Configuring the management concentrator

You can receive notifications by listening to all of the servers directly or through the management concentrator, which is hosted on one or more catalog servers.

About this task

Using the management concentrator reduces the number of JMX listeners to manage. You only need to maintain a listener to the catalog servers in the domain. The management concentrator acquires notifications and relays them to the JMX listener. If the notification traffic is too high between the container servers and catalog servers, the management concentrator can stop listening to notifications. You can filter the set of notifications that are listened for in the catalog server properties file, or you can change the properties at run time using the xscmd utility.

There are two notification groups provided: Basic and All. The Basic notification group includes notification of the following actions:

Server start and stop
Quorum changes
FFDC generation
Warning or Error messages generated in the JVM logs

The All group includes all notifications. You can filter notifications by including or excluding groups of notifications. For example, only include the container reconnect event. If one notification is too noisy, then the include group can be All and the exclude group can be a single notification. See the following notifications that you can filter:

Descriptive name	Filter name	Description
Container reconnect	com.ibm.websphere.objectgrid.server.reconnect	If a container server becomes disconnected from the data grid, WebSphere eXtreme Scale attempts to reconnect those container servers.
Multi-master replication (MMR) replication	com.ibm.websphere.objectgrid.replication.error com.ibm.websphere.objectgrid.replication.recovery	When a primary shard failed to replicate with the foriegn primary shard. The recovery notification event happens when the foreign primary shard did not move, for example in a network outage.
CPU starvation	com.ibm.websphere.objectgrid.server.busy	When a server experiences a CPU starvation event.
Zone failure	com.ibm.websphere.objectgrid.unreachable.zone	When an entire zone fails.
Routing failure events	com.ibm.websphere.objectgrid.shard.unreachable com.ibm.websphere.objectgrid.container.unreachable.clientreport com.ibm.websphere.objectgrid.unreachable.catalogconfirmed com.ibm.websphere.objectgrid.partition.available	When a shard fails and a client receives TargetNotAvailableExceptions.

Procedure

Connect with the management concentrator, which is hosted on the catalog server.

The management concentrator collects all notifications from the container servers. It reduces the number of connections that need to be maintained from JMX listeners.

The management concentrator is available on the catalog servers from the ManagementConcentratorMBean MBean. It can be accessed from the catalog server JMXService port. Use the following code example to connect to the catalog server with the management concentrator:

String serviceURL = "service:jmx:rmi:///jndi/rmi://" + hostName + ":" + JMXPort + "/objectgrid/MBeanServer";
JMXServiceURL jmxUrl = new JMXServiceURL(serviceURL);
JMXConnector jmxCon = JMXConnectorFactory.connect(jmxUrl);
MBeanServerConnection catalogServerConnection = jmxCon.getMBeanServerConnection();

Set<ObjectName> placementSet = catalogServerConnection.queryNames(new ObjectName(
"com.ibm.websphere.objectgrid" + ":*,type=ManagementConcentrator"), null);
ObjectName managementConcentrator = (ObjectName) placementSet.iterator().next();
NotificationFilterSupport filterSupport = null;
catalogServerConnection.addNotificationListener(managementConcentrator, this, filterSupport, null);

Optional: Limit the number of notifications that are received by the management concentrator by setting the include or exclude policy for the management concentrator.
1. Set the includeManagementConcentratorNotifications and excludeManagementConcentratorNotifications properties in the catalog server properties file.
2. Use the xscmd -c setManagementConcentratorState include and exclude options.