chore: Additional upstream metrics Part1

pkg/controllers/metrics/pod/controller.go

@@ -71,6 +71,15 @@ var (
 			Objectives: metrics.SummaryObjectives(),
 		},
 	)
+	podBoundDurationSeconds = prometheus.NewHistogramVec(
+		prometheus.HistogramOpts{
+			Namespace: "karpenter",
+			Subsystem: metrics.PodSubsystem,
+			Name:      "bound_duration_seconds",
+			Help:      "The time from pod creation until the pod is bound.",
+		},
+		labelNames(),
+	)
 )
 
 // Controller for the resource
@@ -82,7 +91,7 @@ type Controller struct {
 }
 
 func init() {
-	crmetrics.Registry.MustRegister(podState, podStartupDurationSeconds)
+	crmetrics.Registry.MustRegister(podState, podStartupDurationSeconds, podBoundDurationSeconds)
 }
 
 func labelNames() []string {
@@ -132,13 +141,19 @@ func (c *Controller) Reconcile(ctx context.Context, req reconcile.Request) (reco
 			Labels:   labels,
 		},
 	})
-	c.recordPodStartupMetric(pod)
+	c.recordPodStartupMetric(pod, labels)
 	return reconcile.Result{}, nil
 }
 
-func (c *Controller) recordPodStartupMetric(pod *corev1.Pod) {
+func (c *Controller) recordPodStartupMetric(pod *corev1.Pod, labels prometheus.Labels) {
 	key := client.ObjectKeyFromObject(pod).String()
 	if pod.Status.Phase == phasePending {
+		cond, ok := lo.Find(pod.Status.Conditions, func(c corev1.PodCondition) bool {
+			return c.Type == corev1.PodScheduled
+		})
+		if ok && cond.Status == corev1.ConditionTrue {
+			podBoundDurationSeconds.With(labels).Observe(cond.LastTransitionTime.Sub(pod.CreationTimestamp.Time).Seconds())
+		}
 		c.pendingPods.Insert(key)
 		return
 	}

pkg/controllers/metrics/pod/suite_test.go

@@ -20,6 +20,8 @@ import (
 	"context"
 	"testing"
 
+	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
+
 	. "github.com/onsi/ginkgo/v2"
 	. "github.com/onsi/gomega"
 	corev1 "k8s.io/api/core/v1"
@@ -84,6 +86,18 @@ var _ = Describe("Pod Metrics", func() {
 		})
 		Expect(found).To(BeTrue())
 	})
+	It("should update the pod bound_duration_seconds metric", func() {
+		p := test.Pod()
+		p.Status.Phase = corev1.PodPending
+		p.Status.Conditions = []corev1.PodCondition{{Type: corev1.PodScheduled, Status: corev1.ConditionTrue, LastTransitionTime: metav1.Now()}}
+		ExpectApplied(ctx, env.Client, p)
+		ExpectReconcileSucceeded(ctx, podController, client.ObjectKeyFromObject(p))
+		_, found := FindMetricWithLabelValues("karpenter_pods_bound_duration_seconds", map[string]string{
+			"name":      p.GetName(),
+			"namespace": p.GetNamespace(),
+		})
+		Expect(found).To(BeTrue())
+	})
 	It("should delete the pod state metric on pod delete", func() {
 		p := test.Pod()
 		ExpectApplied(ctx, env.Client, p)

pkg/controllers/node/termination/controller.go

@@ -129,6 +129,9 @@ func (c *Controller) finalize(ctx context.Context, node *corev1.Node) (reconcile
 
 		return reconcile.Result{RequeueAfter: 1 * time.Second}, nil
 	}
+	NodesDrainedTotal.With(prometheus.Labels{
+		metrics.NodePoolLabel: node.Labels[v1.NodePoolLabelKey],
+	}).Inc()
 	// In order for Pods associated with PersistentVolumes to smoothly migrate from the terminating Node, we wait
 	// for VolumeAttachments of drain-able Pods to be cleaned up before terminating Node and removing its finalizer.
 	// However, if TerminationGracePeriod is configured for Node, and we are past that period, we will skip waiting.

pkg/controllers/node/termination/metrics.go

@@ -28,7 +28,8 @@ import (
 func init() {
 	crmetrics.Registry.MustRegister(
 		TerminationDurationSeconds,
-		NodeLifetimeDurationSeconds)
+		NodeLifetimeDurationSeconds,
+		NodesDrainedTotal)
 }
 
 const dayDuration = time.Hour * 24
@@ -44,6 +45,15 @@ var (
 		},
 		[]string{metrics.NodePoolLabel},
 	)
+	NodesDrainedTotal = prometheus.NewCounterVec(
+		prometheus.CounterOpts{
+			Namespace: metrics.Namespace,
+			Subsystem: metrics.NodeSubsystem,
+			Name:      "drained_total",
+			Help:      "The total number of nodes drained by Karpenter",
+		},
+		[]string{metrics.NodePoolLabel},
+	)
 	NodeLifetimeDurationSeconds = prometheus.NewHistogramVec(
 		prometheus.HistogramOpts{
 			Namespace: metrics.Namespace,

pkg/controllers/node/termination/suite_test.go

@@ -95,6 +95,7 @@ var _ = Describe("Termination", func() {
 		metrics.NodesTerminatedTotal.Reset()
 		termination.TerminationDurationSeconds.Reset()
 		termination.NodeLifetimeDurationSeconds.Reset()
+		termination.NodesDrainedTotal.Reset()
 	})
 
