一句话：主张为 Spark、Flink 等 DISC 框架构建通用可扩展的基于属性的测试框架，以发现静默错误结果。
数据密集型可扩展计算（DISC）框架（如 Apache Spark、Flink、Beam 和 Dask）通过高层编程模型与可扩展运行时支撑现代分析负载，但框架 bug 仍很常见，且许多不会导致崩溃，而是产生静默的错误结果，给测试与调试带来挑战。由于这类系统难以获得 ground-truth 输出，传统 oracle 方法受限。本文是一篇立场论文，主张将基于属性的测试（PBT）作为 DISC 的必要测试范式：用语义契约充当可执行 oracle，并呼吁构建通用、可扩展的 PBT 框架，提供可复用的属性模板，支持在不同 DISC 系统间系统化实例化。文章未报告具体实现或实证评估，而是提出研究议程并论证其必要性。
Data-intensive scalable computing (DISC) frameworks such as Apache Spark, Flink, Beam, and Dask underpin many modern analytics workloads by providing high-level programming models and scalable runtimes. Despite their widespread adoption, framework bugs remain common, and many manifest as silent wrong results rather than crashes. Property-based testing (PBT) has been successful across a range of domains by using semantic contracts as executable oracles when ground-truth outputs are difficult to obtain. We argue that DISC calls for a general and extensible PBT framework that provides reusable property templates and supports systematic instantiation across different DISC systems.
@inproceedings{WuLHGK26,
title = {Why Property-Based Testing is Necessary for Data Intensive Scalable Computing},
author = {Yaoxuan Wu and Ingrid Lee and Ahmad Humayun and Muhammad Ali Gulzar and Miryung Kim},
booktitle = {FSE},
year = {2026},
doi = {10.1145/3803437.3806090}
}