Ai task failure rescheduling method and system, storage medium and electronic device

Abstract

This application provides an AI task fault rescheduling method, an AI task fault rescheduling system, a computer-readable storage medium, and an electronic device, relating to the field of cloud computing technology. The method includes: when a faulty node is detected in the cluster, determining healthy nodes and executing a fault handling process to form a set of AI tasks to be rescheduled; obtaining input parameters including task scheduling priority, health index of healthy nodes, and scheduling latency; determining an optimal scheduling strategy based on a deep reinforcement learning algorithm, with minimizing scheduling latency and minimizing cluster resource imbalance as joint optimization objectives; and executing task rescheduling according to the optimal scheduling strategy. This application can solve the problems of high latency and unbalanced cluster resource utilization in AI task rescheduling after cluster node failure.

Images