Kubernetes Enhanced Stateful Service Migration for ML-Driven Applications in Industry 4.0 Scenarios

State-of-the-art ML-driven applications have shown remarkable effectiveness in enhancing manufacturing processes by detecting operational anomalies in real-time. Nevertheless, these applications place significant demands on computational resources and require constant access to data sources for managing real-time data processing. This often poses challenges in case of service migration between the Cloud and the edge, due to the stateful nature of ML-based software components and their high storage and bandwidth requirements. This paper presents a novel Kubernetes-based framework designed to enable and optimize the migration of stateful services. The framework allows to identify the most critical part of a service component state and to migrate it with very low downtimes that match the strict requirements of ML-driven applications in industrial settings. To demonstrate the practicality and benefits of our approach, we used a real-world ML-driven anomaly detection application as a case study. The efficacy of this migration mechanism is further confirmed through in depth industrial testing, which shows significant reductions in service downtime and improvements in operational continuity. These findings underscore the potential of our enhanced Kubernetes migration framework to tackle the prevalent challenges in contemporary Industry 4.0 environments.