hyperv: Add host CPU ratio metric for monitoring CPU allocation

[Dominik-esb]

Which issue this PR fixes

(optional, in fixes #<issue number>(, fixes #<issue_number>, ...) format, will close that issue when PR gets merged):

N/A - This is a feature enhancement

Special notes for your reviewer

• The metric uses existing performance counters (Hyper-V Hypervisor Logical Processor and Hyper-V Hypervisor Virtual Processor) to calculate the ratio
• The host subcollector is enabled by default as part of the standard hyperv collector configuration
• Includes proper hostname labeling for multi-host monitoring scenarios

Particularly user-facing changes

• New metric: windows_hyperv_host_cpu_ratio with host label
• New subcollector: host (enabled by default in hyperv collector)
• Metric is automatically collected when the hyperv collector is enabled

Checklist

Complete these before marking the PR as ready to review:

• DCO signed
• The PR title has a summary of the changes and the area they affect
• The PR body has a summary to reflect any significant (and particularly user-facing) changes introduced by this PR

[Dominik-esb]

Hi @jkroepke,

could you have a look at this idea?

I’m trying to calculate the ratio between logical CPU cores in a cluster and the total v-cores assigned to VMs (e.g. 100 physical cores / 300 v-cores = 3:1).

round(
  count(count by (vm, core) (windows_hyperv_hypervisor_virtual_processor_time_total{cluster=~"xxx"}))
  / 
  (sum(windows_cpu_logical_processor{cluster=~"xxx"}) / 2)
  * 10
) / 10

This isn’t straightforward, as there is no per-VM metric that exposes the configured core count. Currently, I’m approximating the number of virtual cores using:

count(
  count by (vm, core) (
    windows_hyperv_hypervisor_virtual_processor_time_total{cluster=~"xxx"}
  )
)

This works to some extent, but it doesn’t feel ideal.

While testing, I also noticed that some VMs are missing, causing the v-core count per cluster to differ from what Hyper-V reports. In multiple clusters, 2–6 VMs are missing. This is likely a separate issue, but it directly affects the CPU ratio calculation.

Even if the VM counting issue can be resolved, having a reliable logical-core-to-v-core ratio per host could be useful.

As an alternative it might also be worth considering exposing the configured core count per VM as a dedicated metric

[jkroepke]

This isn’t straightforward, as there is no per-VM metric that exposes the configured core count.

Sometimes, I run additional Win32 APIs calls to gain such informations. Generally, I like to avoid WMI as much as possible. As I know, the Host Compute Service API can be used to query such information. The Host Compute Service API is used by the container collector.