A Study on Emerging Artificial Intelligence for Sustainable Self-Healing Web Applications in the Cloud Era

Cloud-native web applications are required to remain available even in the face of regular faults across distributed environments. However, in practice traditional monitoring pipelines and rule-based recovery mechanisms may times prove fragile, leading to longer outages and dissipated assists. To address this gap, this study presents a hybrid selfhealing framework that combines semantic fault diagnosis, an energy-aware reinforcement learning (RL) actuator and anomaly detection. The framework absorbs traces, metrics and logs, detects anomalies in real time, and maps them into higher-level “failure intents” that leads an RL policy in choosing remediation actions such as restart, reroute, or scale-out. Unlike controllers that upgrade only for uptime, this proposed design clearly factors in energy cost when choosing the recovery strategies. Evaluation on a Kubernetes-based microservice benchmark with six injected fault types shows optimistic results: in more than 30 trials per scenario, the hybrid system cut median mean time to repair (MTTR) by as much as 70% while reducing per-incident energy consumption by 12–20% compared with rule-based baselines. While built on controlled experiments, the findings suggest that AI-driven self-healing can enhance both sustainability and reliability in cloud-native systems.