Broken MariaDB socket after upgrade to 10.4 asterix version

[annc]

Another service broken by DirectAdmin - month after month. This time, MariaDB 10.4.

On CloudLinux 9 - everything works. On CloudLinux 8 - after updating MariaDB 10.4.34 with Custombuild (to the version with an asterisk - it's a complete mystery what that is), a large number of services stopped working.
They returned an authorization error in MySQL, for example:

*ERROR* v1.7.8.8 2026/02/01 - 11:27:24: Link to database cannot be established: SQLSTATE[HY000] [2002] Can't connect to local MySQL server through socket '/var/lib/mysql/mysql.sock' (2) at line 137 in file classes/db/DbPDO.php

These were PrestaShop, WordPress, and Typo3. Not all of them, quite randomly.

Interestingly, shell authentication attempts using both /run/mysql/mysql.sock and /var/lib/mysql.sock worked CORRECTLY.

A moment of debugging confirmed that simply replacing the systemd mariadb.service unit with the one from before the update made everything work.

There was clearly something messed up with the sockets and their permissions.

Honestly? I can't handle this anymore – updating DA or its components means breaking production and having to debug unusual things.

For now, I'm leaving the service with the old systemd unit – at least it works.

Btw, you can't even do a rollback, because it's the version with * .......

 

[fln]

The version with an asterisk makes sure the main MySQL socket file will be in /run/mysql/mysql.sock instead of /var/lib/mysql.sock.

Please open a support ticket, so we could investigate the situation further. Just reverting to the old service file does not address the root cause of the problem. It will likely resurface again.

Because it is being non deterministic I would guess it has something to do with CageFS.