conntrack reconciler must check the dst port

The conntrack reconciler maintains the consistency between the conntrack table on each node and the desired state of Kubernetes UDP services. A valid entry matches a service's ClusterIP, LoadBalancerIP, or ExternalIP and Service port, or any ip matching a NodePort, and has a reverse source IP matching an active endpoint for that service. Other entries are deleted. Services without endpoints and traffic not handled by kube-proxy are ignored Co-authored-by: Daman Arora <aroradaman@gmail.com>
2025-07-23 03:41:45 +00:00 · 2025-02-27 22:14:15 +00:00 · 2025-02-27 22:14:15 +00:00 · 2a60692f5b
commit 2a60692f5b
parent 0821aa2308
3 changed files with 346 additions and 78 deletions
--- a/pkg/proxy/conntrack/cleanup.go
+++ b/pkg/proxy/conntrack/cleanup.go
@ -20,6 +20,8 @@ limitations under the License.
 package conntrack

 import (
+	"net"
+	"strconv"
 	"time"

 	"github.com/vishvananda/netlink"
@ -32,8 +34,14 @@ import (
 	netutils "k8s.io/utils/net"
 )

+// Kubernetes UDP services can be affected by stale conntrack entries.
+// These entries may point to endpoints that no longer exist,
+// leading to packet loss and connectivity problems.
+
 // CleanStaleEntries scans conntrack table and removes any entries
 // for a service that do not correspond to a serving endpoint.
+// List existing conntrack entries and calculate the desired conntrack state
+// based on the current Services and Endpoints.
 func CleanStaleEntries(ct Interface, ipFamily v1.IPFamily,
 	svcPortMap proxy.ServicePortMap, endpointsMap proxy.EndpointsMap) {

@ -46,7 +54,7 @@ func CleanStaleEntries(ct Interface, ipFamily v1.IPFamily,
 		return
 	}

-	// serviceIPEndpointIPs maps service IPs (ClusterIP, LoadBalancerIPs and ExternalIPs)
+	// serviceIPEndpointIPs maps service IPs (ClusterIP, LoadBalancerIPs and ExternalIPs) and Service Port
 	// to the set of serving endpoint IPs.
 	serviceIPEndpointIPs := make(map[string]sets.Set[string])
 	// serviceNodePortEndpointIPs maps service NodePort to the set of serving endpoint IPs.
@ -73,14 +81,28 @@ func CleanStaleEntries(ct Interface, ipFamily v1.IPFamily,
 			}
 		}

-		serviceIPEndpointIPs[svc.ClusterIP().String()] = endpointIPs
+		// a Service without endpoints does not require to clean the conntrack entries associated.
+		if endpointIPs.Len() == 0 {
+			continue
+		}
+
+		// we need to filter entries that are directed to a Service IP:Port frontend
+		// that does not have a backend as part of the endpoints IPs
+		portStr := strconv.Itoa(svc.Port())
+		// clusterIP:Port
+		serviceIPEndpointIPs[net.JoinHostPort(svc.ClusterIP().String(), portStr)] = endpointIPs
+		// loadbalancerIP:Port
 		for _, loadBalancerIP := range svc.LoadBalancerVIPs() {
-			serviceIPEndpointIPs[loadBalancerIP.String()] = endpointIPs
+			serviceIPEndpointIPs[net.JoinHostPort(loadBalancerIP.String(), portStr)] = endpointIPs
 		}
+		// externalIP:Port
 		for _, externalIP := range svc.ExternalIPs() {
-			serviceIPEndpointIPs[externalIP.String()] = endpointIPs
+			serviceIPEndpointIPs[net.JoinHostPort(externalIP.String(), portStr)] = endpointIPs
 		}
+		// we need to filter entries that are directed to a *:NodePort
+		// that does not have a backend as part of the endpoints IPs
 		if svc.NodePort() != 0 {
+			// *:NodePort
 			serviceNodePortEndpointIPs[svc.NodePort()] = endpointIPs
 		}
 	}
@ -92,26 +114,25 @@ func CleanStaleEntries(ct Interface, ipFamily v1.IPFamily,
 			continue
 		}

