Read/Write Splitting with Galera Cluster
Read/Write Splitting with Galera Cluster
Environment & Solution Space
The object of this tutorial is to have a system that appears to the clients of MariaDB MaxScale as if there is a single database behind MariaDB MaxScale. MariaDB MaxScale will split the statements such that write statements will be sent to the current write-master server in the Galera cluster and read statements will be balanced across the rest of the nodes.
Setting up MariaDB MaxScale
The first part of this tutorial is covered in MariaDB MaxScale Tutorial. Please read it and follow the instructions for setting up MariaDB MaxScale with the type of cluster you want to use.
Once you have MariaDB MaxScale installed and the database users created, we can create the configuration file for MariaDB MaxScale.
Creating Your MariaDB MaxScale Configuration
MariaDB MaxScale configuration is held in an ini file that is located in the file maxscale.cnf in the directory /etc, if you have installed in the default location then this file is available in /etc/maxscale.cnf. This is not created as part of the installation process and must be manually created. A template file does exist within the /usr/share/maxscale directory that may be use as a basis for your configuration.
A global, maxscale, section is included within every MariaDB MaxScale configuration file; this is used to set the values of various MariaDB MaxScale wide parameters, perhaps the most important of these is the number of threads that MariaDB MaxScale will use to execute the code that forwards requests and handles responses for clients.
[maxscale] threads=4
The first step is to create a service for our Read/Write Splitter. Create a section in your MariaDB MaxScale configuration file and set the type to service, the section names are the names of the services themselves and should be meaningful to the administrator. Names may contain whitespace.
[Splitter Service] type=service
The router for we need to use for this configuration is the readwritesplit module, also the services should be provided with the list of servers that will be part of the cluster. The server names given here are actually the names of server sections in the configuration file and not the physical hostnames or addresses of the servers.
[Splitter Service] type=service router=readwritesplit servers=dbserv1, dbserv2, dbserv3
The final step in the service sections is to add the username and password that will be used to populate the user data from the database cluster. There are two options for representing the password, either plain text or encrypted passwords may be used. In order to use encrypted passwords a set of keys must be generated that will be used by the encryption and decryption process. To generate the keys use the maxkeys command and pass the name of the secrets file in which the keys are stored.
% maxkeys /var/lib/maxscale/.secrets %
Once the keys have been created the maxpasswd command can be used to generate the encrypted password.
% maxpasswd plainpassword 96F99AA1315BDC3604B006F427DD9484 %
The username and password, either encrypted or plain text, are stored in the service section using the user and passwd parameters.
[Splitter Service] type=service router=readwritesplit servers=dbserv1, dbserv2, dbserv3 user=maxscale passwd=96F99AA1315BDC3604B006F427DD9484
This completes the definitions required by the service, however listening ports must be associated with the service in order to allow network connections. This is done by creating a series of listener sections. This section again is named for the convenience of the administrator and should be of type listener with an entry labeled service which contains the name of the service to associate the listener with. A service may have multiple listeners.
[Splitter Listener] type=listener service=Splitter Service
A listener must also define the protocol module it will use for the incoming network protocol, currently this should be the MariaDBClient protocol for all database listeners. The listener may then supply a network port to listen on and/or a socket within the file system.
[Splitter Listener] type=listener service=Splitter Service protocol=MariaDBClient port=3306 socket=/tmp/ClusterMaster
An address parameter may be given if the listener is required to bind to a particular network address when using hosts with multiple network addresses. The default behavior is to listen on all network interfaces.
The next stage is the configuration is to define the server information. This defines how to connect to each of the servers within the cluster, again a section is created for each server, with the type set to server, the network address and port to connect to and the protocol to use to connect to the server. Currently the protocol module for all database connections in MariaDBBackend.
[dbserv1] type=server address=192.168.2.1 port=3306 protocol=MariaDBBackend [dbserv2] type=server address=192.168.2.2 port=3306 protocol=MariaDBBackend [dbserv3] type=server address=192.168.2.3 port=3306 protocol=MariaDBBackend
In order for MariaDB MaxScale to monitor the servers using the correct monitoring mechanisms a section should be provided that defines the monitor to use and the servers to monitor. Once again a section is created with a symbolic name for the monitor, with the type set to monitor. Parameters are added for the module to use, the list of servers to monitor and the username and password to use when connecting to the the servers with the monitor.
