fix flakes on e2e test TCP CLOSE_WAIT timeout

it turns out that the e2e test was no using the timeout used to
hold the CLOSE_WAIT status, hence the test was flake depending
on how fast it checked the conntrack table.

This PR replaces the dependency on ssh using a pod to check the conntrack
entries on the host in a loop, to make the test more robust
and reduce the flakiness due to race conditions and/or ssh issues.

It also fixes a bug trying to grep the conntrack entry, where
the error was swallowed if a conntrack entry wasn't found..

test/e2e/network/kube_proxy.go

@@ -26,11 +26,11 @@ import (
 
 	v1 "k8s.io/api/core/v1"
 	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
+	"k8s.io/apimachinery/pkg/util/wait"
 
 	"k8s.io/kubernetes/test/e2e/framework"
 	e2enode "k8s.io/kubernetes/test/e2e/framework/node"
 	e2epod "k8s.io/kubernetes/test/e2e/framework/pod"
-	e2essh "k8s.io/kubernetes/test/e2e/framework/ssh"
 	"k8s.io/kubernetes/test/images/agnhost/net/nat"
 	imageutils "k8s.io/kubernetes/test/utils/image"
 
@@ -44,8 +44,8 @@ var _ = SIGDescribe("Network", func() {
 	const (
 		testDaemonHTTPPort     = 11301
 		testDaemonTCPPort      = 11302
-		timeoutSeconds        = 10
+		deadlineTimeoutSeconds = 10
-		postFinTimeoutSeconds = 5
+		postFinTimeoutSeconds  = 30
 	)
 
 	fr := framework.NewDefaultFramework("network")
@@ -81,16 +81,63 @@ var _ = SIGDescribe("Network", func() {
 
 		zero := int64(0)
 
+		// Create a pod to check the conntrack entries on the host node
+		// It mounts the host /proc/net folder to be able to access
+		// the nf_conntrack file with the host conntrack entries
+		privileged := true
+
+		hostExecPod := &v1.Pod{
+			ObjectMeta: metav1.ObjectMeta{
+				Name:      "e2e-net-exec",
+				Namespace: fr.Namespace.Name,
+				Labels:    map[string]string{"app": "e2e-net-exec"},
+			},
+			Spec: v1.PodSpec{
+				HostNetwork: true,
+				NodeName:    clientNodeInfo.name,
+				Containers: []v1.Container{
+					{
+						Name:            "e2e-net-exec",
+						Image:           kubeProxyE2eImage,
+						ImagePullPolicy: "Always",
+						Args:            []string{"pause"},
+						VolumeMounts: []v1.VolumeMount{
+							{
+								Name:      "proc-net",
+								MountPath: "/rootfs/proc/net",
+								ReadOnly:  true,
+							},
+						},
+						SecurityContext: &v1.SecurityContext{
+							Privileged: &privileged,
+						},
+					},
+				},
+				Volumes: []v1.Volume{
+					{
+						Name: "proc-net",
+						VolumeSource: v1.VolumeSource{
+							HostPath: &v1.HostPathVolumeSource{
+								Path: "/proc/net",
+							},
+						},
+					},
+				},
+				TerminationGracePeriodSeconds: &zero,
+			},
+		}
+		fr.PodClient().CreateSync(hostExecPod)
+
 		// Some distributions (Ubuntu 16.04 etc.) don't support the proc file.
-		_, err = e2essh.IssueSSHCommandWithResult(
+		_, err = framework.RunHostCmd(fr.Namespace.Name, "e2e-net-exec",
-			"ls /proc/net/nf_conntrack",
+			"ls /rootfs/proc/net/nf_conntrack")
-			framework.TestContext.Provider,
-			clientNodeInfo.node)
 		if err != nil && strings.Contains(err.Error(), "No such file or directory") {
-			framework.Skipf("The node %s does not support /proc/net/nf_conntrack", clientNodeInfo.name)
+			framework.Skipf("The node %s does not support /proc/net/nf_conntrack",
+				clientNodeInfo.name)
 		}
 		framework.ExpectNoError(err)
 
