Merge pull request #169 from cooperlees/stale_pod_ips

Delete stale pod IPs

Makefile

@@ -1,5 +1,5 @@
 name ?= goldpinger
-version ?= v3.11.1
+version ?= v3.11.2
 bin ?= goldpinger
 pkg ?= "github.com/bloomberg/goldpinger"
 tag = $(name):$(version)

charts/goldpinger/Chart.yaml

@@ -1,7 +1,7 @@
 apiVersion: v1
 name: goldpinger
-appVersion: "3.11.1"
+appVersion: "3.11.2"
-version: 1.1.1
+version: 1.1.2
 description: Goldpinger is a tool to help debug, troubleshoot and visualize network connectivity and slowness issues.
 home: https://github.com/bloomberg/goldpinger
 sources:

pkg/goldpinger/stats.go

@@ -300,6 +300,15 @@ func SetPeerHopCount(hostIP, podIP string, hopCount int32) {
 	).Set(float64(hopCount))
 }
 
+// DeletePeerMetrics removes stale metric labels for a destroyed peer.
+// The response-time histogram is observed with two call_type values:
+// "ping" (continuous from Pinger) and "check" (from CheckAllPods), so both
+// must be pruned. Must be called unconditionally when a peer is removed.
+func DeletePeerMetrics(hostIP, podIP string) {
+	goldpingerResponseTimePeersHistogram.DeleteLabelValues(GoldpingerConfig.Hostname, "ping", hostIP, podIP)
+	goldpingerResponseTimePeersHistogram.DeleteLabelValues(GoldpingerConfig.Hostname, "check", hostIP, podIP)
+}
+
 // DeletePeerUDPMetrics removes stale UDP metric labels for a destroyed peer.
 // This must be kept in sync with all per-peer UDP metrics to avoid stale
 // label sets lingering in /metrics after a pod rolls.
@@ -309,6 +318,7 @@ func DeletePeerUDPMetrics(hostIP, podIP string) {
 	goldpingerPeersUDPRtt.DeleteLabelValues(GoldpingerConfig.Hostname, hostIP, podIP)
 	goldpingerUDPDuplicatesCounter.DeleteLabelValues(GoldpingerConfig.Hostname, hostIP, podIP)
 	goldpingerUDPOutOfOrderCounter.DeleteLabelValues(GoldpingerConfig.Hostname, hostIP, podIP)
+	goldpingerUDPErrorsCounter.DeleteLabelValues(GoldpingerConfig.Hostname, pickPodHostIP(podIP, hostIP))
 }
 
 // ObservePeerUDPRtt records a UDP RTT observation in seconds

pkg/goldpinger/stats_test.go

@@ -7,49 +7,186 @@ import (
 	dto "github.com/prometheus/client_model/go"
 )
 
