lot of anti-wraparound vacuums will mainly be
freezing tuples, rather than removing a lot of dead rows. Which makes it
hard to understand when the failsafe kicks in.

Greetings,

Andres Freund