During start of MariaDB server, slave SQL threads wait for WSREP to reach ready state and only then start applying.

With Galera Cluster and multi-master replication to it set up, multiple threads wait for the state. And then, basically, on MariaDB restarts, only one thread (seems to be more or less random) gets ready state notification. All other slave threads hang waiting (they are retrieving events to relay logs, but SQL threads are shown running and are really not applying anything, exec position does not increase). Any STOP SLAVE command or MariaDB server shutdown hangs then as well waiting for the hung slave threads to stop.