Ibmcloud: add dedicated host support

src/cloud-api-adaptor/install/overlays/ibmcloud/kustomization.yaml

@@ -33,6 +33,8 @@ configMapGenerator:
   - DISABLECVM="true" # Set to false to enable confidential VM
   - INITDATA="" # set default initdata for podvm
   #- TAGS="" # Uncomment and add key1=value1,key2=value2 etc if you want to use specific tags for podvm
+  #- IBMCLOUD_DEDICATED_HOST_IDS="" # Uncomment and set if you want to use dedicated hosts, comma separated list
+  #- IBMCLOUD_DEDICATED_HOST_GROUP_IDS="" # Uncomment and set if you want to use dedicated host groups, comma separated list
   #- PAUSE_IMAGE="" # Uncomment and set if you want to use a specific pause image
   #- TUNNEL_TYPE="" # Uncomment and set if you want to use a specific tunnel type. Defaults to vxlan
   #- VXLAN_PORT="" # Uncomment and set if you want to use a specific vxlan port. Defaults to 4789

src/cloud-api-adaptor/test/provisioner/ibmcloud/provision_common.go

@@ -980,6 +980,8 @@ func (p *IBMCloudProvisioner) GetProperties(ctx context.Context, cfg *envconf.Co
 		"INITDATA":                             IBMCloudProps.InitData,
 		"IBMCLOUD_CLUSTER_ID":                  IBMCloudProps.ClusterID,
 		"TAGS":                                 IBMCloudProps.Tags,
+		"IBMCLOUD_DEDICATED_HOST_IDS":          IBMCloudProps.DedicatedHostIDs,
+		"IBMCLOUD_DEDICATED_HOST_GROUP_IDS":    IBMCloudProps.DedicatedHostGroupIDs,
 	}
 }
 

src/cloud-api-adaptor/test/provisioner/ibmcloud/provision_ibmcloud.properties

@@ -13,6 +13,10 @@ COS_INSTANCE_ID="crn:v1:bluemix:public:cloud-object-storage:global:a/xxxxxxxxxxx
 COS_SERVICE_URL="s3.jp-tok.cloud-object-storage.appdomain.cloud"
 # Resource list -> storage -> a cos service -> instances -> an cos service instance -> service credentials -> apikey
 COS_APIKEY="${MY_COS_SERVICE_APIKEY}"
+# optional
+# DEDICATED_HOST_IDS=""
+# optional
+# DEDICATED_HOST_GROUP_IDS=""
 IS_SELF_MANAGED_CLUSTER="no"
 # bz2-2x8 | bx2-2x8 | bz2e-2x8
 INSTANCE_PROFILE_NAME="bz2-2x8"

src/cloud-api-adaptor/test/provisioner/ibmcloud/provision_initializer.go

@@ -19,42 +19,44 @@ import (
 )
 
 type IBMCloudProperties struct {
-	IBMCloudProvider  string
-	ApiKey            string
-	IamProfileID      string
-	Bucket            string
-	CaaImageTag       string
-	ClusterName       string
-	CosApiKey         string
-	CosInstanceID     string
-	CosServiceURL     string
-	SecurityGroupID   string
-	IamServiceURL     string
-	IksServiceURL     string
-	InitData          string
-	InstanceProfile   string
-	KubeVersion       string
-	PodvmImageID      string
-	PodvmImageArch    string
-	PublicGatewayName string
-	PublicGatewayID   string
-	Region            string
-	ResourceGroupID   string
-	SshKeyContent     string
-	SshKeyID          string
-	SshKeyName        string
-	SubnetName        string
-	SubnetID          string
-	VpcName           string
-	VpcID             string
-	VpcServiceURL     string
-	WorkerFlavor      string
-	WorkerOS          string
-	Zone              string
-	TunnelType        string
-	VxlanPort         string
-	ClusterID         string
-	Tags              string
+	IBMCloudProvider      string
+	ApiKey                string
+	IamProfileID          string
+	Bucket                string
+	CaaImageTag           string
+	ClusterName           string
+	CosApiKey             string
+	CosInstanceID         string
+	CosServiceURL         string
+	SecurityGroupID       string
+	IamServiceURL         string
+	IksServiceURL         string
+	InitData              string
+	InstanceProfile       string
+	KubeVersion           string
+	PodvmImageID          string
+	PodvmImageArch        string
+	PublicGatewayName     string
+	PublicGatewayID       string
+	Region                string
+	ResourceGroupID       string
+	SshKeyContent         string
+	SshKeyID              string
+	SshKeyName            string
+	SubnetName            string
+	SubnetID              string
+	VpcName               string
+	VpcID                 string
+	VpcServiceURL         string
+	WorkerFlavor          string
+	WorkerOS              string
+	Zone                  string
+	TunnelType            string
+	VxlanPort             string
+	ClusterID             string
+	Tags                  string
+	DedicatedHostIDs      string
+	DedicatedHostGroupIDs string
 
