Ensures prometheus metrics associated with a deleted node are no long…

…er reported.

[ upstream commit e9f97cd ]

When a node is deleted from a cluster, metrics associated with that node
are still being exported to prometheus. Short of restarting the agent,
we want to dynamically delete these metrics when a node is removed from the cluster.

This PR ensures node_connectivity_status and node_connectivity_latency
no longer report metrics for nodes that are no longer present on the
cluster.

[ Backporter's notes: Original PR was adapted! ]

The original PR depends (mainly!) on 2 other PRs that haven't been
backported and are fairly substential.
Given this, I've opted to adapt the original implementation to surface the
fix while minimizing impact with 2 updates:
1. pkg/metrics/interfaces did not introduce pkg/metrics/metric wrappers
  as of this release. Hence adapted deletableVec to use the current
implementation. (Referring to commit: 84ea383)
2. pkg/node/manager/manager was adapted to provide for metrics deletion when a
   node is deleted. Subsequent PR refactored the manager metrics structure which
   the original PR used. (Referring to commit: c49ef45)

Signed-off-by: Fernand Galiana <fernand.galiana@isovalent.com>

pkg/metrics/interfaces.go

@@ -27,13 +27,13 @@ type CounterVec interface {
 }
 
 type GaugeVec interface {
+	deletableVec
+
 	WithLabelValues(lvls ...string) prometheus.Gauge
 	prometheus.Collector
 }
 
-type DeletableGaugeVec interface {
-	GaugeVec
-
+type deletableVec interface {
 	Delete(ll prometheus.Labels) bool
 	DeleteLabelValues(lvs ...string) bool
 	DeletePartialMatch(labels prometheus.Labels) int
@@ -44,13 +44,12 @@ var (
 	NoOpMetric    prometheus.Metric    = &metric{}
 	NoOpCollector prometheus.Collector = &collector{}
 
-	NoOpCounter           prometheus.Counter     = &counter{NoOpMetric, NoOpCollector}
-	NoOpCounterVec        CounterVec             = &counterVec{NoOpCollector}
-	NoOpObserver          prometheus.Observer    = &observer{}
-	NoOpObserverVec       prometheus.ObserverVec = &observerVec{NoOpCollector}
-	NoOpGauge             prometheus.Gauge       = &gauge{NoOpMetric, NoOpCollector}
-	NoOpGaugeVec          GaugeVec               = &gaugeVec{NoOpCollector}
-	NoOpGaugeDeletableVec DeletableGaugeVec      = &gaugeDeletableVec{gaugeVec{NoOpCollector}}
+	NoOpCounter     prometheus.Counter     = &counter{NoOpMetric, NoOpCollector}
+	NoOpCounterVec  CounterVec             = &counterVec{NoOpCollector}
+	NoOpObserver    prometheus.Observer    = &observer{}
+	NoOpObserverVec prometheus.ObserverVec = &observerVec{NoOpCollector}
+	NoOpGauge       prometheus.Gauge       = &gauge{NoOpMetric, NoOpCollector}
+	NoOpGaugeVec    GaugeVec               = &gaugeVec{NoOpCollector}
 )
 
 // Metric
@@ -136,28 +135,20 @@ func (g *gauge) SetToCurrentTime() {}
 
 // GaugeVec
 
-type gaugeDeletableVec struct {
-	gaugeVec
+type gaugeVec struct {
+	prometheus.Collector
 }
 
-func (*gaugeDeletableVec) Delete(ll prometheus.Labels) bool {
+func (*gaugeVec) Delete(ll prometheus.Labels) bool {
 	return false
 }
-
-func (*gaugeDeletableVec) DeleteLabelValues(lvs ...string) bool {
+func (*gaugeVec) DeleteLabelValues(lvs ...string) bool {
 	return false
 }
-
-func (*gaugeDeletableVec) DeletePartialMatch(labels prometheus.Labels) int {
+func (*gaugeVec) DeletePartialMatch(labels prometheus.Labels) int {
 	return 0
 }
-
-func (*gaugeDeletableVec) Reset() {}
-
-type gaugeVec struct {
-	prometheus.Collector
-}
-
+func (*gaugeVec) Reset() {}
 func (gv *gaugeVec) WithLabelValues(lvls ...string) prometheus.Gauge {
 	return NoOpGauge
 }

pkg/metrics/metrics.go

@@ -229,11 +229,11 @@ var (
 
 	// NodeConnectivityStatus is the connectivity status between local node to
 	// other node intra or inter cluster.
-	NodeConnectivityStatus = NoOpGaugeDeletableVec
+	NodeConnectivityStatus = NoOpGaugeVec
 
 	// NodeConnectivityLatency is the connectivity latency between local node to
 	// other node intra or inter cluster.
-	NodeConnectivityLatency = NoOpGaugeDeletableVec
+	NodeConnectivityLatency = NoOpGaugeVec
 
 	// Endpoint