The Free Termination Property of Queries Over Time
Conor Power, Paraschos Koutris, Joseph M Hellerstein
TL;DR
Addresses the problem of free termination in coordination-free distributed queries, complementing CALM's soundness with a completeness perspective. Develops a semiautomaton-based model to unify relational transducer reasoning with algebraic frameworks such as CRDTs and pre-semirings used in incremental view maintenance. Characterizes how algebraic properties—acyclicity, partial orders, join-semilattice structure, and group-like updates—govern the existence of free termination states and identifies threshold-like queries that enable termination. Also connects these ideas to distributed protocols (e.g., network flooding) and provides a linear-time procedure for deciding free termination states in finite state spaces, offering a finer granularity of coordination-freeness for specific query-input pairs.
Abstract
Building on prior work on distributed databases and the CALM Theorem, we define and study the question of free termination: in the absence of distributed coordination, what query properties allow nodes in a distributed (database) system to unilaterally terminate execution even though they may receive additional data or messages in the future? This completeness question is complementary to the soundness questions studied in the CALM literature. We also develop a new model based on semiautomata that allows us to bridge from the relational transducer model of the CALM papers to algebraic models that are popular among software engineers (e.g. CRDTs) and of increasing interest to database theory for datalog extensions and incremental view maintenance.
