Previously the mariadb service worked on this assumption:

• if ~/srv/mariadb exists then its contents are in a valid state
• if it doesn't exist then it must be created and initialized

The initialization was first made in a temporary directory, and then /bin/mv was used to atomically move it to ~/srv/mariadb. This way if the initialization fails in the middle we don't end up with a miss-initialized ~/srv/mariadb directory.

However this approach had some issues: mv only works if ~/srv/mariadb is a simple directory: it fails if it's a symlink, a mount point, or a subvolume (btrfs).

Now the mariadb service works on these assumptions instead:

• if ~/srv/mariadb/mysql exists and ~/srv/mariadb/.slapos_initializing does not exists then its in a valid state
• else ~/srv/mariadb must be emptied (or created) and reinitialized

The initialization now starts by emptying (or creating) ~/srv/mariadb and then creating the marker file ~/srv/mariadb/.slapos_initializing. The marker file is atomically removed at the end of a successful initialization using /bin/rm.