Merge pull request kubevirt#13428 from machadovilaca/CNV-51568-51569

Add kubevirt_vmi_migration_(start|end)_time_seconds metrics

docs/observability/metrics.md

@@ -189,12 +189,18 @@ The rate of memory being dirty in the Guest OS. Type: Gauge.
 ### kubevirt_vmi_migration_disk_transfer_rate_bytes
 The rate at which the memory is being transferred. Type: Gauge.
 
+### kubevirt_vmi_migration_end_time_seconds
+The time at which the migration ended. Type: Gauge.
+
 ### kubevirt_vmi_migration_failed
 Indicates if the VMI migration failed. Type: Gauge.
 
 ### kubevirt_vmi_migration_phase_transition_time_from_creation_seconds
 Histogram of VM migration phase transitions duration from creation time in seconds. Type: Histogram.
 
+### kubevirt_vmi_migration_start_time_seconds
+The time at which the migration started. Type: Gauge.
+
 ### kubevirt_vmi_migration_succeeded
 Indicates if the VMI migration succeeded. Type: Gauge.
 

pkg/monitoring/metrics/virt-controller/BUILD.bazel

@@ -31,6 +31,7 @@ go_library(
         "//vendor/k8s.io/api/core/v1:go_default_library",
         "//vendor/k8s.io/apimachinery/pkg/api/resource:go_default_library",
         "//vendor/k8s.io/apimachinery/pkg/apis/meta/v1:go_default_library",
+        "//vendor/k8s.io/apimachinery/pkg/types:go_default_library",
         "//vendor/k8s.io/client-go/tools/cache:go_default_library",
     ],
 )

pkg/monitoring/metrics/virt-controller/vmistats_collector.go

@@ -26,6 +26,7 @@ import (
 	"github.com/machadovilaca/operator-observability/pkg/operatormetrics"
 	k8sv1 "k8s.io/api/core/v1"
 	v1 "k8s.io/apimachinery/pkg/apis/meta/v1"
+	"k8s.io/apimachinery/pkg/types"
 	"k8s.io/client-go/tools/cache"
 
 	"kubevirt.io/client-go/log"
@@ -55,6 +56,8 @@ var (
 			vmiInfo,
 			vmiEvictionBlocker,
 			vmiAddresses,
+			vmiMigrationStartTime,
+			vmiMigrationEndTime,
 		},
 		CollectCallback: vmiStatsCollectorCallback,
 	}
@@ -97,6 +100,22 @@ var (
 		},
 		[]string{"node", "namespace", "name", "network_name", "address", "type"},
 	)
+
+	vmiMigrationStartTime = operatormetrics.NewGaugeVec(
+		operatormetrics.MetricOpts{
+			Name: "kubevirt_vmi_migration_start_time_seconds",
+			Help: "The time at which the migration started.",
+		},
+		[]string{"node", "namespace", "name", "migration_name"},
+	)
+
+	vmiMigrationEndTime = operatormetrics.NewGaugeVec(
+		operatormetrics.MetricOpts{
+			Name: "kubevirt_vmi_migration_end_time_seconds",
+			Help: "The time at which the migration ended.",
+		},
+		[]string{"node", "namespace", "name", "migration_name", "status"},
+	)
 )
 
 func vmiStatsCollectorCallback() []operatormetrics.CollectorResult {
@@ -122,6 +141,7 @@ func reportVmisStats(vmis []*k6tv1.VirtualMachineInstance) []operatormetrics.Col
 		crs = append(crs, collectVMIInfo(vmi))
 		crs = append(crs, getEvictionBlocker(vmi))
 		crs = append(crs, collectVMIInterfacesInfo(vmi)...)
+		crs = append(crs, collectVMIMigrationTime(vmi)...)
 	}
 
