Log-Driven Autonomic Auto-Scaling with LSTM Forecasting: An Industrial Case in On-Premises Containerized Systems

This work presents an autonomic auto-scaling solution for a containerized system deployed on on-premises infrastructure. The solution addresses the lack of native autoscaling — the automatic adjustment of computing resources to match workload demand — in private environments by implementing a self-managing architecture based on the MAPE-K (Monitor-Analyze-Plan-Execute) control loop, an established methodology where the system continuously monitors itself, analyzes changes, plans reactions, and executes them using existing knowledge. It incorporates a Long Short-Term Memory (LSTM) neural network to proactively forecast workload spikes, enabling informed scaling decisions before performance degradation occurs. A key innovation is the non-intrusive, log-driven approach, where system logs serve as the primary knowledge source for analysis and scaling decisions. Preliminary evaluation using a dataset collected from a production deployment at a Brazilian public institution reveals improved system responsiveness and resource utilization during peak demand. These results, derived from a real-world, industrial-scale scenario, demonstrate the practical applicability of the proposed solution and indicate that the LSTM-driven auto-scaler can maintain quality of service under variable workloads while optimizing resource usage.