Set up Galera arbitrator

The size of a cluster increases when a node joins the cluster and decreases when a node leaves. A cluster reacts to replication problems with inconsistency voting. The size of the cluster determines the required votes to achieve a quorum. If a node no longer responds and is disconnected from the cluster the remaining nodes vote. The majority of the nodes that vote are considered to be in the cluster.

The arbitrator is important if you have an even number of nodes remaining in the cluster. The arbitrator keeps the number of nodes as an odd number, which avoids the split-brain situation.

A Galera Arbitrator is a lightweight member of a Percona XtraDB Cluster. This member can vote but does not do any replication and is not included in flow control calculations. The Galera Arbitrator is a separate daemon called garbd. You can start this daemon separately from the cluster and run this daemon either as a service or from the shell. You cannot configure this daemon using the my.cnf file.

Note

For more information on how to set up a cluster you can read in the Configuring Percona XtraDB Cluster on Ubuntu or Configure on Red Hat Enterprise Linux manuals.

Installation

Galera Arbitrator does not need a dedicated server and can be installed on a machine running other applications. The server must have good network connectivity.

Run the following commands as root.

Galera Arbitrator can be installed from Percona’s repository on Debian/Ubuntu distributions with the following command:

apt install percona-xtradb-cluster-garbd

Galera Arbitrator can be installed from Percona’s repository on RedHat or derivative distributions with the following command:

yum install percona-xtradb-cluster-garbd

On RHEL 8 and later, you can use dnf install percona-xtradb-cluster-garbd instead.

Start garbd and configuration

Note

On Percona XtraDB Cluster 8.4, SSL is enabled by default. To run the Galera Arbitrator, you must copy the SSL certificates and configure garbd to use the certificates.

You must specify a cipher. This document uses AES128-SHA256 as an example. Choose a cipher that matches your security policy and OpenSSL configuration. If no cipher is specified, startup can fail with a “Terminate called after throwing an instance of ‘gnu::NotSet’” message.

For more information, see socket.ssl_cipher

When starting from the shell, you can set the parameters from the command line or edit the configuration file. This is an example of starting from the command line:

garbd --group=<CLUSTER_NAME> \
--address="gcomm://<NODE1_IP>:4567,<NODE2_IP>:4567,<NODE3_IP>:4567" \
--option="socket.ssl=YES; socket.ssl_key=<SSL_KEY_PATH>; \
socket.ssl_cert=<SSL_CERT_PATH>; \
socket.ssl_ca=<SSL_CA_PATH>; \
socket.ssl_cipher=AES128-SHA256"

To avoid entering the options each time you start garbd, edit the options in the configuration file. To configure Galera Arbitrator on Ubuntu/Debian, edit the /etc/default/garb file. On RedHat or derivative distributions, the configuration can be found in /etc/sysconfig/garb file.

Replace placeholder values such as <CLUSTER_NAME>, <NODE1_IP>, <NODE2_IP>, <NODE3_IP>, <SSL_KEY_PATH>, <SSL_CERT_PATH>, and <SSL_CA_PATH> with values from your environment.

After installation, before you add your cluster settings, /etc/default/garb (Debian/Ubuntu) from Percona XtraDB Cluster 8.4 packages matches the following template:

# Copyright (C) 2012 Codership Oy
# This config file is to be sourced by garb service script.

# REMOVE THIS AFTER CONFIGURATION

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

# Galera cluster name, should be the same as on the rest of the nodes.
# GALERA_GROUP=""

# Optional Galera internal options string (e.g. SSL settings)
# see http://galeracluster.com/documentation-webpages/galeraparameters.html
# GALERA_OPTIONS=""

# Log file for garbd. Optional, by default logs to syslog
# LOG_FILE=""

Add the parameter information about the cluster. For this document, we use the cluster information from Configuring Percona XtraDB Cluster on Ubuntu.

Note

Please note that you need to remove the # REMOVE THIS AFTER CONFIGURATION line before you can start the service.

This config file is to be sourced by garb service script.
A comma-separated list of node addresses (address[:port]) in the cluster
GALERA_NODES="<NODE1_IP>:4567,<NODE2_IP>:4567,<NODE3_IP>:4567"
Galera cluster name, should be the same as on the rest of the nodes.
GALERA_GROUP="<CLUSTER_NAME>"
Optional Galera internal options string (e.g. SSL settings)
see http://galeracluster.com/documentation-webpages/galeraparameters.html
GALERA_OPTIONS="socket.ssl_key=<SSL_KEY_PATH>;socket.ssl_cert=<SSL_CERT_PATH>;socket.ssl_ca=<SSL_CA_PATH>;socket.ssl_cipher=AES128-SHA256"
Log file for garbd. Optional, by default logs to syslog
LOG_FILE="/var/log/garbd.log"

You can now start the Galera Arbitrator daemon (garbd). Run the following commands as root.

Note

The systemd service name is garb, while the daemon binary name is garbd.

On Debian or UbuntuOn Red Hat Enterprise Linux

systemctl start garb

Expected output

# No output on success

systemctl start garb

Expected output

# No output on success

Additionally, you can check the arbitrator status by running:

On Debian or Ubuntu

systemctl is-active garb

Expected output

active

Verify arbitrator participation

1. Check the arbitrator service state:

   systemctl is-active garb
   

   Expected output

   active
   

2. Check recent arbitrator logs for successful cluster communication:

   journalctl -u garb --no-pager -n 20
   

   Expected output

   ... WSREP: ...
   ... connected ...
   

3. On a Percona XtraDB Cluster node, confirm that the cluster remains in Primary state:

   mysql -e "SHOW STATUS LIKE 'wsrep_cluster_status';"
   

   Expected output

   +----------------------+---------+
   | Variable_name        | Value   |
   +----------------------+---------+
   | wsrep_cluster_status | Primary |
   +----------------------+---------+
   

Troubleshooting

• gnu::NotSet at startup: set socket.ssl_cipher in GALERA_OPTIONS.

• Arbitrator cannot join cluster: verify GALERA_NODES format (IP:4567 entries separated by commas, no missing ports).

• No connectivity to cluster nodes: verify firewall and routing for port 4567 between arbitrator and all cluster nodes.

• Service starts then exits: review journalctl -u garb --no-pager and validate certificate file paths in GALERA_OPTIONS.

On Red Hat Enterprise Linux

systemctl is-active garb

Expected output

active