diff --git a/test/e2e/dra/dra.go b/test/e2e/dra/dra.go
index c27e562e29a..3666fad08fc 100644
--- a/test/e2e/dra/dra.go
+++ b/test/e2e/dra/dra.go
@@ -272,7 +272,7 @@ var _ = framework.SIGDescribe("node")(framework.WithLabel("DRA"), func() {
 		ginkgo.It("must not run a pod if a claim is not ready", func(ctx context.Context) {
 			claim := b.ExternalClaim()
 			b.Create(f.TContext(ctx), claim)
-			pod := b.PodExternal()
+			pod := b.PodExternal(claim.Name)
 
 			// This bypasses scheduling and therefore the pod gets
 			// to run on the node although the claim is not ready.
@@ -295,7 +295,7 @@ var _ = framework.SIGDescribe("node")(framework.WithLabel("DRA"), func() {
 
 		ginkgo.It("must unprepare resources for force-deleted pod", func(ctx context.Context) {
 			claim := b.ExternalClaim()
-			pod := b.PodExternal()
+			pod := b.PodExternal(claim.Name)
 			zero := int64(0)
 			pod.Spec.TerminationGracePeriodSeconds = &zero
 
@@ -361,7 +361,7 @@ var _ = framework.SIGDescribe("node")(framework.WithLabel("DRA"), func() {
 			claimForContainer1.Spec.Devices.Config[1].Opaque.Parameters.Raw = []byte(`{"container1_config1":"true"}`)
 			claimForContainer1.Spec.Devices.Config[2].Opaque.Parameters.Raw = []byte(`{"container1_config2":"true"}`)
 
-			pod := b.PodExternal()
+			pod := b.PodExternal("")
 			pod.Spec.ResourceClaims = []v1.PodResourceClaim{
 				{
 					Name:              "all",
@@ -445,7 +445,7 @@ var _ = framework.SIGDescribe("node")(framework.WithLabel("DRA"), func() {
 			// that the race goes bad (old pod pending shutdown when
 			// new one arrives) and always schedules to the same node.
 			claim := b.ExternalClaim()
-			pod := b.PodExternal()
+			pod := b.PodExternal(claim.Name)
 			node := nodes.NodeNames[0]
 			pod.Spec.NodeSelector = map[string]string{"kubernetes.io/hostname": node}
 			oldClaim := b.Create(f.TContext(ctx), claim, pod)[0].(*resourceapi.ResourceClaim)
@@ -616,8 +616,8 @@ var _ = framework.SIGDescribe("node")(framework.WithLabel("DRA"), func() {
 		})
 
 		ginkgo.It("supports init containers with external claims", func(ctx context.Context) {
-			pod := b.PodExternal()
 			claim := b.ExternalClaim()
+			pod := b.PodExternal(claim.Name)
 			pod.Spec.InitContainers = []v1.Container{pod.Spec.Containers[0]}
 			pod.Spec.InitContainers[0].Name += "-init"
 			// This must succeed for the pod to start.
@@ -628,8 +628,8 @@ var _ = framework.SIGDescribe("node")(framework.WithLabel("DRA"), func() {
 		})
 
 		ginkgo.It("removes reservation from claim when pod is done", func(ctx context.Context) {
-			pod := b.PodExternal()
 			claim := b.ExternalClaim()
+			pod := b.PodExternal(claim.Name)
 			pod.Spec.Containers[0].Command = []string{"true"}
 			b.Create(f.TContext(ctx), claim, pod)
 
@@ -680,7 +680,7 @@ var _ = framework.SIGDescribe("node")(framework.WithLabel("DRA"), func() {
 			b := drautils.NewBuilderNow(tCtx, driver)
 
 			claim := b.ExternalClaim()
-			pod := b.PodExternal()
+			pod := b.PodExternal(claim.Name)
 			b.Create(tCtx, claim, pod)
 
 			// Cannot run pod, no devices.
@@ -725,7 +725,7 @@ var _ = framework.SIGDescribe("node")(framework.WithLabel("DRA"), func() {
 			// Build behaves the same for both driver instances.
 			b := drautils.NewBuilderNow(tCtx, oldDriver)
 			claim := b.ExternalClaim()
-			pod := b.PodExternal()
+			pod := b.PodExternal(claim.Name)
 			b.Create(tCtx, claim, pod)
 			b.TestPod(tCtx, pod)
 
@@ -768,7 +768,7 @@ var _ = framework.SIGDescribe("node")(framework.WithLabel("DRA"), func() {
 			// Build behaves the same for both driver instances.
 			b := drautils.NewBuilderNow(tCtx, oldDriver)
 			claim := b.ExternalClaim()
-			pod := b.PodExternal()
+			pod := b.PodExternal(claim.Name)
 			b.Create(tCtx, claim, pod)
 			b.TestPod(tCtx, pod)
 
