Merge pull request #559 from JoelSpeed/spot-support

Add support for Spot VMs

api/v1alpha2/azuremachine_conversion.go

@@ -36,19 +36,27 @@ func (src *AzureMachine) ConvertTo(dstRaw conversion.Hub) error { // nolint
 	if ok, err := utilconversion.UnmarshalData(src, restored); err != nil || !ok {
 		return err
 	}
-	if restored.Spec.Identity != "" {
-		dst.Spec.Identity = restored.Spec.Identity
+
+	restoreAzureMachineSpec(&restored.Spec, &dst.Spec)
+	return nil
+}
+
+func restoreAzureMachineSpec(restored, dst *infrav1alpha3.AzureMachineSpec) {
+	if restored.Identity != "" {
+		dst.Identity = restored.Identity
 	}
-	if len(restored.Spec.UserAssignedIdentities) > 0 {
-		dst.Spec.UserAssignedIdentities = restored.Spec.UserAssignedIdentities
+	if len(restored.UserAssignedIdentities) > 0 {
+		dst.UserAssignedIdentities = restored.UserAssignedIdentities
 	}
-	if restored.Spec.AcceleratedNetworking != nil {
-		dst.Spec.AcceleratedNetworking = restored.Spec.AcceleratedNetworking
+	if restored.AcceleratedNetworking != nil {
+		dst.AcceleratedNetworking = restored.AcceleratedNetworking
 	}
 
-	dst.Spec.FailureDomain = restored.Spec.FailureDomain
+	dst.FailureDomain = restored.FailureDomain
 
-	return nil
+	if restored.SpotVMOptions != nil {
+		dst.SpotVMOptions = restored.SpotVMOptions.DeepCopy()
+	}
 }
 
 // ConvertFrom converts from the Hub version (v1alpha3) to this version.

api/v1alpha2/azuremachinetemplate_conversion.go

@@ -34,17 +34,8 @@ func (src *AzureMachineTemplate) ConvertTo(dstRaw conversion.Hub) error { // nol
 	if ok, err := utilconversion.UnmarshalData(src, restored); err != nil || !ok {
 		return err
 	}
-	if restored.Spec.Template.Spec.Identity != "" {
-		dst.Spec.Template.Spec.Identity = restored.Spec.Template.Spec.Identity
-	}
-	if len(restored.Spec.Template.Spec.UserAssignedIdentities) > 0 {
-		dst.Spec.Template.Spec.UserAssignedIdentities = restored.Spec.Template.Spec.UserAssignedIdentities
-	}
-	if restored.Spec.Template.Spec.AcceleratedNetworking != nil {
-		dst.Spec.Template.Spec.AcceleratedNetworking = restored.Spec.Template.Spec.AcceleratedNetworking
-	}
-	dst.Spec.Template.Spec.FailureDomain = restored.Spec.Template.Spec.FailureDomain
 
+	restoreAzureMachineSpec(&restored.Spec.Template.Spec, &dst.Spec.Template.Spec)
 	return nil
 }
 

api/v1alpha3/azuremachine_types.go

@@ -87,6 +87,18 @@ type AzureMachineSpec struct {
 	// +kubebuilder:validation:nullable
 	// +optional
 	AcceleratedNetworking *bool `json:"acceleratedNetworking,omitempty"`
+
+	// SpotVMOptions allows the ability to specify the Machine should use a Spot VM
+	// +optional
+	SpotVMOptions *SpotVMOptions `json:"spotVMOptions,omitempty"`
+}
+
+// SpotVMOptions defines the options relevant to running the Machine on Spot VMs
+type SpotVMOptions struct {
+	// MaxPrice defines the maximum price the user is willing to pay for Spot VM instances
+	// +optional
+	// +kubebuilder:validation:Type=number
+	MaxPrice *string `json:"maxPrice,omitempty"`
 }
 
 // AzureMachineStatus defines the observed state of AzureMachine

cloud/services/virtualmachines/virtualmachines.go

@@ -21,6 +21,7 @@ import (
 	"crypto/rand"
 	"encoding/base64"
 	"fmt"
+	"strconv"
 	"strings"
 
 	"github.com/Azure/azure-sdk-for-go/profiles/2019-03-01/authorization/mgmt/authorization"
@@ -49,6 +50,7 @@ type Spec struct {
 	OSDisk                 infrav1.OSDisk
 	CustomData             string
 	UserAssignedIdentities []infrav1.UserAssignedIdentity
+	SpotVMOptions          *infrav1.SpotVMOptions
 }
 