-		origDst := entry.Forward.DstIP.String()
-		origPortDst := int(entry.Forward.DstPort)
-		replySrc := entry.Reverse.SrcIP.String()
+		origDst := entry.Forward.DstIP.String()   // match Service IP
+		origPortDst := int(entry.Forward.DstPort) // match Service Port
+		origPortDstStr := strconv.Itoa(origPortDst)
+		replySrc := entry.Reverse.SrcIP.String() // match Serving Endpoint IP

 		// if the original destination (--orig-dst) of the entry is service IP (ClusterIP,
-		// LoadBalancerIPs or ExternalIPs) and the reply source (--reply-src) is not IP of
-		// any serving endpoint, we clear the entry.
-		if _, ok := serviceIPEndpointIPs[origDst]; ok {
-			if !serviceIPEndpointIPs[origDst].Has(replySrc) {
-				filters = append(filters, filterForNAT(origDst, replySrc, v1.ProtocolUDP))
-			}
+		// LoadBalancerIPs or ExternalIPs) and (--orig-port-dst) of the flow is service Port
+		// and the reply source (--reply-src) is not IP of any serving endpoint, we clear the entry.
+		endpoints, ok := serviceIPEndpointIPs[net.JoinHostPort(origDst, origPortDstStr)]
+		if ok && !endpoints.Has(replySrc) {
+			filters = append(filters, filterForIPPortNAT(origDst, replySrc, entry.Forward.DstPort, v1.ProtocolUDP))
 		}

 		// if the original port destination (--orig-port-dst) of the flow is service
 		// NodePort and the reply source (--reply-src) is not IP of any serving endpoint,
 		// we clear the entry.
-		if _, ok := serviceNodePortEndpointIPs[origPortDst]; ok {
-			if !serviceNodePortEndpointIPs[origPortDst].Has(replySrc) {
-				filters = append(filters, filterForPortNAT(replySrc, origPortDst, v1.ProtocolUDP))
-			}
+		endpoints, ok = serviceNodePortEndpointIPs[origPortDst]
+		if ok && !endpoints.Has(replySrc) {
+			filters = append(filters, filterForPortNAT(replySrc, origPortDst, v1.ProtocolUDP))
 		}
 	}

@ -136,14 +157,16 @@ var protocolMap = map[v1.Protocol]uint8{
 	v1.ProtocolSCTP: unix.IPPROTO_SCTP,
 }

-// filterForNAT returns *conntrackFilter to delete the conntrack entries for connections
-// specified by the destination IP (original direction) and source IP (reply direction).
-func filterForNAT(origin, dest string, protocol v1.Protocol) *conntrackFilter {
-	klog.V(4).InfoS("Adding conntrack filter for cleanup", "org-dst", origin, "reply-src", dest, "protocol", protocol)
+// filterForIPPortNAT returns *conntrackFilter to delete the conntrack entries for connections
+// specified by the destination IP (original direction) and destination port (original direction)
+// and source IP (reply direction).
+func filterForIPPortNAT(origin, dest string, dstPort uint16, protocol v1.Protocol) *conntrackFilter {
+	klog.V(6).InfoS("Adding conntrack filter for cleanup", "org-dst", origin, "reply-src", dest, "protocol", protocol)
 	return &conntrackFilter{
 		protocol: protocolMap[protocol],
 		original: &connectionTuple{
-			dstIP: netutils.ParseIPSloppy(origin),
+			dstIP:   netutils.ParseIPSloppy(origin),
+			dstPort: dstPort,
 		},
 		reply: &connectionTuple{
 			srcIP: netutils.ParseIPSloppy(dest),
--- a/pkg/proxy/conntrack/cleanup_test.go
+++ b/pkg/proxy/conntrack/cleanup_test.go
@ -21,9 +21,11 @@ package conntrack

 import (
 	"fmt"
+	"math/rand"
 	"sort"
 	"testing"

+	"github.com/google/go-cmp/cmp"
 	"github.com/stretchr/testify/require"
 	"github.com/vishvananda/netlink"
 	"golang.org/x/sys/unix"
@ -50,8 +52,11 @@ const (
 	testNonServingEndpointIP = "10.240.1.5"
 	testDeletedEndpointIP    = "10.240.2.6"

