Data set cache processing method, device and system

The invention provides a data set cache processing method, device and system, belongs to the technical field of data processing, and solves the problems that in the prior art, an AIStation V2.1 version data set node cache mode is imperfect in design, so that when multiple users concurrently pull small files and large data sets, the performance is low, and the user experience is poor. And an alarmis generated because the pressure on the system is too large. The method comprises the steps of initializing a container in a Pod, caching data sets according to task requests, judging whether the data sets are caching requests sent by the same user or not, and if the data sets are different data sets cached by the same task of the same user, executing caching in a sequential queuing mode. According to the method, the technical problem of jamming caused by frequent restart of the initcontainer is solved, data set uniqueness verification is increased, performance consumption caused by repeatedcaching of the same data s