e2e: Verify metrics-exporter expose netdevice metrics

Exposed metrics can be verified by scraping the prometheus
endpoint on the `sriov-network-metrics-exporter` pod.
Add a test that spawns an SR-IOV consuming pod and verifies
its receiving counter increase when the interface is pinged from
outside.

Signed-off-by: Andrea Panattoni <apanatto@redhat.com>

go.mod

@@ -27,6 +27,8 @@ require (
 	github.com/pkg/errors v0.9.1
 	github.com/prometheus-operator/prometheus-operator/pkg/apis/monitoring v0.68.0
 	github.com/prometheus-operator/prometheus-operator/pkg/client v0.68.0
+	github.com/prometheus/client_model v0.5.0
+	github.com/prometheus/common v0.45.0
 	github.com/safchain/ethtool v0.3.0
 	github.com/spf13/cobra v1.7.0
 	github.com/stretchr/testify v1.8.4
@@ -116,8 +118,6 @@ require (
 	github.com/peterbourgon/diskv v2.0.1+incompatible // indirect
 	github.com/pmezard/go-difflib v1.0.0 // indirect
 	github.com/prometheus/client_golang v1.17.0 // indirect
-	github.com/prometheus/client_model v0.5.0 // indirect
-	github.com/prometheus/common v0.45.0 // indirect
 	github.com/prometheus/procfs v0.12.0 // indirect
 	github.com/robfig/cron v1.2.0 // indirect
 	github.com/russross/blackfriday/v2 v2.1.0 // indirect

