I'm configuring Percona XtraDB Cluster 8 on two nodes, and use a 3rd node to act as an arbiter.

I cannot make up out of the docs whether to add the IP of the arbiter in the galera nodes addresses of the arbiter’s configuration.

# A comma-separated list of node addresses (address[:port]) in the cluster

Also, do I add the IP of the arbiter in the wsrep_clsuter_address of the Galera configuration? Again, it's not described in the docs.


2 Answers 2


While I too could not find a definite answer in the documentation, the following link states:

While Galera Arbitrator does not participate in replication, it does receive the same data as all other nodes.

Which leads me to believe that yes, you would have to add the Arbitrator IPs to all configuration files as if it were another node.

EDIT: This guide seems to indicate the same, although it is not official documentation.

  • Because it says 'does not participate in replication' I'm more eager to believe it does not has to be included. Also, I quickly scanned through the guide you shared. But I do not see the IP of the arbiter added to a config file.
    – Kevin C
    Commented Jul 23, 2020 at 10:57
  • My understanding is, how else is the arbitrator going to get the data if the other nodes don't know about it. Also, in the guide, the arbitrator is node 3. In the last configuration for node1 the address for node3 is included in the wsrep_cluster_address.
    – Flourid
    Commented Jul 24, 2020 at 12:06
  • I am seeing that node3 is just a node in the cluster performing database queries. This makes me believe the adress isn’t required. Also, I have setup a test myself with 2 nodes and 1 arbiter. I am not using the arbiter’s ip in the config, but all works smoothly. The nodes can talk to each other.
    – Kevin C
    Commented Jul 25, 2020 at 13:41

I would suggest adding the IP and port of the garbd node in the wsrep_cluster_address and in the GALERA_NODES parameter. It is not a strict requirement but considering the garbd is still a member of the cluster albeit it is not a data bearing node it would be best to keep all nodes in the gcomm:// list. However, there is no strict requirement but remember that the garbd node also becomes a "relay" in case of a network connection glitch between two data nodes in a 3-node cluster.

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.