Billing and Usage Metrics

Additional useful metrics for resource usage monitoring and cost estimates

While the standard Kubernetes resource monitoring metrics via the Metrics API may be queried at any time on CoreWeave Cloud to gain more insight on resource usage, CoreWeave provides some additional metrics that may be used to monitor usage for general insights on current resource usage, as well as to generate cost estimates.

The following PromQL metrics may be incorporated into Grafana monitors for better oversight over resource use.

Important

billing metrics are used for estimation purposes only, and to monitor overall use. For actual usage metrics and their billing, access the Billing Portal from the Cloud UI. For an even more fine-tuned monitoring solution, see Grafana.

Compute resource usage metrics

The metrics below return usage values for compute resources (CPU and GPU). These values may be used to monitor usage totals, and to generate estimates for overall costs. For actual billed usage, view your billing statements by accessing the Billing Portal on the Cloud UI.

PromQL metric	Description
billing_mem	Returns the total current memory usage
billing_pod_mem	Returns the total memory usage for each currently running Pod
billing_cpu	Returns the current total CPU usage
billing_pod_cpu	Returns the CPU usage for each currently running CPU Pod
billing_pod_gpu	Returns the GPU usage for each currently running GPU Pod
billing_resource_usage_gpu	Returns the total GPU usage for each currently running GPU

Additional resource usage metrics

The metrics below return usage values for additional resources. These values are estimates of costs based on resource usage, but should not be taken as definitive bills. For actual billed usage, view your billing statements by accessing the Billing Portal on the Cloud UI.

PromQL metric	Description
billing_ip_address	Returns the billing cost for each public IP address in use

Storage and Object Storage usage metrics

The metrics below return usage values given in bytes of storage and Object Storage resources. These values may be used to monitor usage totals, and to generate estimates for overall costs. For actual billed usage, view your billing statements by accessing the Billing Portal on the Cloud UI.

PromQL metric	Description
billing_resource_usage_storage	Returns the total number of bytes used by all storage resources
billing_object_storage_usage_bytes	Returns the total bytes used across all Object Storage resources in a given namespace
billing_object_storage_usage_total_bytes	Returns the total bytes used by all Object Storage resources across all namespaces
billing_object_storage_usage_storageclass_bytes	Returns the total bytes used by all utilized Object Storage classes
billing_object_storage_bucket_usage_bytes	Returns the total bytes used for Object Storage buckets

Cost metrics

The metrics below return cost values in USD. These values are estimates of costs based on resource usage, but should not be taken as definitive bills. For actual billed usage, view your billing statements by accessing the Billing Portal on the Cloud UI.

PromQL metric	Description
billing_pod_cost_total	Returns the total cost for each running Pod
billing_pod_cpu_cost_hourly	Returns the average CPU usage cost per hour
billing_pod_gpu_cost_hourly	Returns the average GPU usage cost per hour
billing_pod_mem_cost_hourly	Returns the average memory usage cost per hour
billing_pod_cost_hourly	Returns the average cost for each Pod per hour

Container metrics via kube-state-metrics

The popular Kubernetes servicekube-state-metrics (KSM) is implemented in CoreWeave infrastructure and can be used to monitor resource usage on all Kubernetes Services, Deployments, jobs, and more.

For a full list of exposed metrics, see the kube-state-metrics documentation on exposed metrics.