From: Tom Lane <tgl@sss.pgh.pa.us>
To: Justin Pryzby <pryzby@telsasoft.com>
cc: Heikki Linnakangas <hlinnaka@iki.fi>,
        "Andrey M. Borodin" <x4mmm@yandex-team.ru>,
        Darafei =?utf-8?Q?Kom=D1=8Fpa?= Praliaskouski <me@komzpa.net>,
        Pavel Borisov <pashkin.elfe@gmail.com>,
        Thomas Munro <thomas.munro@gmail.com>,
        Michael Paquier <michael@paquier.xyz>,
        Alexander Korotkov <a.korotkov@postgrespro.ru>,
        pgsql-hackers@postgresql.org
Subject: Re: Yet another fast GiST build
In-reply-to: <20200920224446.GF30557@telsasoft.com>
References: <7d9158ca-d7cb-88f0-dbbd-1d5cec32b1ae@iki.fi>
 <CAC8Q8t+KG6oeyC697X3agZmHFTMjZYKvg13X2KapbfcTiYvdhw@mail.gmail.com>
 <fdee7ef5-1ea3-e381-0611-19352c4f9d1d@iki.fi>
 <8516B6A8-11B2-41A8-B7EB-D0C38ADD06C3@yandex-team.ru>
 <50e03197-6e75-faec-16dc-41a2a38f080c@iki.fi>
 <123F5F32-9A85-4D0B-9C7A-1686E6BBE15D@yandex-team.ru>
 <85df399a-e8a4-2369-f723-fd1a3a872575@iki.fi>
 <!63596817-db40-1093-c6ce-5dea7dac67c4@iki.fi>
 <884200.1600634472@sss.pgh.pa.us> <891385.1600636205@sss.pgh.pa.us>
 <20200920224446.GF30557@telsasoft.com>
Comments: In-reply-to Justin Pryzby <pryzby@telsasoft.com>
	message dated "Sun, 20 Sep 2020 17:44:46 -0500"
MIME-Version: 1.0
Content-Type: text/plain; charset="us-ascii"
Content-ID: <896112.1600643169.1@sss.pgh.pa.us>
Date: Sun, 20 Sep 2020 19:06:09 -0400
Message-ID: <896113.1600643169@sss.pgh.pa.us>
Precedence: bulk

Justin Pryzby <pryzby@telsasoft.com> writes:
> This also appears to break checksums.

I was wondering about that, because the typical pattern for use of
smgrextend for indexes seems to be

	RelationOpenSmgr(rel);
	PageSetChecksumInplace(page, lastblock);
	smgrextend(rel->rd_smgr, MAIN_FORKNUM, lastblock, zerobuf.data, false);

and gist_indexsortbuild wasn't doing either of the first two things.

gist_indexsortbuild_flush_ready_pages looks like it might be
a few bricks shy of a load too.  But my local CLOBBER_CACHE_ALWAYS
run hasn't gotten to anything except the pretty-trivial index
made in point.sql, so I don't have evidence about it.

Another interesting point is that all the other index AMs seem to WAL-log
the new page before the smgrextend call, whereas this code is doing it
in the other order.  I strongly doubt that both patterns are equally
correct.  Could be that the other AMs are in the wrong though.

			regards, tom lane