IsoPredict: Dynamic Predictive Analysis for Detecting Unserializable Behaviors in Weakly Isolated Data Store Applications

Chujun Geng,Spyros Blanas,Michael D. Bond,Yang Wang

DOI: https://doi.org/10.1145/3656391

2024-04-06

Abstract:This paper presents the first dynamic predictive analysis for data store applications under weak isolation levels, called Isopredict. Given an observed serializable execution of a data store application, Isopredict generates and solves SMT constraints to find an unserializable execution that is a feasible execution of the application. Isopredict introduces novel techniques that handle divergent application behavior; solve mutually recursive sets of constraints; and balance coverage, precision, and performance. An evaluation on four transactional data store benchmarks shows that Isopredict often predicts unserializable behaviors, 99% of which are feasible.

Programming Languages,Databases

What problem does this paper attempt to address?

### What problem does this paper attempt to solve?

This paper aims to address the issue of detecting non-serializable behaviors in distributed data storage systems under weak isolation levels. Specifically:

1. **Background and Motivation**:
- Distributed data storage systems often support weak isolation levels (such as causal consistency or read committed, rc). While these levels improve efficiency, they can also lead to non-serializable execution behaviors.
- Non-serializable execution behaviors are difficult for programmers to understand and often result in errors and failures in real systems.

2. **Main Contributions**:
- Introduced **IsoPredict**, the first dynamic predictive analysis method for detecting non-serializable behaviors in data storage applications under weak isolation levels.
- Given an observed serializable execution, IsoPredict generates and solves SMT constraints to find a possible non-serializable execution.
- By evaluating four transactional data storage benchmarks, results show that IsoPredict can effectively predict non-serializable behaviors, with over 99% of the predictions being validated as feasible executions.

3. **Technical Details**:
- **Dynamic Predictive Analysis**: Analyzes observed program executions and detects alternative feasible non-serializable executions.
- **SMT Constraint Generation**: Encodes execution feasibility, non-serializability, and weak isolation levels into SMT constraints and uses off-the-shelf SMT solvers to solve them.
- **Validation Mechanism**: Optionally verifies whether the predicted non-serializable execution is a feasible execution.

Through these methods, the paper aims to help developers better understand and handle the issue of non-serializable behaviors under weak isolation levels.