The invention relates to a data storage method in a distributed system. The data storage method in the distributed system comprises the steps of determining a storage risk value of each node according to a storage volume, a running load and failure frequency, and grouping multiple nodes, wherein each group comprises a node with a high storage risk value and a node with a low storage risk value, so that an average storage risk value in each group is relatively balanced, the situation that the multiple nodes in which redundant data is distributed are all the nodes where failures frequently occur, and the redundant data is distributed into the groups with balanced risks, which facilitates data maintenance and reduces the risk of losing data.