Performance trade-offs between predictive and reactive autoscaling in stateful microservices

The relevance of studying the trade-offs between reactive and predictive autoscaling of stateful microservices was due to the need to improve stability, resource performance and reliability of cloud systems in dynamic conditions. The aim of the study was to assess the effectiveness of reactive, predictive and hybrid autoscaling strategies in stateful microservice architectures. The methodological basis of the research relied on comparative, content, systems and structural-functional analyses, as well as modelling, which made it possible to comprehensively investigate the performance and stability of different approaches to autoscaling in stateful microservices. Theoretical analysis showed that autoscaling in microservice systems provides adaptive performance, and the choice of strategy depends on the nature of the workload, latency sensitivity and consistency requirements. The review of models demonstrated that reactive autoscaling responds quickly to changes but is accompanied by high latency, performance fluctuations and the risk of state divergence between replicas, whereas predictive autoscaling reduces scaling delay and increases stability and resource efficiency but requires accurate models and additional computational resources. Analysis of the results revealed that hybrid strategies integrate the advantages of both approaches, providing both flexibility and stability, and that system efficiency is determined by a trade-off between latency, stability, state consistency, resource efficiency and financial costs. The results of the study may be useful for developers and architects of cloud microservice systems, IT engineers and DevOps teams for optimising scaling strategies, increasing service stability and ensuring efficient resource use in different cloud environments