diff --git a/test/e2e/storage/testsuites/subpath.go b/test/e2e/storage/testsuites/subpath.go
index 80a1670096a..342374d1159 100644
--- a/test/e2e/storage/testsuites/subpath.go
+++ b/test/e2e/storage/testsuites/subpath.go
@@ -25,8 +25,10 @@ import (
 	v1 "k8s.io/api/core/v1"
 	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
 	"k8s.io/apimachinery/pkg/util/rand"
+	"k8s.io/apimachinery/pkg/util/sets"
 	"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"
 	"k8s.io/kubernetes/test/e2e/framework/volume"
 	"k8s.io/kubernetes/test/e2e/storage/testpatterns"
@@ -88,6 +90,7 @@ func (s *subPathTestSuite) defineTests(driver TestDriver, pattern testpatterns.T
 		config        *PerTestConfig
 		driverCleanup func()
 
+		hostExec          utils.HostExec
 		resource          *genericVolumeTestResource
 		roVolSource       *v1.VolumeSource
 		pod               *v1.Pod
@@ -117,6 +120,7 @@ func (s *subPathTestSuite) defineTests(driver TestDriver, pattern testpatterns.T
 		l.intreeOps, l.migratedOps = getMigrationVolumeOpCounts(f.ClientSet, driver.GetDriverInfo().InTreePluginName)
 		testVolumeSizeRange := s.getTestSuiteInfo().supportedSizeRange
 		l.resource = createGenericVolumeTestResource(driver, l.config, pattern, testVolumeSizeRange)
+		l.hostExec = utils.NewHostExec(f)
 
 		// Setup subPath test dependent resource
 		volType := pattern.VolType
@@ -175,6 +179,10 @@ func (s *subPathTestSuite) defineTests(driver TestDriver, pattern testpatterns.T
 			l.driverCleanup = nil
 		}
 
+		if l.hostExec != nil {
+			l.hostExec.Cleanup()
+		}
+
 		validateMigrationVolumeOpCounts(f.ClientSet, driver.GetDriverInfo().InTreePluginName, l.intreeOps, l.migratedOps)
 	}
 
@@ -333,7 +341,7 @@ func (s *subPathTestSuite) defineTests(driver TestDriver, pattern testpatterns.T
 		init()
 		defer cleanup()
 
-		testSubpathReconstruction(f, l.pod, false)
+		testSubpathReconstruction(f, l.hostExec, l.pod, false)
 	})
 
 	ginkgo.It("should unmount if pod is force deleted while kubelet is down [Disruptive][Slow][LinuxOnly]", func() {
@@ -345,7 +353,7 @@ func (s *subPathTestSuite) defineTests(driver TestDriver, pattern testpatterns.T
 			framework.Skipf("%s volume type does not support reconstruction, skipping", l.resource.volType)
 		}
 
-		testSubpathReconstruction(f, l.pod, true)
+		testSubpathReconstruction(f, l.hostExec, l.pod, true)
 	})
 
 	ginkgo.It("should support readOnly directory specified in the volumeMount", func() {
@@ -878,12 +886,17 @@ func testPodContainerRestart(f *framework.Framework, pod *v1.Pod) {
 	framework.ExpectNoError(err, "while waiting for container to stabilize")
 }
 
-func testSubpathReconstruction(f *framework.Framework, pod *v1.Pod, forceDelete bool) {
+func testSubpathReconstruction(f *framework.Framework, hostExec utils.HostExec, pod *v1.Pod, forceDelete bool) {
 	// This is mostly copied from TestVolumeUnmountsFromDeletedPodWithForceOption()
 
 	// Disruptive test run serially, we can cache all voluem global mount
 	// points and verify after the test that we do not leak any global mount point.
-	mountPoints := utils.FindVolumeGlobalMountPoints(f.ClientSet, pod)
+	nodeList, err := e2enode.GetReadySchedulableNodes(f.ClientSet)
+	framework.ExpectNoError(err, "while listing scheduable nodes")
+	globalMountPointsByNode := make(map[string]sets.String, len(nodeList.Items))
+	for _, node := range nodeList.Items {
+		globalMountPointsByNode[node.Name] = utils.FindVolumeGlobalMountPoints(hostExec, &node)
+	}
 
