Merge pull request kosmos-io#614 from OrangeBao/release-0.4.0-cert

fix: node label lost
OrangeBao · Jun 3, 2024 · 902adfb · 902adfb
2 parents ffd32f7 + 666cd87
commit 902adfb
Show file tree

Hide file tree

Showing 7 changed files with 36 additions and 13 deletions.
diff --git a/cmd/kubenest/operator/app/options/options.go b/cmd/kubenest/operator/app/options/options.go
@@ -58,5 +58,5 @@ func (o *Options) AddFlags(flags *pflag.FlagSet) {
 	flags.BoolVar(&o.KubeNestOptions.ForceDestroy, "kube-nest-force-destroy", false, "Force destroy the node.If it set true.If set to true, Kubernetes will not evict the existing nodes on the node when joining nodes to the tenant's control plane, but will instead force destroy.")
 	flags.StringVar(&o.KubeNestOptions.AnpMode, "kube-nest-anp-mode", "tcp", "kube-apiserver network proxy mode, must be set to tcp or uds. uds mode the replicas for apiserver should be one, and tcp for multi apiserver replicas.")
 	flags.BoolVar(&o.KubeNestOptions.AdmissionPlugins, "kube-nest-admission-plugins", false, "kube-apiserver network disable-admission-plugins, false for - --disable-admission-plugins=License, true for remove the --disable-admission-plugins=License flag .")
-	flags.IntVar(&o.KubeNestOptions.ApiServerReplicas, "kube-nest-apiserver-replicas", 2, "virtual-cluster kube-apiserver replicas. default is 2.")
+	flags.IntVar(&o.KubeNestOptions.ApiServerReplicas, "kube-nest-apiserver-replicas", 1, "virtual-cluster kube-apiserver replicas. default is 2.")
 }
diff --git a/pkg/kubenest/controller/global.node.controller/global_node_controller.go b/pkg/kubenest/controller/global.node.controller/global_node_controller.go
@@ -118,7 +118,7 @@ func (r *GlobalNodeController) SyncTaint(ctx context.Context, globalNode *v1alph
 		err := retry.RetryOnConflict(retry.DefaultRetry, func() error {
 			var targetNode v1.Node
 			if err := r.Get(ctx, types.NamespacedName{Name: globalNode.Name}, &targetNode); err != nil {
-				klog.Errorf("global-node-controller: SyncTaints: can not get global node, err: %s", globalNode.Name)
+				klog.Errorf("global-node-controller: SyncTaints: can not get host node, err: %s", globalNode.Name)
 				return err
 			}
 
@@ -175,15 +175,19 @@ func (r *GlobalNodeController) SyncState(ctx context.Context, globalNode *v1alph
 }
 
 func (r *GlobalNodeController) SyncLabel(ctx context.Context, globalNode *v1alpha1.GlobalNode) error {
+	if globalNode.Spec.State == v1alpha1.NodeInUse {
+		klog.V(4).Infof("global-node-controller: SyncLabel: node is in use %s, skip", globalNode.Name)
+		return nil
+	}
 	err := retry.RetryOnConflict(retry.DefaultRetry, func() error {
 		rootNode, err := r.RootClientSet.CoreV1().Nodes().Get(ctx, globalNode.Name, metav1.GetOptions{})
 		if err != nil {
-			klog.Errorf("global-node-controller: SyncState: can not get root node: %s", globalNode.Name)
+			klog.Errorf("global-node-controller: SyncLabel: can not get root node: %s", globalNode.Name)
 			return err
 		}
 
 		if _, err = r.KosmosClient.KosmosV1alpha1().GlobalNodes().Get(ctx, globalNode.Name, metav1.GetOptions{}); err != nil {
-			klog.Errorf("global-node-controller: SyncState: can not get global node: %s", globalNode.Name)
+			klog.Errorf("global-node-controller: SyncLabel: can not get global node: %s", globalNode.Name)
 			return err
 		}
 
@@ -195,7 +199,7 @@ func (r *GlobalNodeController) SyncLabel(ctx context.Context, globalNode *v1alph
 		updateGlobalNode.Spec.Labels = rootNode.Labels
 
 		if _, err = r.KosmosClient.KosmosV1alpha1().GlobalNodes().Update(ctx, updateGlobalNode, metav1.UpdateOptions{}); err != nil {
-			klog.Errorf("global-node-controller: SyncState: update global node label failed, err: %s", err)
+			klog.Errorf("global-node-controller: SyncLabel: update global node label failed, err: %s", err)
 			return err
 		}
 		return nil
@@ -251,6 +255,10 @@ func (r *GlobalNodeController) Reconcile(ctx context.Context, request reconcile.
 		klog.Errorf("can not get root node: %s", globalNode.Name)
 		return reconcile.Result{RequeueAfter: utils.DefaultRequeueTime}, nil
 	}
+	if globalNode.Spec.State == v1alpha1.NodeInUse {
+		// wait globalNode free
+		return reconcile.Result{RequeueAfter: utils.DefaultRequeueTime}, nil
+	}
 
