Some ideas:

• Use a multithreaded approach similar to some of the $source stages (namely, kafka and change streams) where a separate thread is responsible for flushing documents (in this case, to our kafka topic)
• Allow for more configuration options