Review clustering of JBPM job executor. It is assumed that clustered job executors will not try to execute the same job more than once (to be verified) but can execute differerent jobs in parallel.
The current implementation of background/async WCM submits uses JBPM job executor (as a persistent queue). The serialized nature of submits relies on the the fact that in a non-clustered (single node) env, we only have one job executor thread (which in turn means that long-running submits will block job processing of other tjobs, including WCM content launch and/or other workflow timers).
In a clustered (multi-node) env, it is possible that multiple job executors will try to execute multiple submits in parallel - hence falling back to the previous (before 2.2.3) retry mechanism.
Also requires development and testing of suitable QA scenarios.