Research on Kubernetes-based Preventive Over-commitment Scheduling Algorithm

With the rapid advancement of cloud computing [1] and container cloud technology [2], Kubernetes (K8S) [3] has become the preferred container orchestration and management platform for many enterprises and individuals to deploy online applications. It provides comprehensive cloud computing solutions that greatly facilitate the rapid deployment and operation of applications. However, despite the many advantages K8S brings to microservices architecture, stability issues during microservices updates have gradually emerged in practical applications. Especially the default node scheduling mechanism allows for the phenomenon of resource overcommitment, which poses a significant risk to certain types of applications such as memory-intensive applications. Particularly during batch updates of microservices, excessive allocation of node memory resources may lead to severe overcommitment, which can cause Out of Memory (OOM) errors later in the application’s operation, affecting the stability and user experience of online applications. In light of this, this paper focuses on the research to enhance the stability of online applications during the update process of microservices, aiming to optimize the scheduling algorithm of K8S to ensure efficient utilization of resources while effectively reducing the probability of resource overcommitment. This ensures that online applications can smoothly transition through the update cycle, avoiding short-term request exceptions or service interruptions caused by resource overcommitment, thus safeguarding business continuity.