Support For VMs

[ZebiGoldstone]

Is your feature request related to a problem? Please describe.
KRR currently supports only Kubernetes-native workloads and does not provide recommendations for VMs running under OpenShift Virtualization (KubeVirt).

Describe the solution you'd like
Add support in KRR for generating CPU and memory recommendations for VirtualMachines

Describe alternatives you've considered
Manually analyzing Prometheus metrics

Custom scripts for VM resource tuning
These are time-consuming and inconsistent compared to KRR.

Are you interested in contributing a PR for this?
No, but we’re happy to test and provide feedback.

[aantn]

@ZebiGoldstone I am less familiar with KubeVirt, but if you were to do this manually how would you do so? Which metrics would you look at?

[ZebiGoldstone]

For CPU usage, you can use the same metrics as regular pods (container_cpu_usage_seconds_total) and apply the same calculations to determine the percentile.

However, measuring memory usage is more challenging. The normal metric shows maximum usage at all times, which may not reflect the actual usage. There are a few metrics available, but the one that seems most accurate is kubevirt_vmi_memory_used.

[aantn]

Thanks. This might be doable as a subset of #411 with auto-discovery as a later more advanced feature on top.

[ZebiGoldstone]

Thanks. This might be doable as a subset of #411 with auto-discovery as a later more advanced feature on top.

If I understood that issue correctly, it is to use specific metrics to measure CPU & Ram usage.
What I think would be best in this specific use case is to go over virtual machines as well(just like how KRR runs for deployments/DC/Cronjobs etc.) when running the simple limit, perhaps with an additional flag to enable this feature.