Clustering the Message Broker Profile

The following sections give you information and instructions on how to cluster the message broker profile of WSO2 EI.  

The clustering deployment pattern

This pattern has two WSO2 EI nodes that include the Message Broker profile to serve service requests with high availability and scalability. The following image depicts the sample pattern of this clustering deployment scenario.

Creating the databases

All profiles of WSO2 EI uses a database to store information such as user management details and registry data. All nodes in the cluster must use one central database for config and governance registry mounts. You can create the following databases and associated datasources.

WSO2_USER_DB

In addition to the databases listed above, the Message Broker profile of WSO2 EI requires the following broker-specific data:

MB_DB

Follow the steps given below to create the databases required for the Message Broker profile of WSO2 EI. 

1. Download and install the MySQL Server.

2. Download the MySQL JDBC driver.

3. Download and unzip the WSO2 EI binary distribution. 

   Throughout this guide, <EI_HOME> refers to the extracted directory of the WSO2 EI product distribution.

4. Unzip the downloaded MySQL driver, and copy the MySQL JDBC driver JAR (mysql-connector-java-x.x.xx-bin.jar) to the <EI_HOME>/lib/ directory of both the WSO2 EI nodes.

5. Add the following line to the /etc/hosts file to define the hostname for configuring permissions for the new database: <MYSQL-DB-SERVER-IP> carbondb.mysql-wso2.com

   Do this step only if your database is not on your local machine and on a separate server.

6. Execute the following command in a terminal/command window, where username is the username you want to use to access the databases: mysql -u username -p

7. When prompted, specify the password to access the databases.

8. Create the databases using the following commands:

   mysql> create database WSO2_USER_DB;
   mysql> use WSO2_USER_DB;
   mysql> source <EI_HOME>/dbscripts/mysql.sql;
   mysql> grant all on WSO2_USER_DB.* TO regadmin@"carbondb.mysql-wso2.com" identified by "regadmin";
   
   mysql> create database REGISTRY_DB;
   mysql> use REGISTRY_DB;
   mysql> source <EI_HOME>/dbscripts/mysql.sql;
   mysql> grant all on REGISTRY_DB.* TO regadmin@"carbondb.mysql-wso2.com" identified by "regadmin";
   
   mysql> create database REGISTRY_LOCAL1;
   mysql> use REGISTRY_LOCAL1;
   mysql> source <EI_HOME>/dbscripts/mysql.sql;
   mysql> grant all on REGISTRY_LOCAL1.* TO regadmin@"carbondb.mysql-wso2.com" identified by "regadmin";
    
   mysql> create database REGISTRY_LOCAL2;
   mysql> use REGISTRY_LOCAL2;
   mysql> source <EI_HOME>/dbscripts/mysql.sql;
   mysql> grant all on REGISTRY_LOCAL2.* TO regadmin@"carbondb.mysql-wso2.com" identified 
   by "regadmin";

   About using MySQL in different operating systems

   For users of Microsoft Windows, when creating the database in MySQL, it is important to specify the character set as latin1. Failure to do this may result in an error (error code: 1709) when starting your cluster. This error occurs in certain versions of MySQL (5.6.x) and is related to the UTF-8 encoding. MySQL originally used the latin1 character set by default, which stored characters in a 2-byte sequence. However, in recent versions, MySQL defaults to UTF-8 to be friendlier to international users. Hence, you must use latin1 as the character set as indicated below in the database creation commands to avoid this problem. Note that this may result in issues with non-latin characters (like Hebrew, Japanese, etc.). The following is how your database creation command should look.

   mysql> create database <DATABASE_NAME> character set latin1;

   For users of other operating systems, the standard database creation commands will suffice. For these operating systems, the following is how your database creation command should look.

   mysql> create database <DATABASE_NAME>;

9. Create and configure the MB_DB database, which is specific to the message broker profile using the following commands:

   Note that <EI_HOME> , username, and password are the same as those you specified in the previous steps.

   mysql> create database wso2_mb;
   mysql> use wso2_mb;
   mysql> source <EI_HOME>/wso2/broker/dbscripts/mb-store/mysql-mb.sql;
   mysql> grant all on wso2_mb.* TO regadmin@"carbondb.mysql-wso2.com" identified by "regadmin";

Mounting the registry

Add the following configurations to the <EI_HOME>/wso2/broker/conf/registry.xml file of each WSO2 EI node to configure the shared registry database and mounting details. This ensures that the shared registry for governance and configurations (i.e., the REGISTRY_DB database) mounts on both WSO2 EI nodes. 

