During my bachelor’s thesis, I investigated whether k3s, the lightweight Kubernetes distribution, could improve service continuity in an on-premise industrial containerized environment.
Rather than focusing on Kubernetes theory, this talk presents a practical feasibility study based on real-world inspired architecture. A prototype application was deployed on a multi-node k3s cluster and subjected to controlled failure scenarios, including node shutdowns and pod failures. The system was monitored both from the Kubernetes side and from the user perspective side.
The results show where k3s excels: automatic pod recovery, workload rescheduling, service routing, and self-healing mechanisms. However, the experiments also reveal important limitations. Kubernetes can recover many failures, but not all availability problems disappear automatically. Operational issues, image distribution challenges, and application-level design decisions still play a crucial role.
This talk shares the architecture, experiments, results and lessions learned, providing practical insights for anyone considering k3s as an orschestration solution fro small on-premise deployments.