 // Get provides information about a virtual machine.
@@ -101,6 +103,11 @@ func (s *Service) Reconcile(ctx context.Context, spec interface{}) error {
 	// Set the cloud provider tag
 	additionalTags[infrav1.ClusterAzureCloudProviderTagKey(s.MachineScope.Name())] = string(infrav1.ResourceLifecycleOwned)
 
+	priority, evictionPolicy, billingProfile, err := getSpotVMOptions(vmSpec.SpotVMOptions)
+	if err != nil {
+		return errors.Wrapf(err, "failed to get Spot VM options")
+	}
+
 	virtualMachine := compute.VirtualMachine{
 		Location: to.StringPtr(s.Scope.Location()),
 		Tags: converters.TagsToMap(infrav1.Build(infrav1.BuildParams{
@@ -141,6 +148,9 @@ func (s *Service) Reconcile(ctx context.Context, spec interface{}) error {
 					},
 				},
 			},
+			Priority:       priority,
+			EvictionPolicy: evictionPolicy,
+			BillingProfile: billingProfile,
 		},
 	}
 
@@ -339,6 +349,24 @@ func generateStorageProfile(vmSpec Spec) (*compute.StorageProfile, error) {
 	return storageProfile, nil
 }
 
+func getSpotVMOptions(spotVMOptions *infrav1.SpotVMOptions) (compute.VirtualMachinePriorityTypes, compute.VirtualMachineEvictionPolicyTypes, *compute.BillingProfile, error) {
+	// Spot VM not requested, return zero values to apply defaults
+	if spotVMOptions == nil {
+		return compute.VirtualMachinePriorityTypes(""), compute.VirtualMachineEvictionPolicyTypes(""), nil, nil
+	}
+	var billingProfile *compute.BillingProfile
+	if spotVMOptions.MaxPrice != nil {
+		maxPrice, err := strconv.ParseFloat(*spotVMOptions.MaxPrice, 64)
+		if err != nil {
+			return compute.VirtualMachinePriorityTypes(""), compute.VirtualMachineEvictionPolicyTypes(""), nil, err
+		}
+		billingProfile = &compute.BillingProfile{
+			MaxPrice: &maxPrice,
+		}
+	}
+	return compute.Spot, compute.Deallocate, billingProfile, nil
+}
+
 // GenerateRandomString returns a URL-safe, base64 encoded
 // securely generated random string.
 // It will return an error if the system's secure random