+		// Create the client and server pods
 		clientPodSpec := &v1.Pod{
 			ObjectMeta: metav1.ObjectMeta{
 				Name:      "e2e-net-client",
@@ -105,7 +152,7 @@ var _ = SIGDescribe("Network", func() {
 						Image:           kubeProxyE2eImage,
 						ImagePullPolicy: "Always",
 						Args: []string{
-							"net", "--serve", fmt.Sprintf("0.0.0.0:%d", testDaemonHTTPPort),
+							"net", "--serve", fmt.Sprintf(":%d", testDaemonHTTPPort),
 						},
 					},
 				},
@@ -130,7 +177,7 @@ var _ = SIGDescribe("Network", func() {
 							"net",
 							"--runner", "nat-closewait-server",
 							"--options",
-							fmt.Sprintf(`{"LocalAddr":"0.0.0.0:%v", "PostFindTimeoutSeconds":%v}`,
+							fmt.Sprintf(`{"LocalAddr":":%v", "PostFinTimeoutSeconds":%v}`,
 								testDaemonTCPPort,
 								postFinTimeoutSeconds),
 						},
@@ -166,8 +213,8 @@ var _ = SIGDescribe("Network", func() {
 		options := nat.CloseWaitClientOptions{
 			RemoteAddr: fmt.Sprintf("%v:%v",
 				serverNodeInfo.nodeIP, testDaemonTCPPort),
-			TimeoutSeconds:        timeoutSeconds,
+			TimeoutSeconds:        deadlineTimeoutSeconds,
-			PostFinTimeoutSeconds: 0,
+			PostFinTimeoutSeconds: postFinTimeoutSeconds,
 			LeakConnection:        true,
 		}
 
@@ -179,47 +226,52 @@ var _ = SIGDescribe("Network", func() {
 				`'%v' 2>/dev/null`,
 			testDaemonHTTPPort,
 			string(jsonBytes))
+		// Run the closewait command in a subroutine so it keeps waiting during postFinTimeoutSeconds
+		// otherwise the pod is deleted and the connection is closed loosing the conntrack entry
+		go func() {
 			framework.RunHostCmdOrDie(fr.Namespace.Name, "e2e-net-client", cmd)
+		}()
 
 		<-time.After(time.Duration(1) * time.Second)
 
 		ginkgo.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.
-		e2essh.IssueSSHCommandWithResult(
-			fmt.Sprintf("sudo cat /proc/net/nf_conntrack | grep 'dport=%v'",
-				testDaemonTCPPort),
-			framework.TestContext.Provider,
-			clientNodeInfo.node)
-
-		// Timeout in seconds is available as the fifth column from
-		// /proc/net/nf_conntrack.
-		result, err := e2essh.IssueSSHCommandWithResult(
-			fmt.Sprintf(
-				"sudo cat /proc/net/nf_conntrack "+
-					"| grep 'CLOSE_WAIT.*dst=%v.*dport=%v' "+
-					"| tail -n 1"+
-					"| awk '{print $5}' ",
-				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 = 60
 		const expectedTimeoutSeconds = 60 * 60
-
+		// Obtain the corresponding conntrack entry on the host checking
-		framework.Logf("conntrack entry timeout was: %v, expected: %v",
+		// the nf_conntrack file from the pod e2e-net-exec.
-			timeoutSeconds, expectedTimeoutSeconds)
+		// It retries in a loop if the entry is not found.
-
+		cmd = fmt.Sprintf("cat /rootfs/proc/net/nf_conntrack "+
-		gomega.Expect(math.Abs(float64(timeoutSeconds - expectedTimeoutSeconds))).Should(
+			"| grep -m 1 'CLOSE_WAIT.*dst=%v.*dport=%v' ",
-			gomega.BeNumerically("<", (epsilonSeconds)))
+			serverNodeInfo.nodeIP,
+			testDaemonTCPPort)
+		if err := wait.PollImmediate(5*time.Second, 30*time.Second, func() (bool, error) {
+			result, err := framework.RunHostCmd(fr.Namespace.Name, "e2e-net-exec", cmd)
+			// retry if we can't obtain the conntrack entry
+			if err != nil {
+				framework.Logf("failed to obtain conntrack entry: %v %v", result, err)
+				return false, nil
+			}
+			framework.Logf("conntrack entry for node %v and port %v:  %v", serverNodeInfo.nodeIP, testDaemonTCPPort, result)
+			// Timeout in seconds is available as the fifth column of
+			// the matched entry in /proc/net/nf_conntrack.
+			line := strings.Fields(result)
+			if len(line) < 5 {
+				return false, fmt.Errorf("conntrack entry does not have a timeout field: %v", line)
+			}
+			timeoutSeconds, err := strconv.Atoi(line[4])
+			if err != nil {
+				return false, fmt.Errorf("failed to convert matched timeout %s to integer: %v", line[4], err)
+			}
+			if math.Abs(float64(timeoutSeconds-expectedTimeoutSeconds)) < epsilonSeconds {
+				return true, nil
+			}
+			return false, fmt.Errorf("wrong TCP CLOSE_WAIT timeout: %v expected: %v", timeoutSeconds, expectedTimeoutSeconds)
+		}); err != nil {
+			framework.Failf("no conntrack entry for port %d on node %s", testDaemonTCPPort, serverNodeInfo.nodeIP)
+		}
 	})
 
 	// Regression test for #74839, where: