Merge pull request openshift#1078 from SchSeba/merge-bot-master

OCPBUGS-53346: Merge https://github.com/k8snetworkplumbingwg/sriov-network-operator:master into main

Author: openshift-merge-bot[bot]

Makefile

@@ -68,9 +68,6 @@ clean:
 	@rm -rf $(TARGET_DIR)
 	@rm -rf $(BIN_DIR)
 
-update-codegen:
-	hack/update-codegen.sh
-
 image: ; $(info Building images...)
 	$(IMAGE_BUILDER) build -f $(DOCKERFILE) -t $(IMAGE_TAG) $(CURPATH) $(IMAGE_BUILD_OPTS)
 	$(IMAGE_BUILDER) build -f $(DOCKERFILE_CONFIG_DAEMON) -t $(CONFIG_DAEMON_IMAGE_TAG) $(CURPATH) $(IMAGE_BUILD_OPTS)

bindata/manifests/daemon/daemonset.yaml

@@ -117,7 +117,15 @@ spec:
           command:
             - /bin/bash
             - -c
-            - mkdir -p /host/var/lib/sriov/ && cp /usr/bin/sriov-network-config-daemon /host/var/lib/sriov/sriov-network-config-daemon && chcon -t bin_t /host/var/lib/sriov/sriov-network-config-daemon | true # Allow systemd to run the file, use pipe true to not failed if the system doesn't have selinux or apparmor enabled
+            - |
+              set -e
+              if [ ! -f /host/usr/share/hwdata/pci.ids ]; then # If pci.ids file is missing on the host, config daemon won't be able to discover PCI devices
+                mkdir -p /host/usr/share/hwdata/
+                cp /usr/share/hwdata/pci.ids /host/usr/share/hwdata/pci.ids
+              fi
+              mkdir -p /host/var/lib/sriov/
+              cp /usr/bin/sriov-network-config-daemon /host/var/lib/sriov/sriov-network-config-daemon
+              chcon -t bin_t /host/var/lib/sriov/sriov-network-config-daemon || true # Allow systemd to run the file, use pipe true to not failed if the system doesn't have selinux or apparmor enabled
           securityContext:
             privileged: true
           resources:

bundle/manifests/sriov-network-operator.clusterserviceversion.yaml

@@ -100,7 +100,7 @@ metadata:
     categories: Networking
     certified: "false"
     containerImage: quay.io/openshift/origin-sriov-network-operator:4.19
-    createdAt: "2025-04-06T23:49:43Z"
+    createdAt: "2025-04-11T23:49:10Z"
     description: An operator for configuring SR-IOV components and initializing SRIOV
       network devices in Openshift cluster.
     features.operators.openshift.io/cnf: "false"

cmd/sriov-network-config-daemon/start.go

@@ -40,7 +40,6 @@ import (
 	"sigs.k8s.io/controller-runtime/pkg/metrics/server"
 
 	sriovnetworkv1 "github.com/k8snetworkplumbingwg/sriov-network-operator/api/v1"
-	snclientset "github.com/k8snetworkplumbingwg/sriov-network-operator/pkg/client/clientset/versioned"
 	"github.com/k8snetworkplumbingwg/sriov-network-operator/pkg/consts"
 	"github.com/k8snetworkplumbingwg/sriov-network-operator/pkg/daemon"
 	"github.com/k8snetworkplumbingwg/sriov-network-operator/pkg/featuregate"
@@ -243,7 +242,6 @@ func runStartCmd(cmd *cobra.Command, args []string) error {
 	}
 
 	// create clients
-	snclient := snclientset.NewForConfigOrDie(config)
 	kubeclient := kubernetes.NewForConfigOrDie(config)
 	kClient, err := runtimeclient.New(
 		config,
@@ -254,7 +252,7 @@ func runStartCmd(cmd *cobra.Command, args []string) error {
 		os.Exit(1)
 	}
 
