fix: rename GPUNodePoolIdentifierLabelKey to GPUNodePoolIdentifierLabelFormat

Author: 0x5457

internal/constants/constants.go

@@ -13,7 +13,7 @@ const (
 	LabelKeyOwner = Domain + "/managed-by"
 
 	GPUNodePoolIdentifierLabelPrefix = Domain + "/pool/"
-	GPUNodePoolIdentifierLabelKey    = Domain + "/pool/%s"
+	GPUNodePoolIdentifierLabelFormat = Domain + "/pool/%s"
 	NodeDeletionMark                 = Domain + "/should-delete"
 
 	TensorFusionEnabledLabelKey = Domain + "/enabled"

internal/controller/gpunode_controller.go

@@ -130,7 +130,7 @@ func (r *GPUNodeReconciler) reconcileHypervisorPod(ctx context.Context, node *tf
 					Name:      hypervisorPodName,
 					Namespace: namespace,
 					Labels: map[string]string{
-						fmt.Sprintf(constants.GPUNodePoolIdentifierLabelKey, poolName): "true",
+						fmt.Sprintf(constants.GPUNodePoolIdentifierLabelFormat, poolName): "true",
 					},
 				},
 				Spec: *spec,

internal/controller/gpupool_compaction_controller.go

@@ -49,7 +49,7 @@ func (r *GPUPoolCompactionReconciler) checkNodeCompaction(ctx context.Context, p
 	// Strategy #1, terminate empty node
 	allNodes := &tfv1.GPUNodeList{}
 	if err := r.List(ctx, allNodes, client.MatchingLabels(map[string]string{
-		fmt.Sprintf(constants.GPUNodePoolIdentifierLabelKey, pool.Name): "true",
+		fmt.Sprintf(constants.GPUNodePoolIdentifierLabelFormat, pool.Name): "true",
 	})); err != nil {
 		return fmt.Errorf("failed to list nodes : %w", err)
 	}

internal/controller/gpupool_controller.go

@@ -30,7 +30,9 @@ import (
 	corev1 "k8s.io/api/core/v1"
 	"k8s.io/apimachinery/pkg/api/errors"
 	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
+	"k8s.io/apimachinery/pkg/labels"
 	"k8s.io/apimachinery/pkg/runtime"
+	"k8s.io/apimachinery/pkg/selection"
 	"k8s.io/apimachinery/pkg/types"
 	"k8s.io/client-go/tools/record"
 	schedulingcorev1 "k8s.io/component-helpers/scheduling/corev1"
@@ -123,9 +125,15 @@ func (r *GPUPoolReconciler) startNodeDiscoverys(
 		return fmt.Errorf("unmarshal pod template: %w", err)
 	}
 	// pool.Spec.NodeManagerConfig.NodeSelector
+	selector := labels.NewSelector()
+	poolReq, err := labels.NewRequirement(fmt.Sprintf(constants.GPUNodePoolIdentifierLabelFormat, pool.Name), selection.DoubleEquals, []string{"false"})
+	if err != nil {
+		return fmt.Errorf("new GPUNodePoolIdentifier label seletor: %w", err)
+	}
+	selector = selector.Add(*poolReq)
 	nodes := &tfv1.GPUNodeList{}
-	if err := r.Client.List(ctx, nodes); err != nil {
-		return fmt.Errorf("list nodes: %v", err)
+	if err := r.Client.List(ctx, nodes, &client.ListOptions{LabelSelector: selector}); err != nil {
+		return fmt.Errorf("list gpunodes: %v", err)
 	}
 
 	for _, gpuNode := range nodes.Items {

internal/controller/node_controller.go

@@ -93,7 +93,7 @@ func (r *NodeReconciler) generateGPUNode(ctx context.Context, node *corev1.Node,
 		ObjectMeta: metav1.ObjectMeta{
 			Name: node.Name,
 			Labels: map[string]string{
-				constants.GPUNodePoolIdentifierLabelKey: poolName,
+				fmt.Sprint(constants.GPUNodePoolIdentifierLabelFormat, poolName): "true",
 			},
 		},
 		Spec: tfv1.GPUNodeSpec{