The talk describes the evolution from the classical Chandy—Lamport algorithm for consistent state capture of a distributed system to fault tolerance mechanisms and non-stop scaling of distributed streaming data processing systems. Unsolved problems of elasticity of streaming data processing systems are also presented.