fix bugs (#399)

vesoft-inc · Nov 14, 2023 · 17ace23 · 17ace23
1 parent 8011afa
commit 17ace23
Show file tree

Hide file tree

Showing 14 changed files with 39 additions and 32 deletions.
diff --git a/apis/apps/v1alpha1/nebulacluster_common.go b/apis/apps/v1alpha1/nebulacluster_common.go
@@ -312,24 +312,25 @@ func generateAgentContainer(c NebulaClusterComponent, init bool) corev1.Containe
 		}
 	}
 
-	agentImage := defaultAgentImage
+	container := corev1.Container{
+		Name:    AgentSidecarContainerName,
+		Image:   defaultAgentImage,
+		Command: cmd,
+	}
+	imagePullPolicy := nc.Spec.ImagePullPolicy
+	if imagePullPolicy != nil {
+		container.ImagePullPolicy = *imagePullPolicy
+	}
 	if nc.Spec.Agent != nil {
+		var agentImage string
 		if nc.Spec.Agent.Image != "" {
 			agentImage = nc.Spec.Agent.Image
 		}
 		if nc.Spec.Agent.Version != "" {
 			agentImage = fmt.Sprintf("%s:%s", agentImage, nc.Spec.Agent.Version)
 		}
-	}
-	container := corev1.Container{
-		Name:      AgentSidecarContainerName,
-		Image:     agentImage,
-		Command:   cmd,
-		Resources: nc.Spec.Agent.Resources,
-	}
-	imagePullPolicy := nc.Spec.ImagePullPolicy
-	if imagePullPolicy != nil {
-		container.ImagePullPolicy = *imagePullPolicy
+		container.Image = agentImage
+		container.Resources = nc.Spec.Agent.Resources
 	}
 
 	if nc.IsBREnabled() {

diff --git a/apis/apps/v1alpha1/nebulacluster_types.go b/apis/apps/v1alpha1/nebulacluster_types.go
@@ -100,6 +100,7 @@ type NebulaClusterSpec struct {
 	// +optional
 	EnableAutoFailover *bool `json:"enableAutoFailover,omitempty"`
 
+	// +kubebuilder:default="5m"
 	// +optional
 	FailoverPeriod metav1.Duration `json:"failoverPeriod,omitempty"`
 

diff --git a/apis/autoscaling/v1alpha1/nebulaautoscaler_types.go b/apis/autoscaling/v1alpha1/nebulaautoscaler_types.go
@@ -10,7 +10,7 @@ import (
 
 var (
 	// DefaultPollingPeriod is the default period between 2 autoscaling API polls.
-	DefaultPollingPeriod = 15 * time.Second
+	DefaultPollingPeriod = 30 * time.Second
 )
 
 type NebulaAutoscalerConditionType string

diff --git a/charts/nebula-operator/crds/nebulaclusters.yaml b/charts/nebula-operator/crds/nebulaclusters.yaml
@@ -3187,6 +3187,7 @@ spec:
                 - image
                 type: object
               failoverPeriod:
+                default: 5m
                 type: string
               graphd:
                 properties:

diff --git a/config/crd/bases/apps.nebula-graph.io_nebulaclusters.yaml b/config/crd/bases/apps.nebula-graph.io_nebulaclusters.yaml
@@ -3187,6 +3187,7 @@ spec:
                 - image
                 type: object
               failoverPeriod:
+                default: 5m
                 type: string
               graphd:
                 properties:

diff --git a/pkg/controller/component/graphd_cluster.go b/pkg/controller/component/graphd_cluster.go
@@ -113,7 +113,7 @@ func (c *graphdCluster) syncGraphdWorkload(nc *v1alpha1.NebulaCluster) error {
 	if !notExist {
 		// TODO: validate the timestamp format
 		timestamp, ok := oldWorkload.GetAnnotations()[annotation.AnnRestartTimestamp]
-		if ok {
+		if ok && timestamp != "" {
 			if err := extender.SetTemplateAnnotations(newWorkload,
 				map[string]string{annotation.AnnRestartTimestamp: timestamp}); err != nil {
 				return err

