This document discusses the past, present, and future of Apache Flink. It describes how Flink enables stateful computations over data streams, as well as batch and stream processing. New features in Flink 1.5 include improved integration with resource managers, localized failure recovery, and a 50% rewritten network stack. Upcoming features in Flink 1.6 involve autoscaling, hot-standby replication for high availability, and zero-downtime upgrades. The document concludes that Flink is already the best open-source stream processing system and more exciting features are planned for the future.