 	return crs
@@ -330,3 +350,64 @@ func collectVMIInterfaceInfo(vmi *k6tv1.VirtualMachineInstance, iface k6tv1.Virt
 		Value: 1.0,
 	}
 }
+
+func collectVMIMigrationTime(vmi *k6tv1.VirtualMachineInstance) []operatormetrics.CollectorResult {
+	var cr []operatormetrics.CollectorResult
+	var migrationName string
+
+	if vmi.Status.MigrationState == nil {
+		return cr
+	}
+
+	migrationName = getMigrationNameFromMigrationUID(vmi.Namespace, vmi.Status.MigrationState.MigrationUID)
+
+	if vmi.Status.MigrationState.StartTimestamp != nil {
+		cr = append(cr, operatormetrics.CollectorResult{
+			Metric: vmiMigrationStartTime,
+			Value:  float64(vmi.Status.MigrationState.StartTimestamp.Time.Unix()),
+			Labels: []string{vmi.Status.NodeName, vmi.Namespace, vmi.Name, migrationName},
+		})
+	}
+
+	if vmi.Status.MigrationState.EndTimestamp != nil {
+		cr = append(cr, operatormetrics.CollectorResult{
+			Metric: vmiMigrationEndTime,
+			Value:  float64(vmi.Status.MigrationState.EndTimestamp.Time.Unix()),
+			Labels: []string{vmi.Status.NodeName, vmi.Namespace, vmi.Name, migrationName,
+				calculateMigrationStatus(vmi.Status.MigrationState),
+			},
+		})
+	}
+
+	return cr
+}
+
+func calculateMigrationStatus(migrationState *k6tv1.VirtualMachineInstanceMigrationState) string {
+	if !migrationState.Completed {
+		return ""
+	}
+
+	if migrationState.Failed {
+		return "failed"
+	}
+
+	return "succeeded"
+}
+
+func getMigrationNameFromMigrationUID(namespace string, migrationUID types.UID) string {
+	objs, err := vmiMigrationInformer.GetIndexer().ByIndex(cache.NamespaceIndex, namespace)
+	if err != nil {
+		return none
+	}
+
+	for _, obj := range objs {
+		curMigration := obj.(*k6tv1.VirtualMachineInstanceMigration)
+		if curMigration.UID != migrationUID {
+			continue
+		}
+
+		return curMigration.Name
+	}
+
+	return none
+}

pkg/monitoring/metrics/virt-controller/vmistats_collector_test.go

@@ -403,6 +403,81 @@ var _ = Describe("VMI Stats Collector", func() {
 			Expect(metrics[1].Labels).To(Equal([]string{"testNode", "test-ns", "testvmi", "networkA", "", "InternalIP"}))
 		})
 	})
+
+	Context("VMI migration start and end time metrics", func() {
+		now := metav1.Unix(1000, 0)
+		nowFloatValue := float64(now.Unix())
+
+		Describe("kubevirt_vmi_migration_start_time and kubevirt_vmi_migration_end_time metrics", func() {
+			It("should not create migration metrics for a VMI with no migration state", func() {
+				vmi := &k6tv1.VirtualMachineInstance{
+					ObjectMeta: metav1.ObjectMeta{
+						Namespace: "test-ns",
+						Name:      "testvmi",
+					},
+					Status: k6tv1.VirtualMachineInstanceStatus{
+						NodeName: "testNode",
+					},
+				}
+
+				metrics := collectVMIMigrationTime(vmi)
+				Expect(metrics).To(BeEmpty())
+			})
+
+			It("should create kubevirt_vmi_migration_start_time metric for a migration in progress", func() {
+				vmi := &k6tv1.VirtualMachineInstance{
+					ObjectMeta: metav1.ObjectMeta{
+						Namespace: "test-ns",
+						Name:      "testvmi",
+					},
+					Status: k6tv1.VirtualMachineInstanceStatus{
+						NodeName: "testNode",
+						MigrationState: &k6tv1.VirtualMachineInstanceMigrationState{
+							MigrationUID:   "test-migration-uid",
+							StartTimestamp: &now,
+						},
+					},
+				}
+
+				metrics := collectVMIMigrationTime(vmi)
+				Expect(metrics).To(HaveLen(1))
+
+				Expect(metrics[0].Metric.GetOpts().Name).To(ContainSubstring("kubevirt_vmi_migration_start_time"))
+				Expect(metrics[0].Value).To(BeEquivalentTo(nowFloatValue))
+				Expect(metrics[0].Labels).To(Equal([]string{"testNode", "test-ns", "testvmi", "test-migration"}))
+			})
+
+			It("should create kubevirt_vmi_migration_end_time metric for a completedmigration", func() {
+				vmi := &k6tv1.VirtualMachineInstance{
+					ObjectMeta: metav1.ObjectMeta{
+						Namespace: "test-ns",
+						Name:      "testvmi",
+					},
+					Status: k6tv1.VirtualMachineInstanceStatus{
+						NodeName: "testNode",
+						MigrationState: &k6tv1.VirtualMachineInstanceMigrationState{
+							MigrationUID:   "test-migration-uid",
+							StartTimestamp: &now,
+							EndTimestamp:   &now,
+							Completed:      true,
+							Failed:         false,
+						},
+					},
+				}
+
+				metrics := collectVMIMigrationTime(vmi)
+				Expect(metrics).To(HaveLen(2))
+
+				Expect(metrics[0].Metric.GetOpts().Name).To(ContainSubstring("kubevirt_vmi_migration_start_time"))
+				Expect(metrics[0].Value).To(BeEquivalentTo(nowFloatValue))
+				Expect(metrics[0].Labels).To(Equal([]string{"testNode", "test-ns", "testvmi", "test-migration"}))
+
+				Expect(metrics[1].Metric.GetOpts().Name).To(ContainSubstring("kubevirt_vmi_migration_end_time"))
+				Expect(metrics[1].Value).To(BeEquivalentTo(nowFloatValue))
+				Expect(metrics[1].Labels).To(Equal([]string{"testNode", "test-ns", "testvmi", "test-migration", "succeeded"}))
+			})
+		})
+	})
 })
 
