Add field status.clusterSize and CEL rule for spec.replica transitions

Author: shreyas-s-rao

api/core/v1alpha1/conditions.go

@@ -47,3 +47,13 @@ type Condition struct {
 	// A human-readable message indicating details about the transition.
 	Message string `json:"message"`
 }
+
+// HasConditionWithStatus checks if the Etcd has a condition of the specified type with the given status.
+func (e *Etcd) HasConditionWithStatus(condType ConditionType, status ConditionStatus) bool {
+	for _, condition := range e.Status.Conditions {
+		if condition.Type == condType && condition.Status == status {
+			return true
+		}
+	}
+	return false
+}

api/core/v1alpha1/crds/druid.gardener.cloud_etcds.yaml

@@ -30,8 +30,11 @@ spec:
     - jsonPath: .metadata.creationTimestamp
       name: Age
       type: date
-    - jsonPath: .spec.replicas
+    - jsonPath: .status.clusterSize
       name: Cluster Size
+      type: integer
+    - jsonPath: .spec.replicas
+      name: Desired Replicas
       priority: 1
       type: integer
     - jsonPath: .status.currentReplicas
@@ -648,9 +651,6 @@ spec:
               replicas:
                 format: int32
                 type: integer
-                x-kubernetes-validations:
-                - message: Replicas can either be increased or be downscaled to 0.
-                  rule: 'self==0 ? true : self < oldSelf ? false : true'
               schedulingConstraints:
                 description: |-
                   SchedulingConstraints defines the different scheduling constraints that must be applied to the
@@ -1876,6 +1876,10 @@ spec:
           status:
             description: EtcdStatus defines the observed state of Etcd.
             properties:
+              clusterSize:
+                description: ClusterSize is the last recorded etcd cluster size.
+                format: int32
+                type: integer
               conditions:
                 description: Conditions represents the latest available observations
                   of an etcd's current state.
@@ -2101,6 +2105,11 @@ spec:
                 type: integer
             type: object
         type: object
+        x-kubernetes-validations:
+        - message: etcd.spec.replicas transition is not allowed.
+          rule: 'self.spec.replicas == 0 || (oldSelf.spec.replicas == 0 ? (!has(self.status.clusterSize)
+            || (self.spec.replicas == self.status.clusterSize)) : (self.spec.replicas
+            >= oldSelf.spec.replicas))'
     served: true
     storage: true
     subresources:

api/core/v1alpha1/crds/druid.gardener.cloud_etcds_without_cel.yaml

@@ -30,8 +30,11 @@ spec:
         - jsonPath: .metadata.creationTimestamp
           name: Age
           type: date
-        - jsonPath: .spec.replicas
+        - jsonPath: .status.clusterSize
           name: Cluster Size
+          type: integer
+        - jsonPath: .spec.replicas
+          name: Desired Replicas
           priority: 1
           type: integer
         - jsonPath: .status.currentReplicas
@@ -1741,6 +1744,10 @@ spec:
             status:
               description: EtcdStatus defines the observed state of Etcd.
               properties:
+                clusterSize:
+                  description: ClusterSize is the last recorded etcd cluster size.
+                  format: int32
+                  type: integer
                 conditions:
                   description: Conditions represents the latest available observations of an etcd's current state.
                   items:

api/core/v1alpha1/etcd.go

@@ -48,11 +48,13 @@ const (
 // +kubebuilder:printcolumn:name="All Members Ready",type=string,JSONPath=`.status.conditions[?(@.type=="AllMembersReady")].status`
 // +kubebuilder:printcolumn:name="Backup Ready",type=string,JSONPath=`.status.conditions[?(@.type=="BackupReady")].status`
 // +kubebuilder:printcolumn:name="Age",type=date,JSONPath=`.metadata.creationTimestamp`
-// +kubebuilder:printcolumn:name="Cluster Size",type=integer,JSONPath=`.spec.replicas`,priority=1
+// +kubebuilder:printcolumn:name="Cluster Size",type=integer,JSONPath=`.status.clusterSize`
+// +kubebuilder:printcolumn:name="Desired Replicas",type=integer,JSONPath=`.spec.replicas`,priority=1
 // +kubebuilder:printcolumn:name="Current Replicas",type=integer,JSONPath=`.status.currentReplicas`,priority=1
 // +kubebuilder:printcolumn:name="Ready Replicas",type=integer,JSONPath=`.status.readyReplicas`,priority=1
 
 // Etcd is the Schema for the etcds API
+// +kubebuilder:validation:XValidation:message="etcd.spec.replicas transition is not allowed.",rule="self.spec.replicas == 0 || (oldSelf.spec.replicas == 0 ? (!has(self.status.clusterSize) || (self.spec.replicas == self.status.clusterSize)) : (self.spec.replicas >= oldSelf.spec.replicas))"
 type Etcd struct {
 	metav1.TypeMeta   `json:",inline"`
 	metav1.ObjectMeta `json:"metadata,omitempty"`
@@ -309,7 +311,6 @@ type EtcdSpec struct {
 	// +optional
 	SchedulingConstraints SchedulingConstraints `json:"schedulingConstraints,omitempty"`
 	// +required
-	// +kubebuilder:validation:XValidation:message="Replicas can either be increased or be downscaled to 0.",rule="self==0 ? true : self < oldSelf ? false : true"
 	Replicas int32 `json:"replicas"`
 	// PriorityClassName is the name of a priority class that shall be used for the etcd pods.
 	// +optional
@@ -435,6 +436,9 @@ type EtcdStatus struct {
 	// PeerUrlTLSEnabled captures the state of peer url TLS being enabled for the etcd member(s)
 	// +optional
 	PeerUrlTLSEnabled *bool `json:"peerUrlTLSEnabled,omitempty"`
+	// ClusterSize is the last recorded etcd cluster size.
+	// +optional
+	ClusterSize int32 `json:"clusterSize"`
 }
 
