Research on Kubernetes Scheduler Optimization Based on real Load

With the development of microservices, more and more companies are deploying applications in a containerized manner. Kubernetes is one of the most popular container orchestration tools currently available, which can effectively simplify container management. The scheduler is a core component of Kubernetes, which is responsible for scheduling containers to run on the most appropriate node. The main reference metric for the Kubernetes default scheduler is the resource request rate of the node. It prioritizes scheduling containers to run on nodes with low resource request rates. The resource application rate of a node is calculated by dividing the amount of resources requested by the node by the total amount of resources of the node. This scheduling strategy can meet business requirements in most cases, but it may have problems in certain scenarios. In a certain business scenario, the data processing module of the system needs to run multiple processing tasks on multiple nodes, and the computing resources required for these processing tasks vary. When using the default scheduler, it is difficult for users to predict their actual needs when applying for resources. Therefore, there may be significant differences between the resource application rate of nodes and the actual resource utilization rate of nodes, resulting in significant waste of cluster resources and uneven load. Therefore, this article refers to the relevant research results in the field of Kubernetes scheduling and implements a scheduler based on real load, aiming to solve the possible resource waste and load imbalance when scheduling Pods by default schedulers.