@@ -814,7 +814,7 @@ var _ = framework.SIGDescribe("node")(framework.WithLabel("DRA"), func() {
 			// Build behaves the same for both driver instances.
 			b := drautils.NewBuilderNow(tCtx, oldDriver)
 			claim := b.ExternalClaim()
-			pod := b.PodExternal()
+			pod := b.PodExternal(claim.Name)
 			b.Create(tCtx, claim, pod)
 			b.TestPod(tCtx, pod)
 
@@ -901,10 +901,11 @@ var _ = framework.SIGDescribe("node")(framework.WithLabel("DRA"), func() {
 		ginkgo.It("supports sharing a claim concurrently", func(ctx context.Context) {
 			tCtx := f.TContext(ctx)
 			var objects []klog.KMetadata
-			objects = append(objects, b.ExternalClaim())
+			claim := b.ExternalClaim()
+			objects = append(objects, claim)
 			pods := make([]*v1.Pod, numPods)
 			for i := 0; i < numPods; i++ {
-				pod := b.PodExternal()
+				pod := b.PodExternal(claim.Name)
 				pods[i] = pod
 				objects = append(objects, pod)
 			}
@@ -1015,18 +1016,18 @@ var _ = framework.SIGDescribe("node")(framework.WithLabel("DRA"), func() {
 
 		ginkgo.It("supports simple pod referencing external resource claim", func(ctx context.Context) {
 			tCtx := f.TContext(ctx)
-			pod := b.PodExternal()
 			claim := b.ExternalClaim()
+			pod := b.PodExternal(claim.Name)
 			b.Create(tCtx, claim, pod)
 			b.TestPod(tCtx, pod)
 		})
 
 		ginkgo.It("supports external claim referenced by multiple pods", func(ctx context.Context) {
 			tCtx := f.TContext(ctx)
-			pod1 := b.PodExternal()
-			pod2 := b.PodExternal()
-			pod3 := b.PodExternal()
 			claim := b.ExternalClaim()
+			pod1 := b.PodExternal(claim.Name)
+			pod2 := b.PodExternal(claim.Name)
+			pod3 := b.PodExternal(claim.Name)
 			b.Create(tCtx, claim, pod1, pod2, pod3)
 
 			for _, pod := range []*v1.Pod{pod1, pod2, pod3} {
@@ -1036,10 +1037,10 @@ var _ = framework.SIGDescribe("node")(framework.WithLabel("DRA"), func() {
 
 		ginkgo.It("supports external claim referenced by multiple containers of multiple pods", func(ctx context.Context) {
 			tCtx := f.TContext(ctx)
-			pod1 := b.PodExternalMultiple()
-			pod2 := b.PodExternalMultiple()
-			pod3 := b.PodExternalMultiple()
 			claim := b.ExternalClaim()
+			pod1 := b.PodExternalMultiple(claim.Name)
+			pod2 := b.PodExternalMultiple(claim.Name)
+			pod3 := b.PodExternalMultiple(claim.Name)
 			b.Create(tCtx, claim, pod1, pod2, pod3)
 
 			for _, pod := range []*v1.Pod{pod1, pod2, pod3} {
@@ -1061,8 +1062,8 @@ var _ = framework.SIGDescribe("node")(framework.WithLabel("DRA"), func() {
 
 		ginkgo.It("must deallocate after use", func(ctx context.Context) {
 			tCtx := f.TContext(ctx)
-			pod := b.PodExternal()
 			claim := b.ExternalClaim()
+			pod := b.PodExternal(claim.Name)
 			b.Create(tCtx, claim, pod)
 
 			gomega.Eventually(ctx, func(ctx context.Context) (*resourceapi.ResourceClaim, error) {
@@ -1086,11 +1087,11 @@ var _ = framework.SIGDescribe("node")(framework.WithLabel("DRA"), func() {
 				tCtx := f.TContext(ctx)
 				extendedResourceName := deployDevicePlugin(tCtx, f, nodes.NodeNames[0:1], false)
 
-				pod := b.PodExternal()
+				claim := b.ExternalClaim()
+				pod := b.PodExternal(claim.Name)
 				resources := v1.ResourceList{extendedResourceName: resource.MustParse("1")}
 				pod.Spec.Containers[0].Resources.Requests = resources
 				pod.Spec.Containers[0].Resources.Limits = resources
-				claim := b.ExternalClaim()
 				b.Create(tCtx, claim, pod)
 				b.TestPod(tCtx, pod)
 			})
@@ -1117,13 +1118,12 @@ var _ = framework.SIGDescribe("node")(framework.WithLabel("DRA"), func() {
 			b.Create(tCtx, firstClaim, secondClaim)
 
 			// First pod uses only firstClaim
-			firstPod := b.PodExternal()
+			firstPod := b.PodExternal(firstClaim.Name)
 			b.Create(tCtx, firstPod)
 			b.TestPod(tCtx, firstPod)
 