cloud/services/virtualmachines/virtualmachines_test.go

@@ -343,8 +343,6 @@ func TestGetVM(t *testing.T) {
 }
 
 func TestReconcileVM(t *testing.T) {
-	g := NewWithT(t)
-
 	secret := &corev1.Secret{
 		ObjectMeta: metav1.ObjectMeta{
 			Name: "bootstrap-data",
@@ -368,7 +366,7 @@ func TestReconcileVM(t *testing.T) {
 		machine       clusterv1.Machine
 		machineConfig *infrav1.AzureMachineSpec
 		azureCluster  *infrav1.AzureCluster
-		expect        func(m *mock_virtualmachines.MockClientMockRecorder, mnic *mock_networkinterfaces.MockClientMockRecorder, mpip *mock_publicips.MockClientMockRecorder, mra *mock_roleassignments.MockClientMockRecorder)
+		expect        func(g *WithT, m *mock_virtualmachines.MockClientMockRecorder, mnic *mock_networkinterfaces.MockClientMockRecorder, mpip *mock_publicips.MockClientMockRecorder, mra *mock_roleassignments.MockClientMockRecorder)
 		expectedError string
 	}{
 		{
@@ -417,7 +415,7 @@ func TestReconcileVM(t *testing.T) {
 					},
 				},
 			},
-			expect: func(m *mock_virtualmachines.MockClientMockRecorder, mnic *mock_networkinterfaces.MockClientMockRecorder, mpip *mock_publicips.MockClientMockRecorder, mra *mock_roleassignments.MockClientMockRecorder) {
+			expect: func(g *WithT, m *mock_virtualmachines.MockClientMockRecorder, mnic *mock_networkinterfaces.MockClientMockRecorder, mpip *mock_publicips.MockClientMockRecorder, mra *mock_roleassignments.MockClientMockRecorder) {
 				mnic.Get(gomock.Any(), gomock.Any(), gomock.Any())
 				m.CreateOrUpdate(gomock.Any(), gomock.Any(), gomock.Any(), gomock.Any())
 			},
@@ -470,7 +468,7 @@ func TestReconcileVM(t *testing.T) {
 					},
 				},
 			},
-			expect: func(m *mock_virtualmachines.MockClientMockRecorder, mnic *mock_networkinterfaces.MockClientMockRecorder, mpip *mock_publicips.MockClientMockRecorder, mra *mock_roleassignments.MockClientMockRecorder) {
+			expect: func(g *WithT, m *mock_virtualmachines.MockClientMockRecorder, mnic *mock_networkinterfaces.MockClientMockRecorder, mpip *mock_publicips.MockClientMockRecorder, mra *mock_roleassignments.MockClientMockRecorder) {
 				mnic.Get(gomock.Any(), gomock.Any(), gomock.Any())
 				m.CreateOrUpdate(gomock.Any(), gomock.Any(), gomock.Any(), gomock.Any())
 				mra.Create(gomock.Any(), gomock.Any(), gomock.Any(), gomock.Any())
@@ -525,13 +523,70 @@ func TestReconcileVM(t *testing.T) {
 					},
 				},
 			},
-			expect: func(m *mock_virtualmachines.MockClientMockRecorder, mnic *mock_networkinterfaces.MockClientMockRecorder, mpip *mock_publicips.MockClientMockRecorder, mra *mock_roleassignments.MockClientMockRecorder) {
+			expect: func(g *WithT, m *mock_virtualmachines.MockClientMockRecorder, mnic *mock_networkinterfaces.MockClientMockRecorder, mpip *mock_publicips.MockClientMockRecorder, mra *mock_roleassignments.MockClientMockRecorder) {
 				mnic.Get(gomock.Any(), gomock.Any(), gomock.Any())
 				m.CreateOrUpdate(gomock.Any(), gomock.Any(), gomock.Any(), gomock.Any())
 				mra.Create(gomock.Any(), gomock.Any(), gomock.Any(), gomock.Any())
 			},
 			expectedError: "",
 		},
+		{
+			name: "can create a vm on spot",
+			machine: clusterv1.Machine{
+				ObjectMeta: metav1.ObjectMeta{
+					Labels: map[string]string{"set": "node"},
+				},
+				Spec: clusterv1.MachineSpec{
+					Bootstrap: clusterv1.Bootstrap{
+						Data: to.StringPtr("bootstrap-data"),
+					},
+					Version: to.StringPtr("1.15.7"),
+				},
+			},
+			machineConfig: &infrav1.AzureMachineSpec{
+				VMSize:        "Standard_B2ms",
+				Location:      "eastus",
+				Image:         image,
+				SpotVMOptions: &infrav1.SpotVMOptions{},
+			},
+			azureCluster: &infrav1.AzureCluster{
+				Spec: infrav1.AzureClusterSpec{
+					SubscriptionID: subscriptionID,
+					NetworkSpec: infrav1.NetworkSpec{
+						Subnets: infrav1.Subnets{
+							&infrav1.SubnetSpec{
+								Name: "subnet-1",
+							},
+							&infrav1.SubnetSpec{},
+						},
+					},
+				},
+				Status: infrav1.AzureClusterStatus{
+					Network: infrav1.Network{
+						SecurityGroups: map[infrav1.SecurityGroupRole]infrav1.SecurityGroup{
+							infrav1.SecurityGroupControlPlane: {
+								ID: "1",
+							},
+							infrav1.SecurityGroupNode: {
+								ID: "2",
+							},
+						},
+						APIServerIP: infrav1.PublicIP{
+							DNSName: "azure-test-dns",
+						},
+					},
+				},
+			},
+			expect: func(g *WithT, m *mock_virtualmachines.MockClientMockRecorder, mnic *mock_networkinterfaces.MockClientMockRecorder, mpip *mock_publicips.MockClientMockRecorder, mra *mock_roleassignments.MockClientMockRecorder) {
+				mnic.Get(gomock.Any(), gomock.Any(), gomock.Any())
+				m.CreateOrUpdate(gomock.Any(), gomock.Any(), gomock.Any(), gomock.Any()).Do(func(_, _, _ interface{}, vm compute.VirtualMachine) {
+					g.Expect(vm.Priority).To(Equal(compute.Spot))
+					g.Expect(vm.EvictionPolicy).To(Equal(compute.Deallocate))
+					g.Expect(vm.BillingProfile).To(BeNil())
+				})
+			},
+			expectedError: "",
+		},
 		{
 			name: "vm creation fails",
 			machine: clusterv1.Machine{
@@ -578,7 +633,7 @@ func TestReconcileVM(t *testing.T) {
 					},
 				},
 			},
-			expect: func(m *mock_virtualmachines.MockClientMockRecorder, mnic *mock_networkinterfaces.MockClientMockRecorder, mpip *mock_publicips.MockClientMockRecorder, mra *mock_roleassignments.MockClientMockRecorder) {
+			expect: func(g *WithT, m *mock_virtualmachines.MockClientMockRecorder, mnic *mock_networkinterfaces.MockClientMockRecorder, mpip *mock_publicips.MockClientMockRecorder, mra *mock_roleassignments.MockClientMockRecorder) {
 				mnic.Get(gomock.Any(), gomock.Any(), gomock.Any())
 				m.CreateOrUpdate(gomock.Any(), gomock.Any(), gomock.Any(), gomock.Any()).Return(autorest.NewErrorWithResponse("", "", &http.Response{StatusCode: 500}, "Internal Server Error"))
 			},
@@ -588,6 +643,8 @@ func TestReconcileVM(t *testing.T) {
 
