Configuration Conventions

The GPU, NPU, and HAMi virtual-device examples in this section follow these conventions. Check the values below before applying a sample to another cluster.

Where to apply each object

ObjectCluster and namespace
PrometheusRuleAccelerator cluster, usually kube-system
Collection ConfigMapAccelerator cluster, cpaas-system
Display ConfigMapGlobal/Cost Management Server cluster, kube-public
Cost Model pricesPlatform console for the linked accelerator cluster

Configuration values

  • Verify the source metric names, labels, resource keys, and units in the target environment before applying a sample.
  • Keep category and item values consistent between the collection and display configurations.
  • Use the exact billing label name and value when configuring Price By Label. Label matching is case-sensitive.
  • The prometheus selector and rule namespace must match the active monitoring instance.

Units and refresh time

  • Each resource page defines the unit for its billing method. Do not change a divisor without checking the source unit.
  • The sample collection step is five minutes; usage and bills are not real-time.
  • Allow the collection and billing workers to process samples before checking the console.
  • Changing a Cost Model price applies to subsequently processed data and does not automatically rewrite historical bills.