So if we tie manager2worker_events and worker2manager_events with input-framework, the events won't kick in because cluster communication hasn't been setup while input-framework finishes reading the data.
Due to this, we have to either introduce an artificial latency in reading from input-framework or we have unpredictable data in the tables.
We need is an event cluster::fully_initialized which can substitute for the artificial latency/adhoc timer to trigger reading a file
Use case: reading 1/2 million IPs from a file on manager and sending them to a table on all workers using worker2manager_events
I can provide sample scripts which showcase the problem, if desired.
Just to add to this - having this would make some of the cluster tests so much easier . At the moment, all of them implement exactly this by themselves by counting the number of connected nodes.