metrics: remove deleted peer series (#748)

internal/adapters/metrics.go

@@ -136,3 +136,21 @@ func (m *MetricsServer) UpdatePeerMetrics(peer *domain.Peer, status domain.PeerS
 	m.peerSendBytesTotal.WithLabelValues(labels...).Set(float64(status.BytesTransmitted))
 	m.peerIsConnected.WithLabelValues(labels...).Set(internal.BoolToFloat64(status.IsConnected))
 }
+
+// DeletePeerMetrics removes all Prometheus series for a peer that was deleted
+// from the WireGuard portal. GaugeVecs retain label values until explicitly
+// deleted, so leaving these behind makes a deleted peer indistinguishable from
+// a retained peer that is currently offline.
+func (m *MetricsServer) DeletePeerMetrics(peer domain.Peer) {
+	labels := []string{
+		string(peer.InterfaceIdentifier),
+		peer.Interface.AddressStr(),
+		string(peer.Identifier),
+		peer.DisplayName,
+		string(peer.UserIdentifier),
+	}
+	m.peerLastHandshakeSeconds.DeleteLabelValues(labels...)
+	m.peerReceivedBytesTotal.DeleteLabelValues(labels...)
+	m.peerSendBytesTotal.DeleteLabelValues(labels...)
+	m.peerIsConnected.DeleteLabelValues(labels...)
+}

internal/adapters/metrics_test.go

@@ -0,0 +1,77 @@
+package adapters
+
+import (
+	"testing"
+
+	"github.com/h44z/wg-portal/internal/domain"
+	"github.com/prometheus/client_golang/prometheus"
+)
+
+func TestDeletePeerMetricsRemovesAllPeerSeries(t *testing.T) {
+	labels := []string{"interface", "addresses", "id", "name", "user"}
+	m := &MetricsServer{
+		peerIsConnected:          prometheus.NewGaugeVec(prometheus.GaugeOpts{Name: "peer_up_test"}, labels),
+		peerLastHandshakeSeconds: prometheus.NewGaugeVec(prometheus.GaugeOpts{Name: "peer_handshake_test"}, labels),
+		peerReceivedBytesTotal:   prometheus.NewGaugeVec(prometheus.GaugeOpts{Name: "peer_received_test"}, labels),
+		peerSendBytesTotal:       prometheus.NewGaugeVec(prometheus.GaugeOpts{Name: "peer_sent_test"}, labels),
+	}
+	addr, err := domain.CidrFromString("10.0.0.2/32")
+	if err != nil {
+		t.Fatal(err)
+	}
+	peer := &domain.Peer{
+		Identifier:          "peer-id",
+		InterfaceIdentifier: "wg0",
+		DisplayName:         "peer",
+		UserIdentifier:      "user",
+		Interface:           domain.PeerInterfaceConfig{Addresses: []domain.Cidr{addr}},
+	}
+	m.UpdatePeerMetrics(peer, domain.PeerStatus{PeerId: peer.Identifier, IsConnected: true})
+	m.DeletePeerMetrics(*peer)
+
+	for name, collector := range map[string]prometheus.Collector{
+		"up": m.peerIsConnected, "handshake": m.peerLastHandshakeSeconds,
+		"received": m.peerReceivedBytesTotal, "sent": m.peerSendBytesTotal,
+	} {
+		ch := make(chan prometheus.Metric, 1)
+		collector.Collect(ch)
+		close(ch)
+		if _, ok := <-ch; ok {
+			t.Errorf("%s metric remained after peer deletion", name)
+		}
+	}
+}
+
+func TestDeletePeerMetricsLeavesRetainedPeerSeries(t *testing.T) {
+	labels := []string{"interface", "addresses", "id", "name", "user"}
+	m := &MetricsServer{
+		peerIsConnected:          prometheus.NewGaugeVec(prometheus.GaugeOpts{Name: "peer_up_retained_test"}, labels),
+		peerLastHandshakeSeconds: prometheus.NewGaugeVec(prometheus.GaugeOpts{Name: "peer_handshake_retained_test"}, labels),
+		peerReceivedBytesTotal:   prometheus.NewGaugeVec(prometheus.GaugeOpts{Name: "peer_received_retained_test"}, labels),
+		peerSendBytesTotal:       prometheus.NewGaugeVec(prometheus.GaugeOpts{Name: "peer_sent_retained_test"}, labels),
+	}
+	deletedAddr, err := domain.CidrFromString("10.0.0.2/32")
+	if err != nil {
+		t.Fatal(err)
+	}
+	retainedAddr, err := domain.CidrFromString("10.0.0.3/32")
+	if err != nil {
+		t.Fatal(err)
+	}
+	deleted := &domain.Peer{Identifier: "deleted", InterfaceIdentifier: "wg0", DisplayName: "deleted", UserIdentifier: "user", Interface: domain.PeerInterfaceConfig{Addresses: []domain.Cidr{deletedAddr}}}
+	retained := &domain.Peer{Identifier: "retained", InterfaceIdentifier: "wg0", DisplayName: "retained", UserIdentifier: "user", Interface: domain.PeerInterfaceConfig{Addresses: []domain.Cidr{retainedAddr}}}
+	m.UpdatePeerMetrics(deleted, domain.PeerStatus{PeerId: deleted.Identifier, IsConnected: false})
+	m.UpdatePeerMetrics(retained, domain.PeerStatus{PeerId: retained.Identifier, IsConnected: false})
+	m.DeletePeerMetrics(*deleted)
+
+	ch := make(chan prometheus.Metric, 2)
+	m.peerIsConnected.Collect(ch)
+	close(ch)
+	count := 0
+	for range ch {
+		count++
+	}
+	if count != 1 {
+		t.Fatalf("expected one retained peer series, got %d", count)
+	}
+}

internal/app/wireguard/statistics.go

@@ -31,6 +31,7 @@ type StatisticsDatabaseRepo interface {
 type StatisticsMetricsServer interface {
 	UpdateInterfaceMetrics(status domain.InterfaceStatus)
 	UpdatePeerMetrics(peer *domain.Peer, status domain.PeerStatus)
+	DeletePeerMetrics(peer domain.Peer)
 }
 
 type StatisticsEventBus interface {
@@ -433,6 +434,11 @@ func (c *StatisticsCollector) updatePeerMetrics(ctx context.Context, status doma
 
 func (c *StatisticsCollector) connectToMessageBus() {
 	_ = c.bus.Subscribe(app.TopicPeerIdentifierUpdated, c.handlePeerIdentifierChangeEvent)
+	_ = c.bus.Subscribe(app.TopicPeerDeleted, c.handlePeerDeletedEvent)
+}
+
+func (c *StatisticsCollector) handlePeerDeletedEvent(peer domain.Peer) {
+	c.ms.DeletePeerMetrics(peer)
 }
 
 func (c *StatisticsCollector) handlePeerIdentifierChangeEvent(oldIdentifier, newIdentifier domain.PeerIdentifier) {