 			// Second pod uses firstClaim (already prepared) + secondClaim (new)
-			secondPod := b.PodExternal()
-
+			secondPod := b.PodExternal("")
 			secondPod.Spec.ResourceClaims = []v1.PodResourceClaim{
 				{
 					Name:              "first",
@@ -1362,7 +1362,7 @@ var _ = framework.SIGDescribe("node")(framework.WithLabel("DRA"), func() {
 					},
 				},
 			}
-			pod := b1.PodExternal()
+			pod := b1.PodExternal("")
 			podClaimName := "resource-claim"
 			externalClaimName := "external-multiclaim"
 			pod.Spec.ResourceClaims = []v1.PodResourceClaim{
@@ -1453,7 +1453,7 @@ var _ = framework.SIGDescribe("node")(framework.WithLabel("DRA"), func() {
 					},
 				},
 			}
-			pod := b1.PodExternal()
+			pod := b1.PodExternal("")
 			podClaimName := "resource-claim"
 			externalClaimName := "external-multiclaim"
 			pod.Spec.ResourceClaims = []v1.PodResourceClaim{
@@ -1532,7 +1532,7 @@ var _ = framework.SIGDescribe("node")(framework.WithLabel("DRA"), func() {
 					},
 				},
 			}
-			pod := b1.PodExternal()
+			pod := b1.PodExternal("")
 			podClaimName := "resource-claim"
 			externalClaimName := "external-multiclaim"
 			pod.Spec.ResourceClaims = []v1.PodResourceClaim{
@@ -1750,7 +1750,7 @@ var _ = framework.SIGDescribe("node")(framework.WithLabel("DRA"), func() {
 					},
 				},
 			}
-			pod := b.PodExternal()
+			pod := b.PodExternal("")
 			podClaimName := "resource-claim"
 			pod.Spec.ResourceClaims = []v1.PodResourceClaim{
 				{
@@ -1831,7 +1831,7 @@ var _ = framework.SIGDescribe("node")(framework.WithLabel("DRA"), func() {
 					},
 				},
 			}
-			pod := b.PodExternal()
+			pod := b.PodExternal("")
 			podClaimName := "resource-claim"
 			pod.Spec.ResourceClaims = []v1.PodResourceClaim{
 				{
@@ -1904,7 +1904,7 @@ var _ = framework.SIGDescribe("node")(framework.WithLabel("DRA"), func() {
 			// available, there should be sufficient counters left to allocate
 			// a device.
 			claim := b.ExternalClaim()
-			pod := b.PodExternal()
+			pod := b.PodExternal(claim.Name)
 			pod.Spec.ResourceClaims[0].ResourceClaimName = &claim.Name
 			b.Create(tCtx, claim, pod)
 			b.TestPod(tCtx, pod)
@@ -1912,8 +1912,7 @@ var _ = framework.SIGDescribe("node")(framework.WithLabel("DRA"), func() {
 			// For the second pod, there should not be sufficient counters left, so
 			// it should not succeed. This means the pod should remain in the pending state.
 			claim2 := b.ExternalClaim()
-			pod2 := b.PodExternal()
-			pod2.Spec.ResourceClaims[0].ResourceClaimName = &claim2.Name
+			pod2 := b.PodExternal(claim2.Name)
 			b.Create(tCtx, claim2, pod2)
 
 			gomega.Consistently(ctx, func(ctx context.Context) error {
@@ -1969,8 +1968,7 @@ var _ = framework.SIGDescribe("node")(framework.WithLabel("DRA"), func() {
 					"memory": resource.MustParse("4Gi"),
 				},
 			}
-			pod := b.PodExternal()
-			pod.Spec.ResourceClaims[0].ResourceClaimName = &claim.Name
+			pod := b.PodExternal(claim.Name)
 			b.Create(tCtx, claim, pod)
 			b.TestPod(tCtx, pod)
 
@@ -1981,8 +1979,7 @@ var _ = framework.SIGDescribe("node")(framework.WithLabel("DRA"), func() {
 					"memory": resource.MustParse("8Gi"),
 				},
 			}
-			pod2 := b.PodExternal()
-			pod2.Spec.ResourceClaims[0].ResourceClaimName = &claim2.Name
+			pod2 := b.PodExternal(claim2.Name)
 			b.Create(tCtx, claim2, pod2)
 
 			// The third pod should be able to use the rest 4Gi of the device.
@@ -1992,8 +1989,7 @@ var _ = framework.SIGDescribe("node")(framework.WithLabel("DRA"), func() {
 					"memory": resource.MustParse("4Gi"),
 				},
 			}
-			pod3 := b.PodExternal()
-			pod3.Spec.ResourceClaims[0].ResourceClaimName = &claim3.Name
+			pod3 := b.PodExternal(claim3.Name)
 			b.Create(tCtx, claim3, pod3)
 			b.TestPod(tCtx, pod3)
 
