Prevent populate ig from adding nvidia taint if it has already been set

pkg/testutils/cluster.go

@@ -33,6 +33,9 @@ func BuildMinimalCluster(clusterName string) *kops.Cluster {
 		{Name: "subnet-us-mock-1a", Zone: "us-mock-1a", CIDR: "172.20.1.0/24", Type: kops.SubnetTypePrivate},
 	}
 
+	c.Spec.ContainerRuntime = "containerd"
+	c.Spec.Containerd = &kops.ContainerdConfig{}
+
 	c.Spec.MasterPublicName = fmt.Sprintf("api.%v", clusterName)
 	c.Spec.MasterInternalName = fmt.Sprintf("internal.api.%v", clusterName)
 	c.Spec.KubernetesAPIAccess = []string{"0.0.0.0/0"}

upup/pkg/fi/cloudup/populate_instancegroup_spec.go

@@ -18,6 +18,7 @@ package cloudup
 
 import (
 	"fmt"
+	"strings"
 
 	"github.com/aws/aws-sdk-go/service/ec2"
 	"github.com/blang/semver/v4"
@@ -186,7 +187,15 @@ func PopulateInstanceGroupSpec(cluster *kops.Cluster, input *kops.InstanceGroup,
 					ig.Spec.NodeLabels = make(map[string]string)
 				}
 				ig.Spec.NodeLabels["kops.k8s.io/gpu"] = "1"
-				ig.Spec.Taints = append(ig.Spec.Taints, "nvidia.com/gpu:NoSchedule")
+				hasNvidiaTaint := false
+				for _, taint := range ig.Spec.Taints {
+					if strings.HasPrefix(taint, "nvidia.com/gpu") {
+						hasNvidiaTaint = true
+					}
+				}
+				if !hasNvidiaTaint {
+					ig.Spec.Taints = append(ig.Spec.Taints, "nvidia.com/gpu:NoSchedule")
+				}
 			}
 		}
 	}

upup/pkg/fi/cloudup/populate_instancegroup_spec_test.go

@@ -80,6 +80,49 @@ func TestPopulateInstanceGroup_Image_Required(t *testing.T) {
 	expectErrorFromPopulateInstanceGroup(t, cluster, g, channel, "unable to determine default image for InstanceGroup nodes")
 }
 
+func TestPopulateInstanceGroup_AddTaintsCollision(t *testing.T) {
+	_, cluster := buildMinimalCluster()
+	input := buildMinimalNodeInstanceGroup()
+	input.Spec.Taints = []string{"nvidia.com/gpu:NoSchedule"}
+	input.Spec.MachineType = "g4dn.xlarge"
+	cluster.Spec.Containerd.NvidiaGPU = &kopsapi.NvidiaGPUConfig{Enabled: fi.Bool(true)}
+
+	channel := &kopsapi.Channel{}
+
+	cloud, err := BuildCloud(cluster)
+	if err != nil {
+		t.Fatalf("error from BuildCloud: %v", err)
+	}
+	output, err := PopulateInstanceGroupSpec(cluster, input, cloud, channel)
+	if err != nil {
+		t.Fatalf("error from PopulateInstanceGropuSpec: %v", err)
+	}
+	if len(output.Spec.Taints) != 1 {
+		t.Errorf("Expected only 1 taint, got %d", len(output.Spec.Taints))
+	}
+}
+
+func TestPopulateInstanceGroup_AddTaints(t *testing.T) {
+	_, cluster := buildMinimalCluster()
+	input := buildMinimalNodeInstanceGroup()
+	input.Spec.MachineType = "g4dn.xlarge"
+	cluster.Spec.Containerd.NvidiaGPU = &kopsapi.NvidiaGPUConfig{Enabled: fi.Bool(true)}
+
+	channel := &kopsapi.Channel{}
+
+	cloud, err := BuildCloud(cluster)
+	if err != nil {
+		t.Fatalf("error from BuildCloud: %v", err)
+	}
+	output, err := PopulateInstanceGroupSpec(cluster, input, cloud, channel)
+	if err != nil {
+		t.Fatalf("error from PopulateInstanceGropuSpec: %v", err)
+	}
+	if len(output.Spec.Taints) != 1 {
+		t.Errorf("Expected only 1 taint, got %d", len(output.Spec.Taints))
+	}
+}
+
 func expectErrorFromPopulateInstanceGroup(t *testing.T, cluster *kopsapi.Cluster, g *kopsapi.InstanceGroup, channel *kopsapi.Channel, message string) {
 	cloud, err := BuildCloud(cluster)
 	if err != nil {