OpenBSD

What do you have in the cnf file(s)? Is it reproducible if you run with --no-defaults?
I don't have OpenBSD handy, I tried to run a clean installation of 10.6.7 on FreeBSD 13, didn't get an error, but of course there could be various system- or configuration-related differences.

Nothing special in the default minimal config file that is installed.

[client-server]

#socket=/var/run/mysql/mysql.sock

#port=3306

# This will be passed to all MariaDB clients

[client]

#password=my_password

# The MariaDB server

[mysqld]

# To listen to all network addresses, use "bind-address = *"

bind-address=localhost

# Directory where you want to put your data

#data=/var/mysql

# This is the prefix name to be used for all log, error and replication files

#log-basename=mysqld

# Logging

#general-log

#slow_query_log

Yes, it is still reproducible with --no-defaults. That definitely won't be useful when it comes to trying to reproduce the issue.

I can provide you with an account on one of my build systems if you want to take a look around.

> I can provide you with an account on one of my build systems if you want to take a look around
Yes, please do.

With the provided access, I was able to figure why the problem occurs. It is a regression indeed.

The patch below introduced the usage of "group" option to mysql_install_db

commit 93a5fb00252c54cad2844b385808a1c6ff0037eb

Author: Monty

Date:   Thu Jan 27 14:43:21 2022 +0200

    MDEV-27477 Remaining SUSE patches for 10.2+

    This patch let's you specify not only user to use but also group that

    MariaDB should use.

If the value is set, the script uses it in chown operations, but it also adds it to the bootstrap command line, like it did before for user value, as in --group=<value>.

However, there is no such option as group for mariadbd, so instead it takes it as a prefix of --group-concat-max-len.

By sheer luck, in standard MariaDB installations the value would be either none or mysql. If mysql is provided, it is treated as M suffix, thus the value becomes 0m and then adjusted to 4 (with warnings, but the warnings would be usually silenced).

However, OpenBSD uses group/user _mysql, not mysql, and it even sets it in the modified version of mariadb-install-db by default, so it's always there. The underscore is not a valid suffix for a numeric option, so the server cannot parse the presumed group-concat-max-len value and fails to start.

$ bin/mysqld --group=mysql

2022-02-22 18:47:47 139798922537856 [Note] Using unique option prefix 'group' is error-prone and can break in the future. Please use the full name 'group_concat_max_len' instead.

2022-02-22 18:47:47 139798922537856 [Warning] option 'group_concat_max_len': unsigned value 0 adjusted to 4

$ bin/mysqld --group=_mysql

2022-02-22 18:48:11 139757314038656 [Note] Using unique option prefix 'group' is error-prone and can break in the future. Please use the full name 'group_concat_max_len' instead.

Unknown suffix '_' used for variable 'group_concat_max_len' (value '_mysql')

2022-02-22 18:48:11 139757314038656 [Warning] option 'group_concat_max_len': unsigned value 0 adjusted to 4

2022-02-22 18:48:11 139757314038656 [ERROR] bin/mysqld: Error while setting value '_mysql' to 'group_concat_max_len'

The probable fix is not to add the --group option to the server command line, but instead only use it for ownership operations.

Thank you for looking into this. OpenBSD has had a very minimal patch for group handling which I re-based on that commit, but all it was was just the command-line handling for --group passed to mariadb-install-db and changing the chown operation.

Right. While I didn't try it in your environment, I think you should be able to work around the problem by dropping this part from the script

if test -n "$group"

then

  args="$args --group=$group"

fi

It looks like that piece as well as the line..

GROUP_OPTION="--group=$group"

That was added to mysqld_safe.sh which should be removed. At the moment it does nothing since the shell script was not changed further down where USER_OPTION is handled.

I have now removed sending the --group option to the server.
I have kept the old code (partly in comments) for a day when we want to add a --group option
to the server to force it to use a specific group for created files.

Bug fix was trivial, most time spent on reading up and discussions