Working Principle
Figure 1 shows the working principle of resource monitoring.
NPU Exporter obtains container-related information by calling the standard interface CRI in Kubernetes through the gRPC service, processor network information by calling hccn_tool through exec, and processor information by calling DCMI through dlopen or dlsym. Then, it reports the information to Prometheus.
If you use Telegraf, directly call NPU Exporter to obtain related information.
Parent topic: Resource Monitoring Feature Guide