MIME-Version: 1.0
References: 
 <CAAKRu_ZMw6Npd_qm2KM+FwQ3cMOMx1Dh3VMhp8-V7SOLxdK9-g@mail.gmail.com>
 <CAAKRu_YXP1Us41fkZJeBK5SSj5whNC0C9fZNT6Brdtm+F0agpg@mail.gmail.com>
 <CAAKRu_bAR5uCfjuc06vc_xrZjNCJLs493NgHjTOUDso9qGdE0w@mail.gmail.com>
 <CAAKRu_ac4vUdU-U3ax8j=N8y2OM9-pbFZhafLVfzDOM8MtzvvA@mail.gmail.com>
 <BD5C3345-42F2-4870-89B2-FD8875301714@yandex-team.ru>
 <CAAKRu_a-aVGxNEdkJt+96HGryQXuQNuXe+PhR0KcnUhXSOPBJw@mail.gmail.com>
 <87DD95AA-274F-4F4F-BAD9-7738E5B1F905@yandex-team.ru>
 <CAAKRu_ZH8kL0Zm0j7m7DC9fzk7ru7yf9rm2pEQRvx1iXX25aPQ@mail.gmail.com>
 <CAAKRu_bGCgUuhmy1Mzkw3yCmbGcjNQAjV=OmjuW6hz90PuXKWA@mail.gmail.com>
In-Reply-To: 
 <CAAKRu_bGCgUuhmy1Mzkw3yCmbGcjNQAjV=OmjuW6hz90PuXKWA@mail.gmail.com>
From: Kirill Reshke <reshkekirill@gmail.com>
Date: Tue, 26 Aug 2025 14:58:28 +0500
Message-ID: 
 <CALdSSPhAU56g1gGVT0+wG8RrSWE6qW8TOfNJS1HNAWX6wPgbFA@mail.gmail.com>
Subject: Re: eliminate xl_heap_visible to reduce WAL (and eventually set VM
 on-access)
To: Melanie Plageman <melanieplageman@gmail.com>
Cc: Andrey Borodin <x4mmm@yandex-team.ru>,
	PostgreSQL Hackers <pgsql-hackers@lists.postgresql.org>,
 Andres Freund <andres@anarazel.de>,
	Robert Haas <robertmhaas@gmail.com>
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable
Archived-At: 
 <https://www.postgresql.org/message-id/CALdSSPhAU56g1gGVT0%2BwG8RrSWE6qW8TOfNJS1HNAWX6wPgbFA%40mail.gmail.com>
Precedence: bulk

On Sat, 2 Aug 2025 at 02:36, Melanie Plageman <melanieplageman@gmail.com> w=
rote:
>
> On Thu, Jul 31, 2025 at 6:58=E2=80=AFPM Melanie Plageman
> <melanieplageman@gmail.com> wrote:
> >
> > The patch "Set-pd_prune_xid-on-insert.txt" can be applied as the last
> > patch in the set. It sets pd_prune_xid on insert (so pages filled by
> > COPY or insert can also be set all-visible in the VM before they are
> > vacuumed). I gave it a .txt extension because it currently fails
> > 035_standby_logical_decoding due to a recovery conflict. I need to
> > investigate more to see if this is a bug in my patch set or elsewhere
> > in Postgres.
>
> I figured out that if we set the VM on-access, we need to enable
> hot_standby_feedback in more places in 035_standby_logical_decoding.pl
> to avoid recovery conflicts. I've done that in the attached updated
> version 6. There are a few other issues in
> 035_standby_logical_decoding.pl that I reported here [1]. With these
> changes, setting pd_prune_xid on insert passes tests. Whether or not
> we want to do it (and what the heuristic should be for deciding when
> to do it) is another question.
>
> - Melanie
>
> [1] https://www.postgresql.org/message-id/flat/CAAKRu_YO2mEm%3DZWZKPjTMU%=
3DgW5Y83_KMi_1cr51JwavH0ctd7w%40mail.gmail.com

Hi!

Andrey told me off-list about this thread and I decided to take a look.

I tried to play with each patch in this patchset and find a
corruption, but I was unsuccessful. I will conduct further tests
later. I am not implying that I suspect this patchset causes any
corruption; I am merely attempting to verify it.

I also have few comments and questions. Here is my (very limited)
review of 0001, because I believe that removing xl_heap_visible from
COPY FREEZE is pure win, so this patch can be very beneficial by
itself.

visibilitymap_set_vmbyte is introduced in 0001 and removed in 0012.
This is strange to me, maybe we can avoid visibilitymap_set_vmbyte in
first place?

In 0001:

1)
should we add "Assert(LWLockHeldByMeInMode(BufferDescriptorGetContentLock(b=
ufHdr),
LW_EXCLUSIVE));" in visibilitymap_set_vmbyte?

Also here  `Assert(visibilitymap_pin_ok(BufferGetBlockNumber(buffer),
vmbuffer));` can be beneficial:

>/*
>+ * If we're only adding already frozen rows to a previously empty
>+ * page, mark it as all-frozen and update the visibility map. We're
>+ * already holding a pin on the vmbuffer.
>+ */
>   else if (all_frozen_set)
>+ {
>    PageSetAllVisible(page);
>+ LockBuffer(vmbuffer, BUFFER_LOCK_EXCLUSIVE);
>+ visibilitymap_set_vmbyte(relation,
>+ BufferGetBlockNumber(buffer),
>+ vmbuffer,
>+ VISIBILITYMAP_ALL_VISIBLE |
>+ VISIBILITYMAP_ALL_FROZEN);
>+ }

2)
in heap_xlog_multi_insert:

+
+ visibilitymap_pin(reln, blkno, &vmbuffer);
+ visibilitymap_set_vmbyte(....)

Do we need to pin vmbuffer here? Looks like
XLogReadBufferForRedoExtended already pins vmbuffer. I verified this
with CheckBufferIsPinnedOnce(vmbuffer) just before visibilitymap_pin
and COPY ... WITH (FREEZE true) test.

3)
>+
> +#ifdef TRACE_VISIBILITYMAP
> + elog(DEBUG1, "vm_set %s %d", RelationGetRelationName(rel), heapBlk);
> +#endif

I can see this merely copy-pasted from visibilitymap_set, but maybe
display "flags" also?

4) visibilitymap_set receives  XLogRecPtr recptr parameters, which is
set to WAL record lsn during recovery and to InvalidXLogRecPtr
otherwise. visibilitymap_set manages VM page LSN bases on this recptr
value (inside function logic). visibilitymap_set_vmbyte behaves
vise-versa and makes its caller responsible for page LSN management.
Maybe we should keep these two functions akin to each other?


--=20
Best regards,
Kirill Reshke