Benchmarking Apache Samza: 1.2M messages per second on a single node

I am getting lost between Flink, Samza, and Spark Streaming. The only one I have used in production is Spark streaming (using Kafka) and it worked well for our use case, and these are the types of processes I just don't refine that often (essentially new-school ETL's).

Any reason why I should switch over to Flink/Samza that appear to be more streaming-first oriented? Especially if my goal is that it might need long-term support?