+// TestDeletePeerMetrics_CleansResponseTimeHistogram verifies that
+// DeletePeerMetrics removes the response-time histogram label set for
+// a destroyed peer. This prevents stale pod IPs from lingering in
+// /metrics after rolling updates (see #167).
+func TestDeletePeerMetrics_CleansResponseTimeHistogram(t *testing.T) {
+	tests := []struct {
+		name   string
+		hostIP string
+		podIP  string
+	}{
+		{"IPv4", "10.0.0.1", "10.0.0.2"},
+		{"IPv6", "2001:db8::1", "2001:db8:1::2"},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			origHostname := GoldpingerConfig.Hostname
+			GoldpingerConfig.Hostname = "test-instance"
+			defer func() { GoldpingerConfig.Hostname = origHostname }()
+
+			// Simulate observations for both call_type values the histogram uses
+			goldpingerResponseTimePeersHistogram.WithLabelValues(
+				GoldpingerConfig.Hostname, "ping", tt.hostIP, tt.podIP,
+			).Observe(0.005)
+			goldpingerResponseTimePeersHistogram.WithLabelValues(
+				GoldpingerConfig.Hostname, "check", tt.hostIP, tt.podIP,
+			).Observe(0.010)
+
+			if n := countMetrics(goldpingerResponseTimePeersHistogram); n != 2 {
+				t.Fatalf("response time histogram has %d label sets before cleanup, want 2 — test setup is broken", n)
+			}
+
+			DeletePeerMetrics(tt.hostIP, tt.podIP)
+
+			if n := countMetrics(goldpingerResponseTimePeersHistogram); n != 0 {
+				t.Errorf("response time histogram still has %d label set(s) after DeletePeerMetrics", n)
+			}
+		})
+	}
+}
+
+// TestDeletePeerMetrics_LeavesOtherPeersIntact verifies that pruning
+// metrics for one peer does not affect a different peer's label set.
+func TestDeletePeerMetrics_LeavesOtherPeersIntact(t *testing.T) {
+	tests := []struct {
+		name    string
+		peerA   [2]string // {hostIP, podIP}
+		peerB   [2]string
+	}{
+		{
+			"IPv4",
+			[2]string{"10.0.0.1", "10.0.0.2"},
+			[2]string{"10.0.0.3", "10.0.0.4"},
+		},
+		{
+			"IPv6",
+			[2]string{"2001:db8::1", "2001:db8:1::2"},
+			[2]string{"2001:db8::3", "2001:db8:2::4"},
+		},
+		{
+			"MixedV4DeleteV6Survives",
+			[2]string{"10.0.0.1", "10.0.0.2"},
+			[2]string{"2001:db8::3", "2001:db8:2::4"},
+		},
+		{
+			"MixedV6DeleteV4Survives",
+			[2]string{"2001:db8::1", "2001:db8:1::2"},
+			[2]string{"10.0.0.3", "10.0.0.4"},
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			origHostname := GoldpingerConfig.Hostname
+			GoldpingerConfig.Hostname = "test-instance"
+			defer func() { GoldpingerConfig.Hostname = origHostname }()
+
+			// Peer A — observe both call types
+			goldpingerResponseTimePeersHistogram.WithLabelValues(
+				GoldpingerConfig.Hostname, "ping", tt.peerA[0], tt.peerA[1],
+			).Observe(0.005)
+			goldpingerResponseTimePeersHistogram.WithLabelValues(
+				GoldpingerConfig.Hostname, "check", tt.peerA[0], tt.peerA[1],
+			).Observe(0.006)
+			SetPeerLossPct(tt.peerA[0], tt.peerA[1], 0)
+
+			// Peer B — observe both call types
+			goldpingerResponseTimePeersHistogram.WithLabelValues(
+				GoldpingerConfig.Hostname, "ping", tt.peerB[0], tt.peerB[1],
+			).Observe(0.010)
+			goldpingerResponseTimePeersHistogram.WithLabelValues(
+				GoldpingerConfig.Hostname, "check", tt.peerB[0], tt.peerB[1],
+			).Observe(0.011)
+			SetPeerLossPct(tt.peerB[0], tt.peerB[1], 1.5)
+
+			// Delete peer A only
+			DeletePeerMetrics(tt.peerA[0], tt.peerA[1])
+			DeletePeerUDPMetrics(tt.peerA[0], tt.peerA[1])
+
+			// Peer B's ping and check histogram entries should both survive
+			if n := countMetrics(goldpingerResponseTimePeersHistogram); n != 2 {
+				t.Errorf("response time histogram has %d label set(s), want 2 for peer B (ping+check)", n)
+			}
+			if countMetrics(goldpingerPeersLossPct) == 0 {
+				t.Error("loss pct gauge lost all label sets — peer B should still exist")
+			}
+
+			// Clean up peer B so it doesn't leak into other tests
+			goldpingerResponseTimePeersHistogram.DeleteLabelValues(
+				GoldpingerConfig.Hostname, "ping", tt.peerB[0], tt.peerB[1],
+			)
+			goldpingerResponseTimePeersHistogram.DeleteLabelValues(
+				GoldpingerConfig.Hostname, "check", tt.peerB[0], tt.peerB[1],
+			)
+			goldpingerPeersLossPct.DeleteLabelValues(
+				GoldpingerConfig.Hostname, tt.peerB[0], tt.peerB[1],
+			)
+		})
+	}
+}
+
 // TestDeletePeerUDPMetrics_CleansAllPerPeerMetrics verifies that
 // DeletePeerUDPMetrics removes label sets from every per-peer UDP metric.
 // If a new per-peer metric is added but not cleaned up in
 // DeletePeerUDPMetrics, this test will fail.
 func TestDeletePeerUDPMetrics_CleansAllPerPeerMetrics(t *testing.T) {
-	// Save and restore hostname since we set it for the test
+	tests := []struct {
-	origHostname := GoldpingerConfig.Hostname
+		name   string
-	GoldpingerConfig.Hostname = "test-instance"
+		hostIP string
-	defer func() { GoldpingerConfig.Hostname = origHostname }()
+		podIP  string
-
+	}{
-	hostIP := "10.0.0.1"
+		{"IPv4", "10.0.0.1", "10.0.0.2"},
-	podIP := "10.0.0.2"
+		{"IPv6", "2001:db8::1", "2001:db8:1::2"},
-
-	// Populate all per-peer UDP metrics so they have label values
-	SetPeerLossPct(hostIP, podIP, 5.0)
-	SetPeerHopCount(hostIP, podIP, 2)
-	ObservePeerUDPRtt(hostIP, podIP, 0.001)
-	CountUDPDuplicates(hostIP, podIP, 1)
-	CountUDPOutOfOrder(hostIP, podIP, 1)
-
-	// Verify they exist before cleanup
-	perPeerCollectors := map[string]prometheus.Collector{
-		"goldpinger_peers_loss_pct":         goldpingerPeersLossPct,
-		"goldpinger_peers_hop_count":        goldpingerPeersHopCount,
-		"goldpinger_peers_udp_rtt_s":        goldpingerPeersUDPRtt,
-		"goldpinger_udp_duplicates_total":   goldpingerUDPDuplicatesCounter,
-		"goldpinger_udp_out_of_order_total": goldpingerUDPOutOfOrderCounter,
 	}
 