 	WorkerCount   int
 	IsSelfManaged bool
@@ -68,41 +70,43 @@ var IBMCloudProps = &IBMCloudProperties{}
 
 func InitIBMCloudProperties(properties map[string]string) error {
 	IBMCloudProps = &IBMCloudProperties{
-		IBMCloudProvider:  properties["IBMCLOUD_PROVIDER"],
-		ApiKey:            properties["APIKEY"],
-		IamProfileID:      properties["IAM_PROFILE_ID"],
-		Bucket:            properties["COS_BUCKET"],
-		CaaImageTag:       properties["CAA_IMAGE_TAG"],
-		ClusterName:       properties["CLUSTER_NAME"],
-		CosApiKey:         properties["COS_APIKEY"],
-		CosInstanceID:     properties["COS_INSTANCE_ID"],
-		CosServiceURL:     properties["COS_SERVICE_URL"],
-		IamServiceURL:     properties["IAM_SERVICE_URL"],
-		IksServiceURL:     properties["IKS_SERVICE_URL"],
-		InitData:          properties["INITDATA"],
-		InstanceProfile:   properties["INSTANCE_PROFILE_NAME"],
-		KubeVersion:       properties["KUBE_VERSION"],
-		PodvmImageID:      properties["PODVM_IMAGE_ID"],
-		PodvmImageArch:    properties["PODVM_IMAGE_ARCH"],
-		PublicGatewayName: properties["PUBLIC_GATEWAY_NAME"],
-		Region:            properties["REGION"],
-		ResourceGroupID:   properties["RESOURCE_GROUP_ID"],
-		SshKeyName:        properties["SSH_KEY_NAME"],
-		SshKeyContent:     properties["SSH_PUBLIC_KEY_CONTENT"],
-		SubnetName:        properties["VPC_SUBNET_NAME"],
-		VpcName:           properties["VPC_NAME"],
-		VpcServiceURL:     properties["VPC_SERVICE_URL"],
-		WorkerFlavor:      properties["WORKER_FLAVOR"],
-		WorkerOS:          properties["WORKER_OPERATION_SYSTEM"],
-		Zone:              properties["ZONE"],
-		SshKeyID:          properties["SSH_KEY_ID"],
-		SubnetID:          properties["VPC_SUBNET_ID"],
-		SecurityGroupID:   properties["VPC_SECURITY_GROUP_ID"],
-		VpcID:             properties["VPC_ID"],
-		TunnelType:        properties["TUNNEL_TYPE"],
-		VxlanPort:         properties["VXLAN_PORT"],
-		ClusterID:         properties["CLUSTER_ID"],
-		Tags:              properties["TAGS"],
+		IBMCloudProvider:      properties["IBMCLOUD_PROVIDER"],
+		ApiKey:                properties["APIKEY"],
+		IamProfileID:          properties["IAM_PROFILE_ID"],
+		Bucket:                properties["COS_BUCKET"],
+		CaaImageTag:           properties["CAA_IMAGE_TAG"],
+		ClusterName:           properties["CLUSTER_NAME"],
+		CosApiKey:             properties["COS_APIKEY"],
+		CosInstanceID:         properties["COS_INSTANCE_ID"],
+		CosServiceURL:         properties["COS_SERVICE_URL"],
+		IamServiceURL:         properties["IAM_SERVICE_URL"],
+		IksServiceURL:         properties["IKS_SERVICE_URL"],
+		InitData:              properties["INITDATA"],
+		InstanceProfile:       properties["INSTANCE_PROFILE_NAME"],
+		KubeVersion:           properties["KUBE_VERSION"],
+		PodvmImageID:          properties["PODVM_IMAGE_ID"],
+		PodvmImageArch:        properties["PODVM_IMAGE_ARCH"],
+		PublicGatewayName:     properties["PUBLIC_GATEWAY_NAME"],
+		Region:                properties["REGION"],
+		ResourceGroupID:       properties["RESOURCE_GROUP_ID"],
+		SshKeyName:            properties["SSH_KEY_NAME"],
+		SshKeyContent:         properties["SSH_PUBLIC_KEY_CONTENT"],
+		SubnetName:            properties["VPC_SUBNET_NAME"],
+		VpcName:               properties["VPC_NAME"],
+		VpcServiceURL:         properties["VPC_SERVICE_URL"],
+		WorkerFlavor:          properties["WORKER_FLAVOR"],
+		WorkerOS:              properties["WORKER_OPERATION_SYSTEM"],
+		Zone:                  properties["ZONE"],
+		SshKeyID:              properties["SSH_KEY_ID"],
+		SubnetID:              properties["VPC_SUBNET_ID"],
+		SecurityGroupID:       properties["VPC_SECURITY_GROUP_ID"],
+		VpcID:                 properties["VPC_ID"],
+		TunnelType:            properties["TUNNEL_TYPE"],
+		VxlanPort:             properties["VXLAN_PORT"],
+		ClusterID:             properties["CLUSTER_ID"],
+		Tags:                  properties["TAGS"],
+		DedicatedHostIDs:      properties["DEDICATED_HOST_IDS"],
+		DedicatedHostGroupIDs: properties["DEDICATED_HOST_GROUP_IDS"],
 	}
 