<dbConfig name="sharedregistry">
<dataSource>jdbc/WSO2RegistryDB</dataSource>
</dbConfig> 

<remoteInstance url="https://localhost:9443/registry">
    <id>instanceid</id>
    <dbConfig>sharedregistry</dbConfig>
    <readOnly>false</readOnly>
    <enableCache>true</enableCache>
    <registryRoot>/</registryRoot>
<cacheId>regadmin@jdbc:mysql://carbondb.mysql-wso2.com:3306/REGISTRY_DB?autoReconnect=true</cacheId>
</remoteInstance>
 
<mount path="/_system/config" overwrite="true">
    <instanceId>instanceid</instanceId>
    <targetPath>/_system/mbconfig</targetPath>
</mount>
 
<mount path="/_system/governance" overwrite="true">
    <instanceId>instanceid</instanceId>
    <targetPath>/_system/governance</targetPath>
</mount>

Configuring a message broker profile node

Do the following configurations for all nodes of your cluster.

1. Configure the datasources to point to the REGISTRY_LOCAL1, WSO2_REGISTRY_DB, and WSO2_USER_DB databases as follows in the <EI_HOME>/wso2/broker/conf/datasources/master-datasources.xml file as follows:

   Replace the username, password, and database URL of your MySQL environment accordingly.

2. Remove or comment out the default H2-based WSO2_MB_STORE_DB and WSO2_CARBON_DB datasource configurations from the <EI_HOME>/wso2/broker/conf/datasources/master-datasources.xml file. 

3. Uncomment or add the following WSO2_MB_STORE_DB configuration in the <EI_HOME>/wso2/broker/conf/datasources/master-datasources.xml file based on your DBMS type.

   Update the JDBC URL to correctly point to your database and enter the username and password for your database user with the proper permissions.

4. Add the following configuration in the <EI_HOME>/wso2/broker/conf/user-mgt.xml file of both nodes to configure the user stores. 

   Enter the datasource information for the user store that you configured in the <EI_HOME>/wso2/broker/conf/datasources/master-datasources.xml file. You can change the admin username and password as well. However, you should do this before starting the server.

   <Configuration>
       <AddAdmin>true</AddAdmin>
       <AdminRole>admin</AdminRole>
       <AdminUser>
           <UserName>admin</UserName>
           <Password>admin</Password>
       </AdminUser>
       <EveryOneRoleName>everyone</EveryOneRoleName>
       <Property name="dataSource">jdbc/WSO2UMDB</Property>
   </Configuration>

5. Uncomment or add the following configuration in the <EI_HOME>/wso2/broker/conf/broker.xml file.  (Enter the message store based on your database type and Andes context store).

6. Add the following thrift-related configurations in the <EI_HOME>/wso2/broker/conf/broker.xml file.

   This file is the root configuration file of brokering. Do the changes you do to this file in all the broker profile nodes. Configure the thriftServerHost value to point to the IP address of the message broker profile node.

   <coordination>
       <nodeID>default</nodeID>
       <thriftServerHost>192.168.0.103</thriftServerHost>
       <thriftServerPort>7611</thriftServerPort>
   	<thriftSOTimeout>0</thriftSOTimeout>
   </coordination>

    Click here for more details on the configurations.

