Issues on Performance of Reactive Programming in the Java Ecosystem with Persistent Data Sources

Abstract

The paper presents research on the performance of reactive processing in the Java ecosystem where persistent data sources (both SQL and NoSQL) were used as one single dependency of the node. Several scenarios have been tested, including changing chunks of data stream, different types of database drivers, etc. The results show that in the case of reactive processing of data streams being fetched from single noded persistent data sources it is hard to gain an advantage over nonreactive processing and some advantages began to appear only both with a large number of concurrent users and data streams of significant volume, while for low- and mid-loaded systems reactive processing gives usually unnecessary overhead resulting in degradation of the overall performance. It should be also noticed that we are at the quite early stage of reactive programming development what results sometimes in lack of its support in various layers of technological stacks. This was the case with the PostgreSQL database, where there is no production release of the nonblocking driver, and no support for developing pipeline stored procedures yet.