@@ -2298,7 +2294,7 @@ var _ = framework.SIGDescribe("node")(framework.WithLabel("DRA"), func() {
 			drautils.TestContainerEnv(tCtx, pod, pod.Spec.Containers[0].Name, false, containerEnv...)
 
 			claim := b.ExternalClaim()
-			pod2 := b.PodExternal()
+			pod2 := b.PodExternal(claim.Name)
 			b.Create(tCtx, claim, pod2)
 			b.TestPod(tCtx, pod2)
 
@@ -2832,8 +2828,8 @@ var _ = framework.SIGDescribe("node")(framework.WithLabel("DRA"), func() {
 
 		f.It("must be possible for the driver to update the ResourceClaim.Status.Devices once allocated", f.WithFeatureGate(features.DRAResourceClaimDeviceStatus), func(ctx context.Context) {
 			tCtx := f.TContext(ctx)
-			pod := b.PodExternal()
 			claim := b.ExternalClaim()
+			pod := b.PodExternal(claim.Name)
 			b.Create(tCtx, claim, pod)
 
 			// Waits for the ResourceClaim to be allocated and the pod to be scheduled.
@@ -3073,7 +3069,7 @@ var _ = framework.SIGDescribe("node")(framework.WithLabel("DRA"), func() {
 			claim1b := b1.ExternalClaim()
 			claim2 := b2.ExternalClaim()
 			claim2b := b2.ExternalClaim()
-			pod := b1.PodExternal()
+			pod := b1.PodExternal(claim1.Name)
 			for i, claim := range []*resourceapi.ResourceClaim{claim1b, claim2, claim2b} {
 				pod.Spec.ResourceClaims = append(pod.Spec.ResourceClaims,
 					v1.PodResourceClaim{
diff --git a/test/e2e/dra/utils/builder.go b/test/e2e/dra/utils/builder.go
index 47adf5f0d7f..3404687b49d 100644
--- a/test/e2e/dra/utils/builder.go
+++ b/test/e2e/dra/utils/builder.go
@@ -60,7 +60,7 @@ func (b *Builder) ExtendedResourceName(i int) string {
 	case SingletonIndex:
 		return e2enode.SampleDeviceResourceName
 	default:
-		return b.driver.Name + "/resource" + fmt.Sprintf("-%d", i)
+		return b.Driver.Name + "/resource" + fmt.Sprintf("-%d", i)
 	}
 }
 
@@ -68,7 +68,7 @@ func (b *Builder) ExtendedResourceName(i int) string {
 // namespace.
 type Builder struct {
 	namespace               string
-	driver                  *Driver
+	Driver                  *Driver
 	UseExtendedResourceName bool
 
 	podCounter      int
@@ -79,7 +79,7 @@ type Builder struct {
 
 // ClassName returns the default device class name.
 func (b *Builder) ClassName() string {
-	return b.namespace + b.driver.NameSuffix + "-class"
+	return b.namespace + b.Driver.NameSuffix + "-class"
 }
 
 // SingletonIndex causes Builder.Class and ExtendedResourceName to create a
@@ -115,14 +115,14 @@ func (b *Builder) Class(i int) *resourceapi.DeviceClass {
 	}
 	class.Spec.Selectors = []resourceapi.DeviceSelector{{
 		CEL: &resourceapi.CELDeviceSelector{
-			Expression: fmt.Sprintf(`device.driver == "%s"`, b.driver.Name),
+			Expression: fmt.Sprintf(`device.driver == "%s"`, b.Driver.Name),
 		},
 	}}
 	if b.ClassParameters != "" {
 		class.Spec.Config = []resourceapi.DeviceClassConfiguration{{
 			DeviceConfiguration: resourceapi.DeviceConfiguration{
 				Opaque: &resourceapi.OpaqueDeviceConfiguration{
-					Driver:     b.driver.Name,
+					Driver:     b.Driver.Name,
 					Parameters: runtime.RawExtension{Raw: []byte(b.ClassParameters)},
 				},
 			},
@@ -135,7 +135,7 @@ func (b *Builder) Class(i int) *resourceapi.DeviceClass {
 // that test pods can reference
 func (b *Builder) ExternalClaim() *resourceapi.ResourceClaim {
 	b.claimCounter++
-	name := "external-claim" + b.driver.NameSuffix // This is what podExternal expects.
+	name := "external-claim" + b.Driver.NameSuffix // This is what podExternal expects.
 	if b.claimCounter > 1 {
 		name += fmt.Sprintf("-%d", b.claimCounter)
 	}
@@ -160,7 +160,7 @@ func (b *Builder) claimSpecWithV1beta1() resourcev1beta1.ResourceClaimSpec {
 			Config: []resourcev1beta1.DeviceClaimConfiguration{{
 				DeviceConfiguration: resourcev1beta1.DeviceConfiguration{
 					Opaque: &resourcev1beta1.OpaqueDeviceConfiguration{
-						Driver: b.driver.Name,
+						Driver: b.Driver.Name,
 						Parameters: runtime.RawExtension{
 							Raw: []byte(parameters),
 						},
@@ -188,7 +188,7 @@ func (b *Builder) claimSpecWithV1beta2() resourcev1beta2.ResourceClaimSpec {
 			Config: []resourcev1beta2.DeviceClaimConfiguration{{
 				DeviceConfiguration: resourcev1beta2.DeviceConfiguration{
 					Opaque: &resourcev1beta2.OpaqueDeviceConfiguration{
-						Driver: b.driver.Name,
+						Driver: b.Driver.Name,
 						Parameters: runtime.RawExtension{
 							Raw: []byte(parameters),
 						},
@@ -216,7 +216,7 @@ func (b *Builder) ClaimSpec() resourceapi.ResourceClaimSpec {
 			Config: []resourceapi.DeviceClaimConfiguration{{
 				DeviceConfiguration: resourceapi.DeviceConfiguration{
 					Opaque: &resourceapi.OpaqueDeviceConfiguration{
-						Driver: b.driver.Name,
+						Driver: b.Driver.Name,
 						Parameters: runtime.RawExtension{
 							Raw: []byte(parameters),
 						},
@@ -249,7 +249,7 @@ func (b *Builder) Pod() *v1.Pod {
 	pod.Spec.RestartPolicy = v1.RestartPolicyNever
 	pod.GenerateName = ""
 	b.podCounter++
-	pod.Name = fmt.Sprintf("tester%s-%d", b.driver.NameSuffix, b.podCounter)
+	pod.Name = fmt.Sprintf("tester%s-%d", b.Driver.NameSuffix, b.podCounter)
 	return pod
 }
 
