Recovery of Inference Card Faults

Function Highlights

If an inference NPU resource managed by the cluster scheduling components is faulty, a hot reset is performed on the faulty resource to restore the NPU.

Required Components

• Volcano or other schedulers
• Ascend Device Plugin
• Ascend Docker Runtime
• ClusterD
• NodeD

Instructions

• Refer to Installation and Deployment for component installation.
• Refer to Recovery of Inference Card Faults for feature usage.