maria-discuss team mailing list archive

[Marko Mäkelä <marko.makela@xxxxxxxxxxx>]

Hi Jogchum,

On Wed, Oct 12, 2022 at 4:30 PM Jogchum Reitsma
<jogchum.reitsma@xxxxxxxxx> wrote:
[...]
> When trying to start it manually, I got the message
>
> [ERROR] InnoDB: Upgrade after a crash is not supported. The redo log was
> created with MariaDB 10.5.8.
>
> Searching that message, I saw a post on
> https://bbs.archlinux.org/viewtopic.php?id=259364

The advice given there is plain wrong, because it will cause the
contents of the write-ahead log to be ignored. The 10.8 startup
failure already proves that the log is nonempty, that is, some changes
must be applied. It will result in a database whose state does not
correspond to a single given point of time. Some pages are older, some
newer. That approach could cause various crashes, potentially a long
time afterwards. Many of the crashes on corrupted InnoDB data were
recently fixed in MDEV-13542.

I changed the InnoDB redo log format in MariaDB Server 10.8 in
MDEV-14425. At the same time, in
https://jira.mariadb.org/browse/MDEV-27199 I implemented a check that
prevents corruption if such incorrect and dangerous "advice" is being
followed.

The correct advice is to install MariaDB Server (any version between
10.5 and 10.7 should do), shut down the database, and then upgrade to
10.8.

Before 10.5 (MDEV-12353), the last time the redo log format was
changed was in MariaDB Server 10.2. If you got that error message when
attempting to upgrade from something older than 10.5, then you should
use the exact same major version of MariaDB to perform the shutdown,
because there were some changes to the redo log format in each major
release between 10.1 and 10.5.

I hope that someone can post a correction to that forum.

With best regards,

Marko
-- 
Marko Mäkelä, Lead Developer InnoDB
MariaDB Corporation