The Cluster Edition allows you to configure one or two interconnected networks between instances in the cluster. Although one interconnected network is sufficient, two interconnected networks allow for a redundant and more robust cluster. The interconnected networks form a series of logical links between every instance in the cluster. These links send messages between instances and are monitored by the Cluster Edition. If any failures are detected, the Cluster Edition reroutes traffic over the alternate networks between the instances.
An interconnection link may fail for various reasons: a physical failure such as a cable that is disconnected or broken, a power supply failure, such as a piece of network infrastructure equipment, or a software failure within the networking stack. The Cluster Edition detects these failures by monitoring the traffic flow between instances. Each instance monitors the messages sent over the various links. A link is considered operational as long as there are incoming messages.
An instance may not always send messages because the node has failed or the link is down. If a particular instance in the cluster is quiet, the Cluster Edition initiates an active probing mechanism to determine if the node supporting the instance can be contacted using the network link. This mechanism ensures that periods of inactivity do not falsely trigger a link failure and subsequent switching event.
If a link is flagged as inoperable, periodic attempts are made to establish if the link has been restored so normal operations can be resumed without manual intervention.
Link monitoring is automatically implemented when more
than one network is defined in the Cluster Edition.
