Expose CPU affinity metric

This patch introduces the kubevirt_vmi_cpu_affinity metric which displays the cpu pinning map via boolean labels in the form of vcpu_X_cpu_Y. Signed-off-by: Yuval Turgeman <yturgema@redhat.com>
fossedihelm · Mar 17, 2021 · d5b1ce5 · d5b1ce5
1 parent 74bf53d
commit d5b1ce5
Show file tree

Hide file tree

Showing 7 changed files with 79 additions and 2 deletions.
diff --git a/docs/metrics.md b/docs/metrics.md
@@ -155,7 +155,7 @@ A design proposal and its implementation history can be seen [here](https://docs
 ## kubevirt_vmi_network_transmit_packets_total
 #### HELP kubevirt_vmi_network_transmit_packets_total Network traffic transmit packets
 
- # Other Metrics 
+ # Other Metrics
 ## kubevirt_vmi_memory_actual_balloon_bytes
 #### HELP kubevirt_vmi_memory_actual_balloon_bytes current balloon bytes.
 ## kubevirt_vmi_memory_pgmajfault
@@ -170,3 +170,7 @@ A design proposal and its implementation history can be seen [here](https://docs
 #### HELP kubevirt_vmi_memory_usable_bytes The amount of memory which can be reclaimed by balloon without causing host swapping in bytes.
 ## kubevirt_vmi_memory_used_total_bytes
 #### HELP kubevirt_vmi_memory_used_total_bytes The amount of memory in bytes used by the domain.
+
+ # Other Metrics 
+## kubevirt_vmi_cpu_affinity
+#### HELP kubevirt_vmi_cpu_affinity vcpu affinity details
diff --git a/pkg/monitoring/vms/prometheus/fakeCollector.go b/pkg/monitoring/vms/prometheus/fakeCollector.go
@@ -45,6 +45,7 @@ func (fc fakeCollector) Collect(ch chan<- prometheus.Metric) {
 	out.Memory.UsableSet = true
 	out.Memory.MinorFaultSet = true
 	out.Memory.MajorFaultSet = true
+	out.CPUMapSet = true
 
 	vmi := k6tv1.VirtualMachineInstance{
 		Status: k6tv1.VirtualMachineInstanceStatus{

diff --git a/pkg/monitoring/vms/prometheus/prometheus.go b/pkg/monitoring/vms/prometheus/prometheus.go
@@ -169,6 +169,26 @@ func (metrics *vmiMetrics) updateMemory(mem *stats.DomainStatsMemory) {
 	}
 }
 
+func (metrics *vmiMetrics) updateCPUAffinity(cpuMap [][]bool) {
+	affinityLabels := []string{}
+	affinityValues := []string{}
+
+	for vidx := 0; vidx < len(cpuMap); vidx++ {
+		for cidx := 0; cidx < len(cpuMap[vidx]); cidx++ {
+			affinityLabels = append(affinityLabels, fmt.Sprintf("vcpu_%v_cpu_%v", vidx, cidx))
+			affinityValues = append(affinityValues, fmt.Sprintf("%t", cpuMap[vidx][cidx]))
+		}
+	}
+
+	metrics.pushCustomMetric(
+		"kubevirt_vmi_cpu_affinity",
+		"vcpu affinity details",
+		prometheus.CounterValue, 1,
+		affinityLabels,
+		affinityValues,
+	)
+}
+
 func (metrics *vmiMetrics) updateVcpu(vcpuStats []stats.DomainStatsVcpu) {
 	for vcpuIdx, vcpu := range vcpuStats {
 		stringVcpuIdx := fmt.Sprintf("%d", vcpuIdx)
@@ -585,6 +605,10 @@ func (metrics *vmiMetrics) updateMetrics(vmStats *stats.DomainStats) {
 	metrics.updateVcpu(vmStats.Vcpu)
 	metrics.updateBlock(vmStats.Block)
 	metrics.updateNetwork(vmStats.Net)
+
+	if vmStats.CPUMapSet {
+		metrics.updateCPUAffinity(vmStats.CPUMap)
+	}
 }
 
