fix: type issue

Author: Code2Life

api/v1/gpupool_types.go

@@ -124,7 +124,6 @@ type NodeProvisioner struct {
 	CPULabels map[string]string `json:"cpuNodeLabels,omitempty"`
 
 	// +optional
-	// +kubebuilder:default="100"
 	// NodeProvisioner will start an virtual billing based on public pricing or customized pricing, if the VM's costs exceeded any budget constraints, the new VM will not be created, and alerts will be generated
 	Budget *PeriodicalBudget `json:"budget,omitempty"`
 }
@@ -161,7 +160,6 @@ const (
 )
 
 type Requirement struct {
-	// +kubebuilder:default=""
 	Key NodeRequirementKey `json:"key,omitempty"`
 
 	// +kubebuilder:default="In"

config/crd/bases/tensor-fusion.ai_gpupools.yaml

@@ -256,7 +256,6 @@ spec:
                       NodeSelector is for existing GPUs, NodeProvisioner is for Karpenter-like auto management.
                     properties:
                       budget:
-                        default: "100"
                         description: NodeProvisioner will start an virtual billing
                           based on public pricing or customized pricing, if the VM's
                           costs exceeded any budget constraints, the new VM will not
@@ -286,7 +285,6 @@ spec:
                         items:
                           properties:
                             key:
-                              default: ""
                               enum:
                               - node.kubernetes.io/instance-type
                               - kubernetes.io/arch
@@ -337,7 +335,6 @@ spec:
                         items:
                           properties:
                             key:
-                              default: ""
                               enum:
                               - node.kubernetes.io/instance-type
                               - kubernetes.io/arch

config/crd/bases/tensor-fusion.ai_tensorfusionclusters.yaml

@@ -354,7 +354,6 @@ spec:
                                 NodeSelector is for existing GPUs, NodeProvisioner is for Karpenter-like auto management.
                               properties:
                                 budget:
-                                  default: "100"
                                   description: NodeProvisioner will start an virtual
                                     billing based on public pricing or customized
                                     pricing, if the VM's costs exceeded any budget
@@ -385,7 +384,6 @@ spec:
                                   items:
                                     properties:
                                       key:
-                                        default: ""
                                         enum:
                                         - node.kubernetes.io/instance-type
                                         - kubernetes.io/arch
@@ -436,7 +434,6 @@ spec:
                                   items:
                                     properties:
                                       key:
-                                        default: ""
                                         enum:
                                         - node.kubernetes.io/instance-type
                                         - kubernetes.io/arch

internal/controller/gpupool_controller.go

@@ -84,7 +84,8 @@ func (r *GPUPoolReconciler) Reconcile(ctx context.Context, req ctrl.Request) (ct
 	}
 
 	// For provisioning mode, check if need to scale up GPUNodes upon AvailableCapacity changed
-	if pool.Spec.NodeManagerConfig.NodeProvisioner != nil && pool.Spec.NodeManagerConfig.NodeProvisioner.Mode == tfv1.NodeProvisionerModeNative {
+	isProvisioningMode := pool.Spec.NodeManagerConfig.NodeProvisioner != nil && pool.Spec.NodeManagerConfig.NodeProvisioner.Mode == tfv1.NodeProvisionerModeNative
+	if isProvisioningMode {
 		if err := r.reconcilePoolCapacityWithProvisioner(ctx, pool); err != nil {
 			return ctrl.Result{}, err
 		}
@@ -95,8 +96,10 @@ func (r *GPUPoolReconciler) Reconcile(ctx context.Context, req ctrl.Request) (ct
 
 	// TODO, any GPUNode changes trigger GPUPool reconcile, it should change the status, aggregate the total amount of resources, update current status
 	// THIS NEED TO MOVE INTO GPU NODE CONTROLLER, rather than POOL CONTROLLER
-	if err := r.startNodeDiscoverys(ctx, pool); err != nil {
-		return ctrl.Result{}, err
+	if !isProvisioningMode {
+		if err := r.startNodeDiscoverys(ctx, pool); err != nil {
+			return ctrl.Result{}, err
+		}
 	}
 	// TODO, when componentConfig changed, it should notify corresponding resource to upgrade
 	// eg. when hypervisor changed, should change all owned GPUNode's status.phase to Updating
@@ -111,6 +114,9 @@ func (r *GPUPoolReconciler) startNodeDiscoverys(
 	log := log.FromContext(ctx)
 	log.Info("Starting node node discovery job")
 
+	if pool.Spec.ComponentConfig == nil || pool.Spec.ComponentConfig.NodeDiscovery.PodTemplate == nil {
+		return fmt.Errorf(`missing node discovery pod template in pool spec`)
+	}
 	podTmpl := &corev1.PodTemplate{}
 	err := json.Unmarshal(pool.Spec.ComponentConfig.NodeDiscovery.PodTemplate.Raw, podTmpl)
 	if err != nil {