現今物聯網(IoT)、人工智慧(AI)、大數據等等技術的興起，對於數據處理、自動化的需求也越來越高，因此雲霧協作架構(Cloud Fog Collaboration)也成為其中不可或缺的技術。然而由於其中雲端以及邊緣端在計算環境有很大的差異性，這導致了服務需要在不同的平台上部署與轉移，但是傳統的部署需要在各個環境中進行調整，其中的耦合性導致了部署的難度上升，而微服務架構(Microservices)將應用拆分成多個獨立的輕量化服務，有低耦合度的特性，解決了上述難題。容器化技術則為微服務架構的實現提供了一個適合的方案。然而，透過容器化的形式將微服務部署於雲霧協作環境下，如何維持微服務的高可用性成為了一個重要的議題，雖然現今的容器編排工具如Kubernetes、Docker Swarm都提供了高可用性機制，而Kubernetes針對節點故障、服務故障、服務過載等高可用性機制常面臨問題皆有相關對應手段，但仍存在一些不足，無法很好地滿足微服務的高可用性，除此之外，微服務架構的無狀態(Stateless)特性也給高可用性機制帶來了一定挑戰。因此，本研究設計出一套系統架構，基於Kubernetes和Apache Kafka資料串流平台，透過偵測不同情境執行對應的解決手段，改善Kubernetes的高可用性機制，優化節點、服務故障時的服務恢復時間，並且再過載發生時執行自動擴展以恢復服務的QoS，並且在上述的處理手段中不會影響到原先的任務進度運行。;The rise of technologies such as the Internet of Things (IoT), Artificial Intelligence (AI), and Big Data has led to increasing demands for data processing and automation. As a result, Cloud Fog Collaboration has become an indispensable technology. However, there is significant disparity in the computing environments between the cloud and the edge, leading to the need for services to be deployed and migrated across different platforms. Traditional deployment methods require adjustments in each environment, increasing the complexity of deployment due to coupling. Microservices architecture addresses these challenges by breaking down applications into multiple independent and lightweight services with low coupling. Containerization technology provides a suitable solution for implementing the Microservices architecture.Deploying Microservices in the Cloud Fog Collaboration environment through containerization raises an important issue of how to maintain high availability. Although container orchestration tools like Kubernetes and Docker Swarm provide high availability mechanisms for node failures, service failures, and service overloads, they still have some limitations. Additionally, the stateless nature of Microservices poses challenges to high availability mechanisms. Therefore, in this research, we design a system architecture based on Kubernetes and Apache Kafka data streaming platform. We detect different scenarios and execute corresponding solutions to improve Kubernetes′ high availability mechanism, optimize service recovery time during node and service failures, and perform automatic scaling during service overloads to restore Quality of Service (QoS).
