At current implementation Broker spawns a process for the adapter to serve each event. If the adapter for some reason blocks and never exists, the broker will endlessly wait for it. So, it’s reasonable that broker should limit the duration that an adapter can run in order to process an event. In case that the limit is exceeded, broker should kill the spawned process.
This time limit could be configurable per adapter. I would propose that 60 minutes would be a reasonable duration for this.