Detecting Robustness against MVRC for Transaction Programs with Predicate Reads

TL;DR

This paper introduces a graph-theoretic method to determine if transaction programs with predicate reads are robust against multi-version Read Committed, enabling analysis of more complex workloads for serializability guarantees.

Contribution

It provides a novel characterization for robustness against MVRC for complex transaction programs, including control structures and predicate reads.

Findings

Supports transaction programs with loops, conditionals, and predicate reads

Enables recognition of more workloads as robust than previous methods

Experimental results show improved robustness detection capabilities

Abstract

The transactional robustness problem revolves around deciding whether, for a given workload, a lower isolation level than Serializable is sufficient to guarantee serializability. The paper presents a new characterization for robustness against isolation level (multi-version) Read Committed. It supports transaction programs with control structures (loops and conditionals) and inserts, deletes, and predicate reads -- scenarios that trigger the phantom problem, which is known to be hard to analyze in this context. The characterization is graph-theoretic and not unlike previous decision mechanisms known from the concurrency control literature that database researchers and practicians are comfortable with. We show experimentally that our characterization pushes the frontier in allowing to recognize more and more complex workloads as robust than before.