-	eventRecorder := daemon.NewEventRecorder(snclient, kubeclient, scheme)
+	eventRecorder := daemon.NewEventRecorder(kClient, kubeclient, scheme)
 	defer eventRecorder.Shutdown()
 
 	nodeInfo, err := kubeclient.CoreV1().Nodes().Get(context.Background(), vars.NodeName, v1.GetOptions{})

cmd/sriov-network-operator-config-cleanup/cleanup.go

@@ -4,17 +4,19 @@ import (
 	"context"
 	"time"
 
+	ocpconfigapi "github.com/openshift/api/config/v1"
 	"github.com/spf13/cobra"
+	"k8s.io/apimachinery/pkg/api/errors"
+	"k8s.io/apimachinery/pkg/runtime"
+	utilruntime "k8s.io/apimachinery/pkg/util/runtime"
+	clientgoscheme "k8s.io/client-go/kubernetes/scheme"
 	ctrl "sigs.k8s.io/controller-runtime"
+	"sigs.k8s.io/controller-runtime/pkg/client"
 	"sigs.k8s.io/controller-runtime/pkg/log"
 
+	sriovnetworkv1 "github.com/k8snetworkplumbingwg/sriov-network-operator/api/v1"
 	snolog "github.com/k8snetworkplumbingwg/sriov-network-operator/pkg/log"
-
-	"k8s.io/apimachinery/pkg/api/errors"
-	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
-	"k8s.io/apimachinery/pkg/watch"
-
-	sriovnetworkv1 "github.com/k8snetworkplumbingwg/sriov-network-operator/pkg/client/clientset/versioned/typed/sriovnetwork/v1"
+	"github.com/k8snetworkplumbingwg/sriov-network-operator/pkg/vars"
 )
 
 var (
@@ -25,6 +27,14 @@ var (
 func init() {
 	rootCmd.Flags().StringVarP(&namespace, "namespace", "n", "", "designated SriovOperatorConfig namespace")
 	rootCmd.Flags().IntVarP(&watchTO, "watch-timeout", "w", 10, "sriov-operator config post-delete watch timeout ")
+
+	// Init Scheme
+	newScheme := runtime.NewScheme()
+	utilruntime.Must(clientgoscheme.AddToScheme(newScheme))
+	utilruntime.Must(sriovnetworkv1.AddToScheme(newScheme))
+	utilruntime.Must(ocpconfigapi.AddToScheme(newScheme))
+
+	vars.Scheme = newScheme
 }
 
 func runCleanupCmd(cmd *cobra.Command, args []string) error {
@@ -38,46 +48,42 @@ func runCleanupCmd(cmd *cobra.Command, args []string) error {
 	defer timeoutFunc()
 
 	restConfig := ctrl.GetConfigOrDie()
-	sriovcs, err := sriovnetworkv1.NewForConfig(restConfig)
+	c, err := client.New(restConfig, client.Options{Scheme: vars.Scheme})
 	if err != nil {
 		setupLog.Error(err, "failed to create 'sriovnetworkv1' clientset")
 	}
 
-	err = sriovcs.SriovOperatorConfigs(namespace).Delete(context.Background(), "default", metav1.DeleteOptions{})
+	operatorConfig := &sriovnetworkv1.SriovOperatorConfig{}
+	err = c.Get(ctx, client.ObjectKey{Namespace: namespace, Name: "default"}, operatorConfig)
 	if err != nil {
 		if errors.IsNotFound(err) {
 			return nil
 		}
-		setupLog.Error(err, "failed to delete SriovOperatorConfig")
+		setupLog.Error(err, "failed to get SriovOperatorConfig")
 		return err
 	}
 
-	// watching 'default' config deletion with context timeout, in case sriov-operator fails to delete 'default' config
-	watcher, err := sriovcs.SriovOperatorConfigs(namespace).Watch(ctx, metav1.ListOptions{Watch: true})
+	err = c.Delete(ctx, operatorConfig)
 	if err != nil {
-		setupLog.Error(err, "failed creating 'default' SriovOperatorConfig object watcher")
+		if errors.IsNotFound(err) {
+			return nil
+		}
+		setupLog.Error(err, "failed to delete SriovOperatorConfig")
 		return err
 	}
-	defer watcher.Stop()
-	for {
-		select {
-		case event := <-watcher.ResultChan():
-			if event.Type == watch.Deleted {
-				setupLog.Info("'default' SriovOperatorConfig is deleted")
-				return nil
-			}
 
