Starting with version 0.8.2 Galera comes with an “arbitrator”
garbd, a stateless daemon which acts as a lightweight group member and can serve two purposes:
As a Galera cluster member arbitrator accepts all Galera parameters except those prefixed by “replicator.”.
This is mostly useful for symmetric 2-datacenter setups:
In this example, if one of the data centers fails or loses WAN connection, the node that sees arbitrator (and therefore sees clients) will continue the operation.
Note that at the moment
garbd needs to see all replication traffic (although it does not store it anywhere), so placing it in a location with poor connectivity to the rest of the cluster may lead to poor cluster performance.
Arbitrator failure does not affect cluster operation and a new instance can be reattached to cluster at any time. There can be several arbitrators in the cluster.