 	// Change to busybox
 	pod.Spec.Containers[0].Image = volume.GetTestImage(imageutils.GetE2EImage(imageutils.BusyBox))
@@ -898,7 +911,7 @@ func testSubpathReconstruction(f *framework.Framework, pod *v1.Pod, forceDelete
 
 	ginkgo.By(fmt.Sprintf("Creating pod %s", pod.Name))
 	removeUnusedContainers(pod)
-	pod, err := f.ClientSet.CoreV1().Pods(f.Namespace.Name).Create(pod)
+	pod, err = f.ClientSet.CoreV1().Pods(f.Namespace.Name).Create(pod)
 	framework.ExpectNoError(err, "while creating pod")
 
 	err = e2epod.WaitForPodRunningInNamespace(f.ClientSet, pod)
@@ -907,10 +920,24 @@ func testSubpathReconstruction(f *framework.Framework, pod *v1.Pod, forceDelete
 	pod, err = f.ClientSet.CoreV1().Pods(f.Namespace.Name).Get(pod.Name, metav1.GetOptions{})
 	framework.ExpectNoError(err, "while getting pod")
 
+	var podNode *v1.Node
+	for i := range nodeList.Items {
+		if nodeList.Items[i].Name == pod.Spec.NodeName {
+			podNode = &nodeList.Items[i]
+		}
+	}
+	framework.ExpectNotEqual(podNode, nil, "pod node should exist in scheduable nodes")
+
 	utils.TestVolumeUnmountsFromDeletedPodWithForceOption(f.ClientSet, f, pod, forceDelete, true)
 
-	mountPointsAfter := utils.FindVolumeGlobalMountPoints(f.ClientSet, pod)
-	gomega.Expect(mountPointsAfter).To(gomega.ConsistOf(mountPoints), "Global mount points leaked. Before: %v, After: %v.", mountPoints, mountPointsAfter)
+	if podNode != nil {
+		mountPoints := globalMountPointsByNode[podNode.Name]
+		mountPointsAfter := utils.FindVolumeGlobalMountPoints(hostExec, podNode)
+		s1 := mountPointsAfter.Difference(mountPoints)
+		s2 := mountPoints.Difference(mountPointsAfter)
+		gomega.Expect(s1).To(gomega.BeEmpty(), "global mount points leaked: %v", s1)
+		gomega.Expect(s2).To(gomega.BeEmpty(), "global mount points not found: %v", s2)
+	}
 }
 
 func formatVolume(f *framework.Framework, pod *v1.Pod) {
diff --git a/test/e2e/storage/utils/utils.go b/test/e2e/storage/utils/utils.go
index d4e3767df34..bb3b1918d19 100644
--- a/test/e2e/storage/utils/utils.go
+++ b/test/e2e/storage/utils/utils.go
@@ -32,6 +32,7 @@ import (
 	rbacv1 "k8s.io/api/rbac/v1"
 	apierrs "k8s.io/apimachinery/pkg/api/errors"
 	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
+	"k8s.io/apimachinery/pkg/util/sets"
 	"k8s.io/apimachinery/pkg/util/wait"
 	clientset "k8s.io/client-go/kubernetes"
 	"k8s.io/kubernetes/test/e2e/framework"
@@ -650,13 +651,12 @@ func CheckWriteToPath(pod *v1.Pod, volMode v1.PersistentVolumeMode, path string,
 }
 
 // findMountPoints returns all mount points on given node under specified directory.
-func findMountPoints(nodeIP string, dir string) []string {
-	result, err := e2essh.SSH(fmt.Sprintf(`find %s -type d -exec mountpoint {} \; | grep 'is a mountpoint$'`, dir), nodeIP, framework.TestContext.Provider)
-	e2essh.LogResult(result)
-	framework.ExpectNoError(err, "Encountered SSH error.")
+func findMountPoints(hostExec HostExec, node *v1.Node, dir string) []string {
+	result, err := hostExec.IssueCommandWithResult(fmt.Sprintf(`find %s -type d -exec mountpoint {} \; | grep 'is a mountpoint$' || true`, dir), node)
+	framework.ExpectNoError(err, "Encountered HostExec error.")
 	var mountPoints []string
 	if err != nil {
-		for _, line := range strings.Split(result.Stdout, "\n") {
+		for _, line := range strings.Split(result, "\n") {
 			if line == "" {
 				continue
 			}
@@ -667,9 +667,6 @@ func findMountPoints(nodeIP string, dir string) []string {
 }
 
 // FindVolumeGlobalMountPoints returns all volume global mount points on the node of given pod.
-func FindVolumeGlobalMountPoints(c clientset.Interface, pod *v1.Pod) []string {
-	nodeIP, err := framework.GetHostAddress(c, pod)
-	framework.ExpectNoError(err)
-	nodeIP = nodeIP + ":22"
-	return findMountPoints(nodeIP, "/var/lib/kubelet/plugins")
+func FindVolumeGlobalMountPoints(hostExec HostExec, node *v1.Node) sets.String {
+	return sets.NewString(findMountPoints(hostExec, node, "/var/lib/kubelet/plugins")...)
 }