-		case <-ctx.Done():
-			// check whether object might has been deleted before watch event triggered
-			_, err := sriovcs.SriovOperatorConfigs(namespace).Get(context.Background(), "default", metav1.GetOptions{})
-			if err != nil {
-				if errors.IsNotFound(err) {
-					return nil
-				}
+	for {
+		err = c.Get(ctx, client.ObjectKey{Namespace: namespace, Name: "default"}, operatorConfig)
+		if err != nil {
+			if errors.IsNotFound(err) {
+				break
 			}
-			err = ctx.Err()
-			setupLog.Error(err, "timeout has occurred for 'default' SriovOperatorConfig deletion")
+			setupLog.Error(err, "failed to check sriovOperatorConfig exist")
 			return err
 		}
+		time.Sleep(100 * time.Millisecond)
 	}
+
+	setupLog.Info("'default' SriovOperatorConfig is deleted")
+	return nil
 }

controllers/drain_controller.go

@@ -21,6 +21,7 @@ import (
 	"fmt"
 	"sync"
 
+	"github.com/go-logr/logr"
 	corev1 "k8s.io/api/core/v1"
 	"k8s.io/apimachinery/pkg/api/errors"
 	"k8s.io/apimachinery/pkg/runtime"
@@ -76,8 +77,8 @@ func NewDrainReconcileController(client client.Client, Scheme *runtime.Scheme, r
 // For more details, check Reconcile and its Result here:
 // - https://pkg.go.dev/sigs.k8s.io/[email protected]/pkg/reconcile
 func (dr *DrainReconcile) Reconcile(ctx context.Context, req ctrl.Request) (ctrl.Result, error) {
-	reqLogger := log.FromContext(ctx)
-	reqLogger.Info("Reconciling Drain")
+	ctx = context.WithValue(ctx, "logger", log.FromContext(ctx))
+	reqLogger := log.FromContext(ctx).WithName("Drain Reconcile")
 
 	req.Namespace = vars.Namespace
 
@@ -149,14 +150,14 @@ func (dr *DrainReconcile) Reconcile(ctx context.Context, req ctrl.Request) (ctrl
 		//  doesn't need to drain anymore, so we can stop the drain
 		if nodeStateDrainAnnotationCurrent == constants.DrainComplete ||
 			nodeStateDrainAnnotationCurrent == constants.Draining {
-			return dr.handleNodeIdleNodeStateDrainingOrCompleted(ctx, &reqLogger, node, nodeNetworkState)
+			return dr.handleNodeIdleNodeStateDrainingOrCompleted(ctx, node, nodeNetworkState)
 		}
 	}
 
 	// this cover the case a node request to drain or reboot
 	if nodeDrainAnnotation == constants.DrainRequired ||
 		nodeDrainAnnotation == constants.RebootRequired {
-		return dr.handleNodeDrainOrReboot(ctx, &reqLogger, node, nodeNetworkState, nodeDrainAnnotation, nodeStateDrainAnnotationCurrent)
+		return dr.handleNodeDrainOrReboot(ctx, node, nodeNetworkState, nodeDrainAnnotation, nodeStateDrainAnnotationCurrent)
 	}
 
 	reqLogger.Error(nil, "unexpected node drain annotation")
@@ -209,7 +210,17 @@ func (dr *DrainReconcile) SetupWithManager(mgr ctrl.Manager) error {
 	nodeStatePredicates := builder.WithPredicates(DrainStateAnnotationPredicate{})
 
