GoogleCloudPlatform / opentelemetry-operations-collector

Apache License 2.0
41 stars 29 forks source link

Implement an initial set of uniform DCGM GPU metrics in `dcgmreceiver`. #219

Closed igorpeshansky closed 2 months ago

igorpeshansky commented 3 months ago

This PR implements an initial set of DCGM GPU metrics from the uniform set defined in go/gce-gke-gpu-dcgm-metrics in dcgmreceiver.

In the process, it changes the GPU device-identifying metric attributes into resource attributes, adds support for older (and newer) GPU devices, and removes the assumption of 1:1 metric/DCGM field correspondence.