The second node can't join the first node because SST will get killed by systemd after the default timeout hits.
systemctl show mariadb.service | grep Timeout will show timeout set to 1m 30s for startup, but an SST can last hours with large dataset and/or slow disks and/or slow networks.
In fact, it is common for an SST to take several hours in production.
Setting TimeoutSec=0 under Services in the mariadb.service config file under systemd fixes the problem.
Right now, it is impossible to deploy Galera Cluster under 10.2.13 and CentOS 7 unless the above workaround is in place.