Good news: Galera cluster backups can be performed just like regular MySQL backups, using your favourite backup script. And since all the nodes are identical, backuping one node means backuping the whole cluster.
In order to associate global transaction ID with the backup and avoid cluster stalling, the backup should be performed in the same manner as state snapshot transfer between the nodes. For that
For example, running the following command
/usr/bin/garbd --address gcomm://<donor node address>?gmcast.listen_addr=tcp://0.0.0.0:4444 --group <wsrep_cluster_name> --donor <wsrep_node_name on donor> --sst backup
will cause the chosen donor node to run wsrep_sst_backup script and pass the corresponding global transaction ID to it. See SST script interface specification for details.
part is an arbitrary listen socket address that
garbd will have to open for communicating with the cluster. Needs to be specified only if default socket address (0.0.0.0:4567) is already busy.
Currently (2012.09.01) there is a usability issue when garbd exits immediately after making successful SST request with confusing diagnostic that makes it look like a failure. This is not a failure: the backup script is being run by the donor mysqld. Its progress should be monitored in the donor mysqld error log and/or in the script log