DRA: add device taint eviction controller

The controller is derived from the node taint eviction controller.
In contrast to that controller it tracks the UID of pods to prevent
deleting the wrong pod when it got replaced.

Author: pohly

cmd/kube-controller-manager/app/controllermanager.go

@@ -588,6 +588,7 @@ func NewControllerDescriptors() map[string]*ControllerDescriptor {
 	// feature gated
 	register(newStorageVersionGarbageCollectorControllerDescriptor())
 	register(newResourceClaimControllerDescriptor())
+	register(newDeviceTaintEvictionControllerDescriptor())
 	register(newLegacyServiceAccountTokenCleanerControllerDescriptor())
 	register(newValidatingAdmissionPolicyStatusControllerDescriptor())
 	register(newTaintEvictionControllerDescriptor())

cmd/kube-controller-manager/app/controllermanager_test.go

@@ -93,6 +93,7 @@ func TestControllerNamesDeclaration(t *testing.T) {
 		names.EphemeralVolumeController,
 		names.StorageVersionGarbageCollectorController,
 		names.ResourceClaimController,
+		names.DeviceTaintEvictionController,
 		names.LegacyServiceAccountTokenCleanerController,
 		names.ValidatingAdmissionPolicyStatusController,
 		names.ServiceCIDRController,

cmd/kube-controller-manager/app/core.go

@@ -41,6 +41,7 @@ import (
 	"k8s.io/klog/v2"
 	"k8s.io/kubernetes/cmd/kube-controller-manager/names"
 	pkgcontroller "k8s.io/kubernetes/pkg/controller"
+	"k8s.io/kubernetes/pkg/controller/devicetainteviction"
 	endpointcontroller "k8s.io/kubernetes/pkg/controller/endpoint"
 	"k8s.io/kubernetes/pkg/controller/garbagecollector"
 	namespacecontroller "k8s.io/kubernetes/pkg/controller/namespace"
@@ -231,6 +232,32 @@ func startTaintEvictionController(ctx context.Context, controllerContext Control
 	return nil, true, nil
 }
 
+func newDeviceTaintEvictionControllerDescriptor() *ControllerDescriptor {
+	return &ControllerDescriptor{
+		name:     names.DeviceTaintEvictionController,
+		initFunc: startDeviceTaintEvictionController,
+		requiredFeatureGates: []featuregate.Feature{
+			// TODO update app.TestFeatureGatedControllersShouldNotDefineAliases when removing these feature gates.
+			features.DynamicResourceAllocation,
+			features.DRADeviceTaints,
+		},
+	}
+}
+
+func startDeviceTaintEvictionController(ctx context.Context, controllerContext ControllerContext, controllerName string) (controller.Interface, bool, error) {
+	deviceTaintEvictionController := devicetainteviction.New(
+		controllerContext.ClientBuilder.ClientOrDie(names.DeviceTaintEvictionController),
+		controllerContext.InformerFactory.Core().V1().Pods(),
+		controllerContext.InformerFactory.Resource().V1beta1().ResourceClaims(),
+		controllerContext.InformerFactory.Resource().V1beta1().ResourceSlices(),
+		controllerContext.InformerFactory.Resource().V1alpha3().DeviceTaintRules(),
+		controllerContext.InformerFactory.Resource().V1beta1().DeviceClasses(),
+		controllerName,
+	)
+	go deviceTaintEvictionController.Run(ctx)
+	return nil, true, nil
+}
+
 func newCloudNodeLifecycleControllerDescriptor() *ControllerDescriptor {
 	return &ControllerDescriptor{
 		name:                      cpnames.CloudNodeLifecycleController,

cmd/kube-controller-manager/names/controller_names.go

@@ -69,6 +69,7 @@ const (
 	NodeIpamController                                 = "node-ipam-controller"
 	NodeLifecycleController                            = "node-lifecycle-controller"
 	TaintEvictionController                            = "taint-eviction-controller"
+	DeviceTaintEvictionController                      = "device-taint-eviction-controller"
 	PersistentVolumeBinderController                   = "persistentvolume-binder-controller"
 	PersistentVolumeAttachDetachController             = "persistentvolume-attach-detach-controller"
 	PersistentVolumeExpanderController                 = "persistentvolume-expander-controller"

hack/verify-prometheus-imports.sh

@@ -37,6 +37,7 @@ source "${KUBE_ROOT}/hack/lib/util.sh"
 # See: https://github.com/kubernetes/kubernetes/issues/89267
 allowed_prometheus_importers=(
   ./cluster/images/etcd-version-monitor/etcd-version-monitor.go
+  ./pkg/controller/devicetainteviction/device_taint_eviction_test.go
   ./staging/src/k8s.io/component-base/metrics/prometheusextension/timing_histogram.go
   ./staging/src/k8s.io/component-base/metrics/prometheusextension/timing_histogram_test.go
   ./staging/src/k8s.io/component-base/metrics/prometheusextension/timing_histogram_vec.go