-	testPort     = 8000
-	testNodePort = 32000
+	testServicePort     = 8000
+	testServiceNodePort = 32000
+	// testNonServicePort is used to mock conntrack flow entries which are not owned by
+	// kube-proxy and reconciler should not consider these for cleanup
+	testNonServicePort = 3000
 )

 func TestCleanStaleEntries(t *testing.T) {
@ -74,19 +79,19 @@ func TestCleanStaleEntries(t *testing.T) {
 			Ports: []v1.ServicePort{
 				{
 					Name:     "test-tcp",
-					Port:     testPort,
+					Port:     testServicePort,
 					Protocol: v1.ProtocolTCP,
 				},
 				{
 					Name:     "test-udp",
-					Port:     testPort,
-					NodePort: testNodePort,
+					Port:     testServicePort,
+					NodePort: testServiceNodePort,
 					Protocol: v1.ProtocolUDP,
 				},
 				{
 					Name:     "test-sctp",
-					Port:     testPort,
-					NodePort: testNodePort,
+					Port:     testServicePort,
+					NodePort: testServiceNodePort,
 					Protocol: v1.ProtocolSCTP,
 				},
 			},
@ -126,17 +131,17 @@ func TestCleanStaleEntries(t *testing.T) {
 		Ports: []discovery.EndpointPort{
 			{
 				Name:     ptr.To("test-tcp"),
-				Port:     ptr.To(int32(testPort)),
+				Port:     ptr.To(int32(testServicePort)),
 				Protocol: ptr.To(v1.ProtocolTCP),
 			},
 			{
 				Name:     ptr.To("test-udp"),
-				Port:     ptr.To(int32(testPort)),
+				Port:     ptr.To(int32(testServicePort)),
 				Protocol: ptr.To(v1.ProtocolUDP),
 			},
 			{
 				Name:     ptr.To("test-sctp"),
-				Port:     ptr.To(int32(testPort)),
+				Port:     ptr.To(int32(testServicePort)),
 				Protocol: ptr.To(v1.ProtocolSCTP),
 			},
 		},
@ -212,35 +217,47 @@ func TestCleanStaleEntries(t *testing.T) {
 		}
 	}

