Validate support for GPU instance types

integration/tests/instance_selector/instance_selector_test.go

@@ -9,12 +9,12 @@ import (
 	. "github.com/onsi/ginkgo/extensions/table"
 	. "github.com/onsi/gomega"
 	"github.com/pkg/errors"
-	"github.com/weaveworks/eksctl/pkg/utils"
 
 	"github.com/weaveworks/eksctl/integration/tests"
 	api "github.com/weaveworks/eksctl/pkg/apis/eksctl.io/v1alpha5"
 	"github.com/weaveworks/eksctl/pkg/eks"
 	"github.com/weaveworks/eksctl/pkg/testutils"
+	instanceutils "github.com/weaveworks/eksctl/pkg/utils/instance"
 
 	. "github.com/onsi/ginkgo"
 )
@@ -58,7 +58,7 @@ var _ = Describe("(Integration) [Instance Selector test]", func() {
 	},
 		Entry("non-GPU instances", func(instanceTypes []string) {
 			for _, instanceType := range instanceTypes {
-				Expect(instanceType).ToNot(Satisfy(utils.IsGPUInstanceType))
+				Expect(instanceType).ToNot(Satisfy(instanceutils.IsGPUInstanceType))
 			}
 		}, "--instance-selector-vcpus=8",
 			"--instance-selector-memory=32",

pkg/actions/nodegroup/create.go

@@ -14,7 +14,7 @@ import (
 	"github.com/weaveworks/eksctl/pkg/eks"
 	"github.com/weaveworks/eksctl/pkg/kubernetes"
 	"github.com/weaveworks/eksctl/pkg/printers"
-	"github.com/weaveworks/eksctl/pkg/utils"
+	instanceutils "github.com/weaveworks/eksctl/pkg/utils/instance"
 	"github.com/weaveworks/eksctl/pkg/utils/tasks"
 	"github.com/weaveworks/eksctl/pkg/vpc"
 
@@ -278,7 +278,7 @@ func (m *Manager) checkARMSupport(ctl *eks.ClusterProvider, clientSet kubernetes
 	if err != nil {
 		return err
 	}
-	if api.ClusterHasInstanceType(cfg, utils.IsARMInstanceType) {
+	if api.ClusterHasInstanceType(cfg, instanceutils.IsARMInstanceType) {
 		upToDate, err := defaultaddons.DoAddonsSupportMultiArch(clientSet, rawClient, kubernetesVersion, ctl.Provider.Region())
 		if err != nil {
 			return err

pkg/ami/auto_resolver.go

@@ -6,7 +6,7 @@ import (
 	"github.com/aws/aws-sdk-go/service/ec2/ec2iface"
 	"github.com/kris-nova/logger"
 	api "github.com/weaveworks/eksctl/pkg/apis/eksctl.io/v1alpha5"
-	"github.com/weaveworks/eksctl/pkg/utils"
+	instanceutils "github.com/weaveworks/eksctl/pkg/utils/instance"
 )
 
 const (
@@ -71,7 +71,7 @@ func (r *AutoResolver) Resolve(region, version, instanceType, imageFamily string
 
 	imageClasses := MakeImageSearchPatterns(version)[imageFamily]
 	namePattern := imageClasses[ImageClassGeneral]
-	if utils.IsGPUInstanceType(instanceType) {
+	if instanceutils.IsGPUInstanceType(instanceType) {
 		var ok bool
 		namePattern, ok = imageClasses[ImageClassGPU]
 		if !ok {
@@ -80,7 +80,7 @@ func (r *AutoResolver) Resolve(region, version, instanceType, imageFamily string
 		}
 	}
 
-	if utils.IsARMInstanceType(instanceType) {
+	if instanceutils.IsARMInstanceType(instanceType) {
 		var ok bool
 		namePattern, ok = imageClasses[ImageClassARM]
 		if !ok {

pkg/ami/auto_resolver_test.go

@@ -8,7 +8,6 @@ import (
 	"github.com/stretchr/testify/mock"
 	. "github.com/weaveworks/eksctl/pkg/ami"
 	api "github.com/weaveworks/eksctl/pkg/apis/eksctl.io/v1alpha5"
-	"github.com/weaveworks/eksctl/pkg/eks"
 	"github.com/weaveworks/eksctl/pkg/testutils/mockprovider"
 )
 
@@ -77,7 +76,7 @@ var _ = Describe("AMI Auto Resolution", func() {
 					BeforeEach(func() {
 						imageState = "available"
 
-						_, p = createProviders()
+						p = mockprovider.NewMockProvider()
 						addMockDescribeImages(p, "amazon-eks-node-1.15-v*", expectedAmi, imageState, "2018-08-20T23:25:53.000Z", api.NodeImageFamilyAmazonLinux2)
 						resolver := NewAutoResolver(p.MockEC2())
 						resolvedAmi, err = resolver.Resolve(region, version, instanceType, imageFamily)
@@ -101,7 +100,7 @@ var _ = Describe("AMI Auto Resolution", func() {
 						imageState = "available"
 						imageFamily = "Ubuntu1804"
 