 // LastOperationType is a string alias representing type of the last operation.

docs/api-reference/etcd-druid-api.md

@@ -428,6 +428,7 @@ _Appears in:_
 | `labelSelector` _[LabelSelector](https://kubernetes.io/docs/reference/generated/kubernetes-api/v1.29/#labelselector-v1-meta)_ | LabelSelector is a label query over pods that should match the replica count.<br />It must match the pod template's labels.<br />Deprecated: this field will be removed in the future. |  |  |
 | `members` _[EtcdMemberStatus](#etcdmemberstatus) array_ | Members represents the members of the etcd cluster |  |  |
 | `peerUrlTLSEnabled` _boolean_ | PeerUrlTLSEnabled captures the state of peer url TLS being enabled for the etcd member(s) |  |  |
+| `clusterSize` _integer_ | ClusterSize is the last recorded etcd cluster size. |  |  |
 
 
 #### GarbageCollectionPolicy

internal/controller/etcd/reconcile_status.go

@@ -27,8 +27,9 @@ func (r *Reconciler) reconcileStatus(ctx component.OperatorContext, etcdObjectKe
 	sLog := r.logger.WithValues("etcd", etcdObjectKey, "operation", "reconcileStatus").WithValues("runID", ctx.RunID)
 	originalEtcd := etcd.DeepCopy()
 	mutateETCDStatusStepFns := []mutateEtcdStatusFn{
-		r.mutateETCDStatusWithMemberStatusAndConditions,
-		r.inspectStatefulSetAndMutateETCDStatus,
+		r.mutateEtcdStatusWithMemberStatusAndConditions,
+		r.inspectStatefulSetAndMutateEtcdStatus,
+		r.mutateEtcdStatusWithClusterSize,
 	}
 	for _, fn := range mutateETCDStatusStepFns {
 		if stepResult := fn(ctx, etcd, sLog); ctrlutils.ShortCircuitReconcileFlow(stepResult) {
@@ -42,7 +43,7 @@ func (r *Reconciler) reconcileStatus(ctx component.OperatorContext, etcdObjectKe
 	return ctrlutils.ContinueReconcile()
 }
 
-func (r *Reconciler) mutateETCDStatusWithMemberStatusAndConditions(ctx component.OperatorContext, etcd *druidv1alpha1.Etcd, logger logr.Logger) ctrlutils.ReconcileStepResult {
+func (r *Reconciler) mutateEtcdStatusWithMemberStatusAndConditions(ctx component.OperatorContext, etcd *druidv1alpha1.Etcd, logger logr.Logger) ctrlutils.ReconcileStepResult {
 	statusCheck := status.NewChecker(r.client, r.config.EtcdMember.NotReadyThreshold, r.config.EtcdMember.UnknownThreshold)
 	if err := statusCheck.Check(ctx, logger, etcd); err != nil {
 		logger.Error(err, "Error executing status checks to update member status and conditions")
@@ -51,7 +52,7 @@ func (r *Reconciler) mutateETCDStatusWithMemberStatusAndConditions(ctx component
 	return ctrlutils.ContinueReconcile()
 }
 
-func (r *Reconciler) inspectStatefulSetAndMutateETCDStatus(ctx component.OperatorContext, etcd *druidv1alpha1.Etcd, _ logr.Logger) ctrlutils.ReconcileStepResult {
+func (r *Reconciler) inspectStatefulSetAndMutateEtcdStatus(ctx component.OperatorContext, etcd *druidv1alpha1.Etcd, _ logr.Logger) ctrlutils.ReconcileStepResult {
 	sts, err := kubernetes.GetStatefulSet(ctx, r.client, etcd)
 	if err != nil {
 		return ctrlutils.ReconcileWithError(err)
@@ -79,3 +80,11 @@ func (r *Reconciler) inspectStatefulSetAndMutateETCDStatus(ctx component.Operato
 	}
 	return ctrlutils.ContinueReconcile()
 }
+
+func (r *Reconciler) mutateEtcdStatusWithClusterSize(_ component.OperatorContext, etcd *druidv1alpha1.Etcd, _ logr.Logger) ctrlutils.ReconcileStepResult {
+	if etcd.Spec.Replicas > etcd.Status.ClusterSize && etcd.HasConditionWithStatus(druidv1alpha1.ConditionTypeAllMembersReady, druidv1alpha1.ConditionTrue) {
+		etcd.Status.ClusterSize = etcd.Spec.Replicas
+	}
+
+	return ctrlutils.ContinueReconcile()
+}