-	for name, collector := range perPeerCollectors {
+	for _, tt := range tests {
-		if countMetrics(collector) == 0 {
+		t.Run(tt.name, func(t *testing.T) {
-			t.Fatalf("metric %s has no label values before cleanup — test setup is broken", name)
+			origHostname := GoldpingerConfig.Hostname
-		}
+			origUseHostIP := GoldpingerConfig.UseHostIP
-	}
+			GoldpingerConfig.Hostname = "test-instance"
+			GoldpingerConfig.UseHostIP = false
+			defer func() {
+				GoldpingerConfig.Hostname = origHostname
+				GoldpingerConfig.UseHostIP = origUseHostIP
+			}()
 
-	// Run cleanup
+			// Populate all per-peer UDP metrics so they have label values
-	DeletePeerUDPMetrics(hostIP, podIP)
+			SetPeerLossPct(tt.hostIP, tt.podIP, 5.0)
+			SetPeerHopCount(tt.hostIP, tt.podIP, 2)
+			ObservePeerUDPRtt(tt.hostIP, tt.podIP, 0.001)
+			CountUDPDuplicates(tt.hostIP, tt.podIP, 1)
+			CountUDPOutOfOrder(tt.hostIP, tt.podIP, 1)
+			CountUDPError(tt.podIP) // UseHostIP=false so target=podIP
 
-	// Verify all per-peer metrics are cleaned up
+			// Verify they exist before cleanup
-	for name, collector := range perPeerCollectors {
+			perPeerCollectors := map[string]prometheus.Collector{
-		if n := countMetrics(collector); n != 0 {
+				"goldpinger_peers_loss_pct":         goldpingerPeersLossPct,
-			t.Errorf("metric %s still has %d label set(s) after DeletePeerUDPMetrics — add it to the cleanup function", name, n)
+				"goldpinger_peers_hop_count":        goldpingerPeersHopCount,
-		}
+				"goldpinger_peers_udp_rtt_s":        goldpingerPeersUDPRtt,
+				"goldpinger_udp_duplicates_total":   goldpingerUDPDuplicatesCounter,
+				"goldpinger_udp_out_of_order_total": goldpingerUDPOutOfOrderCounter,
+				"goldpinger_udp_errors_total":       goldpingerUDPErrorsCounter,
+			}
+
+			for name, collector := range perPeerCollectors {
+				if countMetrics(collector) == 0 {
+					t.Fatalf("metric %s has no label values before cleanup — test setup is broken", name)
+				}
+			}
+
+			// Run cleanup
+			DeletePeerUDPMetrics(tt.hostIP, tt.podIP)
+
+			// Verify all per-peer metrics are cleaned up
+			for name, collector := range perPeerCollectors {
+				if n := countMetrics(collector); n != 0 {
+					t.Errorf("metric %s still has %d label set(s) after DeletePeerUDPMetrics — add it to the cleanup function", name, n)
+				}
+			}
+		})
 	}
 }
 

pkg/goldpinger/updater.go

@@ -137,7 +137,9 @@ func destroyPingers(pingers map[string]*Pinger, deletedPods map[string]*Goldping
 		// Close the channel to stop pinging
 		close(pinger.stopChan)
 
-		// Clean up stale UDP metric labels for this peer
+		// Clean up stale metric labels for this peer so defunct pod IPs
+		// don't linger in /metrics after rolling updates (see #167)
+		DeletePeerMetrics(pod.HostIP, pod.PodIP)
 		if GoldpingerConfig.UDPEnabled {
 			DeletePeerUDPMetrics(pod.HostIP, pod.PodIP)
 		}