 	if len(IBMCloudProps.IBMCloudProvider) <= 0 {

src/cloud-api-adaptor/test/provisioner/ibmcloud/provision_kustomize.go

@@ -67,6 +67,10 @@ func isKustomizeConfigMapKey(key string) bool {
 		return true
 	case "TAGS":
 		return true
+	case "IBMCLOUD_DEDICATED_HOST_IDS":
+		return true
+	case "IBMCLOUD_DEDICATED_HOST_GROUP_IDS":
+		return true
 	default:
 		return false
 	}

src/cloud-providers/ibmcloud/manager.go

@@ -45,6 +45,8 @@ func (_ *Manager) ParseCmd(flags *flag.FlagSet) {
 	reg.CustomTypeWithEnv(&ibmcloudVPCConfig.InstanceProfiles, "profile-list", "", "IBMCLOUD_PODVM_INSTANCE_PROFILE_LIST", "List of instance profile names to be used for the Pod VMs, comma separated")
 	reg.CustomTypeWithEnv(&ibmcloudVPCConfig.Images, "image-id", "", "IBMCLOUD_PODVM_IMAGE_ID", "List of Image IDs, comma separated", provider.Required())
 	reg.CustomTypeWithEnv(&ibmcloudVPCConfig.Tags, "tags", "", "TAGS", "List of tags to attach to the Pod VMs, comma separated")
+	reg.CustomTypeWithEnv(&ibmcloudVPCConfig.DedicatedHostIDs, "dedicated-host-ids", "", "IBMCLOUD_DEDICATED_HOST_IDS", "List of Dedicated Host IDs, provide one from each Zone")
+	reg.CustomTypeWithEnv(&ibmcloudVPCConfig.DedicatedHostGroupIDs, "dedicated-host-group-ids", "", "IBMCLOUD_DEDICATED_HOST_GROUP_IDS", "List of Dedicated Host Group IDs, provide one from each Zone")
 }
 
 func (_ *Manager) LoadEnv() {

src/cloud-providers/ibmcloud/provider.go

@@ -136,6 +136,37 @@ func NewProvider(config *Config) (provider.Provider, error) {
 		}
 	}
 