 	for _, tc := range testcases {
 		t.Run(tc.name, func(t *testing.T) {
+			g := NewWithT(t)
+
 			mockCtrl := gomock.NewController(t)
 			vmMock := mock_virtualmachines.NewMockClient(mockCtrl)
 			interfaceMock := mock_networkinterfaces.NewMockClient(mockCtrl)
@@ -639,7 +696,7 @@ func TestReconcileVM(t *testing.T) {
 			g.Expect(err).NotTo(HaveOccurred())
 
 			machineScope.AzureMachine.Spec = *tc.machineConfig
-			tc.expect(vmMock.EXPECT(), interfaceMock.EXPECT(), publicIPMock.EXPECT(), roleAssignmentMock.EXPECT())
+			tc.expect(g, vmMock.EXPECT(), interfaceMock.EXPECT(), publicIPMock.EXPECT(), roleAssignmentMock.EXPECT())
 
 			clusterScope, err := scope.NewClusterScope(scope.ClusterScopeParams{
 				AzureClients: scope.AzureClients{
@@ -661,13 +718,14 @@ func TestReconcileVM(t *testing.T) {
 			}
 
 			vmSpec := &Spec{
-				Name:       machineScope.Name(),
-				NICName:    "test-nic",
-				SSHKeyData: "fake-key",
-				Size:       machineScope.AzureMachine.Spec.VMSize,
-				OSDisk:     machineScope.AzureMachine.Spec.OSDisk,
-				Image:      machineScope.AzureMachine.Spec.Image,
-				CustomData: *machineScope.Machine.Spec.Bootstrap.Data,
+				Name:          machineScope.Name(),
+				NICName:       "test-nic",
+				SSHKeyData:    "fake-key",
+				Size:          machineScope.AzureMachine.Spec.VMSize,
+				OSDisk:        machineScope.AzureMachine.Spec.OSDisk,
+				Image:         machineScope.AzureMachine.Spec.Image,
+				CustomData:    *machineScope.Machine.Spec.Bootstrap.Data,
+				SpotVMOptions: machineScope.AzureMachine.Spec.SpotVMOptions,
 			}
 			err = s.Reconcile(context.TODO(), vmSpec)
 			if tc.expectedError != "" {

config/crd/bases/infrastructure.cluster.x-k8s.io_azuremachines.yaml

@@ -390,6 +390,15 @@ spec:
                 description: ProviderID is the unique identifier as specified by the
                   cloud provider.
                 type: string
+              spotVMOptions:
+                description: SpotVMOptions allows the ability to specify the Machine
+                  should use a Spot VM
+                properties:
+                  maxPrice:
+                    description: MaxPrice defines the maximum price the user is willing
+                      to pay for Spot VM instances
+                    type: number
+                type: object
               sshPublicKey:
                 type: string
               userAssignedIdentities:

config/crd/bases/infrastructure.cluster.x-k8s.io_azuremachinetemplates.yaml

@@ -334,6 +334,15 @@ spec:
                         description: ProviderID is the unique identifier as specified
                           by the cloud provider.
                         type: string
+                      spotVMOptions:
+                        description: SpotVMOptions allows the ability to specify the
+                          Machine should use a Spot VM
+                        properties:
+                          maxPrice:
+                            description: MaxPrice defines the maximum price the user
+                              is willing to pay for Spot VM instances
+                            type: number
+                        type: object
                       sshPublicKey:
                         type: string
                       userAssignedIdentities:

controllers/azuremachine_reconciler.go

@@ -271,6 +271,7 @@ func (s *azureMachineService) reconcileVirtualMachine(ctx context.Context, nicNa
 		Zone:                   vmZone,
 		Identity:               s.machineScope.AzureMachine.Spec.Identity,
 		UserAssignedIdentities: s.machineScope.AzureMachine.Spec.UserAssignedIdentities,
+		SpotVMOptions:          s.machineScope.AzureMachine.Spec.SpotVMOptions,
 	}
 
 	err = s.virtualMachinesSvc.Reconcile(ctx, vmSpec)