metrics: add cilium_datapath_nat_gc_entries

Signed-off-by: ArthurChiao <arthurchiao@hotmail.com>

pkg/maps/ctmap/ctmap.go

@@ -91,6 +91,9 @@ const (
 
 	metricsAlive   = "alive"
 	metricsDeleted = "deleted"
+
+	metricsIngress = "ingress"
+	metricsEgress  = "egress"
 )
 
 type action int
@@ -555,7 +558,13 @@ func PurgeOrphanNATEntries(ctMapTCP, ctMapAny *Map) *NatGCStats {
 		return nil
 	}
 
-	stats := newNatGCStats(natMap)
+	family := gcFamilyIPv4
+	if ctMapTCP.mapType.isIPv6() {
+		family = gcFamilyIPv6
+	}
+	stats := newNatGCStats(natMap, family)
+	defer stats.finish()
+
 	cb := func(key bpf.MapKey, value bpf.MapValue) {
 		natKey := key.(nat.NatKey)
 		natVal := value.(nat.NatEntry)

pkg/maps/ctmap/metrics.go

@@ -126,15 +126,26 @@ func (s *gcStats) finish() {
 type NatGCStats struct {
 	*bpf.DumpStats
 
+	// family is the address family
+	Family gcFamily
+
 	IngressAlive   uint32
 	IngressDeleted uint32
 	EgressDeleted  uint32
 	// It's not possible with the current PurgeOrphanNATEntries implementation
 	// to correctly count EgressAlive, so skip it
 }
 
-func newNatGCStats(m NatMap) NatGCStats {
+func newNatGCStats(m NatMap, family gcFamily) NatGCStats {
 	return NatGCStats{
 		DumpStats: m.DumpStats(),
+		Family:    family,
 	}
 }
+
+func (s *NatGCStats) finish() {
+	family := s.Family.String()
+	metrics.NatGCSize.WithLabelValues(family, metricsIngress, metricsAlive).Set(float64(s.IngressAlive))
+	metrics.NatGCSize.WithLabelValues(family, metricsIngress, metricsDeleted).Set(float64(s.IngressDeleted))
+	metrics.NatGCSize.WithLabelValues(family, metricsEgress, metricsDeleted).Set(float64(s.EgressDeleted))
+}

pkg/metrics/metrics.go

@@ -314,6 +314,9 @@ var (
 	// ConntrackGCSize the number of entries in the conntrack table
 	ConntrackGCSize = NoOpGaugeVec
 
+	// NatGCSize the number of entries in the nat table
+	NatGCSize = NoOpGaugeVec
+
 	// ConntrackGCDuration the duration of the conntrack GC process in milliseconds.
 	ConntrackGCDuration = NoOpObserverVec
 
@@ -898,6 +901,16 @@ func CreateConfiguration(metricsEnabled []string) (Configuration, []prometheus.C
 			collectors = append(collectors, ConntrackGCSize)
 			c.ConntrackGCSizeEnabled = true
 
+			NatGCSize = prometheus.NewGaugeVec(prometheus.GaugeOpts{
+				Namespace: Namespace,
+				Subsystem: SubsystemDatapath,
+				Name:      "nat_gc_entries",
+				Help: "The number of alive and deleted nat entries at the end " +
+					"of a garbage collector run labeled by datapath family.",
+			}, []string{LabelDatapathFamily, LabelDirection, LabelStatus})
+
+			collectors = append(collectors, NatGCSize)
+
 		case Namespace + "_" + SubsystemDatapath + "_conntrack_gc_duration_seconds":
 			ConntrackGCDuration = prometheus.NewHistogramVec(prometheus.HistogramOpts{
 				Namespace: Namespace,