@@ -313,12 +313,11 @@ func (b *Builder) PodInlineMultiple() (*v1.Pod, *resourceapi.ResourceClaimTempla
 	return pod, template
 }
 
-// PodExternal adds a pod that references external resource claim with default class name and parameters.
-func (b *Builder) PodExternal() *v1.Pod {
+// PodExternal adds a pod that references the named resource claim.
+func (b *Builder) PodExternal(externalClaimName string) *v1.Pod {
 	pod := b.Pod()
 	pod.Spec.Containers[0].Name = "with-resource"
 	podClaimName := "resource-claim"
-	externalClaimName := "external-claim" + b.driver.NameSuffix
 	pod.Spec.ResourceClaims = []v1.PodResourceClaim{
 		{
 			Name:              podClaimName,
@@ -329,9 +328,9 @@ func (b *Builder) PodExternal() *v1.Pod {
 	return pod
 }
 
-// podShared returns a pod with 3 containers that reference external resource claim with default class name and parameters.
-func (b *Builder) PodExternalMultiple() *v1.Pod {
-	pod := b.PodExternal()
+// podShared returns a pod with 3 containers that reference the named external resource claim.
+func (b *Builder) PodExternalMultiple(externalClaimName string) *v1.Pod {
+	pod := b.PodExternal(externalClaimName)
 	pod.Spec.Containers = append(pod.Spec.Containers, *pod.Spec.Containers[0].DeepCopy(), *pod.Spec.Containers[0].DeepCopy())
 	pod.Spec.Containers[1].Name += "-1"
 	pod.Spec.Containers[2].Name += "-2"
@@ -417,7 +416,7 @@ func (b *Builder) DeletePodAndWaitForNotFound(tCtx ktesting.TContext, pod *v1.Po
 func (b *Builder) TestPod(tCtx ktesting.TContext, pod *v1.Pod, env ...string) {
 	tCtx.Helper()
 
-	if !b.driver.WithKubelet {
+	if !b.Driver.WithKubelet {
 		// Less testing when we cannot rely on the kubelet to actually run the pod.
 		err := e2epod.WaitForPodScheduled(tCtx, tCtx.Client(), pod.Namespace, pod.Name)
 		tCtx.ExpectNoError(err, "schedule pod")
@@ -474,7 +473,7 @@ func TestContainerEnv(tCtx ktesting.TContext, pod *v1.Pod, containerName string,
 }
 
 func NewBuilder(f *framework.Framework, driver *Driver) *Builder {
-	b := &Builder{driver: driver}
+	b := &Builder{Driver: driver}
 	ginkgo.BeforeEach(func() {
 		b.setUp(f.TContext(context.Background()))
 	})
@@ -482,7 +481,7 @@ func NewBuilder(f *framework.Framework, driver *Driver) *Builder {
 }
 
 func NewBuilderNow(tCtx ktesting.TContext, driver *Driver) *Builder {
-	b := &Builder{driver: driver}
+	b := &Builder{Driver: driver}
 	b.setUp(tCtx)
 	return b
 }
@@ -516,7 +515,7 @@ func (b *Builder) tearDown(tCtx ktesting.TContext) {
 		}
 		tCtx.Logf("Deleting %T %s", &pod, klog.KObj(&pod))
 		options := metav1.DeleteOptions{}
-		if !b.driver.WithRealNodes {
+		if !b.Driver.WithRealNodes {
 			// Force-delete, no kubelet.
 			options.GracePeriodSeconds = ptr.To(int64(0))
 		}
@@ -542,7 +541,7 @@ func (b *Builder) tearDown(tCtx ktesting.TContext) {
 		}
 	}
 