test/conformance/tests/test_exporter_metrics.go

@@ -0,0 +1,188 @@
+package tests
+
+import (
+	"context"
+	"fmt"
+	"strings"
+
+	"github.com/k8snetworkplumbingwg/sriov-network-operator/pkg/utils"
+	"github.com/k8snetworkplumbingwg/sriov-network-operator/test/util/cluster"
+	"github.com/k8snetworkplumbingwg/sriov-network-operator/test/util/discovery"
+	"github.com/k8snetworkplumbingwg/sriov-network-operator/test/util/namespaces"
+	"github.com/k8snetworkplumbingwg/sriov-network-operator/test/util/network"
+	"github.com/k8snetworkplumbingwg/sriov-network-operator/test/util/pod"
+
+	dto "github.com/prometheus/client_model/go"
+	"github.com/prometheus/common/expfmt"
+
+	corev1 "k8s.io/api/core/v1"
+	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
+
+	. "github.com/onsi/ginkgo/v2"
+	. "github.com/onsi/gomega"
+)
+
+var _ = FDescribe("[sriov] Metrics Exporter", Ordered, func() {
+
+	BeforeAll(func() {
+		if cluster.VirtualCluster() {
+			Skip("IGB driver does not support VF statistics")
+		}
+
+		err := namespaces.Create(namespaces.Test, clients)
+		Expect(err).ToNot(HaveOccurred())
+
+		err = namespaces.Clean(operatorNamespace, namespaces.Test, clients, discovery.Enabled())
+		Expect(err).ToNot(HaveOccurred())
+
+		featureFlagInitialValue := isFeatureFlagEnabled("metricsExporter")
+		DeferCleanup(func() {
+			By("Restoring initial feature flag value")
+			setFeatureFlag("metricsExporter", featureFlagInitialValue)
+		})
+
+		By("Enabling `metricsExporter` feature flag")
+		setFeatureFlag("metricsExporter", true)
+
+		By("Adding ")
+		err = utils.AddLabelToNamespace(context.Background(), operatorNamespace, "openshift.io/cluster-monitoring", "true", clients)
+		Expect(err).ToNot(HaveOccurred())
+
+		WaitForSRIOVStable()
+	})
+
+	It("collects metrics regarding receiving traffic via VF", func() {
+		sriovInfos, err := cluster.DiscoverSriov(clients, operatorNamespace)
+		Expect(err).ToNot(HaveOccurred())
+
+		node, nic, err := sriovInfos.FindOneSriovNodeAndDevice()
+		Expect(err).ToNot(HaveOccurred())
+		By("Using device " + nic.Name + " on node " + node)
+
+		_, err = network.CreateSriovPolicy(clients, "test-me-policy-", operatorNamespace, nic.Name, node, 2, "metricsResource", "netdevice")
+		Expect(err).ToNot(HaveOccurred())
+
+		err = network.CreateSriovNetwork(clients, nic, "test-me-network", namespaces.Test, operatorNamespace, "metricsResource", ipamIpv4)
+		Expect(err).ToNot(HaveOccurred())
+		waitForNetAttachDef("test-me-network", namespaces.Test)
+
+		pod := createTestPod(node, []string{"test-me-network"})
+
+		ips, err := network.GetSriovNicIPs(pod, "net1")
+		Expect(err).ToNot(HaveOccurred())
+		Expect(ips).NotTo(BeNil(), "No sriov network interface found.")
+		Expect(len(ips)).Should(Equal(1))
+
+		initialMetrics := getMetricsForNode(node)
+		initialRxBytes := getCounterForPod(initialMetrics, pod, "sriov_vf_rx_bytes")
+		initialRxPackets := getCounterForPod(initialMetrics, pod, "sriov_vf_rx_packets")
+
+		for _, ip := range ips {
+			pingPod(ip, node, "test-me-network")
+		}
+
+		finalMetrics := getMetricsForNode(node)
+		finalRxBytes := getCounterForPod(finalMetrics, pod, "sriov_vf_rx_bytes")
+		finalRxPackets := getCounterForPod(finalMetrics, pod, "sriov_vf_rx_packets")
+
+		Expect(finalRxBytes).Should(BeNumerically(">", initialRxBytes))
+		Expect(finalRxPackets).Should(BeNumerically(">", initialRxPackets+3))
+	})
+
+})
+
+func getMetricsForNode(nodeName string) map[string]*dto.MetricFamily {
+	metricsExporterPods, err := clients.Pods(operatorNamespace).List(context.Background(), metav1.ListOptions{
+		LabelSelector: "app=sriov-network-metrics-exporter",
+		FieldSelector: "spec.nodeName=" + nodeName,
+	})
+	ExpectWithOffset(1, err).ToNot(HaveOccurred())
+	ExpectWithOffset(1, metricsExporterPods.Items).ToNot(HaveLen(0), "At least one operator pod expected")
+
+	metricsExporterPod := metricsExporterPods.Items[0]
+
+	command := []string{"curl", "http://127.0.0.1:9110/metrics"}
+	stdout, stderr, err := pod.ExecCommand(clients, &metricsExporterPod, command...)
+	Expect(err).ToNot(HaveOccurred(),
+		"pod: [%s/%s] command: [%v]\nstdout: %s\nstderr: %s", metricsExporterPod.Namespace, metricsExporterPod.Name, command, stdout, stderr)
+
+	// Clean the scraped output from carriage returns
+	stdout = strings.ReplaceAll(stdout, "\r", "")
+
+	var parser expfmt.TextParser
+	mf, err := parser.TextToMetricFamilies(strings.NewReader(stdout))
+	Expect(err).ToNot(HaveOccurred())
+
+	return mf
+}
+
+func getCounterForPod(mf map[string]*dto.MetricFamily, p *corev1.Pod, metricName string) float64 {
+	pciAddress := findPciAddressForPod(mf, p)
+	return findCounterForPciAddr(mf, pciAddress, metricName)
+}
+
+func findPciAddressForPod(mf map[string]*dto.MetricFamily, p *corev1.Pod) string {
+	kubePodDeviceMetric := findKubePodDeviceMetric(mf, p)
+	for _, labelPair := range kubePodDeviceMetric.Label {
+		if labelPair.GetName() == "pciAddr" {
+			return *labelPair.Value
+		}
+	}
+
+	Fail(fmt.Sprintf("Can't find PCI Address for pod  [%s/%s] in metrics %+v", p.Name, p.Namespace, mf))
+	return ""
+}
+
+func findKubePodDeviceMetric(mf map[string]*dto.MetricFamily, pod *corev1.Pod) *dto.Metric {
+	metricFamily, ok := mf["sriov_kubepoddevice"]
+	Expect(ok).To(BeTrue(), "sriov_kubepoddevice metric not found: %+v", mf)
+
+	kubePodDeviceMetric := findMetricForPod(metricFamily.Metric, pod)
+	Expect(kubePodDeviceMetric).ToNot(BeNil(), "sriov_kubepoddevice metric for pod [%s/%s]  not found: %+v", pod.Name, pod.Namespace, mf)
+
+	return kubePodDeviceMetric
+}
+
+func findCounterForPciAddr(mf map[string]*dto.MetricFamily, pciAddress string, metricName string) float64 {
+	metricFamily, ok := mf[metricName]
+	Expect(ok).To(BeTrue(), "metric %s not found: %+v", metricName, mf)
+
+	metric := findMetricFor(metricFamily.Metric, map[string]string{
+		"pciAddr": pciAddress,
+	})
+	Expect(metric).ToNot(BeNil(), "metric %s for pciAddr %s not found: %+v", metricName, pciAddress, mf)
+
+	return *metric.GetCounter().Value
+}
+
+func findMetricForPod(metrics []*dto.Metric, pod *corev1.Pod) *dto.Metric {
+	return findMetricFor(metrics, map[string]string{
+		"pod":       pod.Name,
+		"namespace": pod.Namespace,
+	})
+}
+
+func findMetricFor(metrics []*dto.Metric, labelsToMatch map[string]string) *dto.Metric {
+	for _, metric := range metrics {
+		if areLabelsMatching(metric.Label, labelsToMatch) {
+			return metric
+		}
+	}
+
+	return nil
+}
+
+func areLabelsMatching(labels []*dto.LabelPair, labelsToMatch map[string]string) bool {
+	for _, labelPair := range labels {
+		valueToMatch, ok := labelsToMatch[labelPair.GetName()]
+		if !ok {
+			continue
+		}
+
+		if *labelPair.Value != valueToMatch {
+			return false
+		}
+	}
+
+	return true
+}