 	Context("Reconciliation", func() {
@@ -841,6 +842,7 @@ var _ = Describe("Termination", func() {
 			node = ExpectNodeExists(ctx, env.Client, node.Name)
 			// Reconcile twice, once to set the NodeClaim to terminating, another to check the instance termination status (and delete the node).
 			ExpectObjectReconciled(ctx, env.Client, terminationController, node)
+			ExpectMetricCounterValue(termination.NodesDrainedTotal, 1, map[string]string{"nodepool": node.Labels[v1.NodePoolLabelKey]})
 			ExpectObjectReconciled(ctx, env.Client, terminationController, node)
 
 			m, ok := FindMetricWithLabelValues("karpenter_nodes_terminated_total", map[string]string{"nodepool": node.Labels[v1.NodePoolLabelKey]})

pkg/controllers/provisioning/provisioner.go

@@ -162,6 +162,7 @@ func (p *Provisioner) GetPendingPods(ctx context.Context) ([]*corev1.Pod, error)
 	pods = lo.Reject(pods, func(po *corev1.Pod, _ int) bool {
 		if err := p.Validate(ctx, po); err != nil {
 			log.FromContext(ctx).WithValues("Pod", klog.KRef(po.Namespace, po.Name)).V(1).Info(fmt.Sprintf("ignoring pod, %s", err))
+			metrics.IgnoredPodTotal.Inc()
 			return true
 		}
 		return false
@@ -346,6 +347,7 @@ func (p *Provisioner) Schedule(ctx context.Context) (scheduler.Results, error) {
 	results := s.Solve(ctx, pods).TruncateInstanceTypes(scheduler.MaxInstanceTypes)
 	if len(results.NewNodeClaims) > 0 {
 		log.FromContext(ctx).WithValues("Pods", pretty.Slice(lo.Map(pods, func(p *corev1.Pod, _ int) string { return klog.KRef(p.Namespace, p.Name).String() }), 5), "duration", time.Since(start)).Info("found provisionable pod(s)")
+		scheduler.PodsNominatedTotal.With(prometheus.Labels{scheduler.ControllerLabel: injection.GetControllerName(ctx)}).Add(float64(len(pods)))
 	}
 	results.Record(ctx, p.recorder, p.cluster)
 	return results, nil

pkg/controllers/provisioning/scheduling/metrics.go

@@ -24,7 +24,7 @@ import (
 )
 
 func init() {
-	crmetrics.Registry.MustRegister(SchedulingDurationSeconds, QueueDepth)
+	crmetrics.Registry.MustRegister(SchedulingDurationSeconds, QueueDepth, PodsNominatedTotal)
 }
 
 const (
@@ -58,4 +58,15 @@ var (
 			schedulingIDLabel,
 		},
 	)
+	PodsNominatedTotal = prometheus.NewCounterVec(
+		prometheus.CounterOpts{
+			Namespace: metrics.Namespace,
+			Subsystem: schedulerSubsystem,
+			Name:      "pods_nominated_total",
+			Help:      "The number of pods that the scheduler has processed and made a decision for.",
+		},
+		[]string{
+			ControllerLabel,
+		},
+	)
 )

pkg/controllers/provisioning/suite_test.go

@@ -22,6 +22,8 @@ import (
 	"testing"
 	"time"
 
+	schedulingMetrics "sigs.k8s.io/karpenter/pkg/controllers/provisioning/scheduling"
+
 	. "github.com/onsi/ginkgo/v2"
 	. "github.com/onsi/gomega"
 	"github.com/samber/lo"
@@ -97,13 +99,15 @@ var _ = AfterEach(func() {
 	ExpectCleanedUp(ctx, env.Client)
 	cloudProvider.Reset()
 	cluster.Reset()
+	schedulingMetrics.PodsNominatedTotal.Reset()
 })
 
 var _ = Describe("Provisioning", func() {
 	It("should provision nodes", func() {
 		ExpectApplied(ctx, env.Client, test.NodePool())
 		pod := test.UnschedulablePod()
 		ExpectProvisioned(ctx, env.Client, cluster, cloudProvider, prov, pod)
+		ExpectMetricCounterValue(schedulingMetrics.PodsNominatedTotal, 1, nil)
 		nodes := &corev1.NodeList{}
 		Expect(env.Client.List(ctx, nodes)).To(Succeed())
 		Expect(len(nodes.Items)).To(Equal(1))

pkg/metrics/metrics.go

@@ -89,9 +89,16 @@ var (
 			NodePoolLabel,
 		},
 	)
+	IgnoredPodTotal = prometheus.NewGauge(
+		prometheus.GaugeOpts{
+			Namespace: Namespace,
+			Name:      "ignored_pod_total",
+			Help:      "Number of pods ignored during scheduling by Karpenter",
+		},
+	)
 )
 
 func init() {
 	crmetrics.Registry.MustRegister(NodeClaimsCreatedTotal, NodeClaimsTerminatedTotal, NodeClaimsDisruptedTotal,
-		NodesCreatedTotal, NodesTerminatedTotal)
+		NodesCreatedTotal, NodesTerminatedTotal, IgnoredPodTotal)
 }