一句话：提出基于 eBPF 与预测模型的 Kubernetes 自动扩缩容工具链，兼顾安全隔离与流量突增鲁棒性。
针对 Kubernetes 中传统 HPA 等反应式扩缩容难以快速应对流量尖峰、且无法区分正常突发流量与 DDoS 攻击的问题，本文提出开源工具链 SPARK。其方法是在内核层利用 eBPF 网络层执行安全策略，同时基于预测模型编排扩缩容决策，从而在提供流量感知扩缩容的同时实现 L7 安全隔离与网络快速收敛。评估表明，与标准反应式扩缩容相比，预测式方法在突发流量激增期间将超时错误减少 32%。
Achieving high availability and robust security in Kubernetes requires more than reactive scaling and standard perimeter firewalls. Traditional autoscalers, such as HPA, often fail to react quickly to traffic spikes and cannot distinguish between legitimate flash crowds and DDoS attacks. We present an open-source toolchain to provide a traffic-aware autoscaling approach that utilizes an eBPF-based networking layer to enforce security policies at the kernel level while orchestrating scaling decisions based on predictive models. Our results demonstrate that the predictive approach reduces timeout errors by 32% during sudden traffic surges compared to standard reactive scaling, while ensuring immediate network convergence and layer 7 security isolation for newly scaled pods.