In cluster mode, dynamic updates via the request_key function do not work.
The reason is, that, at the moment, in cluster mode the function is defined only on the manager.
I addressed this in the branch topic/bernhard/ticket1177. Basically, the function request_key is now ignored on worker nodes.
This fixes the problem (if you have a non-cluster-aware script file, that calls the function on both, the master and the worker nodes it will now work flawlessly). However, it is not entirely pretty - there might be use cases where we actually want to allow just worker nodes to trigger data collection. That case just silently fails at the moment.
Not entirely pretty, but it should work for now. If anyone has a better solution, I am open for suggestions
Seth, are you going to merge this?