-						_, p = createProviders()
+						p = mockprovider.NewMockProvider()
 						addMockDescribeImages(p, "ubuntu-eks/k8s_1.15/images/*18.04*", expectedAmi, imageState, "2018-08-20T23:25:53.000Z", api.NodeImageFamilyUbuntu1804)
 
 						resolver := NewAutoResolver(p.MockEC2())
@@ -125,7 +124,7 @@ var _ = Describe("AMI Auto Resolution", func() {
 					BeforeEach(func() {
 						imageState = "pending"
 
-						_, p = createProviders()
+						p = mockprovider.NewMockProvider()
 						addMockDescribeImagesMultiple(p, "amazon-eks-node-1.15-v*", []returnAmi{})
 
 						resolver := NewAutoResolver(p.MockEC2())
@@ -150,7 +149,7 @@ var _ = Describe("AMI Auto Resolution", func() {
 						imageState = "available"
 						expectedAmi = "ami-5678"
 
-						_, p = createProviders()
+						p = mockprovider.NewMockProvider()
 						images := []returnAmi{
 							{
 								createdDate: "2018-08-20T23:25:53.000Z",
@@ -193,7 +192,7 @@ var _ = Describe("AMI Auto Resolution", func() {
 					BeforeEach(func() {
 						imageState = "available"
 
-						_, p = createProviders()
+						p = mockprovider.NewMockProvider()
 						addMockDescribeImages(p, "amazon-eks-gpu-node-1.15-*", expectedAmi, imageState, "2018-08-20T23:25:53.000Z", api.NodeImageFamilyAmazonLinux2)
 						resolver := NewAutoResolver(p.MockEC2())
 						resolvedAmi, err = resolver.Resolve(region, version, instanceType, imageFamily)
@@ -216,16 +215,6 @@ var _ = Describe("AMI Auto Resolution", func() {
 	})
 })
 
-func createProviders() (*eks.ClusterProvider, *mockprovider.MockProvider) {
-	p := mockprovider.NewMockProvider()
-
-	c := &eks.ClusterProvider{
-		Provider: p,
-	}
-
-	return c, p
-}
-
 func addMockDescribeImages(p *mockprovider.MockProvider, expectedNamePattern string, amiID string, amiState string, createdDate string, instanceFamily string) {
 	p.MockEC2().On("DescribeImages",
 		mock.MatchedBy(func(input *ec2.DescribeImagesInput) bool {

pkg/ami/ssm_resolver.go

@@ -10,8 +10,8 @@ import (
 	"github.com/kris-nova/logger"
 
 	api "github.com/weaveworks/eksctl/pkg/apis/eksctl.io/v1alpha5"
-
 	"github.com/weaveworks/eksctl/pkg/utils"
+	instanceutils "github.com/weaveworks/eksctl/pkg/utils/instance"
 )
 
 // SSMResolver resolves the AMI to the defaults for the region
@@ -86,19 +86,19 @@ func MakeManagedSSMParameterName(version, imageFamily, amiType string) (string,
 // instanceEC2ArchName returns the name of the architecture as used by EC2
 // resources.
 func instanceEC2ArchName(instanceType string) string {
-	if utils.IsARMInstanceType(instanceType) {
+	if instanceutils.IsARMInstanceType(instanceType) {
 		return "arm64"
 	}
 	return "x86_64"
 }
 
 func imageType(imageFamily, instanceType string) string {
 	family := utils.ToKebabCase(imageFamily)
-	if utils.IsGPUInstanceType(instanceType) {
+	if instanceutils.IsGPUInstanceType(instanceType) {
 		return family + "-gpu"
 	}
 
-	if utils.IsARMInstanceType(instanceType) {
+	if instanceutils.IsARMInstanceType(instanceType) {
 		return family + "-arm64"
 	}
 	return family

pkg/ami/ssm_resolver_test.go

@@ -42,7 +42,7 @@ var _ = Describe("AMI Auto Resolution", func() {
 				Context("and AL2 ami is available", func() {
 					BeforeEach(func() {
 
-						_, p = createProviders()
+						p = mockprovider.NewMockProvider()
 						addMockGetParameter(p, "/aws/service/eks/optimized-ami/1.12/amazon-linux-2/recommended/image_id", expectedAmi)
 						resolver := NewSSMResolver(p.MockSSM())
 						resolvedAmi, err = resolver.Resolve(region, version, instanceType, imageFamily)
@@ -64,7 +64,7 @@ var _ = Describe("AMI Auto Resolution", func() {
 				Context("and ami is NOT available", func() {
 					BeforeEach(func() {
 
