Method and apparatus for reducing a likelihood of a cascade failure in a multi-device array. The array preferably comprises a controller and a plurality of storage devices to define a memory space across which data are stored in accordance with a selected RAID configuration. The controller operates to sever an operational connection between the storage devices and a host device in relation to a detected temperature of at least one storage device of the array. Preferably, when a selected device reaches a first threshold temperature level, the controller arms for a potential shutdown. When a selected device reaches a second higher threshold temperature, the controller preferably powers down all of the devices and executes a self-reboot operation. The controller preferably monitors a temperature of the array while the devices remain powered down, after which the storage devices are powered up and data reconstruction operations take place as required.