 	return ctrl.NewControllerManagedBy(mgr).
-		WithOptions(controller.Options{MaxConcurrentReconciles: 50}).
+		WithOptions(controller.Options{
+			MaxConcurrentReconciles: 50,
+			LogConstructor: func(request *reconcile.Request) logr.Logger {
+				logger := mgr.GetLogger().WithValues("Function", "Drain")
+				// Inspired by https://github.com/kubernetes-sigs/controller-runtime/blob/52b17917caa97ec546423867d9637f1787830f3e/pkg/builder/controller.go#L447
+				if req, ok := any(request).(*reconcile.Request); ok && req != nil {
+					logger = logger.WithValues("node", request.Name)
+				}
+				return logger
+			},
+		}).
 		For(&corev1.Node{}, nodePredicates).
 		Watches(&sriovnetworkv1.SriovNetworkNodeState{}, createUpdateEnqueue, nodeStatePredicates).
 		Complete(dr)

controllers/drain_controller_helper.go

@@ -11,7 +11,6 @@ import (
 	"k8s.io/apimachinery/pkg/labels"
 	ctrl "sigs.k8s.io/controller-runtime"
 	"sigs.k8s.io/controller-runtime/pkg/client"
-	"sigs.k8s.io/controller-runtime/pkg/log"
 	"sigs.k8s.io/controller-runtime/pkg/reconcile"
 
 	sriovnetworkv1 "github.com/k8snetworkplumbingwg/sriov-network-operator/api/v1"
@@ -21,9 +20,9 @@ import (
 )
 
 func (dr *DrainReconcile) handleNodeIdleNodeStateDrainingOrCompleted(ctx context.Context,
-	reqLogger *logr.Logger,
 	node *corev1.Node,
 	nodeNetworkState *sriovnetworkv1.SriovNetworkNodeState) (ctrl.Result, error) {
+	reqLogger := ctx.Value("logger").(logr.Logger).WithName("handleNodeIdleNodeStateDrainingOrCompleted")
 	completed, err := dr.drainer.CompleteDrainNode(ctx, node)
 	if err != nil {
 		reqLogger.Error(err, "failed to complete drain on node")
@@ -61,11 +60,11 @@ func (dr *DrainReconcile) handleNodeIdleNodeStateDrainingOrCompleted(ctx context
 }
 
 func (dr *DrainReconcile) handleNodeDrainOrReboot(ctx context.Context,
-	reqLogger *logr.Logger,
 	node *corev1.Node,
 	nodeNetworkState *sriovnetworkv1.SriovNetworkNodeState,
 	nodeDrainAnnotation,
 	nodeStateDrainAnnotationCurrent string) (ctrl.Result, error) {
+	reqLogger := ctx.Value("logger").(logr.Logger).WithName("handleNodeDrainOrReboot")
 	// nothing to do here we need to wait for the node to move back to idle
 	if nodeStateDrainAnnotationCurrent == constants.DrainComplete {
 		reqLogger.Info("node requested a drain and nodeState is on drain completed nothing todo")
@@ -139,9 +138,7 @@ func (dr *DrainReconcile) handleNodeDrainOrReboot(ctx context.Context,
 }
 
 func (dr *DrainReconcile) tryDrainNode(ctx context.Context, node *corev1.Node) (*reconcile.Result, error) {
-	// configure logs
-	reqLogger := log.FromContext(ctx)
-	reqLogger.Info("checkForNodeDrain():")
+	reqLogger := ctx.Value("logger").(logr.Logger).WithName("tryDrainNode")
 
 	//critical section we need to check if we can start the draining
 	dr.drainCheckMutex.Lock()
@@ -169,7 +166,7 @@ func (dr *DrainReconcile) tryDrainNode(ctx context.Context, node *corev1.Node) (
 		err = dr.Get(ctx, client.ObjectKey{Name: nodeObj.GetName(), Namespace: vars.Namespace}, snns)
 		if err != nil {
 			if errors.IsNotFound(err) {
-				reqLogger.V(2).Info("node doesn't have a sriovNetworkNodePolicy")
+				reqLogger.V(2).Info("node doesn't have a sriovNetworkNodeState, skipping")
 				continue
 			}
 			return nil, err
@@ -211,8 +208,7 @@ func (dr *DrainReconcile) tryDrainNode(ctx context.Context, node *corev1.Node) (
 }
 
