With Galera, transactions are confirmed across all nodes before they are committed. If a transaction fails on a node, that node is immediately removed from the cluster. In other words, Galera replication is synchronous. You never lose a transaction – there is no lag (and Galera’s row-based replication is around 5x faster).

Galera Cluster is an insider

MySQL MMM is an outsider – it’s “dumb” as far as what’s actually going on with the server. It does one kind of check and that’s all it knows how to react to.

Galera Cluster is an “insider” and can therefore be smarter about the internal state of each node and do the right thing without manual intervention (a node being synced or unsynced, for example, becoming a donor, etc – it’s all automatic).

via From MySQL+MMM to MariaDB+Galera Cluster: A High Availability Makeover.