Improved Fault-tolerance and Zero Data Loss in Apache Spark Streaming - The Databricks Blog