Cluster Models

Message Queue supports two clustering models both of which provide a scalable message service, but with each providing a different level of message service availability:

• Conventional broker clusters. A conventional broker cluster provides for service availability. When a broker or a connection fails, clients connected to the failed broker reconnect to another broker in the cluster. However, messages and state information stored in the failed broker cannot be recovered until the failed broker is brought back online. The broker or connection failure can therefore result in a significant delay and in messages being delivered out of order.

• Enhanced broker clusters. An enhanced broker cluster provides for data availability in addition to service availability. When a broker or a connection fails, another broker takes over the pending work of the failed broker. The failover broker has access to the failed broker's messages and state information. Clients connected to the failed broker reconnect to the failover broker. In an enhanced cluster, as compared to a conventional cluster, a broker or connection failure rarely results in significant delays in message delivery and messages are always delivered in order.

Despite the message service availability offered by both conventional and enhanced broker clusters, they do not provide a guarantee against failure and the possibility that certain failures, for example in the middle of a transaction, could require that some operations be repeated. It is the responsibility of the messaging application (both producers and consumers) to handle and respond appropriately to failure notifications from the messaging service.

Conventional and enhanced broker clusters are built on the same underlying infrastructure and message delivery mechanisms. They differ in how brokers in the cluster are synchronized with one another and in how the cluster detects and responds to failures.

The sections that follow first describe the infrastructure and delivery mechanisms common to both clustering models, after which the unique aspects of each model is explained.