Content-Type: text/plain;
	charset=utf-8
Mime-Version: 1.0 (Mac OS X Mail 13.4 \(3608.120.23.2.7\))
Subject: Re: reporting TID/table with corruption error
From: Mark Dilger <mark.dilger@enterprisedb.com>
In-Reply-To: <924415DA-A514-4B67-AC66-E3D6C5F13F89@yandex-team.ru>
Date: Thu, 19 Aug 2021 10:49:09 -0700
Cc: Peter Geoghegan <pg@bowt.ie>,
 Alvaro Herrera <alvherre@alvh.no-ip.org>,
 Pg Hackers <pgsql-hackers@lists.postgresql.org>
Content-Transfer-Encoding: quoted-printable
Message-Id: <F412B035-AF83-40C6-BA0A-4C7AA810091E@enterprisedb.com>
References: <202108191637.oqyzrdtnheir@alvherre.pgsql>
 <CAH2-WzmCLNK09r_C+A9ZxmmdZ2AjgDGyCJjVYWBun2LUws_gJw@mail.gmail.com>
 <924415DA-A514-4B67-AC66-E3D6C5F13F89@yandex-team.ru>
To: Andrey Borodin <x4mmm@yandex-team.ru>
Archived-At: 
 <https://www.postgresql.org/message-id/F412B035-AF83-40C6-BA0A-4C7AA810091E%40enterprisedb.com>
Precedence: bulk


> On Aug 19, 2021, at 10:17 AM, Andrey Borodin <x4mmm@yandex-team.ru> =
wrote:
>=20
> It would be great to see relation, block, offset, xmin\xmax and, =
probably, flags whenever ERRCODE_DATA_CORRUPTED\ERRCODE_INDEX_CORRUPTED =
is used. Iif it's possible to extract this information, of cause. This =
is needed especially in amcheck functions.

blockno, offnum and attnum are already included in every result for =
amcheck functions over heap relations, though attnum may be null if the =
corruption is not specific to any particular column.

xmin, xmax and various flags may occur in the corruption message if they =
are relevant, but they are not always present.

There was some disagreement during the development of verify_heapam on =
this point.  We went with the idea that the user could find and inspect =
the corrupt data with another tool if they had the (blockno, offnum, =
attnum) information.  As such, it wasn't necessary to include all the =
data in the output.

It shouldn't be too complicated to have a second interface that returns =
all of the 23 byte main table tuple header information and also the 23 =
byte toast tuple header (when relevant) along with the corruption =
message.  The guts of the corruption check would be shared between the =
two interfaces.  I haven't tried writing a patch yet, but it seems the =
patch shouldn't be terribly complicated.

=E2=80=94
Mark Dilger
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company