   Configuration	Description
   coordination	This configuration is related to brokering thrift communications.
   nodeID	The node ID of each member in a cluster must be unique. In a clustered deployment, an ID is assigned to each MB node via the cluster node identifier. The node ID specified by this parameter can be used to override the cluster node identifier for this MB node. If the node ID is set to 'default', the node ID will be retrieved using the Hazelcast member ID that is specified in the axis2.xml file as shown below. <parameter name="localMemberHost">192.168.0.102</parameter> Note that when this parameter is set to 'default', the node ID is made up of the 'localMemberHost' and the 'localMemberPort' (defined in the axis2.xml) as shown below. The 'localMemberHost' will be the hostname of the server if the hostname can be mapped with the provided IP address. Otherwise, the 'localMemberHost' will always be the provided IP address. NODE:localMemberHost:localMemberPort
   thriftServerHost	This is a sub-element of the <coordination> tag. WSO2 EI uses Apache Thrift for communications relating to message delivery. Therefore, an Apache Thrift server is started in each brokering node in a clustered deployment. This element should point to the IP address of the Apache Thrift server. This should point to the IP address of the brokering node that hosts the thrift server. The default value for this is localhost. For example, if you are configuring a brokering node hosted in 192.168.0.102 as the thrift server, this value should be 192.168.0.102.
   thriftServerPort	This is another sub-element of the <coordination> tag. This should point to the port of the thrift server in the message broker profile. What is the effective thrift server port ? The broker.xml file indicates 7611 as the default thrift server port. However, the default thrift server port that is effective in the MB profile of WSO2 EI is 7614 (7611 + 3). This is because, by default, the server port of the MB profile of WSO2 EI is offset by 3 in the <EI_HOME>/wso2/broker/conf/carbon. xml file. It is important to note that the effective thrift server port depends upon the default server port offset in the carbon.xml file as well as the thriftServerPort value in the broker.xml file. Be sure to keep the correct thrift server port open in your system to avoid errors. It is recommended to use the same thrift server port for all broker nodes in your cluster.
   thriftSOTimeout	This is used to handle half-open TCP connections between the broker nodes in a cluster. In such situations, the socket may need to have a timeout value to invalidate the connection (in milliseconds). A timeout of zero is interpreted as an infinite timeout. Be sure to set this value to 180000 milliseconds as shown below in order to handle a high number of half-open TCP connections. <thriftSOTimeout>180000</thriftSOTimeout>

7. Edit the <EI_HOME>/wso2/broker/conf/axis2/axis2.xml file as follows to set up the cluster configurations.

   • Enable clustering for this node as follows: <clustering class="org.wso2.carbon.core.clustering.hazelcast.HazelcastClusteringAgent" enable="true">
   • Set the membership scheme to "wka" to enable the well-known address registration method. (This node sends cluster initiation messages to the WKA members): 
     <parameter name="membershipScheme">wka</parameter>
   • Specify the name of the cluster this node will join as follows: <parameter name="domain">wso2.ei.domain</parameter>


   • Specify the hosts to communicate cluster messages as follows: <parameter name="localMemberHost">xxx.xxx.xxx.xx2</parameter>

   • Specify the port to communicate cluster messages as follows: <parameter name="localMemberPort">4100</parameter>

     This port number is not affected by the port offset value specified in the <EI_HOME>/wso2/broker/conf/carbon. xml file. If this port number is already assigned to another server, the clustering framework automatically increments this port number. However, if there are two servers running on the same machine, ensure that a unique port is set for each server.

   • Specify the well-known members as follows: (The port value for the WKA node must be the same value as it's localMemberPort (in this case it is 4100).

     You can also use IP address ranges for the hostname (e.g., 192.168.1.2-10). However, you can define a range only for the last portion of the IP address. Smaller the range, the faster the time it takes to discover members since each node has to scan a lesser number of potential members. The best practice is to add all the members (including itself) in all the nodes to avoid any conflicts in configurations.

     <members>    
         <member>
             <hostName>xxx.xxx.xxx.xx1</hostName>
             <port>4100</port>
         </member>
         <member>
             <hostName>xxx.xxx.xxx.xx2</hostName>
             <port>4200</port>
         </member>
     </members>

8. Enable heartbeat messaging for each of the broker nodes by following the steps given below. This is required for handling TCP connections between the broker nodes and client systems (publishers and subscribers).

   What is heartbeat messaging?

    Click here for more information.

   When a client is connected to the broker, both the broker and the client should be able to detect problem situations where the TCP connection is half open or where the connecting client/broker is unresponsive. This can be achieved by enabling heartbeat messaging between the broker and the client.

   The heartbeat messaging configuration allows both the broker and the client to verify whether the connection is inactive or whether the connecting system (broker or client) is inactive by periodically sending messages to each other.