-	for host, plugin := range b.driver.Nodes {
+	for host, plugin := range b.Driver.Nodes {
 		tCtx.Logf("Waiting for resources on %s to be unprepared", host)
 		tCtx.Eventually(func(ktesting.TContext) []app.ClaimID { return plugin.GetPreparedResources() }).WithTimeout(time.Minute).Should(gomega.BeEmpty(), "prepared claims on host %s", host)
 	}
diff --git a/test/e2e/dra/utils/deploy.go b/test/e2e/dra/utils/deploy.go
index becde2315f8..5e0b16bb777 100644
--- a/test/e2e/dra/utils/deploy.go
+++ b/test/e2e/dra/utils/deploy.go
@@ -1038,6 +1038,7 @@ func (d *Driver) TearDown(tCtx ktesting.TContext) {
 //
 // Only use this in tests where kubelet support for DRA is guaranteed.
 func (d *Driver) IsGone(tCtx ktesting.TContext) {
+	tCtx.Helper()
 	tCtx.Logf("Waiting for ResourceSlices of driver %s to be removed...", d.Name)
 	tCtx.Eventually(d.NewGetSlices()).WithTimeout(2 * time.Minute).Should(gomega.HaveField("Items", gomega.BeEmpty()))
 }
diff --git a/test/e2e_dra/coredra_test.go b/test/e2e_dra/coredra_test.go
index 293014a03bf..2bc1d483d7d 100644
--- a/test/e2e_dra/coredra_test.go
+++ b/test/e2e_dra/coredra_test.go
@@ -31,7 +31,7 @@ import (
 func coreDRA(tCtx ktesting.TContext, b *drautils.Builder) upgradedTestFunc {
 	namespace := tCtx.Namespace()
 	claim := b.ExternalClaim()
-	pod := b.PodExternal()
+	pod := b.PodExternal(claim.Name)
 	b.Create(tCtx, claim, pod)
 	b.TestPod(tCtx, pod)
 
@@ -43,7 +43,7 @@ func coreDRA(tCtx ktesting.TContext, b *drautils.Builder) upgradedTestFunc {
 
