Kubernetes scheduling method and device and storage medium

Abstract

The invention provides a Kubernetes scheduling method and device and a storage medium. The method comprises, removing Nodes which do not meet requirements by filtering according to a pre-selection algorithm; setting a queue for each Node according to the resource type; dividing priorities for a Pod to be scheduled; analyzing the state of the Pod to be scheduled; when the cluster resources are met, calculating the similarity of the Pod according to a first rule, and searching a first Node for scheduling; when cluster resources are not satisfied, carrying gout preemption scheduling, setting first rule calculation similarity and second rule priority preemption rules, and searching Nodes for preemption scheduling; and setting the maximum cycle index for the Pod which cannot be scheduled, wherein the Pod is scheduled preferentially when the maximum cycle index is reached. The problems that the priority is set by neglecting the actual condition of cluster resources, the resources cannot be reasonably arranged in the preemption process, and scheduling cannot be completed all the time are solved. According to the invention, the task operation integrity and the load balance of the whole cluster are improved, generation of resource fragments is reduced, the scheduling timeliness is improved, and the resource scheduling efficiency in a cluster management system is enhanced.

Description

technical field [0001] The invention relates to the technical field of container arrangement, in particular to a method for realizing a Kubernetes scheduling strategy. Background technique [0002] Kubernetes is an open source system, mainly used for the management of integrated applications across hosts in the cluster. A core feature of Kubernetes is that it dynamically manages containers to ensure that the state of the cluster meets user needs. Pod is the smallest unit for scheduling and management in the cluster. All containers run in the Pod, and each Pod can run one or more containers. Each cluster consists of control node Master and Node nodes ( figure 1 ). The core of the Scheduler (scheduler) is how to select the appropriate Node node from the cluster to assign to the Pod to be scheduled ( figure 2 ). [0003] Most of the current kubernetes resource scheduling algorithms score and prioritize Node nodes, such as the built-in LeastRequested and Priority algorithm...

AI Technical Summary

Problems solved by technology

[0004] In order to solve the problem that the scheduling strategy in the prior art ignores the actual situation of cluster resources and sets the priority, and the resources cannot be reasonably arranged during the preemption process, the present invention provides a Kubernetes scheduling method, including:

Images