   For example, you can set the heartbeat delay time to 30 seconds. This means that if the broker (or the client) does not receive the expected response from the other system within 30 seconds, it will send a heartbeat message to check if the connection is inactive or if the other system is inactive. If the connection is inactive or is half-open, there will be a clear indication of connection failure. However, if the connection is active but the other system is unresponsive, the broker or the client will continue to send heartbeat messages (every 30 seconds). By default, the broker and clients are configured to terminate (timeout) the connection after sending two heartbeat messages; however, this time out configuration can be changed.

   Note: In a scenario where the subscriber client is sending the heartbeat message to the broker node, if the connection is found to be broken or if the broker node is inactive, the connection will failover to another broker node in the cluster.

   1. Open the  qpid-config.xml file that is stored in the  <EI_HOME>/wso2/broker/conf/advanced directory.

   2. Set the heartbeat delay as shown below. The recommended heartbeat delay is 30 seconds.

      <heartbeat>
          <delay>30</delay>
          <timeoutFactor>2.0</timeoutFactor>
      </heartbeat>

      The elements in the above configuration are explained below.

      delay	The <delay> element specifies the time interval between heartbeat messages. This is the time that the broker or client will wait to receive a response from the other party. If the response is not received within this time, a heartbeat message is triggered.
      timeoutFactor	The number of heartbeat messages the broker will send to the client before terminating the connection. That is, if the timeoutFactor is 2, the broker will send heartbeat messages every 30 seconds twice, and if a response is not received from the client, the connection will be terminated.

Optional: Configuring Hazelcast properties

WSO2 products use Hazelcast as its default clustering engine. You can configure the hazelcast properties for the product nodes by following the steps given below.

1. Add the following property configurations to the hazelcast.properties file stored in the <EI_HOME>/wso2/broker/conf/ directory.

   #Disabling the hazelcast shutdown hook
   hazelcast.shutdownhook.enabled=false
   #Setting the hazelcast logging type to log4j
   hazelcast.logging.type=log4j

   The above configurations are explained below.

   • Hazelcast shutdown hook: This configuration disables the shutdown hook in hazelcast, which ensures that the hazelcast instance shuts down gracefully whenever the product node shuts down. If the hazelcast shutdown hook is enabled (which is the default behavior of a product), you will see errors such as "Hazelcast instance is not active!" at the time of shutting down the product node: This is because the hazelcast instance shuts down too early when the shutdown hook is enabled.
   • Hazelcast logging type: This configuration sets the hazelcast logging type to log4j, which allows hazelcast logs to be written to the wso2carbon.log file.

2. If you have enabled log4j for hazelcast logging as explained above, be sure to enter the configuration shown below in the log4j.properties file (stored in the  <EI_HOME>/wso2/broker/conf/  directory). This can be used to configure the log level for hazelcast logging. For a clustered production environment, it is recommended to use INFO as the log level as shown below.

   log4j.logger.com.hazelcast=INFO

Optional: Configuring cluster coordination

If rdbmsBasedCoordination enabled is set to true, the Message Broker profile in WSO2 EI introduces cluster coordination through an RDBMS. This means that the coordination between the nodes in a cluster can be managed through an RDBMS, just as message persistence. Therefore, the RDBMS that is connected to the MB nodes in the cluster is, by default, used for message persistence as well as cluster coordination.

Shown below is the configuration in the broker.xml file (stored in the <EI_HOME>/wso2/broker/conf/ directory), which enables RDBMS-based cluster coordination. If required, you can disable the following configuration, which will allow the hazelcast engine to manage cluster coordination. However, note that you need to configure the cluster to handle network partitioning when hazelcast-based cluster coordination is used. 