-						_, p = createProviders()
+						p = mockprovider.NewMockProvider()
 						addMockFailedGetParameter(p, "/aws/service/eks/optimized-ami/1.12/amazon-linux-2/recommended/image_id")
 
 						resolver := NewSSMResolver(p.MockSSM())
@@ -94,7 +94,7 @@ var _ = Describe("AMI Auto Resolution", func() {
 				Context("and ami is available", func() {
 					BeforeEach(func() {
 
-						_, p = createProviders()
+						p = mockprovider.NewMockProvider()
 						addMockGetParameter(p, "/aws/service/eks/optimized-ami/1.12/amazon-linux-2-gpu/recommended/image_id", expectedAmi)
 						resolver := NewSSMResolver(p.MockSSM())
 						resolvedAmi, err = resolver.Resolve(region, version, instanceType, imageFamily)
@@ -122,7 +122,7 @@ var _ = Describe("AMI Auto Resolution", func() {
 				Context("and ami is available", func() {
 					BeforeEach(func() {
 						version = "1.14"
-						_, p = createProviders()
+						p = mockprovider.NewMockProvider()
 					})
 
 					It("should return a valid Full image for 1.14", func() {
@@ -160,7 +160,7 @@ var _ = Describe("AMI Auto Resolution", func() {
 
 			Context("and Ubuntu family", func() {
 				BeforeEach(func() {
-					_, p = createProviders()
+					p = mockprovider.NewMockProvider()
 				})
 
 				It("should return an error", func() {
@@ -180,7 +180,7 @@ var _ = Describe("AMI Auto Resolution", func() {
 
 				Context("and ami is available", func() {
 					BeforeEach(func() {
-						_, p = createProviders()
+						p = mockprovider.NewMockProvider()
 						addMockGetParameter(p, "/aws/service/bottlerocket/aws-k8s-1.15/x86_64/latest/image_id", expectedAmi)
 						resolver := NewSSMResolver(p.MockSSM())
 						resolvedAmi, err = resolver.Resolve(region, version, instanceType, imageFamily)
@@ -201,7 +201,7 @@ var _ = Describe("AMI Auto Resolution", func() {
 
 				Context("and ami is NOT available", func() {
 					BeforeEach(func() {
-						_, p = createProviders()
+						p = mockprovider.NewMockProvider()
 						addMockFailedGetParameter(p, "/aws/service/bottlerocket/aws-k8s-1.15/x86_64/latest/image_id")
 
 						resolver := NewSSMResolver(p.MockSSM())
@@ -229,7 +229,7 @@ var _ = Describe("AMI Auto Resolution", func() {
 
 					Context("and ami is available", func() {
 						BeforeEach(func() {
-							_, p = createProviders()
+							p = mockprovider.NewMockProvider()
 							addMockGetParameter(p, "/aws/service/bottlerocket/aws-k8s-1.15/arm64/latest/image_id", expectedAmi)
 							resolver := NewSSMResolver(p.MockSSM())
 							resolvedAmi, err = resolver.Resolve(region, version, instanceType, imageFamily)
@@ -250,7 +250,7 @@ var _ = Describe("AMI Auto Resolution", func() {
 
 					Context("and ami is NOT available", func() {
 						BeforeEach(func() {
-							_, p = createProviders()
+							p = mockprovider.NewMockProvider()
 							addMockFailedGetParameter(p, "/aws/service/bottlerocket/aws-k8s-1.15/arm64/latest/image_id")
 
 							resolver := NewSSMResolver(p.MockSSM())
@@ -278,7 +278,7 @@ var _ = Describe("AMI Auto Resolution", func() {
 	})
 })
 
-func addMockGetParameter(p *mockprovider.MockProvider, name string, amiID string) {
+func addMockGetParameter(p *mockprovider.MockProvider, name, amiID string) {
 	p.MockSSM().On("GetParameter",
 		mock.MatchedBy(func(input *ssm.GetParameterInput) bool {
 			return *input.Name == name

