Dynamic Plan Migration for Snapshot-Equivalent Continuous Queries in Data Stream Systems

A data stream management system executes a large number of continuous queries in parallel. As stream characteristics and query workload change over time, the plan initially installed for a continuous query may become inefficient. As a consequence, the query optimizer will re-optimize this plan based on the current statistics. The replacement of the running plan with a more efficient but semantically equivalent plan at runtime is called dynamic plan migration. In order to have a sound semantic foundation for query optimization, we investigate dynamic plan migration for snapshot-equivalent plans. We develop a general method for dynamic plan migration that treats the old and new plan as snapshotequivalent black boxes. This enables the query optimizer to apply the conventional transformation rules during re-optimization. As a consequence, our approach supports the dynamic optimization of arbitrary continuous queries expressible in CQL, whereas existing solutions are limited in their scope.