From eb2787abc8e18505c5cd4869c9d59d39336760ee Mon Sep 17 00:00:00 2001
From: Kenji Kaneda <kenji.kaneda@gmail.com>
Date: Fri, 3 Jan 2025 13:09:24 -0800
Subject: [PATCH] feat(engine): ignore cordoned GPU nodes from cluster status

---
 dispatcher/internal/clusterstatus/manager.go  |  5 +++++
 .../internal/clusterstatus/manager_test.go    | 21 +++++++++++++++++++
 2 files changed, 26 insertions(+)

diff --git a/dispatcher/internal/clusterstatus/manager.go b/dispatcher/internal/clusterstatus/manager.go
index 616ac52..cc1a6c7 100644
--- a/dispatcher/internal/clusterstatus/manager.go
+++ b/dispatcher/internal/clusterstatus/manager.go
@@ -152,6 +152,11 @@ func toProvisionableResource(np krpv1.NodePool) *v1.ProvisionableResource {
 }
 
 func toGPUNode(node corev1.Node, logger logr.Logger) (*v1.GpuNode, bool) {
+	// Ignore cordoned nodes.
+	if node.Spec.Unschedulable {
+		return nil, false
+	}
+
 	// TODO(kenji): Support other accelerator types.
 	rs := map[corev1.ResourceName]bool{
 		nvidiaGPU: true,
diff --git a/dispatcher/internal/clusterstatus/manager_test.go b/dispatcher/internal/clusterstatus/manager_test.go
index ad843bc..c15bb2b 100644
--- a/dispatcher/internal/clusterstatus/manager_test.go
+++ b/dispatcher/internal/clusterstatus/manager_test.go
@@ -79,6 +79,27 @@ func TestManager(t *testing.T) {
 				GpuNodes: []*v1.GpuNode{},
 			},
 		},
+		{
+			name: "cordoned gpu node",
+			objs: []runtime.Object{
+				&corev1.Node{
+					ObjectMeta: metav1.ObjectMeta{
+						Name: "node1",
+					},
+					Spec: corev1.NodeSpec{
+						Unschedulable: true,
+					},
+					Status: corev1.NodeStatus{
+						Allocatable: corev1.ResourceList{
+							nvidiaGPU: resource.MustParse("1"),
+						},
+					},
+				},
+			},
+			want: &v1.ClusterStatus{
+				GpuNodes: []*v1.GpuNode{},
+			},
+		},
 		{
 			name: "provisionable resources of instance type",
 			objs: []runtime.Object{