A system and method for recovering from a failure in a clustered environment sharing a common nonvolatile storage device. The first computer reserves the device by identifying itself in a reservation key associated with the device. The first device sends a signal to one or more backup computers informing the backup computers that the first computer is operational. When the first computer fails to operate, one of the backup computers breaks the first computer's reservation to the storage device and writes a new reservation key to the reservation storage area identifying the backup computer as the reservation holder of the storage device. When the first computer is reinitialized, it reads the reservation storage area and does not attempt to break the reservation made by the backup computer. One embodiment provides for the first computer to request control from the backup computer upon being reinitialized whereupon the backup computer releases its reservation and relinquishes control. Another embodiment provides for the first computer to act as a backup computer. Writing to the nonvolatile storage device simultaneously from multiple paths, such as from a NUMA machine, is further disclosed using a reservation key identifying the computer.
展开▼