 		// Create another claim and pod, this time using the latest Kubernetes.
 		claim = b.ExternalClaim()
-		pod = b.PodExternal()
+		pod = b.PodExternal(claim.Name)
 		pod.Spec.ResourceClaims[0].ResourceClaimName = &claim.Name
 		b.Create(tCtx, claim, pod)
 		b.TestPod(tCtx, pod)
diff --git a/test/e2e_dra/devicetaints_test.go b/test/e2e_dra/devicetaints_test.go
new file mode 100644
index 00000000000..9dc0d55c050
--- /dev/null
+++ b/test/e2e_dra/devicetaints_test.go
@@ -0,0 +1,136 @@
+/*
+Copyright 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 e2edra
+
+import (
+	"time"
+
+	"github.com/onsi/gomega"
+	resourceapi "k8s.io/api/resource/v1"
+	resourcealpha "k8s.io/api/resource/v1alpha3"
+	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
+	drautils "k8s.io/kubernetes/test/e2e/dra/utils"
+	e2epod "k8s.io/kubernetes/test/e2e/framework/pod"
+	"k8s.io/kubernetes/test/utils/ktesting"
+	"k8s.io/utils/ptr"
+)
+
+// deviceTaints checks that:
+// - A pod which gets scheduled on the previous release because of a toleration is kept running after an upgrade.
+// - A DeviceTaintRule created to evict the pod before a downgrade prevents pod scheduling after a downgrade.
+func deviceTaints(tCtx ktesting.TContext, b *drautils.Builder) upgradedTestFunc {
+	namespace := tCtx.Namespace()
+	taintKey := "devicetaints"
+	taintValueFromSlice := "from-slice"
+	taintValueFromRule := "from-rule"
+	taintedDevice := "tainted-device"
+
+	// We need additional devices which are only used by this test.
+	// We achieve that with cluster-scoped devices that start out with
+	// a taint.
+	slice := &resourceapi.ResourceSlice{
+		ObjectMeta: metav1.ObjectMeta{
+			Name: "devicetaints",
+		},
+		Spec: resourceapi.ResourceSliceSpec{
+			Driver: b.Driver.Name,
+			Pool: resourceapi.ResourcePool{
+				Name:               "devicetaints",
+				ResourceSliceCount: 1,
+			},
+			AllNodes: ptr.To(true),
+			Devices: []resourceapi.Device{{
+				Name: taintedDevice,
+				Attributes: map[resourceapi.QualifiedName]resourceapi.DeviceAttribute{
+					"example.com/type": {
+						StringValue: ptr.To("devicetaints"),
+					},
+				},
+				Taints: []resourceapi.DeviceTaint{{
+					Key:    taintKey,
+					Value:  taintValueFromSlice,
+					Effect: resourceapi.DeviceTaintEffectNoSchedule,
+				}},
+			}},
+		},
+	}
+	_, err := tCtx.Client().ResourceV1().ResourceSlices().Create(tCtx, slice, metav1.CreateOptions{})
+	tCtx.ExpectNoError(err)
+
+	tCtx.Log("The pod wants exactly the tainted device -> not schedulable.")
+	claim := b.ExternalClaim()
+	pod := b.PodExternal(claim.Name)
+	claim.Spec.Devices.Requests[0].Exactly.Selectors = []resourceapi.DeviceSelector{{
+		CEL: &resourceapi.CELDeviceSelector{
+			Expression: `device.attributes["example.com"].?type.orValue("") == "devicetaints"`,
+		},
+	}}
+	b.Create(tCtx, claim, pod)
+	tCtx.ExpectNoError(e2epod.WaitForPodNameUnschedulableInNamespace(tCtx, tCtx.Client(), pod.Name, namespace))
+
+	tCtx.Log("Adding a toleration makes the pod schedulable.")
+	claim.Spec.Devices.Requests[0].Exactly.Tolerations = []resourceapi.DeviceToleration{{
+		Key:    taintKey,
+		Value:  taintValueFromSlice,
+		Effect: resourceapi.DeviceTaintEffectNoSchedule,
+	}}
+	tCtx.ExpectNoError(tCtx.Client().ResourceV1().ResourceClaims(namespace).Delete(tCtx, claim.Name, metav1.DeleteOptions{}))
+	_, err = tCtx.Client().ResourceV1().ResourceClaims(namespace).Create(tCtx, claim, metav1.CreateOptions{})
+	tCtx.ExpectNoError(err)
+	b.TestPod(tCtx, pod)
+
+	return func(tCtx ktesting.TContext) downgradedTestFunc {
+		tCtx.Log("Pod running consistently after upgrade.")
+		tCtx.Consistently(func(tCtx ktesting.TContext) error {
+			return e2epod.WaitForPodRunningInNamespace(tCtx, tCtx.Client(), pod)
+		}).WithTimeout(30 * time.Second).WithPolling(5 * time.Second).Should(gomega.Succeed())
+
+		tCtx.Logf("Evict pod through DeviceTaintRule.")
+		rule := &resourcealpha.DeviceTaintRule{
+			ObjectMeta: metav1.ObjectMeta{
+				Name: "device-taint-rule",
+			},
+			Spec: resourcealpha.DeviceTaintRuleSpec{
+				DeviceSelector: &resourcealpha.DeviceTaintSelector{
+					Driver: &b.Driver.Name,
+					Pool:   &slice.Spec.Pool.Name,
+					Device: &taintedDevice,
+				},
+				Taint: resourcealpha.DeviceTaint{
+					Key:    taintKey,
+					Value:  taintValueFromRule,
+					Effect: resourcealpha.DeviceTaintEffectNoExecute,
+				},
+			},
+		}
+		_, err := tCtx.Client().ResourceV1alpha3().DeviceTaintRules().Create(tCtx, rule, metav1.CreateOptions{})
+		tCtx.ExpectNoError(err)
+		tCtx.ExpectNoError(e2epod.WaitForPodNotFoundInNamespace(tCtx, tCtx.Client(), pod.Name, namespace, 5*time.Minute))
+
+		return func(tCtx ktesting.TContext) {
+			tCtx.Log("DeviceTaintRule still in effect.")
+			b.Create(tCtx, pod)
+			tCtx.ExpectNoError(e2epod.WaitForPodNameUnschedulableInNamespace(tCtx, tCtx.Client(), pod.Name, namespace))
+
+			// We must clean up manually, otherwise the code which checks for ResourceSlice deletion after
+			// driver removal gets stuck waiting for the removal of special ResourceSlice.
+			// This cannot be scheduled via tCtx.Cleanup after creating it because then it would be removed
+			// after the first sub-test.
+			tCtx.ExpectNoError(tCtx.Client().ResourceV1().ResourceSlices().Delete(tCtx, slice.Name, metav1.DeleteOptions{}))
+		}
+	}
+}
diff --git a/test/e2e_dra/upgradedowngrade_test.go b/test/e2e_dra/upgradedowngrade_test.go
index b53c06090f5..2abc6e77893 100644
--- a/test/e2e_dra/upgradedowngrade_test.go
+++ b/test/e2e_dra/upgradedowngrade_test.go
@@ -66,7 +66,8 @@ func init() {
 // sub-test. That function then returns the next piece of code, which then
 // returns the final code. Each callback function is executed as a sub-test.
 // The builder is configured to not delete objects when that sub-test ends,
-// so objects persist until the entire test is done.
+// so objects persist until the entire test is done. The same DRA driver
+// is used for all sub-tests.
 //
 // Each sub-test must be self-contained. They intentionally run in a random
 // order. However, they share the same cluster and the 8 devices which are
@@ -74,6 +75,7 @@ func init() {
 var subTests = map[string]initialTestFunc{
 	"core DRA":                    coreDRA,
 	"ResourceClaim device status": resourceClaimDeviceStatus,
+	"DeviceTaints":                deviceTaints,
 }
 