 	if err := r.SyncLabel(ctx, &globalNode); err != nil {
 		klog.Warningf("sync label %s error: %v", request.NamespacedName, err)

diff --git a/pkg/kubenest/controller/virtualcluster.node.controller/env/env.go b/pkg/kubenest/controller/virtualcluster.node.controller/env/env.go
@@ -123,3 +123,15 @@ func GetControlPlaneLabel() string {
 	}
 	return controllPlaneLabel
 }
+
+func GetWaitNodeReadTime() int {
+	readTimeSeconds := os.Getenv("WAIT_NODE_READ_TIME")
+	if len(readTimeSeconds) == 0 {
+		readTimeSeconds = "30"
+	}
+	num, err := strconv.Atoi(readTimeSeconds)
+	if err != nil {
+		klog.Fatalf("convert WAIT_NODE_READ_TIME failed, err: %s", err)
+	}
+	return num
+}
diff --git a/pkg/kubenest/controller/virtualcluster.node.controller/workflow/task/task.go b/pkg/kubenest/controller/virtualcluster.node.controller/workflow/task/task.go
@@ -271,8 +271,8 @@ func NewWaitNodeReadyTask(isHost bool) Task {
 		Run: func(ctx context.Context, to TaskOpt, _ interface{}) (interface{}, error) {
 			isReady := false
 
-			waitFunc := func() {
-				waitCtx, cancel := context.WithTimeout(ctx, 30*time.Second) // total waiting time
+			waitFunc := func(timeout time.Duration) {
+				waitCtx, cancel := context.WithTimeout(ctx, timeout) // total waiting time
 				defer cancel()
 				wait.UntilWithContext(waitCtx, func(ctx context.Context) {
 					client := to.VirtualK8sClient
@@ -295,7 +295,7 @@ func NewWaitNodeReadyTask(isHost bool) Task {
 				}, 10*time.Second) // Interval time
 			}
 
-			waitFunc()
+			waitFunc(time.Duration(env.GetWaitNodeReadTime()) * time.Second)
 
 			if isReady {
 				return nil, nil
@@ -322,7 +322,7 @@ func NewWaitNodeReadyTask(isHost bool) Task {
 			}
 
 			klog.V(4).Infof("wait for the node to be ready again. %s", to.NodeInfo.Name)
-			waitFunc()
+			waitFunc(time.Duration(env.GetWaitNodeReadTime()*2) * time.Second)
 
 			if isReady {
 				return nil, nil

diff --git a/pkg/kubenest/controller/virtualcluster.node.controller/workflow/workflow.go b/pkg/kubenest/controller/virtualcluster.node.controller/workflow/workflow.go
@@ -2,6 +2,7 @@ package workflow
 
 import (
 	"context"
+	"time"
 
 	"k8s.io/klog/v2"
 
@@ -11,7 +12,7 @@ import (
 
 const (
 	retryCount = 0
-	maxRetries = 3
+	maxRetries = 5
 )
 
 type WorkflowData struct {
@@ -27,7 +28,9 @@ func RunWithRetry(ctx context.Context, task task.Task, opt task.TaskOpt, preArgs
 			if !task.Retry {
 				break
 			}
-			klog.V(4).Infof("work flow retry %d, task name: %s, err: %s", i, task.Name, err)
+			waitTime := 3 * (i + 1)
+			klog.V(4).Infof("work flow retry %d after %ds, task name: %s, err: %s", i, waitTime, task.Name, err)
+			time.Sleep(time.Duration(waitTime) * time.Second)
 		} else {
 			break
 		}

diff --git a/pkg/kubenest/tasks/coredns.go b/pkg/kubenest/tasks/coredns.go
@@ -280,7 +280,7 @@ func applyYMLTemplate(dynamicClient dynamic.Interface, manifestGlob string, temp
 		}
 
 		err = apiclient.TryRunCommand(func() error {
-			return util.CreateObject(dynamicClient, obj.GetNamespace(), obj.GetName(), &obj)
+			return util.ApplyObject(dynamicClient, &obj)
 		}, 3)
 		if err != nil {
 			return errors.Wrapf(err, "Create object error")

diff --git a/pkg/kubenest/tasks/manifests_components.go b/pkg/kubenest/tasks/manifests_components.go
@@ -163,7 +163,7 @@ func applyTemplatedManifests(component string, dynamicClient dynamic.Interface,
 			obj = unstructured.Unstructured{Object: res}
 		}
 		err = apiclient.TryRunCommand(func() error {
-			return util.CreateObject(dynamicClient, obj.GetNamespace(), obj.GetName(), &obj)
+			return util.ApplyObject(dynamicClient, &obj)
 		}, 3)
 		if err != nil {
 			return errors.Wrapf(err, "Create object error")