Message-ID: <f2d1846e-04ac-43f5-9fdb-80d8e7daf49d@iki.fi>
Date: Fri, 13 Mar 2026 16:32:10 +0200
MIME-Version: 1.0
User-Agent: Mozilla Thunderbird
Subject: Re: Rework SLRU I/O errors handle
From: Heikki Linnakangas <hlinnaka@iki.fi>
To: Maxim Orlov <orlovmg@gmail.com>
Cc: =?UTF-8?Q?=C3=81lvaro_Herrera?= <alvherre@kurilemu.de>,
 Postgres hackers <pgsql-hackers@lists.postgresql.org>
References: <202603061507.6ynh6qimlcma@alvherre.pgsql>
 <74f5dc14-29fd-46d9-ba8b-2844328d7bdf@iki.fi>
 <aa7908db-b88e-4bb6-a2bc-5b3cee9a28cb@iki.fi>
 <a965875a-34ec-4145-98f1-3e27b8ca028f@iki.fi>
 <CACG=ezaUzQqyXO4Oy6_=7cs4svbL2VZqa5VB0v3Yv81_9Y0Z7g@mail.gmail.com>
 <511e3a69-da70-4bb2-9185-c93d3af8fdf5@iki.fi>
Content-Language: en-US
In-Reply-To: <511e3a69-da70-4bb2-9185-c93d3af8fdf5@iki.fi>
Content-Type: text/plain; charset=UTF-8; format=flowed
Content-Transfer-Encoding: 8bit
Archived-At: 
 <https://www.postgresql.org/message-id/f2d1846e-04ac-43f5-9fdb-80d8e7daf49d%40iki.fi>
Precedence: bulk

On 11/03/2026 12:51, Heikki Linnakangas wrote:
> On 06/03/2026 19:08, Maxim Orlov wrote:
>> I don't know what's happening with my mail, but I haven't
>> received the previous letters.
>>
>> Anyway, v4 looks good to me.
>> Perhaps the extra double line following clog_errdetail_for_io_error
>> is unnecessary? But as always, to your taste.
> 
> Thanks. I did one more iteration on this: I realized that the error we 
> now printed for errors on pg_multixact/members always printed the 
> failing offset, whereas before this patch we usually printed the failing 
> *multixid* that the member is part of. Printing the multixid might 
> actually be more useful; the offset can more easily be deduced from the 
> segment filename and physical offset that is printed anyway, but it's 
> harder to know which multixid it belongs to. This printing the 
> originating multixid seems useful. If things go badly wrong and you need 
> to do manual debugging of a corrupted database, the multixid can more 
> easily be compared with the xmax in the heap and with pg_waldump output, 
> for example.
> 
> We can print both, per attached, which is even better. This is perhaps 
> overkill, but then again, if you hit an error like this, you really 
> appreciate any extra information you can get.

I hear no objections, so committed that. Thanks!

- Heikki