In an event driven system you can start with some safe assumptions:

* Events always publish or they error, so always plan on a global and robust error handling approach.

* Just because an event, or rather its assigned handler, always publishes does not mean it’s either predictable, sequential, or free of side effects. You have to let that tight control go and instead test for results, not execution.

* There are, roughly speaking, only two kinds of events: local and network. Local events may originate from the user or the system but their stack trace is generally reliable and complete, but not so with network events. This becomes more true when the given application is connected peer to peer and thus roles/identity are not clear/distinct.