-	// mock existing entries before cleanup
-	// we create 36 fake flow entries ( 3 Endpoints * 3 Protocols * ( 3 (ServiceIPs) + 1 (NodePort))
-	var mockEntries []*netlink.ConntrackFlow
-	// expectedEntries are the entries on which we will assert the cleanup logic
-	var expectedEntries []*netlink.ConntrackFlow
+	// The following mock conntrack flow entries `entriesBeforeCleanup` and `entriesAfterCleanup`
+	// represent conntrack flow entries before and after reconciler cleanup loop. Before cleanup,
+	// reconciler lists the conntrack flows, receiving `entriesBeforeCleanup` and after cleanup,
+	// we list the conntrack flows and assert them to match with `entriesAfterCleanup`.
+	// {entriesBeforeCleanup} - {entriesAfterCleanup} = entries cleared by conntrack reconciler
+	var entriesBeforeCleanup []*netlink.ConntrackFlow
+	// entriesBeforeCleanup - entriesAfterCleanup = entries cleared by conntrack reconciler
+	var entriesAfterCleanup []*netlink.ConntrackFlow
+
+	// we create 63 fake flow entries ( 3 Endpoints * 3 Protocols * ( 3 (ServiceIP:ServicePort) + 3 (ServiceIP:NonServicePort) + 1 (NodePort))
 	for _, dnatDest := range []string{testServingEndpointIP, testNonServingEndpointIP, testDeletedEndpointIP} {
 		for _, proto := range []uint8{unix.IPPROTO_TCP, unix.IPPROTO_UDP, unix.IPPROTO_SCTP} {
+
 			for _, origDest := range []string{testClusterIP, testLoadBalancerIP, testExternalIP} {
-				entry := &netlink.ConntrackFlow{
-					FamilyType: unix.AF_INET,
-					Forward: netlink.IPTuple{
-						DstIP:    netutils.ParseIPSloppy(origDest),
-						Protocol: proto,
-					},
-					Reverse: netlink.IPTuple{
-						Protocol: proto,
-						SrcIP:    netutils.ParseIPSloppy(dnatDest),
-					},
-				}
-				mockEntries = append(mockEntries, entry)
-				// we do not expect deleted or non-serving UDP endpoints flows to be present after cleanup
-				if !(proto == unix.IPPROTO_UDP && (dnatDest == testNonServingEndpointIP || dnatDest == testDeletedEndpointIP)) {
-					expectedEntries = append(expectedEntries, entry)
+				for _, port := range []uint16{testServicePort, testNonServicePort} {
+					entry := &netlink.ConntrackFlow{
+						FamilyType: unix.AF_INET,
+						Forward: netlink.IPTuple{
+							DstIP:    netutils.ParseIPSloppy(origDest),
+							DstPort:  port,
+							Protocol: proto,
+						},
+						Reverse: netlink.IPTuple{
+							Protocol: proto,
+							SrcIP:    netutils.ParseIPSloppy(dnatDest),
+						},
+					}
+					entriesBeforeCleanup = append(entriesBeforeCleanup, entry)
+					if proto == unix.IPPROTO_UDP && port == testServicePort && dnatDest != testServingEndpointIP {
+						// we do not expect UDP entries with destination port `testServicePort` and DNATed destination
+						// address not an address of serving endpoint to be present after cleanup.
+					} else {
+						entriesAfterCleanup = append(entriesAfterCleanup, entry)
+					}
 				}
 			}
+
 			entry := &netlink.ConntrackFlow{
 				FamilyType: unix.AF_INET,
 				Forward: netlink.IPTuple{
-					DstPort:  testNodePort,
+					DstPort:  testServiceNodePort,
 					Protocol: proto,
 				},
 				Reverse: netlink.IPTuple{
@ -248,50 +265,264 @@ func TestCleanStaleEntries(t *testing.T) {
 					SrcIP:    netutils.ParseIPSloppy(dnatDest),
 				},
 			}
-			mockEntries = append(mockEntries, entry)
-			// we do not expect deleted or non-serving UDP endpoints entries to be present after cleanup
-			if !(proto == unix.IPPROTO_UDP && (dnatDest == testNonServingEndpointIP || dnatDest == testDeletedEndpointIP)) {
-				expectedEntries = append(expectedEntries, entry)
+			entriesBeforeCleanup = append(entriesBeforeCleanup, entry)
+			if proto == unix.IPPROTO_UDP && dnatDest != testServingEndpointIP {
+				// we do not expect UDP entries with DNATed destination address not
+				// an address of serving endpoint to be present after cleanup.
+			} else {
+				entriesAfterCleanup = append(entriesAfterCleanup, entry)
 			}
 		}
 	}

-	// add some non-DNATed mock entries which should be cleared up by reconciler
+	// add 6 non-DNATed mock entries which should be cleared up by reconciler
 	// These will exist if the proxy don't have DROP/REJECT rule for service with
 	// no endpoints, --orig-dst and --reply-src will be same for these entries.
 	for _, ip := range []string{testClusterIP, testLoadBalancerIP, testExternalIP} {
-		entry := &netlink.ConntrackFlow{
-			FamilyType: unix.AF_INET,
-			Forward: netlink.IPTuple{
-				DstIP:    netutils.ParseIPSloppy(ip),
-				Protocol: unix.IPPROTO_UDP,
-			},
-			Reverse: netlink.IPTuple{
-				Protocol: unix.IPPROTO_UDP,
-				SrcIP:    netutils.ParseIPSloppy(ip),
-			},
+		for _, port := range []uint16{testServicePort, testNonServicePort} {
+			entry := &netlink.ConntrackFlow{
+				FamilyType: unix.AF_INET,
+				Forward: netlink.IPTuple{
+					DstIP:    netutils.ParseIPSloppy(ip),
+					DstPort:  port,
+					Protocol: unix.IPPROTO_UDP,
+				},
+				Reverse: netlink.IPTuple{
+					Protocol: unix.IPPROTO_UDP,
+					SrcIP:    netutils.ParseIPSloppy(ip),
+				},
+			}
+
+			entriesBeforeCleanup = append(entriesBeforeCleanup, entry)
+			// we do not expect entries with destination port `testServicePort` to be
+			// present after cleanup.
+			if port != testServicePort {
+				entriesAfterCleanup = append(entriesAfterCleanup, entry)
+			}
+
 		}
-		mockEntries = append(mockEntries, entry)
 	}

+	t.Logf("entries before cleanup %d after cleanup %d", len(entriesBeforeCleanup), len(entriesAfterCleanup))
 	fake := NewFake()
-	fake.entries = mockEntries
+	fake.entries = entriesBeforeCleanup
 	CleanStaleEntries(fake, testIPFamily, svcPortMap, endpointsMap)

 	actualEntries, _ := fake.ListEntries(ipFamilyMap[testIPFamily])
-	require.Equal(t, len(expectedEntries), len(actualEntries))
+	require.Equal(t, len(entriesAfterCleanup), len(actualEntries))

 	// sort the actual flows before comparison
 	sort.Slice(actualEntries, func(i, j int) bool {
 		return actualEntries[i].String() < actualEntries[j].String()
 	})
 	// sort the expected flows before comparison
-	sort.Slice(expectedEntries, func(i, j int) bool {
-		return expectedEntries[i].String() < expectedEntries[j].String()
+	sort.Slice(entriesAfterCleanup, func(i, j int) bool {
+		return entriesAfterCleanup[i].String() < entriesAfterCleanup[j].String()
 	})

-	for i := 0; i < len(expectedEntries); i++ {
-		require.Equal(t, expectedEntries[i], actualEntries[i])
+	if diff := cmp.Diff(entriesAfterCleanup, actualEntries); len(diff) > 0 {
+		t.Errorf("unexpected entries after cleanup: %s", diff)
+	}
+}
+
+func TestPerformanceCleanStaleEntries(t *testing.T) {
+	sct := proxy.NewServiceChangeTracker(nil, v1.IPv4Protocol, nil, nil)
+	svc := &v1.Service{
+		ObjectMeta: metav1.ObjectMeta{
+			Name:      testServiceName,
+			Namespace: testServiceNamespace,
+		},
+		Spec: v1.ServiceSpec{
+			ClusterIP:   testClusterIP,
+			ExternalIPs: []string{testExternalIP},
+			Ports: []v1.ServicePort{
+				{
+					Name:     "test-udp",
+					Port:     testServicePort,
+					Protocol: v1.ProtocolUDP,
+				},
+			},
+		},
+		Status: v1.ServiceStatus{
+			LoadBalancer: v1.LoadBalancerStatus{
+				Ingress: []v1.LoadBalancerIngress{{
+					IP: testLoadBalancerIP,
+				}},
+			},
+		},
+	}
+
+	sct.Update(nil, svc)
+	svcPortMap := make(proxy.ServicePortMap)
+	_ = svcPortMap.Update(sct)
+
+	ect := proxy.NewEndpointsChangeTracker("test-worker", nil, v1.IPv4Protocol, nil, nil)
+	eps := &discovery.EndpointSlice{
+		TypeMeta:    metav1.TypeMeta{},
+		AddressType: discovery.AddressTypeIPv4,
+		ObjectMeta: metav1.ObjectMeta{
+			Name:      fmt.Sprintf("%s-0", testServiceName),
+			Namespace: testServiceNamespace,
+			Labels:    map[string]string{discovery.LabelServiceName: testServiceName},
+		},
+		Endpoints: []discovery.Endpoint{
+			{
+				Addresses:  []string{testServingEndpointIP},
+				Conditions: discovery.EndpointConditions{Serving: ptr.To(true)},
+			},
+			{
+				Addresses:  []string{testNonServingEndpointIP},
+				Conditions: discovery.EndpointConditions{Serving: ptr.To(false)},
+			},
+		},
+		Ports: []discovery.EndpointPort{
+			{
+				Name:     ptr.To("test-udp"),
+				Port:     ptr.To(int32(testServicePort)),
+				Protocol: ptr.To(v1.ProtocolUDP),
+			},
+		},
+	}
+
+	ect.EndpointSliceUpdate(eps, false)
+	endpointsMap := make(proxy.EndpointsMap)
+	_ = endpointsMap.Update(ect)
+
+	genConntrackEntry := func(dstPort uint16, revSrc string) *netlink.ConntrackFlow {
+		return &netlink.ConntrackFlow{
+			FamilyType: unix.AF_INET,
+			Forward: netlink.IPTuple{
+				DstIP:    netutils.ParseIPSloppy(testExternalIP),
+				DstPort:  dstPort,
+				Protocol: unix.IPPROTO_UDP,
+			},
+			Reverse: netlink.IPTuple{
+				Protocol: unix.IPPROTO_UDP,
+				SrcIP:    netutils.ParseIPSloppy(revSrc),
+			},
+		}
+	}
+
+	fake := NewFake()
+	// 1 valid entry
+	fake.entries = append(fake.entries, genConntrackEntry(uint16(testServicePort), testServingEndpointIP))
+	expectedEntries := 1
+	// 1 stale entry
+	fake.entries = append(fake.entries, genConntrackEntry(uint16(testServicePort), testDeletedEndpointIP))
+	expectedDeleted := 1
+	// 1 M to the Service IP with random ports
+	for i := 0; i < 1000*1000; i++ {
+		port := uint16(rand.Intn(65535))
+		if port == testServicePort {
+			expectedDeleted++
+		} else {
+			expectedEntries++
+		}
+		fake.entries = append(fake.entries, genConntrackEntry(port, testDeletedEndpointIP))
+	}
+
+	CleanStaleEntries(fake, testIPFamily, svcPortMap, endpointsMap)
+	actualEntries, _ := fake.ListEntries(ipFamilyMap[testIPFamily])
+	if len(actualEntries) != expectedEntries {
+		t.Errorf("unexpected number of entries, got %d expected %d", len(actualEntries), expectedEntries)
+	}
+	// expected conntrack entries
+	// 1 for CleanStaleEntries + 1 for ListEntries + 1 for ConntrackDeleteFilters to dump the conntrack table
+	// n for the expected deleted stale entries
+	t.Logf("expected deleted %d", expectedDeleted)
+	if fake.netlinkRequests != 3+expectedDeleted {
+		t.Errorf("expected %d netlink requests, got %d", 3+expectedDeleted, fake.netlinkRequests)
+	}
+}
+
+func TestServiceWithoutEndpoints(t *testing.T) {
+	sct := proxy.NewServiceChangeTracker(nil, v1.IPv4Protocol, nil, nil)
+	svc := &v1.Service{
+		ObjectMeta: metav1.ObjectMeta{
+			Name:      testServiceName,
+			Namespace: testServiceNamespace,
+		},
+		Spec: v1.ServiceSpec{
+			ClusterIP:   testClusterIP,
+			ExternalIPs: []string{testExternalIP},
+			Ports: []v1.ServicePort{
+				{
+					Name:     "test-udp",
+					Port:     testServicePort,
+					Protocol: v1.ProtocolUDP,
+				},
+			},
+		},
+		Status: v1.ServiceStatus{
+			LoadBalancer: v1.LoadBalancerStatus{
+				Ingress: []v1.LoadBalancerIngress{{
+					IP: testLoadBalancerIP,
+				}},
+			},
+		},
+	}
+
+	sct.Update(nil, svc)
+	svcPortMap := make(proxy.ServicePortMap)
+	_ = svcPortMap.Update(sct)
+
+	ect := proxy.NewEndpointsChangeTracker("test-worker", nil, v1.IPv4Protocol, nil, nil)
+	eps := &discovery.EndpointSlice{
+		TypeMeta:    metav1.TypeMeta{},
+		AddressType: discovery.AddressTypeIPv4,
+		ObjectMeta: metav1.ObjectMeta{
+			Name:      fmt.Sprintf("%s-0", testServiceName),
+			Namespace: testServiceNamespace,
+			Labels:    map[string]string{discovery.LabelServiceName: "non-existing-service"},
+		},
+		Endpoints: []discovery.Endpoint{
+			{
+				Addresses:  []string{testServingEndpointIP},
+				Conditions: discovery.EndpointConditions{Serving: ptr.To(true)},
+			},
+			{
+				Addresses:  []string{testNonServingEndpointIP},
+				Conditions: discovery.EndpointConditions{Serving: ptr.To(false)},
+			},
+		},
+		Ports: []discovery.EndpointPort{
+			{
+				Name:     ptr.To("test-udp"),
+				Port:     ptr.To(int32(testServicePort)),
+				Protocol: ptr.To(v1.ProtocolUDP),
+			},
+		},
+	}
+
+	ect.EndpointSliceUpdate(eps, false)
+	endpointsMap := make(proxy.EndpointsMap)
+	_ = endpointsMap.Update(ect)
+
+	genConntrackEntry := func(dstPort uint16, revSrc string) *netlink.ConntrackFlow {
+		return &netlink.ConntrackFlow{
+			FamilyType: unix.AF_INET,
+			Forward: netlink.IPTuple{
+				DstIP:    netutils.ParseIPSloppy(testExternalIP),
+				DstPort:  dstPort,
+				Protocol: unix.IPPROTO_UDP,
+			},
+			Reverse: netlink.IPTuple{
+				Protocol: unix.IPPROTO_UDP,
+				SrcIP:    netutils.ParseIPSloppy(revSrc),
+			},
+		}
+	}
+
+	fake := NewFake()
+	// 1 valid entry
+	fake.entries = append(fake.entries, genConntrackEntry(uint16(testServicePort), testServingEndpointIP))
+	// 1 stale entry
+	fake.entries = append(fake.entries, genConntrackEntry(uint16(testServicePort), testDeletedEndpointIP))
+
+	CleanStaleEntries(fake, testIPFamily, svcPortMap, endpointsMap)
+	actualEntries, _ := fake.ListEntries(ipFamilyMap[testIPFamily])
+	if len(actualEntries) != 2 {
+		t.Errorf("unexpected number of entries, got %d expected %d", len(actualEntries), 2)
 	}
 }

@ -300,6 +531,7 @@ func TestFilterForNAT(t *testing.T) {
 		name           string
 		orig           string
 		dest           string
+		dstPort        uint16
 		protocol       v1.Protocol
 		expectedFilter *conntrackFilter
 	}{
@ -329,7 +561,7 @@ func TestFilterForNAT(t *testing.T) {

 	for _, tc := range testCases {
 		t.Run(tc.name, func(t *testing.T) {
-			require.Equal(t, tc.expectedFilter, filterForNAT(tc.orig, tc.dest, tc.protocol))
+			require.Equal(t, tc.expectedFilter, filterForIPPortNAT(tc.orig, tc.dest, tc.dstPort, tc.protocol))
 		})
 	}
 }
--- a/pkg/proxy/conntrack/fake.go
+++ b/pkg/proxy/conntrack/fake.go
@ -25,7 +25,8 @@ import (

 // FakeInterface implements Interface by just recording entries that have been cleared.
 type FakeInterface struct {
-	entries []*netlink.ConntrackFlow
+	entries         []*netlink.ConntrackFlow
+	netlinkRequests int // try to get the estimated number of netlink request
 }

 var _ Interface = &FakeInterface{}
@ -37,18 +38,30 @@ func NewFake() *FakeInterface {

 // ListEntries is part of Interface
 func (fake *FakeInterface) ListEntries(_ uint8) ([]*netlink.ConntrackFlow, error) {
-	return fake.entries, nil
+	entries := make([]*netlink.ConntrackFlow, len(fake.entries))
+	copy(entries, fake.entries)
+	// 1 netlink request to dump the table
+	// https://github.com/vishvananda/netlink/blob/0af32151e72b990c271ef6268e8aadb7e015f2bd/conntrack_linux.go#L93-L94
+	fake.netlinkRequests++
+	return entries, nil
 }

 // ClearEntries is part of Interface
 func (fake *FakeInterface) ClearEntries(_ uint8, filters ...netlink.CustomConntrackFilter) (int, error) {
 	var flows []*netlink.ConntrackFlow
 	before := len(fake.entries)
+	// 1 netlink request to dump the table
+	// https://github.com/vishvananda/netlink/blob/0af32151e72b990c271ef6268e8aadb7e015f2bd/conntrack_linux.go#L163
+	fake.netlinkRequests++
+
 	for _, flow := range fake.entries {
 		var matched bool
 		for _, filter := range filters {
 			matched = filter.MatchConntrackFlow(flow)
 			if matched {
+				// 1 netlink request to delete the flow
+				// https://github.com/vishvananda/netlink/blob/0af32151e72b990c271ef6268e8aadb7e015f2bd/conntrack_linux.go#L182
+				fake.netlinkRequests++
 				break
 			}
 		}