We are developing an application that will be deployed to a WAS cluster. The app consumes messages off a number of JMS (Websphere MQ) queues and writes results to a single database.
However, we have a requirement that deep within the app all messages for a given ID are processed one at a time. We currently enforce this using a synchronised collection but this will not work in the cluster.
So is there a way of configuring the application/cluster/activation-specification to achieve this?
The first thing to realize is that applications must be designed to eliminate message affinities in order to scale horizontally. There's a limit to how much the messaging transport can be expected to do to overcome the application's design constraints. Processing which alters message delivery routes, sequence or destination based on message payload contents is properly a job for an ESB rather than a transport. The optimal solution is to redesign the application with an architecture appropriate for horizontal scaling.
That said, one way to do this is to use message groups. The application instance looking for a new group should ignore any messages in which the JMSXGroupID is not 1. The applications sending messages must be prepared to set the group ID and sequence numbers accordingly.
Another way to accomplish this is by intercept-and-dispatch. A program reads the main queue looking for new instances of an ID. When it sees a new ID, it moves that one message to a separate dispatch queue. The consuming application instances all read the dispatch queue looking for work. Since the IDs in this queue are guaranteed to be unique, each Id is assigned to a specific app instance. The app then opens the primary queue using a selector for the ID to be processed. Producer applications must set the ID as a message property or in the Correlation ID field to make the selection possible but they do not need to manage groups.
I can tell that both you and the previous responder know a lot more about this than I do, but just in case this helps...
We have an MDB that we want to process only one request at a time across the whole cluster, so we set the "Maximum concurrent endpoints" in the Activation Specification to 1.
So maybe you could use a different destination for this particular ID or have the initial destination's logic re-queue the ID to this second destination?
