Hey. I want build a recommendation system. My idea is to use kafka Spark + Streaming + Spark.MLlib. But some colleagues argue that sparkstreaming is micro-batch and that used Storm, would actually streaming and without lag..
What do you think? Storm in speed layer or sparkStreaming? In my view, an recommendation system needs statefull state, right? In this case we would have to use Storm Trident, which will give the same as SparkStreaming.
I do not know if my ideia is correct. I really need help :)