A method for processing a fault of a lock server in a distributed system is disclosed, where the distributed system includes m lock servers, each lock server locally stores same lock server takeover relationship information, and m is a natural number greater than 2. Lock servers in the distributed system that are not faulty receive a first notification message, where the first notification message carries information about a fault of a first lock server; after receiving the first notification message, a second lock server in the distributed system determines that the second lock server is a takeover lock server of the first lock server according to locally stored lock server takeover relationship information, and the takeover lock server enters a silent state; after receiving the first notification message, a third lock server in the distributed system determines that the third lock server is not the takeover lock server of the first lock server according to locally stored lock server takeover relationship information; and after receiving a locking request, the third lock server allocates lock permission information according to the locking request. In the present invention, an affected range of a faulty lock server can be minimized and stability of a distributed system is enhanced.