diff --git a/pkg/controller/component/metad_cluster.go b/pkg/controller/component/metad_cluster.go
@@ -120,7 +120,7 @@ func (c *metadCluster) syncMetadWorkload(nc *v1alpha1.NebulaCluster) error {
 
 	if !notExist {
 		timestamp, ok := oldWorkload.GetAnnotations()[annotation.AnnRestartTimestamp]
-		if ok {
+		if ok && timestamp != "" {
 			if err := extender.SetTemplateAnnotations(newWorkload,
 				map[string]string{annotation.AnnRestartTimestamp: timestamp}); err != nil {
 				return err

diff --git a/pkg/controller/component/storaged_cluster.go b/pkg/controller/component/storaged_cluster.go
@@ -133,7 +133,7 @@ func (c *storagedCluster) syncStoragedWorkload(nc *v1alpha1.NebulaCluster) error
 
 	if !notExist {
 		timestamp, ok := oldWorkload.GetAnnotations()[annotation.AnnRestartTimestamp]
-		if ok {
+		if ok && timestamp != "" {
 			if err := extender.SetTemplateAnnotations(newWorkload,
 				map[string]string{annotation.AnnRestartTimestamp: timestamp}); err != nil {
 				return err
@@ -288,6 +288,10 @@ func (c *storagedCluster) syncNebulaClusterStatus(
 		_ = metaClient.Disconnect()
 	}()
 
+	if !nc.IsAutoFailoverEnabled() {
+		return syncComponentStatus(nc.StoragedComponent(), &nc.Status.Storaged.ComponentStatus, oldWorkload)
+	}
+
 	hostItems, err := metaClient.ListHosts(meta.ListHostType_STORAGE)
 	if err != nil {
 		return err

diff --git a/pkg/controller/component/storaged_failover.go b/pkg/controller/component/storaged_failover.go
@@ -41,7 +41,7 @@ import (
 
 const (
 	PVCProtectionFinalizer = "kubernetes.io/pvc-protection"
-	RestartTolerancePeriod = time.Minute * 1
+	RestartTolerancePeriod = time.Minute * 2
 )
 
 var (

diff --git a/pkg/controller/nebulacluster/nebula_cluster_condition_updater.go b/pkg/controller/nebulacluster/nebula_cluster_condition_updater.go
@@ -63,15 +63,15 @@ func (u *nebulaClusterConditionUpdater) updateReadyCondition(nc *v1alpha1.Nebula
 	case !allWorkloadsAreUpToDate(nc):
 		reason = condition.WorkloadNotUpToDate
 		message = "Workload is in progress"
-	case !nc.GraphdComponent().IsReady():
-		reason = condition.GraphdUnhealthy
-		message = "Graphd is not healthy"
 	case !nc.MetadComponent().IsReady():
 		reason = condition.MetadUnhealthy
 		message = "Metad is not healthy"
 	case !nc.IsStoragedAvailable():
 		reason = condition.StoragedUnhealthy
 		message = "Storaged is not healthy"
+	case !nc.GraphdComponent().IsReady():
+		reason = condition.GraphdUnhealthy
+		message = "Graphd is not healthy"
 	default:
 		status = corev1.ConditionTrue
 		reason = condition.WorkloadReady

diff --git a/pkg/controller/nebulacluster/nebula_cluster_control.go b/pkg/controller/nebulacluster/nebula_cluster_control.go
@@ -86,16 +86,15 @@ func (c *defaultNebulaClusterControl) UpdateNebulaCluster(nc *v1alpha1.NebulaClu
 	}
 
 	c.conditionUpdater.Update(nc)
-
-	if apiequality.Semantic.DeepEqual(&nc.Status, oldStatus) && nc.IsConditionReady() {
-		return errorutils.NewAggregate(errs)
-	}
-
 	nc.Status.ObservedGeneration = nc.Generation
 	if err := c.nebulaClient.UpdateNebulaClusterStatus(nc.DeepCopy()); err != nil {
 		errs = append(errs, err)
 	}
 