 func (metrics *vmiMetrics) newPrometheusDesc(name string, help string, customLabels []string) *prometheus.Desc {

diff --git a/pkg/monitoring/vms/prometheus/prometheus_test.go b/pkg/monitoring/vms/prometheus/prometheus_test.go
@@ -20,6 +20,8 @@
 package prometheus
 
 import (
+	"fmt"
+
 	"github.com/prometheus/client_golang/prometheus"
 	io_prometheus_client "github.com/prometheus/client_model/go"
 	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
@@ -932,6 +934,39 @@ var _ = Describe("Prometheus", func() {
 			Expect(result).ToNot(BeNil())
 			Expect(result.Desc().String()).To(ContainSubstring("kubevirt_vmi_vcpu_wait_seconds"))
 		})
+
+		It("should expose vcpu to cpu pinning metric", func() {
+			ch := make(chan prometheus.Metric, 1)
+			defer close(ch)
+
+			ps := prometheusScraper{ch: ch}
+
+			vmStats := &stats.DomainStats{
+				Cpu:       &stats.DomainStatsCPU{},
+				Memory:    &stats.DomainStatsMemory{},
+				Net:       []stats.DomainStatsNet{},
+				Vcpu:      []stats.DomainStatsVcpu{},
+				CPUMapSet: true,
+				CPUMap:    [][]bool{{true, false, true}},
+			}
+
+			vmi := k6tv1.VirtualMachineInstance{}
+			ps.Report("test", &vmi, vmStats)
+
+			result := <-ch
+			dto := &io_prometheus_client.Metric{}
+			result.Write(dto)
+
+			Expect(result).ToNot(BeNil())
+			Expect(result.Desc().String()).To(ContainSubstring("kubevirt_vmi_cpu_affinity"))
+			s := ""
+			for _, lp := range dto.GetLabel() {
+				s += fmt.Sprintf("%v=%v ", lp.GetName(), lp.GetValue())
+			}
+			Expect(s).To(ContainSubstring("vcpu_0_cpu_0=true"))
+			Expect(s).To(ContainSubstring("vcpu_0_cpu_1=false"))
+			Expect(s).To(ContainSubstring("vcpu_0_cpu_2=true"))
+		})
 	})
 })
 

diff --git a/pkg/virt-launcher/virtwrap/cli/libvirt.go b/pkg/virt-launcher/virtwrap/cli/libvirt.go
@@ -302,6 +302,14 @@ func (l *LibvirtConnection) GetDomainStats(statsTypes libvirt.DomainStatsTypes,
 			return list, err
 		}
 
+		cpuMap, err := domStat.Domain.GetVcpuPinInfo(libvirt.DOMAIN_AFFECT_CURRENT)
+		if err != nil {
+			return list, err
+		}
+
+		stat.CPUMap = cpuMap
+		stat.CPUMapSet = true
+
 		list = append(list, stat)
 	}
 

diff --git a/pkg/virt-launcher/virtwrap/stats/types.go b/pkg/virt-launcher/virtwrap/stats/types.go
@@ -58,6 +58,9 @@ type DomainStats struct {
 	Net   []DomainStatsNet
 	Block []DomainStatsBlock
 	// omitted from libvirt-go: Perf
+	// extra stats
+	CPUMapSet bool
+	CPUMap    [][]bool
 }
 
 type DomainStatsCPU struct {

diff --git a/pkg/virt-launcher/virtwrap/statsconv/util/domstats_utils.go b/pkg/virt-launcher/virtwrap/statsconv/util/domstats_utils.go
@@ -240,7 +240,9 @@ var Testdataexpected = `{
        "WaitSet": true,
        "Wait": 1500
      }
-   ]
+   ],
+   "CPUMapSet": false,
+   "CPUMap": null
  }`
 
 func LoadStats() ([]libvirt.DomainStats, error) {