Performance Evaluation of Kafka Clients Using a Reactive API

This thesis evaluates the performance of a core-banking application that consumes messages from a Kafka topic, and after performing validation and a static transformation stores the results in another Kafka topic. Performances are evaluated on two different versions of the application: traditional and reactive. The focus is on the way the two applications interact with Kafka. Both versions are developed using Spring: the traditional version uses Spring MVC while the reactive one uses Spring WebFlux. Spring WebFlux internally uses Project Reactor. Both versions consume and produce messages from/to Apache Kafka, however in the reactive version messages are consumed and produced using functional APIs provided by Reactor Kafka. Internal metrics of Kafka clients are collected through the JMX reporter. The obtained results show that the reactive implementation is able to achieve higher throughput and use fewer resources with respect to the traditional implementation. The implication of this result is that a company that develops software using a reactive approach is able to accommodate a higher number of requests with equal resources poured into the system. Alternatively, an organization could achieve the same throughput but use fewer resources.