基于负载预测的K8s水平自动伸缩优化方法研究OA

Kubernetes(K8s),a mainstream container orchestration platform,has been widely adopted in key areas such as cloud native and distributed systems.However,its native horizontal pod automatic scaling mechanism responds passively based on thresholds,resulting in significant scaling delays in the face of traffic bursts.This delay can easily lead to application performance bottlenecks and even service degradation,making it difficult to meet high availability and immediate response requirements.To effectively overcome this latency bottleneck and improve the proactive nature of automatic scaling,we propose a horizontal pod automatic scaling method based on load prediction for K8s.This method introduces an auxiliary container that deploys an LSTM model,working in conjunction with the target pod.This container processes the Prometheus monitoring metric stream in real time and uses LSTM to dynamically predict future short-term load trends.When the predicted value exceeds the warning threshold,the system proactively triggers scaling.Experimental results show that compared to the native HPA,the proposed method more accurately captures load changes,enabling proactive scaling and reduction.This mechanism significantly shortens application response time,improves system scaling efficiency and responsiveness,and avoids wasting server resources.