GraphQL-Aware Healing in Service-Oriented Architectures via Multi-Signal Learning

This paper introduces an adaptive test and runtime healing approach that delivers resolver-level resilience for GraphQL service-oriented architectures by unifying three telemetry streams: semantic log embeddings obtained from large language models, structural dependencies encoded via graph neural networks, and statistically grounded operational metrics. These signals are fused into a single reinforcement learning state vector, enabling a deep Q-network to learn context-aware recovery actions including selective retry, safe skip, dependency reordering, and escalation without obscuring root causes. The approach is evaluated in a production-grade case study involving a real-world lifestyle coaching platform used by thousands of active users. The application's asynchronous, cloud-native architecture with complex resolver interactions and AI-powered personalization provides a realistic and challenging environment for assessing the system's robustness. Across more than one thousand simulated failure episodes that inject realistic cloud uncertainty, the approach improves test and runtime success rates from 68.7% to 92%, reduces mean-time-to-recovery from 687 ms to 203 ms, and trims CI compute time by 61% using a KL-stability early-stop rule. It also preserves tail-latency accuracy within a 5% error bound while incurring only 11.8 ms median inference overhead per healed request. These results demonstrate that statistically principled, reinforcement-learning-driven healing offers a practical, fine-grained self-recovery solution for serviceoriented systems deployed in modern, real-world cloud applications.