Towards execution guarantees for stream queries

The unbounded nature of data streams and the low-latency requirements of stream processing present interesting challenges in Data Stream Management System (DSMS) design. Streaming query operators are typically designed to produce results with low latency, as well as to efficiently manage their state. Stream-progress delimitation techniques, such as punctuation, can help query operators achieve these goals. In this work, we look at deriving execution guarantees with respect to result production and state management for complete queries over punctuated streams. These guarantees are derived before query execution. We formalize notions of successful stream processing at an operator level, and extend these definitions to stream queries as a whole. We introduce a framework, punctuation contracts, for analyzing data processing and punctuation propagation from input to output on individual operators. We then use our framework to analyze complete queries and determine, prior to execution, if every valid input is eventually emitted, and no item remains in operator state indefinitely. Finally, we discuss extensions needed to bound query memory requirements; we describe four stream properties that can be used to help understand and quantify memory and CPU usage.