Yes, we're doing this when all MariaDB containers are down.
But we still curious about this problem. Ordering to shutdown then boot up the MariaDB cluster still can caused this issue.
My initial guess is that the docker is startup earlier than network, caused they can't connect each other.
But it still a guess because the connection usually back if restart container manually. At least this is what we solve if the service can't connect to DB or AMQP but both them are fine.
Perhaps I may try this. For now, it seems like using mariadb_recovery is the only way to let MariaDB back online if reboot the whole cluster right?