+	if apiequality.Semantic.DeepEqual(&nc.Status, oldStatus) && nc.IsReady() {
+		return errorutils.NewAggregate(errs)
+	}
+
 	if !nc.IsConditionReady() {
 		errs = append(errs, utilerrors.ReconcileErrorf("waiting for nebulacluster ready"))
 	}

diff --git a/pkg/controller/nebularestore/nebula_restore_manager.go b/pkg/controller/nebularestore/nebula_restore_manager.go
@@ -112,6 +112,10 @@ func (rm *restoreManager) syncRestoreProcess(rt *v1alpha1.NebulaRestore) error {
 	if rt.Spec.BR.ClusterNamespace != nil {
 		ns = *rt.Spec.BR.ClusterNamespace
 	}
+	original, err := rm.clientSet.NebulaCluster().GetNebulaCluster(ns, originalName)
+	if err != nil {
+		return err
+	}
 
 	restoredName, err := rm.getRestoredName(rt)
 	if err != nil {
@@ -125,6 +129,7 @@ func (rm *restoreManager) syncRestoreProcess(rt *v1alpha1.NebulaRestore) error {
 		}
 		nc.Annotations = nil
 		nc.Spec.Storaged.EnableForceUpdate = nil
+		nc.Spec.EnableAutoFailover = original.Spec.EnableAutoFailover
 		if err := rm.clientSet.NebulaCluster().UpdateNebulaCluster(nc); err != nil {
 			return fmt.Errorf("remove cluster [%s/%s] annotations failed: %v", ns, restoredName, err)
 		}
@@ -141,11 +146,6 @@ func (rm *restoreManager) syncRestoreProcess(rt *v1alpha1.NebulaRestore) error {
 			})
 	}
 
-	original, err := rm.clientSet.NebulaCluster().GetNebulaCluster(ns, originalName)
-	if err != nil {
-		return err
-	}
-
 	options, err := nebula.ClientOptions(original, nebula.SetIsMeta(true))
 	if err != nil {
 		return err
@@ -343,6 +343,7 @@ func (rm *restoreManager) genNebulaCluster(restoredName string, rt *v1alpha1.Neb
 	nc.Spec.Storaged.InitContainers = append(nc.Spec.Storaged.InitContainers, v1alpha1.GenerateInitAgentContainer(nc.StoragedComponent()))
 
 	nc.Spec.Storaged.EnableForceUpdate = pointer.Bool(true)
+	nc.Spec.EnableAutoFailover = nil
 	if rt.Spec.NodeSelector != nil {
 		nc.Spec.NodeSelector = rt.Spec.NodeSelector
 	}

diff --git a/pkg/kube/nebulacluster.go b/pkg/kube/nebulacluster.go
@@ -118,7 +118,7 @@ func (c *nebulaClusterClient) UpdateNebulaClusterStatus(nc *v1alpha1.NebulaClust
 			return err
 		}
 
-		if reflect.DeepEqual(status, ncClone.Status) {
+		if reflect.DeepEqual(*status, ncClone.Status) {
 			return nil
 		}
 

diff --git a/pkg/util/extender/unstructured.go b/pkg/util/extender/unstructured.go
@@ -20,7 +20,6 @@ import (
 	"encoding/json"
 	"strconv"
 	"strings"
-	"time"
 
 	apiequality "k8s.io/apimachinery/pkg/api/equality"
 	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
@@ -303,7 +302,7 @@ func SetRestartTimestamp(obj *unstructured.Unstructured) error {
 	for k, v := range obj.GetAnnotations() {
 		annotations[k] = v
 	}
-	annotations[annotation.AnnRestartTimestamp] = time.Now().Format(time.RFC3339)
+	annotations[annotation.AnnRestartTimestamp] = ""
 	obj.SetAnnotations(annotations)
 	return nil
 }
-Original file line number
+Diff line change
@@ Expand Up / @@ -41,7 +41,7 @@ import ( @@
     const (
     	PVCProtectionFinalizer = "kubernetes.io/pvc-protection"
-    	RestartTolerancePeriod = time.Minute * 1
+    	RestartTolerancePeriod = time.Minute * 2
     )
     var (
@@ Expand Down @@