+	// Return error early
+	if config.ZoneName == "" {
+		return nil, fmt.Errorf("zone was not provided and could not detect automatically")
+	}
+
+	if len(config.DedicatedHostIDs) > 0 {
+		selected, err := pickIDInZone(
+			config.DedicatedHostIDs,
+			config.ZoneName,
+			func(id string) (string, error) { return getDedicatedHostZone(vpcV1, id) },
+			"Dedicated Host",
+		)
+		if err != nil {
+			return nil, err
+		}
+		config.selectedDedicatedHostID = selected
+	}
+
+	if len(config.DedicatedHostGroupIDs) > 0 {
+		selected, err := pickIDInZone(
+			config.DedicatedHostGroupIDs,
+			config.ZoneName,
+			func(id string) (string, error) { return getDedicatedHostGroupZone(vpcV1, id) },
+			"Dedicated Host Group",
+		)
+		if err != nil {
+			return nil, err
+		}
+		config.selectedDedicatedHostGroupID = selected
+	}
+
 	gTaggingV1, err := globaltaggingv1.NewGlobalTaggingV1(
 		&globaltaggingv1.GlobalTaggingV1Options{
 			Authenticator: authenticator,
@@ -209,6 +240,59 @@ func fetchVPCDetails(vpcV1 *vpcv1.VpcV1, subnetID string) (vpcID string, resourc
 	return
 }
 
+func getDedicatedHostZone(vpcV1 *vpcv1.VpcV1, dedicatedHostID string) (string, error) {
+	dedicatedHostOptions := vpcv1.GetDedicatedHostOptions{
+		ID: &dedicatedHostID,
+	}
+	dedicatedHost, response, err := vpcV1.GetDedicatedHost(&dedicatedHostOptions)
+	if err != nil {
+		return "", fmt.Errorf("VPC error with:\n %w\nfurther details:\n %v", err, response)
+	}
+
+	return *dedicatedHost.Zone.Name, nil
+}
+
+func getDedicatedHostGroupZone(vpcV1 *vpcv1.VpcV1, dedicatedHostGroupID string) (string, error) {
+	dedicatedHostGroupOptions := vpcv1.GetDedicatedHostGroupOptions{
+		ID: &dedicatedHostGroupID,
+	}
+	dedicatedHostGroup, response, err := vpcV1.GetDedicatedHostGroup(&dedicatedHostGroupOptions)
+	if err != nil {
+		return "", fmt.Errorf("VPC error with:\n %w\nfurther details:\n %v", err, response)
+	}
+
+	return *dedicatedHostGroup.Zone.Name, nil
+}
+
+// pickIDInZone finds the first ID whose zone equals zoneName.
+// If multiple IDs match the zone, it logs a warning and returns the first.
+// If none match, it returns a descriptive error.
+func pickIDInZone(ids []string, zoneName string, getZone func(string) (string, error), resourceLabel string) (string, error) {
+	var selected string
+
+	for _, id := range ids {
+		zone, err := getZone(id)
+		if err != nil {
+			return "", fmt.Errorf("couldn't get %s %s's zone: %w", id, resourceLabel, err)
+		}
+		if zone == zoneName {
+			if selected != "" && logger != nil {
+				logger.Printf("warning: multiple %ss were provided in zone %s; only one will be used",
+					resourceLabel, zoneName)
+				// Continue to keep the first match as the selected one.
+				continue
+			}
+			selected = id
+		}
+	}
+
+	if selected == "" {
+		return "", fmt.Errorf("no %s in zone %s was provided; please provide a %s in the specified zone for High Availability",
+			resourceLabel, zoneName, resourceLabel)
+	}
+	return selected, nil
+}
+
 func (p *ibmcloudVPCProvider) getAttachTagOptions(vpcInstanceCRN *string) (*globaltaggingv1.AttachTagOptions, error) {
 	if vpcInstanceCRN == nil {
 		return nil, fmt.Errorf("missing vpc instance crn, can't create attach tag options")
@@ -277,6 +361,15 @@ func (p *ibmcloudVPCProvider) getInstancePrototype(instanceName, userData, insta
 		}
 	}
 
