diff --git a/test/e2e/cluster_upgrade.go b/test/e2e/cluster_upgrade.go
index 5d4f6a12009..2a2cbe8e902 100644
--- a/test/e2e/cluster_upgrade.go
+++ b/test/e2e/cluster_upgrade.go
@@ -52,6 +52,7 @@ var _ = framework.KubeDescribe("Upgrade [Feature:Upgrade]", func() {
 				testConfigMapsDuringUpgrade(f, sem)
 				testGuestbookApplicationDuringUpgrade(f, sem)
 				testDaemonSetDuringUpgrade(f, sem)
+				testJobsDuringUpgrade(f, sem)
 			})
 			cm.Do()
 		})
@@ -72,6 +73,7 @@ var _ = framework.KubeDescribe("Upgrade [Feature:Upgrade]", func() {
 				testConfigMapsBeforeAndAfterUpgrade(f, sem)
 				testGuestbookApplicationBeforeAndAfterUpgrade(f, sem)
 				testDaemonSetBeforeAndAfterUpgrade(f, sem)
+				testJobsBeforeAndAfterUpgrade(f, sem)
 			})
 			cm.Do()
 		})
@@ -90,6 +92,7 @@ var _ = framework.KubeDescribe("Upgrade [Feature:Upgrade]", func() {
 				testConfigMapsDuringUpgrade(f, sem)
 				testGuestbookApplicationDuringUpgrade(f, sem)
 				testDaemonSetDuringUpgrade(f, sem)
+				testJobsDuringUpgrade(f, sem)
 			})
 			cm.Do()
 		})
@@ -112,6 +115,7 @@ var _ = framework.KubeDescribe("Upgrade [Feature:Upgrade]", func() {
 				testConfigMapsBeforeAndAfterUpgrade(f, sem)
 				testGuestbookApplicationBeforeAndAfterUpgrade(f, sem)
 				testDaemonSetBeforeAndAfterUpgrade(f, sem)
+				testJobsBeforeAndAfterUpgrade(f, sem)
 			})
 			cm.Do()
 		})
@@ -132,6 +136,7 @@ var _ = framework.KubeDescribe("Upgrade [Feature:Upgrade]", func() {
 				testConfigMapsDuringUpgrade(f, sem)
 				testGuestbookApplicationDuringUpgrade(f, sem)
 				testDaemonSetDuringUpgrade(f, sem)
+				testJobsDuringUpgrade(f, sem)
 			})
 			cm.Do()
 		})
@@ -404,3 +409,43 @@ func testDaemonSet(f *framework.Framework, sem *chaosmonkey.Semaphore, testDurin
 	By("teardown daemonset")
 	TestDaemonSetWithNodeAffinityTeardown(f, dsName)
 }
+
+func testJobsBeforeAndAfterUpgrade(f *framework.Framework, sem *chaosmonkey.Semaphore) {
+	testJobs(f, sem, false)
+}
+
+func testJobsDuringUpgrade(f *framework.Framework, sem *chaosmonkey.Semaphore) {
+	testJobs(f, sem, true)
+}
+
+func testJobs(f *framework.Framework, sem *chaosmonkey.Semaphore, testDuringDisruption bool) {
+	parallelism := int32(2)
+	completions := int32(4)
+
+	// Setup
+	By("setup job")
+	job := TestJobsSetup("randomlySucceedOrFail", "rand-non-local", api.RestartPolicyNever, parallelism, completions)
+	// Validate
+	By("validate job before upgrade")
+	TestJobsValidate(f, job, completions)
+
+	sem.Ready()
+
+	if testDuringDisruption {
+		// Continuously validate
+		wait.Until(func() {
+			By("validate job during upgrade")
+			TestJobsValidate(f, job, completions)
+		}, framework.Poll, sem.StopCh)
+	} else {
+		// Block until chaosmonkey is done
+		By("waiting for upgrade to finish without validating job")
+		<-sem.StopCh
+	}
+
+	// Validate after upgrade
+	By("validate job after upgrade")
+	TestJobsValidate(f, job, completions)
+
+	// Teardown
+}
diff --git a/test/e2e/job.go b/test/e2e/job.go
index 05bea4bd323..aa08071c028 100644
--- a/test/e2e/job.go
+++ b/test/e2e/job.go
@@ -79,20 +79,13 @@ var _ = framework.KubeDescribe("Job", func() {
 
 	// Pods sometimes fail, but eventually succeed, after pod restarts
 	It("should run a job to completion when tasks sometimes fail and are not locally restarted", func() {
-		By("Creating a job")
 		// 50% chance of container success, local restarts.
 		// Can't use the failOnce approach because that relies
 		// on an emptyDir, which is not preserved across new pods.
 		// Worst case analysis: 15 failures, each taking 1 minute to
 		// run due to some slowness, 1 in 2^15 chance of happening,
 		// causing test flake.  Should be very rare.
-		job := newTestJob("randomlySucceedOrFail", "rand-non-local", api.RestartPolicyNever, parallelism, completions)
-		job, err := createJob(f.ClientSet, f.Namespace.Name, job)
-		Expect(err).NotTo(HaveOccurred())
-
-		By("Ensuring job reaches completions")
-		err = waitForJobFinish(f.ClientSet, f.Namespace.Name, job.Name, completions)
-		Expect(err).NotTo(HaveOccurred())
+		TestJobs(f, "randomlySucceedOrFail", "rand-non-local", api.RestartPolicyNever, parallelism, completions)
 	})
 
 	It("should keep restarting failed pods", func() {
@@ -339,3 +332,26 @@ func newBool(val bool) *bool {
 	*p = val
 	return p
 }
+
+func TestJobsSetup(behavior, name string, rPol api.RestartPolicy, parallelism, completions int32) *batch.Job {
+	return newTestJob(behavior, name, rPol, parallelism, completions)
+}
+
+func TestJobsValidate(f *framework.Framework, job *batch.Job, completions int32) {
+	By("Creating a job")
+	job, err := createJob(f.ClientSet, f.Namespace.Name, job)
+	Expect(err).NotTo(HaveOccurred())
+
+	By("Ensuring job reaches completions")
+	err = waitForJobFinish(f.ClientSet, f.Namespace.Name, job.Name, completions)
+	Expect(err).NotTo(HaveOccurred())
+
+	By("Delete the job")
+	err = deleteJob(f.ClientSet, f.Namespace.Name, job.ObjectMeta.Name)
+	Expect(err).NotTo(HaveOccurred())
+}
+
+func TestJobs(f *framework.Framework, behavior, name string, rPol api.RestartPolicy, parallelism, completions int32) {
+	job := TestJobsSetup(behavior, name, rPol, parallelism, completions)
+	TestJobsValidate(f, job, completions)
+}