 type initialTestFunc func(tCtx ktesting.TContext, builder *drautils.Builder) upgradedTestFunc
@@ -210,8 +212,8 @@ func testUpgradeDowngrade(tCtx ktesting.TContext) {
 	tCtx.Step(fmt.Sprintf("bring up v%d.%d", major, previousMinor), func(tCtx ktesting.TContext) {
 		cluster = localupcluster.New(tCtx)
 		localUpClusterEnv := map[string]string{
-			"RUNTIME_CONFIG": "resource.k8s.io/v1beta1,resource.k8s.io/v1beta2",
-			"FEATURE_GATES":  "DynamicResourceAllocation=true",
+			"RUNTIME_CONFIG": "resource.k8s.io/v1beta1,resource.k8s.io/v1beta2,resource.k8s.io/v1alpha3",
+			"FEATURE_GATES":  "DynamicResourceAllocation=true,DRADeviceTaintRules=true,DRADeviceTaints=true",
 			// *not* needed because driver will run in "local filesystem" mode (= driver.IsLocal): "ALLOW_PRIVILEGED": "1",
 		}
 		cluster.Start(tCtx, binDir, localUpClusterEnv)
@@ -247,6 +249,7 @@ func testUpgradeDowngrade(tCtx ktesting.TContext) {
 			})
 		}
 	})
+	numSlices := len(driver.NewGetSlices()(tCtx).Items)
 
 	// We could split this up into first updating the apiserver, then control plane components, then restarting kubelet.
 	// For the purpose of this test here we we primarily care about full before/after comparisons, so not done yet.
@@ -255,7 +258,7 @@ func testUpgradeDowngrade(tCtx ktesting.TContext) {
 
 	// The kubelet wipes all ResourceSlices on a restart because it doesn't know which drivers were running.
 	// Wait for the ResourceSlice controller in the driver to notice and recreate the ResourceSlices.
-	tCtx.WithStep("wait for ResourceSlices").Eventually(driver.NewGetSlices()).WithTimeout(5 * time.Minute).Should(gomega.HaveField("Items", gomega.HaveLen(len(nodes.NodeNames))))
+	tCtx.WithStep("wait for ResourceSlices").Eventually(driver.NewGetSlices()).WithTimeout(5 * time.Minute).Should(gomega.HaveField("Items", gomega.HaveLen(numSlices)))
 
 	downgradedTestFuncs := make(map[string]downgradedTestFunc, len(subTests))
 	tCtx.Run("after-cluster-upgrade", func(tCtx ktesting.TContext) {
diff --git a/test/integration/dra/resourceclaim_test.go b/test/integration/dra/resourceclaim_test.go
index d57c60ba1ab..27a3097a849 100644
--- a/test/integration/dra/resourceclaim_test.go
+++ b/test/integration/dra/resourceclaim_test.go
@@ -57,7 +57,8 @@ func testShareResourceClaimSequentially(tCtx ktesting.TContext) {
 	b := drautils.NewBuilderNow(tCtx, driver)
 
 	var objects []klog.KMetadata
-	objects = append(objects, b.ExternalClaim())
+	claim := b.ExternalClaim()
+	objects = append(objects, claim)
 
 	// This test used to test usage of the claim by one pod
 	// at a time. After removing the "not sharable"
@@ -67,7 +68,7 @@ func testShareResourceClaimSequentially(tCtx ktesting.TContext) {
 	tCtx.Logf("Creating %d pods sharing the same claim", numMaxPods)
 	pods := make([]*v1.Pod, numMaxPods)
 	for i := range numMaxPods {
-		pod := b.PodExternal()
+		pod := b.PodExternal(claim.Name)
 		pods[i] = pod
 		objects = append(objects, pod)
 	}
@@ -130,7 +131,7 @@ func testShareResourceClaimSequentially(tCtx ktesting.TContext) {
 	morePods := make([]*v1.Pod, numMorePods)
 	objects = nil
 	for i := range numMorePods {
-		pod := b.PodExternal()
+		pod := b.PodExternal(claim.Name)
 		morePods[i] = pod
 		objects = append(objects, pod)
 	}