Chris Ho

After testing, when operating about 100 VMs, Kubevirt components' CPU utilization surge, as below:

> After testing, when operating about 100 VMs, Kubevirt components' CPU utilization surge, as below: ... According to the above, the CPU is not found to be too high.

**Latest tests:** 100 VMS, 1c1g, 30 minutes restart continuously. ### CPU **Query** ``` sum(irate(container_cpu_usage_seconds_total{container="virt-api"}[5m]) * 100) by (pod) sum(irate(container_cpu_usage_seconds_total{container="virt-controller"}[5m]) * 100) by (pod) sum(irate(container_cpu_usage_seconds_total{container="virt-handler"}[5m]) * 100) by (pod) ``` Component...

Use kubevirt's cluster-profiler to profile the control-plane https://github.com/harvester/harvester/issues/2729#issuecomment-1254981568 Shows the virt components Flame Graph. virt-api: virt-controller: virt-handler(among the nodes):

Each period is 1 minute, so if you want to see the statistics, wait at least 2 minutes.

thanks @rthallisey I recently did some CPU benchmark tests on virt components, when operating about 100 VMs, as below:    Can you provide pprof for virt components? I...

After re-verification, CPU performance is fine, the PromQL statement refers to the CPU percentage ratio and not the number of cores used. https://github.com/harvester/harvester/pull/2774#issuecomment-1294426262 In addition, I also analyzed kubevirt components...

Look into the Kubevirt components CPU spike issue

> Can you share some detail about the pressure you're applying to the KubeVirt control-plane, e.g. creating 100 vmis at once? I have tried to restart 100 vmis at once,...

Performance test report: https://github.com/harvester/harvester/pull/2774#issuecomment-1294426262