 func (dr *DrainReconcile) findNodePoolConfig(ctx context.Context, node *corev1.Node) (*sriovnetworkv1.SriovNetworkPoolConfig, []corev1.Node, error) {
-	logger := log.FromContext(ctx)
-	logger.Info("findNodePoolConfig():")
+	logger := ctx.Value("logger").(logr.Logger).WithName("findNodePoolConfig")
 	// get all the sriov network pool configs
 	npcl := &sriovnetworkv1.SriovNetworkPoolConfigList{}
 	err := dr.List(ctx, npcl)

controllers/sriovoperatorconfig_controller.go

@@ -85,7 +85,7 @@ func (r *SriovOperatorConfigReconciler) Reconcile(ctx context.Context, req ctrl.
 	if err != nil {
 		if apierrors.IsNotFound(err) {
 			logger.Info("default SriovOperatorConfig object not found. waiting for creation.")
-			return reconcile.Result{}, err
+			return reconcile.Result{}, nil
 		}
 		// Error reading the object - requeue the request.
 		logger.Error(err, "Failed to get default SriovOperatorConfig object")
@@ -99,12 +99,9 @@ func (r *SriovOperatorConfigReconciler) Reconcile(ctx context.Context, req ctrl.
 		// The object is being deleted
 		return r.handleSriovOperatorConfigDeletion(ctx, defaultConfig, logger)
 	}
-	// add finalizer if needed
-	if !sriovnetworkv1.StringInArray(sriovnetworkv1.OPERATORCONFIGFINALIZERNAME, defaultConfig.ObjectMeta.Finalizers) {
-		defaultConfig.ObjectMeta.Finalizers = append(defaultConfig.ObjectMeta.Finalizers, sriovnetworkv1.OPERATORCONFIGFINALIZERNAME)
-		if err := r.Update(ctx, defaultConfig); err != nil {
-			return reconcile.Result{}, err
-		}
+
+	if err = r.syncOperatorConfigFinalizers(ctx, defaultConfig, logger); err != nil {
+		return reconcile.Result{}, err
 	}
 
 	r.FeatureGate.Init(defaultConfig.Spec.FeatureGates)
@@ -448,6 +445,29 @@ func (r *SriovOperatorConfigReconciler) syncOpenShiftSystemdService(ctx context.
 	return r.setLabelInsideObject(ctx, cr, objs)
 }
 
+func (r SriovOperatorConfigReconciler) syncOperatorConfigFinalizers(ctx context.Context, defaultConfig *sriovnetworkv1.SriovOperatorConfig, logger logr.Logger) error {
+	if sriovnetworkv1.StringInArray(sriovnetworkv1.OPERATORCONFIGFINALIZERNAME, defaultConfig.ObjectMeta.Finalizers) {
+		return nil
+	}
+
+	newObj := defaultConfig.DeepCopyObject().(client.Object)
+	newObj.SetFinalizers(
+		append(newObj.GetFinalizers(), sriovnetworkv1.OPERATORCONFIGFINALIZERNAME),
+	)
+
+	logger.WithName("syncOperatorConfigFinalizers").
+		Info("Adding finalizer", "key", sriovnetworkv1.OPERATORCONFIGFINALIZERNAME)
+
+	patch := client.MergeFrom(defaultConfig)
+	err := r.Patch(ctx, newObj, patch)
+	if err != nil {
+		return fmt.Errorf("can't patch SriovOperatorConfig to add finalizer [%s]: %w", sriovnetworkv1.OPERATORCONFIGFINALIZERNAME, err)
+	}
+
+	// Refresh the defaultConfig object with the latest changes
+	return r.Get(ctx, types.NamespacedName{Namespace: defaultConfig.Namespace, Name: defaultConfig.Name}, defaultConfig)
+}
+
 func (r *SriovOperatorConfigReconciler) handleSriovOperatorConfigDeletion(ctx context.Context,
 	defaultConfig *sriovnetworkv1.SriovOperatorConfig, logger logr.Logger) (ctrl.Result, error) {
 	var err error