[Galera Monitor] type=monitor module=galeramon servers=dbserv1, dbserv2, dbserv3 user=maxscale passwd=96F99AA1315BDC3604B006F427DD9484
As with the password definition in the server either plain text or encrypted passwords may be used.
This monitor module will assign one node within the Galera Cluster as the current master and other nodes as slave. Only those nodes that are active members of the cluster are considered when making the choice of master node. Normally the master node will be the node with the lowest value of the status variable, WSREP_LOCAL_INDEX. When cluster membership changes a new master may be elected. In order to prevent changes of the node that is currently master, a parameter can be added to the monitor that will result in the current master remaining as master even if a node with a lower value of WSREP_LOCAL_INDEX joins the cluster. This parameter is called disable_master_failback.
[Galera Monitor] type=monitor module=galeramon disable_master_failback=1 servers=dbserv1, dbserv2, dbserv3 user=maxscale passwd=96F99AA1315BDC3604B006F427DD9484
Using this option the master node will only change if there is a problem with the current master and never because other nodes have joined the cluster.
The final stage in the configuration is to add the option service which is used by the maxadmin command to connect to MariaDB MaxScale for monitoring and administration purposes. This creates a service section and a listener section.
[CLI] type=service router=cli [CLI Listener] type=listener service=CLI protocol=maxscaled socket=default
Starting MariaDB MaxScale
Upon completion of the configuration process MariaDB MaxScale is ready to be started for the first time. This may either be done manually by running the maxscale command or via the service interface.
% maxscale
or
% service maxscale start
Check the error log in /var/log/maxscale to see if any errors are detected in the configuration file and to confirm MariaDB MaxScale has been started. Also the maxadmin command may be used to confirm that MariaDB MaxScale is running and the services, listeners etc have been correctly configured.
% maxadmin list services Services. --------------------------+----------------------+--------+--------------- Service Name | Router Module | #Users | Total Sessions --------------------------+----------------------+--------+--------------- Splitter Service | readwritesplit | 1 | 1 CLI | cli | 2 | 2 --------------------------+----------------------+--------+--------------- % maxadmin list servers Servers. -------------------+-----------------+-------+-------------+-------------------- Server | Address | Port | Connections | Status -------------------+-----------------+-------+-------------+-------------------- dbserv1 | 192.168.2.1 | 3306 | 0 | Running, Synced, Master dbserv2 | 192.168.2.2 | 3306 | 0 | Running, Synced, Slave dbserv3 | 192.168.2.3 | 3306 | 0 | Running, Synced, Slave -------------------+-----------------+-------+-------------+--------------------
A Galera Cluster is a multi-master clustering technology, however the monitor is able to impose false notions of master and slave roles within a Galera Cluster in order to facilitate the use of Galera as if it were a standard MariaDB Replication setup. This is merely an internal MariaDB MaxScale convenience and has no impact on the behavior of the cluster but does allow the monitor to create these pseudo roles which are utilized by the Read/Write Splitter.
You can control which Galera node is the master server by using the priority mechanism of the Galera Monitor module. For more details, read the Galera Monitor documentation.
% maxadmin list listeners Listeners. ---------------------+--------------------+-----------------+-------+-------- Service Name | Protocol Module | Address | Port | State ---------------------+--------------------+-----------------+-------+-------- Splitter Service | MariaDBClient | * | 3306 | Running CLI | maxscaled | localhost | 6603 | Running ---------------------+--------------------+-----------------+-------+--------
MariaDB MaxScale is now ready to start accepting client connections and routing them to the master or slaves within your cluster. Other configuration options are available that can alter the criteria used for routing, these include monitoring the replication lag within the cluster and routing only to slaves that are within a predetermined delay from the current master or using weights to obtain unequal balancing operations. These options may be found in the MariaDB MaxScale Configuration Guide. More details on the use of maxadmin can be found in the document MaxAdmin - The MariaDB MaxScale Administration & Monitoring Client Application.
