Intelligent API Health Monitoring and Auto-Healing System with Dynamic Scaling for Cloud-Native Microservices

Abstract

APIs support communication between services in systems that utilize a microservices architecture, so they must always be operational for the systems to function properly. Traditional API monitoring solutions can produce alerts to notify administrators of API issues, but they rely on human intervention to restore API operation, leading to gradual downtime. This paper describes an Intelligent API Health Monitoring and Auto-Healing System (IAHMAS), which is a cloud-native implementation of an integrated system capable of real-time health monitoring, automated failure detection, self-healing and automated scaling. To implement IAHMAS, you need to perform periodic heartbeat checks on monitored API endpoints, use a customizable failure threshold detection mechanism to trigger automated failover logic and service instance restarts, and apply Kubernetes Horizontal Pod Autoscaler (HPA)-based compute resource scaling when service degradation occurs due to load. The three quantitative metrics defined to measure the performance of the system are the System Availability Rate (SAR), defined as the fraction of the total observation time during which all the APIs that are subject to observation remain available; the Mean Time to Recovery (MTTR), defined as the average recovery time from the API's unavailability to its recovery; and the Auto-Healing Efficiency (AHE), defined as the fraction of unsuccessful incidents that lead to auto-healing. The system was tested on five enterprise API services over 30 days, including 50 failure scenarios. It achieved a SAR of 99.99%, an MTTR of 9.9 seconds, and an AHE of 96.0%. The improvement is significant compared to the baseline of a SAR of 99.54% and an MTTR of 28.4 minutes when failure recovery is performed manually. Over 99% mean recovery time improvement is achieved over existing approaches, as well as high-availability service levels consistent with enterprise-grade SLA requirements, and we show that our approach is both practical and deployable for automated API reliability management in cloud-native systems.