Move primary IP count to separate metric to not affect the IPAM math (#1579)

move primary IP count to separate metric to not affect the ipam math

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

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

Author: rbtr

cns/ipampool/metrics.go

@@ -78,6 +78,14 @@ var (
 		},
 		[]string{subnetLabel, subnetCIDRLabel, podnetARMIDLabel},
 	)
+	ipamPrimaryIPCount = prometheus.NewGaugeVec(
+		prometheus.GaugeOpts{
+			Name:        "cx_ipam_primary_ips",
+			Help:        "NC Primary IP count.",
+			ConstLabels: prometheus.Labels{customerMetricLabel: customerMetricLabelValue},
+		},
+		[]string{subnetLabel, subnetCIDRLabel, podnetARMIDLabel},
+	)
 	ipamRequestedIPConfigCount = prometheus.NewGaugeVec(
 		prometheus.GaugeOpts{
 			Name:        "cx_ipam_requested_ips",
@@ -106,6 +114,7 @@ func init() {
 		ipamMaxIPCount,
 		ipamPendingProgramIPCount,
 		ipamPendingReleaseIPCount,
+		ipamPrimaryIPCount,
 		ipamRequestedIPConfigCount,
 		ipamTotalIPCount,
 	)
@@ -120,6 +129,7 @@ func observeIPPoolState(state ipPoolState, meta metaState, labels []string) {
 	ipamMaxIPCount.WithLabelValues(labels...).Set(float64(meta.max))
 	ipamPendingProgramIPCount.WithLabelValues(labels...).Set(float64(state.pendingProgramming))
 	ipamPendingReleaseIPCount.WithLabelValues(labels...).Set(float64(state.pendingRelease))
+	ipamPrimaryIPCount.WithLabelValues(labels...).Set(float64(len(meta.primaryIPAddresses)))
 	ipamRequestedIPConfigCount.WithLabelValues(labels...).Set(float64(state.requestedIPs))
 	ipamTotalIPCount.WithLabelValues(labels...).Set(float64(state.totalIPs))
 }

cns/ipampool/monitor.go

@@ -178,9 +178,9 @@ type ipPoolState struct {
 	totalIPs int64
 }
 
-func buildIPPoolState(ips map[string]cns.IPConfigurationStatus, spec v1alpha.NodeNetworkConfigSpec, primaryIPAddresses map[string]struct{}) ipPoolState {
+func buildIPPoolState(ips map[string]cns.IPConfigurationStatus, spec v1alpha.NodeNetworkConfigSpec) ipPoolState {
 	state := ipPoolState{
-		totalIPs:     int64(len(primaryIPAddresses) + len(ips)),
+		totalIPs:     int64(len(ips)),
 		requestedIPs: spec.RequestedIPCount,
 	}
 	for _, v := range ips {
@@ -205,22 +205,22 @@ var statelogDownsample int
 func (pm *Monitor) reconcile(ctx context.Context) error {
 	allocatedIPs := pm.httpService.GetPodIPConfigState()
 	meta := pm.metastate
-	state := buildIPPoolState(allocatedIPs, pm.spec, meta.primaryIPAddresses)
+	state := buildIPPoolState(allocatedIPs, pm.spec)
 	observeIPPoolState(state, meta, []string{subnet, subnetCIDR, subnetARMID})
 
+	// log every 30th reconcile to reduce the AI load. we will always log when the monitor
+	// changes the pool, below.
+	if statelogDownsample = (statelogDownsample + 1) % 30; statelogDownsample == 0 { //nolint:gomnd //downsample by 30
+		logger.Printf("ipam-pool-monitor state: %+v, meta: %+v", state, meta)
+	}
+
 	// if the subnet is exhausted, overwrite the batch/minfree/maxfree in the meta copy for this iteration
 	if meta.exhausted {
 		meta.batch = 1
 		meta.minFreeCount = 1
 		meta.maxFreeCount = 2
 	}
 
-	// log every 30th reconcile to reduce the AI load. we will always log when the monitor
-	// changes the pool, below.
-	if statelogDownsample = (statelogDownsample + 1) % 30; statelogDownsample == 0 { //nolint:gomnd //downsample by 30
-		logger.Printf("ipam-pool-monitor state %+v", state)
-	}
-
 	switch {
 	// pod count is increasing
 	case state.expectedAvailableIPs < meta.minFreeCount: