Adds TCPCloseWaitTimeout option to kube-proxy for sysctl nf_conntrack_tcp_timeout_time_wait

cmd/kube-proxy/app/conntrack.go

@@ -27,24 +27,33 @@ import (
 	"k8s.io/kubernetes/pkg/util/sysctl"
 )
 
+// Conntracker is an interface to the global sysctl. Descriptions of the various
+// sysctl fields can be found here:
+//
+// https://www.kernel.org/doc/Documentation/networking/nf_conntrack-sysctl.txt
 type Conntracker interface {
+	// SetMax adjusts nf_conntrack_max.
 	SetMax(max int) error
+	// SetTCPEstablishedTimeout adjusts nf_conntrack_tcp_timeout_established.
 	SetTCPEstablishedTimeout(seconds int) error
+	// SetTCPCloseWaitTimeout nf_conntrack_tcp_timeout_close_wait.
+	SetTCPCloseWaitTimeout(seconds int) error
 }
 
 type realConntracker struct{}
 
 var readOnlySysFSError = errors.New("readOnlySysFS")
 
-func (realConntracker) SetMax(max int) error {
-	glog.Infof("Setting nf_conntrack_max to %d", max)
-	if err := sysctl.New().SetSysctl("net/netfilter/nf_conntrack_max", max); err != nil {
+func (rct realConntracker) SetMax(max int) error {
+	if err := rct.setIntSysCtl("nf_conntrack_max", max); err != nil {
 		return err
 	}
-	// sysfs is expected to be mounted as 'rw'. However, it may be unexpectedly mounted as
-	// 'ro' by docker because of a known docker issue (https://github.com/docker/docker/issues/24000).
-	// Setting conntrack will fail when sysfs is readonly. When that happens, we don't set conntrack
-	// hashsize and return a special error readOnlySysFSError here. The caller should deal with
+	// sysfs is expected to be mounted as 'rw'. However, it may be
+	// unexpectedly mounted as 'ro' by docker because of a known docker
+	// issue (https://github.com/docker/docker/issues/24000). Setting
+	// conntrack will fail when sysfs is readonly. When that happens, we
+	// don't set conntrack hashsize and return a special error
+	// readOnlySysFSError here. The caller should deal with
 	// readOnlySysFSError differently.
 	writable, err := isSysFSWritable()
 	if err != nil {
@@ -58,9 +67,22 @@ func (realConntracker) SetMax(max int) error {
 	return writeIntStringFile("/sys/module/nf_conntrack/parameters/hashsize", max/4)
 }
 
-func (realConntracker) SetTCPEstablishedTimeout(seconds int) error {
-	glog.Infof("Setting nf_conntrack_tcp_timeout_established to %d", seconds)
-	return sysctl.New().SetSysctl("net/netfilter/nf_conntrack_tcp_timeout_established", seconds)
+func (rct realConntracker) SetTCPEstablishedTimeout(seconds int) error {
+	return rct.setIntSysCtl("nf_conntrack_tcp_timeout_established", seconds)
+}
+
+func (rct realConntracker) SetTCPCloseWaitTimeout(seconds int) error {
+	return rct.setIntSysCtl("nf_conntrack_tcp_timeout_close_wait", seconds)
+}
+
+func (realConntracker) setIntSysCtl(name string, value int) error {
+	entry := "net/netfilter/" + name
+
+	glog.Infof("Set sysctl '%v' to %v", entry, value)
+	if err := sysctl.New().SetSysctl(entry, value); err != nil {
+		return err
+	}
+	return nil
 }
 
 // isSysFSWritable checks /proc/mounts to see whether sysfs is 'rw' or not.
@@ -73,18 +95,23 @@ func isSysFSWritable() (bool, error) {
 		glog.Errorf("failed to list mount points: %v", err)
 		return false, err
 	}
+
 	for _, mountPoint := range mountPoints {
+		const sysfsDevice = "sysfs"
 		if mountPoint.Device != sysfsDevice {
 			continue
 		}
 		// Check whether sysfs is 'rw'
+		const permWritable = "rw"
 		if len(mountPoint.Opts) > 0 && mountPoint.Opts[0] == permWritable {
 			return true, nil
 		}
-		glog.Errorf("sysfs is not writable: %+v", mountPoint)
-		break
+		glog.Errorf("sysfs is not writable: %+v (mount options are %v)",
+			mountPoint, mountPoint.Opts)
+		return false, readOnlySysFSError
 	}
-	return false, nil
+
+	return false, errors.New("No sysfs mounted")
 }
 
 func writeIntStringFile(filename string, value int) error {

cmd/kube-proxy/app/options/options.go

@@ -94,5 +94,10 @@ func (s *ProxyServerConfig) AddFlags(fs *pflag.FlagSet) {
 	fs.Int32Var(&s.ConntrackMin, "conntrack-min", s.ConntrackMin,
 		"Minimum number of conntrack entries to allocate, regardless of conntrack-max-per-core (set conntrack-max-per-core=0 to leave the limit as-is).")
 	fs.DurationVar(&s.ConntrackTCPEstablishedTimeout.Duration, "conntrack-tcp-timeout-established", s.ConntrackTCPEstablishedTimeout.Duration, "Idle timeout for established TCP connections (0 to leave as-is)")
+	fs.DurationVar(
+		&s.ConntrackTCPCloseWaitTimeout.Duration, "conntrack-tcp-timeout-close-wait",
+		s.ConntrackTCPCloseWaitTimeout.Duration,
+		"NAT timeout for TCP connections in the CLOSE_WAIT state")
+
 	config.DefaultFeatureGate.AddFlag(fs)
 }

cmd/kube-proxy/app/server.go

@@ -316,12 +316,22 @@ func (s *ProxyServer) Run() error {
 				// administrator should decide whether and how to handle this issue,
 				// whether to drain the node and restart docker.
 				// TODO(random-liu): Remove this when the docker bug is fixed.
-				const message = "DOCKER RESTART NEEDED (docker issue #24000): /sys is read-only: can't raise conntrack limits, problems may arise later."
+				const message = "DOCKER RESTART NEEDED (docker issue #24000): /sys is read-only: " +
+					"cannot modify conntrack limits, problems may arise later."
 				s.Recorder.Eventf(s.Config.NodeRef, api.EventTypeWarning, err.Error(), message)
 			}
 		}
+
 		if s.Config.ConntrackTCPEstablishedTimeout.Duration > 0 {
-			if err := s.Conntracker.SetTCPEstablishedTimeout(int(s.Config.ConntrackTCPEstablishedTimeout.Duration / time.Second)); err != nil {
+			timeout := int(s.Config.ConntrackTCPEstablishedTimeout.Duration / time.Second)
+			if err := s.Conntracker.SetTCPEstablishedTimeout(timeout); err != nil {
+				return err
+			}
+		}
+
+		if s.Config.ConntrackTCPCloseWaitTimeout.Duration > 0 {
+			timeout := int(s.Config.ConntrackTCPCloseWaitTimeout.Duration / time.Second)
+			if err := s.Conntracker.SetTCPCloseWaitTimeout(timeout); err != nil {
 				return err
 			}
 		}

hack/verify-flags/known-flags.txt

@@ -90,21 +90,22 @@ concurrent-gc-syncs
 concurrent-namespace-syncs
 concurrent-replicaset-syncs
 concurrent-resource-quota-syncs
-concurrent-service-syncs
 concurrent-serviceaccount-token-syncs
+concurrent-service-syncs
 config-sync-period
 configure-cloud-routes
 conntrack-max
 conntrack-max-per-core
 conntrack-min
+conntrack-tcp-timeout-close-wait
 conntrack-tcp-timeout-established
 consumer-port
 consumer-service-name
 consumer-service-namespace
-contain-pod-resources
 container-port
 container-runtime
 container-runtime-endpoint
+contain-pod-resources
 controller-start-interval
 cors-allowed-origins
 cpu-cfs-quota

pkg/apis/componentconfig/types.go

@@ -75,8 +75,12 @@ type KubeProxyConfiguration struct {
 	// regardless of conntrackMaxPerCore (set conntrackMaxPerCore=0 to leave the limit as-is).
 	ConntrackMin int32 `json:"conntrackMin"`
 	// conntrackTCPEstablishedTimeout is how long an idle TCP connection will be kept open
-	// (e.g. '250ms', '2s').  Must be greater than 0.
+	// (e.g. '2s').  Must be greater than 0.
 	ConntrackTCPEstablishedTimeout unversioned.Duration `json:"conntrackTCPEstablishedTimeout"`
+	// conntrackTCPCloseWaitTimeout is how long an idle conntrack entry
+	// in CLOSE_WAIT state will remain in the conntrack
+	// table. (e.g. '60s'). Must be greater than 0 to set.
+	ConntrackTCPCloseWaitTimeout unversioned.Duration `json:"conntrackTCPCloseWaitTimeout"`
 }
 
 // Currently two modes of proxying are available: 'userspace' (older, stable) or 'iptables'

pkg/apis/componentconfig/v1alpha1/defaults.go

@@ -101,6 +101,9 @@ func SetDefaults_KubeProxyConfiguration(obj *KubeProxyConfiguration) {
 	if obj.ConntrackTCPEstablishedTimeout == zero {
 		obj.ConntrackTCPEstablishedTimeout = unversioned.Duration{Duration: 24 * time.Hour} // 1 day (1/5 default)
 	}
+	if obj.ConntrackTCPCloseWaitTimeout == zero {
+		obj.ConntrackTCPCloseWaitTimeout = unversioned.Duration{Duration: 1 * time.Hour}
+	}
 }
 
 func SetDefaults_KubeSchedulerConfiguration(obj *KubeSchedulerConfiguration) {

pkg/apis/componentconfig/v1alpha1/types.go

@@ -71,9 +71,13 @@ type KubeProxyConfiguration struct {
 	// conntrackMin is the minimum value of connect-tracking records to allocate,
 	// regardless of conntrackMaxPerCore (set conntrackMaxPerCore=0 to leave the limit as-is).
 	ConntrackMin int32 `json:"conntrackMin"`
-	// conntrackTCPEstablishedTimeout is how long an idle TCP connection will be kept open
-	// (e.g. '250ms', '2s').  Must be greater than 0.
+	// conntrackTCPEstablishedTimeout is how long an idle TCP connection
+	// will be kept open (e.g. '2s').  Must be greater than 0.
 	ConntrackTCPEstablishedTimeout unversioned.Duration `json:"conntrackTCPEstablishedTimeout"`
+	// conntrackTCPCloseWaitTimeout is how long an idle conntrack entry
+	// in CLOSE_WAIT state will remain in the conntrack
+	// table. (e.g. '60s'). Must be greater than 0 to set.
+	ConntrackTCPCloseWaitTimeout unversioned.Duration `json:"conntrackTCPCloseWaitTimeout"`
 }
 
 // Currently two modes of proxying are available: 'userspace' (older, stable) or 'iptables'

pkg/apis/componentconfig/v1alpha1/zz_generated.conversion.go

@@ -80,6 +80,7 @@ func autoConvert_v1alpha1_KubeProxyConfiguration_To_componentconfig_KubeProxyCon
 	out.ConntrackMaxPerCore = in.ConntrackMaxPerCore
 	out.ConntrackMin = in.ConntrackMin
 	out.ConntrackTCPEstablishedTimeout = in.ConntrackTCPEstablishedTimeout
+	out.ConntrackTCPCloseWaitTimeout = in.ConntrackTCPCloseWaitTimeout
 	return nil
 }
 
@@ -107,6 +108,7 @@ func autoConvert_componentconfig_KubeProxyConfiguration_To_v1alpha1_KubeProxyCon
 	out.ConntrackMaxPerCore = in.ConntrackMaxPerCore
 	out.ConntrackMin = in.ConntrackMin
 	out.ConntrackTCPEstablishedTimeout = in.ConntrackTCPEstablishedTimeout
+	out.ConntrackTCPCloseWaitTimeout = in.ConntrackTCPCloseWaitTimeout
 	return nil
 }
 

pkg/apis/componentconfig/v1alpha1/zz_generated.deepcopy.go

@@ -83,6 +83,7 @@ func DeepCopy_v1alpha1_KubeProxyConfiguration(in interface{}, out interface{}, c
 		out.ConntrackMaxPerCore = in.ConntrackMaxPerCore
 		out.ConntrackMin = in.ConntrackMin
 		out.ConntrackTCPEstablishedTimeout = in.ConntrackTCPEstablishedTimeout
+		out.ConntrackTCPCloseWaitTimeout = in.ConntrackTCPCloseWaitTimeout
 		return nil
 	}
 }

pkg/apis/componentconfig/zz_generated.deepcopy.go

@@ -171,6 +171,7 @@ func DeepCopy_componentconfig_KubeProxyConfiguration(in interface{}, out interfa
 		out.ConntrackMaxPerCore = in.ConntrackMaxPerCore
 		out.ConntrackMin = in.ConntrackMin
 		out.ConntrackTCPEstablishedTimeout = in.ConntrackTCPEstablishedTimeout
+		out.ConntrackTCPCloseWaitTimeout = in.ConntrackTCPCloseWaitTimeout
 		return nil
 	}
 }

pkg/generated/openapi/zz_generated.openapi.go

@@ -1989,12 +1989,18 @@ var OpenAPIDefinitions *common.OpenAPIDefinitions = &common.OpenAPIDefinitions{
 					},
 					"conntrackTCPEstablishedTimeout": {
 						SchemaProps: spec.SchemaProps{
-							Description: "conntrackTCPEstablishedTimeout is how long an idle TCP connection will be kept open (e.g. '250ms', '2s').  Must be greater than 0.",
+							Description: "conntrackTCPEstablishedTimeout is how long an idle TCP connection will be kept open (e.g. '2s').  Must be greater than 0.",
+							Ref:         spec.MustCreateRef("#/definitions/unversioned.Duration"),
+						},
+					},
+					"conntrackTCPCloseWaitTimeout": {
+						SchemaProps: spec.SchemaProps{
+							Description: "conntrackTCPCloseWaitTimeout is how long an idle conntrack entry in CLOSE_WAIT state will remain in the conntrack table. (e.g. '60s'). Must be greater than 0 to set.",
 							Ref:         spec.MustCreateRef("#/definitions/unversioned.Duration"),
 						},
 					},
 				},
-				Required: []string{"TypeMeta", "bindAddress", "clusterCIDR", "healthzBindAddress", "healthzPort", "hostnameOverride", "iptablesMasqueradeBit", "iptablesSyncPeriodSeconds", "kubeconfigPath", "masqueradeAll", "master", "oomScoreAdj", "mode", "portRange", "resourceContainer", "udpTimeoutMilliseconds", "conntrackMax", "conntrackMaxPerCore", "conntrackMin", "conntrackTCPEstablishedTimeout"},
+				Required: []string{"TypeMeta", "bindAddress", "clusterCIDR", "healthzBindAddress", "healthzPort", "hostnameOverride", "iptablesMasqueradeBit", "iptablesSyncPeriodSeconds", "kubeconfigPath", "masqueradeAll", "master", "oomScoreAdj", "mode", "portRange", "resourceContainer", "udpTimeoutMilliseconds", "conntrackMax", "conntrackMaxPerCore", "conntrackMin", "conntrackTCPEstablishedTimeout", "conntrackTCPCloseWaitTimeout"},
 			},
 		},
 		Dependencies: []string{
@@ -13717,12 +13723,18 @@ var OpenAPIDefinitions *common.OpenAPIDefinitions = &common.OpenAPIDefinitions{
 					},
 					"conntrackTCPEstablishedTimeout": {
 						SchemaProps: spec.SchemaProps{
-							Description: "conntrackTCPEstablishedTimeout is how long an idle TCP connection will be kept open (e.g. '250ms', '2s').  Must be greater than 0.",
+							Description: "conntrackTCPEstablishedTimeout is how long an idle TCP connection will be kept open (e.g. '2s').  Must be greater than 0.",
+							Ref:         spec.MustCreateRef("#/definitions/unversioned.Duration"),
+						},
+					},
+					"conntrackTCPCloseWaitTimeout": {
+						SchemaProps: spec.SchemaProps{
+							Description: "conntrackTCPCloseWaitTimeout is how long an idle conntrack entry in CLOSE_WAIT state will remain in the conntrack table. (e.g. '60s'). Must be greater than 0 to set.",
 							Ref:         spec.MustCreateRef("#/definitions/unversioned.Duration"),
 						},
 					},
 				},
-				Required: []string{"TypeMeta", "bindAddress", "clusterCIDR", "healthzBindAddress", "healthzPort", "hostnameOverride", "iptablesMasqueradeBit", "iptablesSyncPeriodSeconds", "kubeconfigPath", "masqueradeAll", "master", "oomScoreAdj", "mode", "portRange", "resourceContainer", "udpTimeoutMilliseconds", "conntrackMax", "conntrackMaxPerCore", "conntrackMin", "conntrackTCPEstablishedTimeout"},
+				Required: []string{"TypeMeta", "bindAddress", "clusterCIDR", "healthzBindAddress", "healthzPort", "hostnameOverride", "iptablesMasqueradeBit", "iptablesSyncPeriodSeconds", "kubeconfigPath", "masqueradeAll", "master", "oomScoreAdj", "mode", "portRange", "resourceContainer", "udpTimeoutMilliseconds", "conntrackMax", "conntrackMaxPerCore", "conntrackMin", "conntrackTCPEstablishedTimeout", "conntrackTCPCloseWaitTimeout"},
 			},
 		},
 		Dependencies: []string{

test/e2e/BUILD

@@ -58,6 +58,7 @@ go_library(
         "initial_resources.go",
         "job.go",
         "kibana_logging.go",
+        "kube_proxy.go",
         "kubectl.go",
         "kubelet.go",
         "kubelet_perf.go",
@@ -171,6 +172,7 @@ go_library(
         "//test/e2e/chaosmonkey:go_default_library",
         "//test/e2e/common:go_default_library",
         "//test/e2e/framework:go_default_library",
+        "//test/images/net/nat:go_default_library",
         "//test/utils:go_default_library",
         "//vendor:github.com/aws/aws-sdk-go/aws",
         "//vendor:github.com/aws/aws-sdk-go/aws/awserr",

test/e2e/framework/util.go

@@ -3332,7 +3332,7 @@ func LogSSHResult(result SSHResult) {
 	Logf("ssh %s: exit code: %d", remote, result.Code)
 }
 
-func IssueSSHCommand(cmd, provider string, node *api.Node) error {
+func IssueSSHCommandWithResult(cmd, provider string, node *api.Node) (*SSHResult, error) {
 	Logf("Getting external IP address for %s", node.Name)
 	host := ""
 	for _, a := range node.Status.Addresses {
@@ -3341,15 +3341,34 @@ func IssueSSHCommand(cmd, provider string, node *api.Node) error {
 			break
 		}
 	}
+
 	if host == "" {
-		return fmt.Errorf("couldn't find external IP address for node %s", node.Name)
+		return nil, fmt.Errorf("couldn't find external IP address for node %s", node.Name)
 	}
-	Logf("Calling %s on %s(%s)", cmd, node.Name, host)
+
+	Logf("SSH %q on %s(%s)", cmd, node.Name, host)
 	result, err := SSH(cmd, host, provider)
 	LogSSHResult(result)
+
 	if result.Code != 0 || err != nil {
-		return fmt.Errorf("failed running %q: %v (exit code %d)", cmd, err, result.Code)
+		return nil, fmt.Errorf("failed running %q: %v (exit code %d)",
+			cmd, err, result.Code)
 	}
+
+	return &result, nil
+}
+
+func IssueSSHCommand(cmd, provider string, node *api.Node) error {
+	result, err := IssueSSHCommandWithResult(cmd, provider, node)
+	if result != nil {
+		LogSSHResult(*result)
+	}
+
+	if result.Code != 0 || err != nil {
+		return fmt.Errorf("failed running %q: %v (exit code %d)",
+			cmd, err, result.Code)
+	}
+
 	return nil
 }
 

test/e2e/kube_proxy.go

+/*
+Copyright 2016 The Kubernetes Authors.
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+*/
+
+package e2e
+
+import (
+	"encoding/json"
+	"fmt"
+	"math"
+	"strconv"
+	"strings"
+	"time"
+
+	"k8s.io/kubernetes/pkg/api"
+
+	"k8s.io/kubernetes/test/e2e/framework"
+	"k8s.io/kubernetes/test/images/net/nat"
+
+	. "github.com/onsi/ginkgo"
+	. "github.com/onsi/gomega"
+)
+
+const kubeProxyE2eImage = "gcr.io/google_containers/e2e-net-amd64:1.0"
+
+var _ = framework.KubeDescribe("Network", func() {
+	const (
+		testDaemonHttpPort    = 11301
+		testDaemonTcpPort     = 11302
+		timeoutSeconds        = 10
+		postFinTimeoutSeconds = 5
+	)
+
+	fr := framework.NewDefaultFramework("network")
+
+	It("should set TCP CLOSE_WAIT timeout", func() {
+		nodes := framework.GetReadySchedulableNodesOrDie(fr.ClientSet)
+		ips := collectAddresses(nodes, api.NodeInternalIP)
+
+		if len(nodes.Items) < 2 {
+			framework.Skipf(
+				"Test requires >= 2 Ready nodes, but there are only %v nodes",
+				len(nodes.Items))
+		}
+
+		type NodeInfo struct {
+			node   *api.Node
+			name   string
+			nodeIp string
+		}
+
+		clientNodeInfo := NodeInfo{
+			node:   &nodes.Items[0],
+			name:   nodes.Items[0].Name,
+			nodeIp: ips[0],
+		}
+
+		serverNodeInfo := NodeInfo{
+			node:   &nodes.Items[1],
+			name:   nodes.Items[1].Name,
+			nodeIp: ips[1],
+		}
+
+		zero := int64(0)
+
+		clientPodSpec := &api.Pod{
+			ObjectMeta: api.ObjectMeta{
+				Name:      "e2e-net-client",
+				Namespace: fr.Namespace.Name,
+				Labels:    map[string]string{"app": "e2e-net-client"},
+			},
+			Spec: api.PodSpec{
+				NodeName: clientNodeInfo.name,
+				Containers: []api.Container{
+					{
+						Name:            "e2e-net-client",
+						Image:           kubeProxyE2eImage,
+						ImagePullPolicy: "Always",
+						Command: []string{
+							"/net", "-serve", fmt.Sprintf("0.0.0.0:%d", testDaemonHttpPort),
+						},
+					},
+				},
+				TerminationGracePeriodSeconds: &zero,
+			},
+		}
+
+		serverPodSpec := &api.Pod{
+			ObjectMeta: api.ObjectMeta{
+				Name:      "e2e-net-server",
+				Namespace: fr.Namespace.Name,
+				Labels:    map[string]string{"app": "e2e-net-server"},
+			},
+			Spec: api.PodSpec{
+				NodeName: serverNodeInfo.name,
+				Containers: []api.Container{
+					{
+						Name:            "e2e-net-server",
+						Image:           kubeProxyE2eImage,
+						ImagePullPolicy: "Always",
+						Command: []string{
+							"/net",
+							"-runner", "nat-closewait-server",
+							"-options",
+							fmt.Sprintf(`{"LocalAddr":"0.0.0.0:%v", "PostFindTimeoutSeconds":%v}`,
+								testDaemonTcpPort,
+								postFinTimeoutSeconds),
+						},
+						Ports: []api.ContainerPort{
+							{
+								Name:          "tcp",
+								ContainerPort: testDaemonTcpPort,
+								HostPort:      testDaemonTcpPort,
+							},
+						},
+					},
+				},
+				TerminationGracePeriodSeconds: &zero,
+			},
+		}
+
+		By(fmt.Sprintf(
+			"Launching a server daemon on node %v (node ip: %v, image: %v)",
+			serverNodeInfo.name,
+			serverNodeInfo.nodeIp,
+			kubeProxyE2eImage))
+		fr.PodClient().CreateSync(serverPodSpec)
+
+		By(fmt.Sprintf(
+			"Launching a client daemon on node %v (node ip: %v, image: %v)",
+			clientNodeInfo.name,
+			clientNodeInfo.nodeIp,
+			kubeProxyE2eImage))
+		fr.PodClient().CreateSync(clientPodSpec)
+
+		By("Make client connect")
+
+		options := nat.CloseWaitClientOptions{
+			RemoteAddr: fmt.Sprintf("%v:%v",
+				serverNodeInfo.nodeIp, testDaemonTcpPort),
+			TimeoutSeconds:        timeoutSeconds,
+			PostFinTimeoutSeconds: 0,
+			LeakConnection:        true,
+		}
+
+		jsonBytes, err := json.Marshal(options)
+		cmd := fmt.Sprintf(
+			`curl -X POST http://localhost:%v/run/nat-closewait-client -d `+
+				`'%v' 2>/dev/null`,
+			testDaemonHttpPort,
+			string(jsonBytes))
+		framework.RunHostCmdOrDie(fr.Namespace.Name, "e2e-net-client", cmd)
+
+		<-time.After(time.Duration(1) * time.Second)
+
+		By("Checking /proc/net/nf_conntrack for the timeout")
+		// If test flakes occur here, then this check should be performed
+		// in a loop as there may be a race with the client connecting.
+		framework.IssueSSHCommandWithResult(
+			fmt.Sprintf("sudo cat /proc/net/ip_conntrack | grep 'dport=%v'",
+				testDaemonTcpPort),
+			framework.TestContext.Provider,
+			clientNodeInfo.node)
+
+		// Timeout in seconds is available as the third column from
+		// /proc/net/ip_conntrack.
+		result, err := framework.IssueSSHCommandWithResult(
+			fmt.Sprintf(
+				"sudo cat /proc/net/ip_conntrack "+
+					"| grep 'CLOSE_WAIT.*dst=%v.*dport=%v' "+
+					"| awk '{print $3}'",
+				serverNodeInfo.nodeIp,
+				testDaemonTcpPort),
+			framework.TestContext.Provider,
+			clientNodeInfo.node)
+		framework.ExpectNoError(err)
+
+		timeoutSeconds, err := strconv.Atoi(strings.TrimSpace(result.Stdout))
+		framework.ExpectNoError(err)
+
+		// These must be synchronized from the default values set in
+		// pkg/apis/../defaults.go ConntrackTCPCloseWaitTimeout. The
+		// current defaults are hidden in the initialization code.
+		const epsilonSeconds = 10
+		const expectedTimeoutSeconds = 60 * 60
+
+		framework.Logf("conntrack entry timeout was: %v, expected: %v",
+			timeoutSeconds, expectedTimeoutSeconds)
+
+		Expect(math.Abs(float64(timeoutSeconds - expectedTimeoutSeconds))).Should(
+			BeNumerically("<", (epsilonSeconds)))
+	})
+})

test/test_owners.csv

@@ -261,6 +261,7 @@ Network Partition should come back up if node goes down,foxish,0
 Network Partition should create new pods when node is partitioned,foxish,0
 Network Partition should eagerly create replacement pod during network partition when termination grace is non-zero,foxish,0
 Network Partition should not reschedule pets if there is a network partition,foxish,0
+Network should set TCP CLOSE_WAIT timeout,bowei,0
 Networking Granular Checks: Pods should function for intra-pod communication: http,stts,0
 Networking Granular Checks: Pods should function for intra-pod communication: udp,freehan,0
 Networking Granular Checks: Pods should function for node-pod communication: http,spxtr,1