pkg/apis/eksctl.io/v1alpha5/gpu_validation_test.go

@@ -0,0 +1,91 @@
+package v1alpha5_test
+
+import (
+	"fmt"
+
+	. "github.com/onsi/ginkgo"
+	. "github.com/onsi/ginkgo/extensions/table"
+	. "github.com/onsi/gomega"
+
+	api "github.com/weaveworks/eksctl/pkg/apis/eksctl.io/v1alpha5"
+)
+
+var _ = Describe("GPU instance support", func() {
+
+	type gpuInstanceEntry struct {
+		gpuInstanceType string
+		amiFamily       string
+
+		expectUnsupportedErr bool
+	}
+
+	assertValidationError := func(e gpuInstanceEntry, err error) {
+		if e.expectUnsupportedErr {
+			Expect(err).To(HaveOccurred())
+			Expect(err).To(MatchError(ContainSubstring(fmt.Sprintf("GPU instance types are not supported for %s", e.amiFamily))))
+			return
+		}
+		Expect(err).ToNot(HaveOccurred())
+	}
+
+	DescribeTable("managed nodegroups", func(e gpuInstanceEntry) {
+		mng := api.NewManagedNodeGroup()
+		mng.InstanceType = e.gpuInstanceType
+		mng.AMIFamily = e.amiFamily
+		mng.InstanceSelector = &api.InstanceSelector{}
+		assertValidationError(e, api.ValidateManagedNodeGroup(mng, 0))
+	},
+		Entry("AL2", gpuInstanceEntry{
+			gpuInstanceType: "asdf",
+			amiFamily:       api.NodeImageFamilyAmazonLinux2,
+		}),
+		Entry("Bottlerocket", gpuInstanceEntry{
+			amiFamily:            api.NodeImageFamilyBottlerocket,
+			gpuInstanceType:      "g4dn.xlarge",
+			expectUnsupportedErr: true,
+		}),
+		Entry("Ubuntu2004", gpuInstanceEntry{
+			amiFamily:            api.NodeImageFamilyUbuntu2004,
+			gpuInstanceType:      "g4dn.xlarge",
+			expectUnsupportedErr: true,
+		}),
+	)
+
+	DescribeTable("unmanaged nodegroups", func(e gpuInstanceEntry) {
+		ng := api.NewNodeGroup()
+		ng.InstanceType = e.gpuInstanceType
+		ng.AMIFamily = e.amiFamily
+		assertValidationError(e, api.ValidateNodeGroup(0, ng))
+
+	},
+		Entry("AL2", gpuInstanceEntry{
+			gpuInstanceType: "g4dn.xlarge",
+			amiFamily:       api.NodeImageFamilyAmazonLinux2,
+		}),
+		Entry("Bottlerocket", gpuInstanceEntry{
+			amiFamily:            api.NodeImageFamilyBottlerocket,
+			gpuInstanceType:      "g4dn.xlarge",
+			expectUnsupportedErr: true,
+		}),
+		Entry("Ubuntu2004", gpuInstanceEntry{
+			amiFamily:            api.NodeImageFamilyUbuntu2004,
+			gpuInstanceType:      "g4dn.xlarge",
+			expectUnsupportedErr: true,
+		}),
+		Entry("Windows2004", gpuInstanceEntry{
+			amiFamily:            api.NodeImageFamilyWindowsServer2004CoreContainer,
+			gpuInstanceType:      "p2.xlarge",
+			expectUnsupportedErr: true,
+		}),
+		Entry("Windows2019Core", gpuInstanceEntry{
+			amiFamily:            api.NodeImageFamilyWindowsServer2019CoreContainer,
+			gpuInstanceType:      "g3.8xlarge",
+			expectUnsupportedErr: true,
+		}),
+		Entry("Windows2019Full", gpuInstanceEntry{
+			amiFamily:            api.NodeImageFamilyWindowsServer2019FullContainer,
+			gpuInstanceType:      "p3.2xlarge",
+			expectUnsupportedErr: true,
+		}),
+	)
+})

pkg/apis/eksctl.io/v1alpha5/instance.go

@@ -0,0 +1,32 @@
+package v1alpha5
+
+import (
+	instanceutils "github.com/weaveworks/eksctl/pkg/utils/instance"
+)
+
+// SelectInstanceType determines which instanceType is relevant for selecting an AMI
+// If the nodegroup has mixed instances it will prefer a GPU instance type over a general class one
+// This is to make sure that the AMI that is selected later is valid for all the types
+func SelectInstanceType(np NodePool) string {
+	var instanceTypes []string
+	switch ng := np.(type) {
+	case *NodeGroup:
+		if ng.InstancesDistribution != nil {
+			instanceTypes = ng.InstancesDistribution.InstanceTypes
+		}
+	case *ManagedNodeGroup:
+		instanceTypes = ng.InstanceTypes
+	}
+
+	hasMixedInstances := len(instanceTypes) > 0
+	if hasMixedInstances {
+		for _, instanceType := range instanceTypes {
+			if instanceutils.IsGPUInstanceType(instanceType) {
+				return instanceType
+			}
+		}
+		return instanceTypes[0]
+	}
+
+	return np.BaseNodeGroup().InstanceType
+}

pkg/apis/eksctl.io/v1alpha5/types.go

@@ -822,7 +822,6 @@ func NewManagedNodeGroup() *ManagedNodeGroup {
 			SSH: &NodeGroupSSH{
 				Allow:         Disabled(),
 				PublicKeyName: &publicKey,
-				EnableSSM:     Disabled(),
 			},
 			IAM: &NodeGroupIAM{
 				WithAddonPolicies: NodeGroupIAMAddonPolicies{