Conjunctive queries with predicates in the form of comparisons that span multiple relations have regained interest recently, due to their relevance in OLAP queries, spatiotemporal databases, and machine learning over relational data. The standard technique, predicate pushdown, has limited efficacy on such comparisons. A technique by Willard can be used to process short comparisons that are adjacent in the join tree in time linear in the input size plus output size. In this paper, we describe a new algorithm for evaluating conjunctive queries with both short and long comparisons, and identify an acyclic condition under which linear time can be achieved. We have also implemented the new algorithm on top of Spark, and our experimental results demonstrate order-of-magnitude speedups over SparkSQL on a variety of graph patterns and analytical queries.

由于它们在 OLAP 查询、时空数据库和关系数据的机器学习中的相关性，以跨越多个关系的比较形式的谓词联合查询最近重新引起了人们的兴趣。标准技术，谓词下推，在此类比较中的效果有限。Willard 的一项技术可用于处理在时间上与输入大小和输出大小呈线性关系的连接树中相邻的短比较。在本文中，我们描述了一种用于评估具有短比较和长比较的联合查询的新算法，并确定了可以实现线性时间的非循环条件。我们还在 Spark 之上实施了新算法，我们的实验结果证明了在各种图形模式和分析查询上比 SparkSQL 有数量级的加速。