There is always one observer which upon a predefined event performs some tasks.
In your case, that observer is started, earlier than it should, or at least before all the actual requirements are fulfilled.
I would search to find which is that observer (the stack trace would easily reveal that) and I would try to express the dependency on that observer (e.g. by having it observer the configuration object). If this doesn't solve the issue I would create and fire a new event that will signal that all requirements are fullfilled.
In cube kubernetes, we do something similar: "When the configuration and the logger are available we produce a session and then we fire the start event".
Custom events and observes sounds like a feasible thing to do.
In order to see exact events/observers flow you can also enable debug mode and it will be logged at runtime. Simply set -Darquillian.debug=true