<rdbmsBasedCoordination enabled="true">
    <!-- Heartbeat interval used in the RDBMS base coordination algorithm in milliseconds -->
    <heartbeatInterval>5000</heartbeatInterval>
    <!-- Time to wait before informing others about coordinator change in milliseconds. This value should be
            larger than a database read time including network latency and should be less than heartbeatInterval -->
    <coordinatorEntryCreationWaitTime>3000</coordinatorEntryCreationWaitTime>
    <!-- Time interval used to poll database for membership related events in milliseconds. -->
    <eventPollingInterval>4000</eventPollingInterval>
</rdbmsBasedCoordination>
<!-- Enabling this will make the cluster notifications such as Queue changes(additions and deletions),
        Subscription changes, etc. sent within the cluster be synchronized using RDBMS. If set to false, Hazelcast
        will be used for this purpose.-->
<RDBMSBasedClusterEventSynchronization enabled="true">
    <!--Specifies the interval at which, the cluster events will be read from the database. Needs to be
            declared in milliseconds. Setting this to a very low value could downgrade the performance where as
            setting this to a large value could increase the time taken for a cluster event to be synchronized in
            all the nodes in a cluster.-->
    <eventSyncInterval>1000</eventSyncInterval>
</RDBMSBasedClusterEventSynchronization>

Optional: Handling network partitioning in Hazelcast-based clustering

Network partitioning in the network used for cluster coordination can sometimes disrupt the cluster. The Message Broker profile of WSO2 EI introduces a new configuration to handle network partitioning that may occur when cluster coordination is managed by the hazelcast engine (as opposed to the RDBMS as explained above).

For example, consider a cluster of three MB nodes that uses two separate networks for cluster coordination and message persistence. The network with the RDBMS will persist messages, subscriptions and queues etc., and the cluster network communicates information about subscriptions, queue additions, and to decide on the coordinator of the cluster.  If one of the nodes in the cluster disconnects from the network used for cluster coordination, that node will separate from the cluster and function as a separate node/cluster (separate partition). That is, the three-node cluster will now be working as two separate clusters. However, the disconnected node will still be up and running, and message persistence will continue uninterrupted through the message persistence network. This will cause inconsistencies in the system because message persistence and cluster coordination will not be synchronized. In such a situation, it is necessary to stop the disconnected node from accepting messages.

The above situation can be prevented in the Message Broker profile by configuring the minimum node count of the cluster.  For example, if the cluster size is 5 and we have configured the minimum node count to 3 nodes, during a network partition (or when nodes crash or shut down) any partition that has 3 or more nodes will keep functioning, while the other partitions will stop processing messages.

Follow the steps given below to enable this feature.

1. Open the <EI_HOME>/wso2/broker/repository/conf/broker.xml file and set the <networkPartitionsDetection> element shown below to enabled="true" . Note that this configuration is disabled by default as shown below.

   <?xml version="1.0"?>
   <networkPartitionsDetection enabled="false">
       <minimumClusterSize>1</minimumClusterSize>
   </networkPartitionsDetection>

2. Update the <minimumClusterSize> property in the above configuration to specify the minimum node count the cluster should maintain in order to operate. Note that if this value should be at least two for cluster coordination to work.  If the number of nodes in the cluster becomes less than the configured value, the cluster will not accept any incoming traffic. That is, all subscriptions will be disconnected.

Testing the cluster

Follow the steps below to test the cluster.

1. Navigate to <EI_HOME>/bin/ via the terminal  and start the Message broker profile:

2. Check for ‘member joined’ log messages in all consoles.

   Additional information on logs and new nodes

   When you terminate one node, all nodes identify that the node has left the cluster. The same applies when a new node joins the cluster. If you want to add another new node, copy existing node without any changes if you are running it on a new server (such as xxx.xxx.xxx.184). If you intend to use the new node on a server where another WSO2 product is running, use a copy of node and change the port offset accordingly in the <EI_HOME>/wso2/broker/conf/carbon.xml file. You also have to change localMemberPort in the <EI_HOME>/wso2/broker/conf/axis2/axis2.xml file if that product has clustering enabled. Also, map all hostnames to the relevant IP addresses when creating a new node. The log messages indicate if the new node joins the cluster.

3. Access the Management Console using the following URL:  https://xxx.xxx.xxx.xx2:443/carbon
4. Sign into the management console using admin as the username and password.
5. Click Node List.
   
   You see that the 2 nodes are up and running successfully. The message store health needs to be Healthy and only one node can be the Coordinator Node. 
   

What's next

If you want to performance tune the WSO2 Message Broker (MB) profile, see Tuning the Message Broker Profile.