...
This schema presents a problem if all database servers are down. In this case one server (preferrebly the last one to go down) should be forced to start first, and then the rest will load their databases from this node when they start. To force a node to start you should edit "/etc/mysql/my.cnf" so that the line listing other nodes in the cluster lists no nodes: "wsrep_cluster_address = gcomm://192.168.254.2,192.168.254.8,192.168.254.9,192.168.254.10 -> wsrep_cluster_address = gcomm://". After this edit a normal "systemctl start mysql" will work on this node. Next you should start the rest of the servers before you run puppet on the first node to revert your configchange.
Backups
The mysql servers are regularly dumping the content of their databases to "/var/backups/MysqlDump*". Old backups are automaticly deleted to prevent full disks.