But if you look at the picture at the first page, you see the whole principle. There is a job queue, where all applications can register new jobs (with a timestamp, an identification and a “worker queue”). The job gets stored and an EJB polls this table. If a job is expired, it send it to its “worker queue”. The MDB at this queue is able to read the identification and can perform a task for it. That’s it

The second part covers the ability for a failover in a cluster environment.