fix(notifier): wip

Signed-off-by: machine424 <ayoubmrini424@gmail.com>

notifier/notifier.go

@@ -196,6 +196,7 @@ func newAlertMetrics(r prometheus.Registerer, queueCap int, queueLen, alertmanag
 	m.queueCapacity.Set(float64(queueCap))
 
 	if r != nil {
+
 		r.MustRegister(
 			m.latency,
 			m.errors,
@@ -267,6 +268,15 @@ func (n *Manager) ApplyConfig(conf *config.Config) error {
 		amSets[k] = ams
 	}
 
+	// Drop series of the previous alertmanagers.
+	// TODO: This may result in unneeded
+	 resets? (if nothing changes?)
+	for _, ams := range n.alertmanagers {
+		for _, am := range ams.ams {
+			ams.dropMetrics(am.url().String())
+		}
+	}
+
 	n.alertmanagers = amSets
 
 	return nil
@@ -752,6 +762,18 @@ func newAlertmanagerSet(cfg *config.AlertmanagerConfig, logger log.Logger, metri
 	return s, nil
 }
 
+func (s *alertmanagerSet) initializeMetrics(lvs ...string) {
+	// This will initialize the Counters for the AM to 0.
+	s.metrics.sent.WithLabelValues(lvs...)
+	s.metrics.errors.WithLabelValues(lvs...)
+}
+
+func (s *alertmanagerSet) dropMetrics(lvs ...string) {
+	s.metrics.latency.DeleteLabelValues(lvs...)
+	s.metrics.sent.DeleteLabelValues(lvs...)
+	s.metrics.errors.DeleteLabelValues(lvs...)
+}
+
 // sync extracts a deduplicated set of Alertmanager endpoints from a list
 // of target groups definitions.
 func (s *alertmanagerSet) sync(tgs []*targetgroup.Group) {
@@ -782,11 +804,7 @@ func (s *alertmanagerSet) sync(tgs []*targetgroup.Group) {
 		if _, ok := seen[us]; ok {
 			continue
 		}
-
-		// This will initialize the Counters for the AM to 0.
-		s.metrics.sent.WithLabelValues(us)
-		s.metrics.errors.WithLabelValues(us)
-
+		s.initializeMetrics(us)
 		seen[us] = struct{}{}
 		s.ams = append(s.ams, am)
 	}
@@ -796,9 +814,7 @@ func (s *alertmanagerSet) sync(tgs []*targetgroup.Group) {
 		if _, ok := seen[us]; ok {
 			continue
 		}
-		s.metrics.latency.DeleteLabelValues(us)
-		s.metrics.sent.DeleteLabelValues(us)
-		s.metrics.errors.DeleteLabelValues(us)
+		s.dropMetrics(us)
 		seen[us] = struct{}{}
 	}
 }

notifier/notifier_test.go

@@ -1017,3 +1017,85 @@ func TestStop_DrainingEnabled(t *testing.T) {
 
 	require.Equal(t, int64(2), alertsReceived.Load())
 }
+
+func metricsWithStringAsLabelValue(g prometheus.Gatherer, s string) ([]string, error) {
+	families, err := g.Gather()
+	if err != nil {
+		return nil, err
+	}
+
+	metrics := []string{}
+	for _, f := range families {
+		for _, m := range f.GetMetric() {
+			for _, v := range m.GetLabel() {
+				if v.GetValue() == s {
+					metrics = append(metrics, f.GetName())
+				}
+			}
+		}
+	}
+	return metrics, nil
+}
+
+func TestAlertMetrics(t *testing.T) {
+	targetGroup := func(s string) *targetgroup.Group {
+		return &targetgroup.Group{
+			Targets: []model.LabelSet{
+				{
+					"__address__": model.LabelValue(s),
+				},
+			},
+		}
+	}
+	alertmanagerURL := func(s string) string {
+		return fmt.Sprintf("http://%s/api/v2/alerts", s)
+	}
+
+	reg := prometheus.NewRegistry()
+	n := NewManager(&Options{Registerer: reg}, nil)
+	cfg := &config.Config{}
+	s := `
+alerting:
+  alertmanagers:
+  - static_configs:
+`
+
+	targetURL1 := "alertmanager:9093"
+	require.NoError(t, yaml.UnmarshalStrict([]byte(s), cfg))
+	require.Len(t, cfg.AlertingConfig.AlertmanagerConfigs, 1)
+
+	require.NoError(t, n.ApplyConfig(cfg))
+	tgs := map[string][]*targetgroup.Group{"config-0": {targetGroup(targetURL1)}}
+	n.reload(tgs)
+
+	metrics, err := metricsWithStringAsLabelValue(reg, alertmanagerURL(targetURL1))
+	require.NoError(t, err)
+	// Corresponds to:
+	// metrics.sent
+	// metrics.errors
+	require.Len(t, metrics, 2)
+
+	// The alertmanager targer gets changed.
+	targetURL2 := "alertmanager:9094"
+	tgs = map[string][]*targetgroup.Group{"config-0": {targetGroup(targetURL2)}}
+	n.reload(tgs)
+
+	// targetURL1 related series were dropped.
+	metrics, err = metricsWithStringAsLabelValue(reg, alertmanagerURL(targetURL1))
+	require.NoError(t, err)
+	require.Len(t, metrics, 0)
+
+	s = `
+alerting:
+  alertmanagers:
+`
+	// Drop the config.
+	require.NoError(t, yaml.UnmarshalStrict([]byte(s), cfg))
+	require.Len(t, cfg.AlertingConfig.AlertmanagerConfigs, 0)
+
+	require.NoError(t, n.ApplyConfig(cfg))
+	// targetURL2 related series were dropped.
+	metrics, err = metricsWithStringAsLabelValue(reg, alertmanagerURL(targetURL2))
+	require.NoError(t, err)
+	require.Len(t, metrics, 0)
+}