Hi, I currently run 1 bolt per executor but kafka producer is IO bound and does not need much cpu. I believe Kafka API creates one thread pool per JVM so all communication is asynchronous. In this scenario, the only logic that should execute on the Storm thread is tuple to kafka message mapping and callbacks. So would it be better to scale by increasing the number of tasks per bolt and have 1 bolt per worker?
Please share your experience. Thank you, Yarik.
