Cluster Display

Performance Manager displays clusters using the new auto-discovery feature.

Display representation of clusters

When monitoring a cluster, Performance Manager accesses the members of the cluster from the Connection Manager on the cluster. When the membership changes, Performance Manager adjusts its representation of the cluster as follows:

Auto-discovery for clusters

When you add a node, the node checks to see if it belongs to a cluster. If it does, then it creates and names a cluster node using the director name for the cluster. If the create succeeds or the cluster node already exists, the cluster node determines which nodes are in the cluster and creates them. If the node already exists it makes them children of itself.

The cluster node then requests cluster information (director name and cluster members) from one of its children. It saves the name of this node when the session is saved. If the cluster node fails to get the cluster information from this child node, it will remove it as described below, and switch to another child for the information. If there are no other children it will still remove the failing child, but continue to attempt to get the information from that node, using the saved node name. Occasionally a node removed from the cluster node in this way is still actually part of the cluster. In that case the cluster node will bring it back in when it gets an updated list of members from one of the other nodes.

If membership changes, the cluster node changes its children as follows:

If the director name changes, the cluster node changes its name to match the new director name. This changes all uses of the old name to the new name in displays and thresholds. Note that this means cluster nodes defined in old sessions will have their names changed to match the director name.

Possible Anomalies:

Director name changes may result in two cluster nodes for the same cluster appearing in Performance Manager. This may happen if attempts to get cluster information from a node occur during the change and a node is removed from the cluster as described above. If the node(s) removed from the cluster notices the new director name before the cluster node notices it, the removed node will create a new cluster node with the new name.

Usually the pre-existing cluster node notices the director name change, and also notices there is already a cluster node with the same name. In that case it does the following:

If the pre-existing node removed all of its children because it could not get information from them, it will continue asking for information from the last node that it polled. If this node never responds, this cluster node will continue to exist without children even if a new cluster node has been created based on information from the other nodes.

Note: To avoid conflicts between group names and cluster node (director) names, do not give group nodes the same names as cluster director names. This interferes with cluster auto-discovery.

For example, if you give a group the same name as a cluster director when a corresponding cluster node does not exist in the session, and then add nodes from that cluster to the session, the cluster nodes will not be created.


Back to the User's Guide home page
Return to:
        Overview         Node Management