+	// When both dedicated host id and group id provided the (more specific) dedicated host id will be used as the placement target
+	if p.serviceConfig.selectedDedicatedHostGroupID != "" {
+		prototype.PlacementTarget = &vpcv1.InstancePlacementTargetPrototypeDedicatedHostGroupIdentityDedicatedHostGroupIdentityByID{ID: &p.serviceConfig.selectedDedicatedHostGroupID}
+	}
+
+	if p.serviceConfig.selectedDedicatedHostID != "" {
+		prototype.PlacementTarget = &vpcv1.InstancePlacementTargetPrototypeDedicatedHostGroupIdentityDedicatedHostGroupIdentityByID{ID: &p.serviceConfig.selectedDedicatedHostID}
+	}
+
 	return prototype
 }
 
@@ -345,9 +438,8 @@ func (p *ibmcloudVPCProvider) CreateInstance(ctx context.Context, podName, sandb
 
 	logger.Printf("CreateInstance: name: %q", instanceName)
 
-	vpcInstance, resp, err := p.vpc.CreateInstanceWithContext(ctx, &vpcv1.CreateInstanceOptions{InstancePrototype: prototype})
+	vpcInstance, err := p.createInstanceWithFallback(ctx, prototype)
 	if err != nil {
-		logger.Printf("failed to create an instance : %v and the response is %s", err, resp)
 		return nil, err
 	}
 
@@ -390,7 +482,7 @@ func (p *ibmcloudVPCProvider) CreateInstance(ctx context.Context, podName, sandb
 		return instance, fmt.Errorf("failed to get attach tag options: %w", err)
 	}
 
-	_, resp, err = p.globalTagging.AttachTagWithContext(ctx, options)
+	_, resp, err := p.globalTagging.AttachTagWithContext(ctx, options)
 	if err != nil {
 		return instance, fmt.Errorf("failed to attach tags: %w and the response is %s", err, resp)
 	}
@@ -399,6 +491,44 @@ func (p *ibmcloudVPCProvider) CreateInstance(ctx context.Context, podName, sandb
 	return instance, nil
 }
 
+func (p *ibmcloudVPCProvider) createInstanceWithFallback(ctx context.Context, prototype *vpcv1.InstancePrototype) (*vpcv1.Instance, error) {
+
+	dedicatedHostID := p.serviceConfig.selectedDedicatedHostID
+	dedicatedHostGroupID := p.serviceConfig.selectedDedicatedHostGroupID
+
+	inst, resp, err := p.vpc.CreateInstanceWithContext(ctx, &vpcv1.CreateInstanceOptions{
+		InstancePrototype: prototype,
+	})
+	if err == nil {
+		return inst, nil
+	}
+
+	// Fallback if both IDs exist
+	if dedicatedHostID != "" && dedicatedHostGroupID != "" {
+		logger.Printf("creation failed on dedicated host %q: %v; retrying on dedicated host group %q", dedicatedHostID, err, dedicatedHostGroupID)
+
+		prototype.PlacementTarget =
+			&vpcv1.InstancePlacementTargetPrototypeDedicatedHostGroupIdentityDedicatedHostGroupIdentityByID{
+				ID: &dedicatedHostGroupID,
+			}
+
+		inst2, resp2, err2 := p.vpc.CreateInstanceWithContext(ctx, &vpcv1.CreateInstanceOptions{
+			InstancePrototype: prototype,
+		})
+		if err2 == nil {
+			return inst2, nil
+		}
+
+		// Return both errors for context.
+		return nil, errors.Join(
+			fmt.Errorf("instance creation on dedicated host %q failed: %w and the response is %s", dedicatedHostID, err, resp),
+			fmt.Errorf("fallback instance creation on dedicated host group %q failed: %w and the response is %s", dedicatedHostGroupID, err2, resp2),
+		)
+	}
+
+	return nil, fmt.Errorf("failed to create an instance: %w and the response is %s", err, resp)
+}
+
 // Select an instance profile based on the memory and vcpu requirements
 func (p *ibmcloudVPCProvider) selectInstanceProfile(ctx context.Context, spec provider.InstanceTypeSpec) (string, error) {