Not saying your wrong about what he meant, but Spark can also keep state across the Cluster, not just data. Check out the "UpdateStateByKey Operation" in the following page.

http://spark.apache.org/docs/latest/streaming-programming-gu...