feat: add metrics for count of known NNC and NCs (#3389)

Signed-off-by: Evan Baker <[email protected]>

Author: rbtr

cns/kubecontroller/nodenetworkconfig/metrics.go

@@ -24,12 +24,26 @@ var (
 			Help: "Unused IP count.",
 		},
 	)
+	hasNNC = prometheus.NewGauge(
+		prometheus.GaugeOpts{
+			Name: "nnc_has_nodenetworkconfig",
+			Help: "Has received a NodeNetworkConfig",
+		},
+	)
+	ncs = prometheus.NewGauge(
+		prometheus.GaugeOpts{
+			Name: "nnc_ncs",
+			Help: "Network Container count in the NodeNetworkConfig",
+		},
+	)
 )
 
 func init() {
 	metrics.Registry.MustRegister(
 		allocatedIPs,
 		requestedIPs,
 		unusedIPs,
+		hasNNC,
+		ncs,
 	)
 }

cns/kubecontroller/nodenetworkconfig/reconciler.go

@@ -64,21 +64,24 @@ func (r *Reconciler) Reconcile(ctx context.Context, req reconcile.Request) (reco
 	nnc, err := r.nnccli.Get(ctx, req.NamespacedName)
 	if err != nil {
 		if apierrors.IsNotFound(err) {
+			hasNNC.Set(0)
 			logger.Printf("[cns-rc] CRD not found, ignoring %v", err)
 			return reconcile.Result{}, errors.Wrapf(client.IgnoreNotFound(err), "NodeNetworkConfig %v not found", req.NamespacedName)
 		}
 		logger.Errorf("[cns-rc] Error retrieving CRD from cache : %v", err)
 		return reconcile.Result{}, errors.Wrapf(err, "failed to get NodeNetworkConfig %v", req.NamespacedName)
 	}
-
+	hasNNC.Set(1)
 	logger.Printf("[cns-rc] CRD Spec: %+v", nnc.Spec)
 
 	ipAssignments := 0
 
 	// during node upgrades, an nnc may be updated with new ncs. at any given time, only the ncs
 	// that exist in the nnc are valid. any others that may have been previously created and no
 	// longer exist in the nnc should be considered stale.
-	validNCIDs := make([]string, len(nnc.Status.NetworkContainers))
+	ncCount := len(nnc.Status.NetworkContainers)
+	ncs.Set(float64(ncCount))
+	validNCIDs := make([]string, ncCount)
 	for i := range nnc.Status.NetworkContainers {
 		validNCIDs[i] = nnc.Status.NetworkContainers[i].ID
 	}

cns/restserver/internalapi.go

@@ -225,6 +225,7 @@ func (service *HTTPRestService) syncHostNCVersion(ctx context.Context, channelMo
 	for _, nc := range ncVersionListResp.Containers {
 		nmaNCs[strings.ToLower(nc.NetworkContainerID)] = nc.Version
 	}
+	hasNC.Set(float64(len(nmaNCs)))
 	for ncID := range outdatedNCs {
 		nmaNCVersionStr, ok := nmaNCs[ncID]
 		if !ok {

cns/restserver/metrics.go

@@ -65,6 +65,12 @@ var (
 		},
 		[]string{"ok"},
 	)
+	hasNC = prometheus.NewGauge(
+		prometheus.GaugeOpts{
+			Name: "has_networkcontainer",
+			Help: "Number of Network Containers retrieved from NMA",
+		},
+	)
 	allocatedIPCount = prometheus.NewGaugeVec(
 		prometheus.GaugeOpts{
 			Name:        "cx_allocated_ips_v2",
@@ -114,6 +120,7 @@ func init() {
 		ipConfigStatusStateTransitionTime,
 		syncHostNCVersionCount,
 		syncHostNCVersionLatency,
+		hasNC,
 		allocatedIPCount,
 		assignedIPCount,
 		availableIPCount,