Support for Schema Evolution in Data Stream Management Systems

Unlike Database Management Systems (DBMSs), Data Stream Management Systems (DSMSs) do not evaluate queries over static data sets — rather, they continuously produce result streams to standing queries, and often operate in a context where any interruption can lead to data loss. Support for schema evolution in such an environment is currently unaddressed. In this work we address evolution in DSMSs by introducing a new element to streams, called an accent, that precedes and describes an evolution. We characterize how a subset of commonly used query operators in DSMS act on and propagate accents with respect to three evolution primitives: Add Attribute, Drop Attribute, and Alter Data.