Re: Orphan page in _bt_split

From: Konstantin Knizhnik <knizhnik(at)garret(dot)ru>
To: Michael Paquier <michael(at)paquier(dot)xyz>
Cc: Peter Geoghegan <pg(at)bowt(dot)ie>, PostgreSQL Hackers <pgsql-hackers(at)lists(dot)postgresql(dot)org>
Subject: Re: Orphan page in _bt_split
Date: 2025-09-25 06:41:00
Message-ID: d3c77645-156d-4049-9013-082e34c6f21e@garret.ru
Views: Whole Thread | Raw Message | Download mbox | Resend email
Thread:
Lists: pgsql-hackers


On 25/09/2025 7:35 AM, Michael Paquier wrote:
> On Mon, Sep 22, 2025 at 07:44:18PM +0300, Константин Книжник wrote:
>> Attached please find rebased version of the patch with fixed mistypings.
> I have looked at v3.
>
> - leftpage = PageGetTempPage(origpage);
> + leftpage = leftpage_buf.data;
> + memcpy(leftpage, origpage, BLCKSZ);
> _bt_pageinit(leftpage, BufferGetPageSize(buf));
>
> What's the point of copying the contents of origpage to leftpage?
> _bt_pageinit() is going to initialize leftpage (plus a few more things
> set like the page LSN, etc.), so the memcpy should not be necessary,
> no?
>
> + rightpage = BufferGetPage(rbuf);
> + memcpy(rightpage, rightpage_buf.data, BLCKSZ);
> + ropaque = BTPageGetOpaque(rightpage);
>
> When we reach this state of the logic, aka at the beginning of the
> critical section, the right and left pages are in a correct state,
> and your code is OK because we copy the contents of the right page
> back into its legitimate place. What is not OK to me is that the
> copy of the "temporary" right page back to "rbuf" is not explained. I
> think that this deserves a comment, especially the part about ropaque
> which is set once by this proposal, then manipulated while in the
> critical section.
> --
> Michael

Sorry, I have attached wrong patch.

Attachment Content-Type Size
v5_bt_split.patch text/plain 5.8 KB

In response to

Responses

Browse pgsql-hackers by date

  From Date Subject
Next Message Michael Paquier 2025-09-25 06:42:33 Re: Get rid of pgstat_count_backend_io_op*() functions
Previous Message Dilip Kumar 2025-09-25 06:23:45 Re: Proposal: Conflict log history table for Logical Replication