The invention discloses a method for managing cache data of a hyper-converged system. The method comprises the following steps: dynamically saving an LRU queue when the host of a virtual machine runs normally, wherein one LRU queue element corresponds to one cache record, and the cache record is generated when the initial data is cached from a low-speed storage medium to a high-speed storage medium; and when the virtual machine is restarted on a new host after the host of the virtual machine has failure, obtaining the dynamically-saved LRU queue, and sequentially removing the cache records in the dynamically-saved LRU queue to read the corresponding initial data from the low-speed storage medium and cache the data to the high-speed storage medium of the new host according to the cache records. The invention also discloses a device for managing the cache data of the hyper-converged system. According to the method and device disclosed by the invention, the beneficial effect that the data reading performance can be rapidly improved even after the virtual machine is restarted in the hyper-converged system can be achieved.