mysql-cluster

Installs and configures a MySQL mult-master cluster using Percona XtraDB Cluster. Currently implemented as a wrapper of the percona cookbook, provided as a simplified LWRP.

LWRPs

This cookbook is intended to be consumed through its LWRP, and therefore doesn't include any recipes. Here is an overview of the LWRP provided:

mysql_cluster

Attributes

Name	Description	Type	Required	Default
cluster_name	Name of the db cluster	String	true	N/A
node_ips	Array of db nodes that make up the cluster	Array	true	N/A
bind_address	Address to bind db to.	String	true	N/A
bind_interface	Interface to bind the db server to	String	true	N/A
root_password	Root password for the db	String	true	N/A
debian_password	Debian password for the db	String	true	N/A
bootstrapping	Whether or not to treat the run as bootstrapping	Boolean	false	false
enable_myisam_replication	Whether or not to replicate MyISAM tables	Boolean	false	true

Example:

mysql_cluster 'cluster01' do
  node_ips ['192.168.10.1', '192.168.10.2', '192.168.10.3']
  bind_address '192.168.10.1'
  bind_interface 'eth0'
  root_password 'mysecurepass'
  debian_password 'mysecurepass'
  bootstrapping true
end

Note on Bootstrapping: in order to make bootstrapping easy, we have included a bootstrapping attribute. This affects the node_ips that are used in the cluster address. By default, all node ips are included. If you are bootstrapping, however, the first node will fail to startup because the other nodes aren't running. Normally you would use the bootstrap-pxc option to startup the db service (instead of start):

$ sudo service mysql bootstrap-pxc

You can do this manually and always keep the bootstrapping attribute to false, or you can set it to true and it will use a cluster address that only includes the previous nodes in the cluster:

db01 -> gcomm://
db02 -> gcomm://db01
db03 -> gcomm://db01,db02

As long as you converge the nodes in the order of the node_ips, it should bootstrap successfully. Once you have the whole cluster configured, you will want to run Chef again on all nodes with bootstrapping set to false. This will restore the default behavior, and all nodes will have a full cluster address with all node ips. If you don't do this, you risk nodes failing to rejoin the cluster when they restart (mainly the first node, which has an empty cluster address).