VILNIUS TECH Library invites you to follow the published new dissertations. The dissertation „Increasing availability of stateful micro-services in orchestrated container systems“ prepared at VILNIUS TECH by Kęstutis Pakrijauskas. The dissertation was prepared in 2020–2026. Scientific consultant – Prof. Dr Dalius Mažeika.
The dissertation was defended at the public meeting of the Dissertation Defense Council of the Scientific Field of Informatics Engineering in the Aula Doctoralis Meeting Hall of Vilnius Gediminas Technical University at 10 a.m. on 21 May 2026.
Stateful microservices are essential components in modern cloud computing environments. This research delves into the complexities and challenges of ensuring the reliability and availability of stateful microservices in container orchestration systems such as Kubernetes and Docker Swarm. Unlike stateless microservices, stateful microservices handle critical data, making their management and recovery more intricate and resource-intensive. The research highlights various existing methodologies to enhance the reliability of these services, including resource optimization, machine learning-based predictions, and custom schedulers. A significant focus of the research is the impact that maintenance activities have on the availability of stateful microservices. The dissertation proposes a method for loosely coupled, transparent failover, which minimizes downtime by observing database connection activity and terminating idle client connections, thereby redirecting requests with minimal client impact. This method was validated through experiments, achieving near-zero downtime during failover operations. Another critical aspect explored is burst tolerance in stateful microservices. The proposed rule-based method enhances burst tolerance through write-scaling and load balancing, distributing burst workloads across multiple nodes. This approach reduces failure rates and extends operational time under burst conditions, significantly improving service availability. Experimental results demonstrate that this method allows stateful microservices to sustain burst loads for nearly twice as long as traditional methods. Additionally, the research investigates the influence of various factors such as load intensity, load balancer type, request length, and connection type on the availability of database clusters during failover. The findings indicate that load balancers operating at the 1st OSI level provide higher availability than those operating at the 4th OSI level. The dissertation results were published in four scientific publications, two of which were in reviewed scientific journals indexed in the Web of Science database, and presented at two international conferences.
Doctoral dissertation readers can search via VILNIUS TECH Virtual Library.