 func interfacesFor(values [][]string) []k6tv1.VirtualMachineInstanceNetworkInterface {
@@ -447,6 +522,7 @@ func setupTestCollector() {
 	preferenceInformer, _ = testutils.NewFakeInformerFor(&instancetypev1beta1.VirtualMachinePreference{})
 	clusterPreferenceInformer, _ = testutils.NewFakeInformerFor(&instancetypev1beta1.VirtualMachineClusterPreference{})
 	kvPodInformer, _ = testutils.NewFakeInformerFor(&k8sv1.Pod{})
+	vmiMigrationInformer, _ = testutils.NewFakeInformerFor(&k6tv1.VirtualMachineInstanceMigration{})
 
 	_ = instanceTypeInformer.GetStore().Add(&instancetypev1beta1.VirtualMachineInstancetype{
 		ObjectMeta: newObjectMetaForInstancetypes("i-managed", "kubevirt.io"),
@@ -476,6 +552,14 @@ func setupTestCollector() {
 	_ = kvPodInformer.GetStore().Add(&k8sv1.Pod{
 		ObjectMeta: newPodMetaForInformer("virt-launcher-testpod", "test-ns", "test-vmi-uid"),
 	})
+
+	_ = vmiMigrationInformer.GetStore().Add(&k6tv1.VirtualMachineInstanceMigration{
+		ObjectMeta: metav1.ObjectMeta{
+			Name:      "test-migration",
+			Namespace: "test-ns",
+			UID:       "test-migration-uid",
+		},
+	})
 }
 
 func newObjectMetaForInstancetypes(name, vendor string) metav1.ObjectMeta {

tests/monitoring/metrics.go

@@ -86,6 +86,8 @@ var _ = Describe("[sig-monitoring]Metrics", decorators.SigMonitoring, func() {
 			"kubevirt_vmi_migration_dirty_memory_rate_bytes":                     true,
 			"kubevirt_vmi_migration_disk_transfer_rate_bytes":                    true,
 			"kubevirt_vmi_migration_data_total_bytes":                            true,
+			"kubevirt_vmi_migration_start_time_seconds":                          true,
+			"kubevirt_vmi_migration_end_time_seconds":                            true,
 		}
 
 		It("should contain virt components metrics", func() {