Fix GPU cluster creation and resize (#182)

* Fix GPU cluster creation and resize

* Update gcore_ai_cluster docs

* Prevent resize to 0 instances

* Update gcorelabscloud-go to v0.11.0

docs/resources/ai_cluster.md

@@ -23,7 +23,8 @@ resource "gcore_ai_cluster" "cluster1" {
   image_id = "f6aa6e75-ab88-4c19-889d-79133366cb83"
   cluster_name      = "cluster1"
   keypair_name = "front"
-
+  instances_count = 1
+  
   volume {
     source     = "image"
     image_id = "f6aa6e75-ab88-4c19-889d-79133366cb83"
@@ -69,6 +70,7 @@ resource "gcore_ai_cluster" "cluster1" {
 - `user_data` (String) String in base64 format. Must not be passed together with 'username' or 'password'. Examples of the user_data: https://cloudinit.readthedocs.io/en/latest/topics/examples.html
 - `username` (String) A name of a new user in the Linux instance. It may be passed with a 'password' parameter
 - `volume` (Block Set) List of volumes attached to the cluster (see [below for nested schema](#nestedblock--volume))
+- `instances_count` (Number) Number of instances in the cluster
 
 ### Read-Only
 

gcore/resource_gcore_ai_cluster.go

@@ -29,8 +29,9 @@ const (
 	AIClusterCreatingTimeout int = 1200
 	AIClusterSuspendTimeout  int = 300
 
-	AIClusterPoint = "ai/clusters"
-	TaskPoint      = "tasks"
+	AIClusterPoint    = "ai/clusters"
+	AIClusterGPUPoint = "ai/clusters/gpu"
+	TaskPoint         = "tasks"
 )
 
 const (
@@ -109,6 +110,11 @@ func resourceAICluster() *schema.Resource {
 				Description: "AI Cluster Name",
 				Required:    true,
 			},
+			"instances_count": {
+				Type:        schema.TypeInt,
+				Description: "Number of instances to create",
+				Optional:    true,
+			},
 			"cluster_status": {
 				Type:        schema.TypeString,
 				Description: "AI Cluster status",
@@ -190,7 +196,6 @@ func resourceAICluster() *schema.Resource {
 						"size": {
 							Type:        schema.TypeInt,
 							Description: "Volume size, GiB",
-							Computed:    true,
 							Optional:    true,
 						},
 						"created_at": {
@@ -212,7 +217,7 @@ func resourceAICluster() *schema.Resource {
 						"image_id": {
 							Type:        schema.TypeString,
 							Description: "Volume ID. Mandatory if volume is pre-existing volume",
-							Optional:    true,
+							Required:    true,
 						},
 						"attachments": {
 							Type:        schema.TypeSet,
@@ -538,9 +543,6 @@ func checkAIClusterStatus(client *gcorecloud.ServiceClient, clusterID string, de
 }
 
 func validateCreateOpts(createOpts *ai.CreateOpts) error {
-	if isBmFlavor(createOpts.Flavor) && len(createOpts.Volumes) > 0 {
-		return errors.New("volumes are not supported for baremetal poplar servers")
-	}
 	if !isBmFlavor(createOpts.Flavor) && len(createOpts.Volumes) == 0 {
 		return errors.New("at least one image volume is required for vm poplar cluster")
 	}
@@ -590,7 +592,7 @@ func resourceAIClusterCreate(ctx context.Context, d *schema.ResourceData, m inte
 	config := m.(*Config)
 	provider := config.Provider
 
-	client, err := CreateClient(provider, d, AIClusterPoint, versionPointV1)
+	client, err := CreateClient(provider, d, AIClusterGPUPoint, versionPointV1)
 	if err != nil {
 		return diag.FromErr(err)
 	}
@@ -609,6 +611,10 @@ func resourceAIClusterCreate(ctx context.Context, d *schema.ResourceData, m inte
 	createOpts.Keypair = d.Get("keypair_name").(string)
 	createOpts.ImageID = d.Get("image_id").(string)
 
+	if instancesCount, ok := d.GetOk("instances_count"); ok {
+		createOpts.InstancesCount = instancesCount.(int)
+	}
+
 	if userData, ok := d.GetOk("userdata"); ok {
 		createOpts.UserData = userData.(string)
 	}
@@ -658,10 +664,8 @@ func resourceAIClusterCreate(ctx context.Context, d *schema.ResourceData, m inte
 		if err != nil {
 			return nil, fmt.Errorf("cannot get task with ID: %s. Error: %w", task, err)
 		}
-		clusterID, err := ai.ExtractAIClusterIDFromTask(taskInfo)
-		if err != nil {
-			return nil, fmt.Errorf("cannot retrieve AI cluster ID from task info: %w", err)
-		}
+		// on create task, the cluster_id matches the task_id
+		clusterID := taskInfo.ID
 		return clusterID, nil
 	},
 	)
@@ -805,7 +809,7 @@ func getDetachOptions(instanceInterfaces []instances.Interface, detachIface ai.A
 	}
 }
 
-var IsResize bool = false
+var IsResize = false
 
 func resourceAIClusterUpdate(ctx context.Context, d *schema.ResourceData, m interface{}) diag.Diagnostics {
 	log.Println("[DEBUG] Start AI cluster updating")
@@ -815,6 +819,10 @@ func resourceAIClusterUpdate(ctx context.Context, d *schema.ResourceData, m inte
 	if err != nil {
 		return diag.FromErr(err)
 	}
+	clientV1GPU, err := CreateClient(provider, d, AIClusterGPUPoint, versionPointV1)
+	if err != nil {
+		return diag.FromErr(err)
+	}
 	clientV2, err := CreateClient(provider, d, AIClusterPoint, versionPointV2)
 	if err != nil {
 		return diag.FromErr(err)
@@ -866,61 +874,26 @@ func resourceAIClusterUpdate(ctx context.Context, d *schema.ResourceData, m inte
 	}
 
 	// Make resize
-	if d.HasChanges("flavor", "image_id", "keypair_name", "user_data", "username", "password") || (d.HasChanges("interface") && isBmFlavor(d.Get("flavor").(string))) {
+	if d.HasChanges("instances_count") {
 		IsResize = true
 		_, newSGs := d.GetChange("security_group")
 		securityGroupList := newSGs.(*schema.Set).List()
 		securityGroupIDs := make([]gcorecloud.ItemID, len(securityGroupList))
 		for sgIndex, sgID := range securityGroupList {
 			securityGroupIDs[sgIndex] = gcorecloud.ItemID{ID: sgID.(map[string]interface{})["id"].(string)}
 		}
-		_, flavor := d.GetChange("flavor")
-		_, image_id := d.GetChange("image_id")
-		_, keypairName := d.GetChange("keypair_name")
-		_, userData := d.GetChange("user_data")
-		_, username := d.GetChange("username")
-		_, password := d.GetChange("password")
 
-		resizeOpts := ai.ResizeAIClusterOpts{
-			Flavor:         flavor.(string),
-			ImageID:        image_id.(string),
-			Interfaces:     []instances.InterfaceInstanceCreateOpts{},
-			Volumes:        []instances.CreateVolumeOpts{},
-			SecurityGroups: securityGroupIDs,
-			Keypair:        keypairName.(string),
-			Password:       password.(string),
-			Username:       username.(string),
-			UserData:       userData.(string),
-			Metadata:       map[string]string{},
+		instancesCount, ok := d.GetOk("instances_count")
+		if !ok || instancesCount.(int) == 0 {
+			// if the number of instances has been specified before, then it cannot be removed or set to 0
+			return diag.FromErr(errors.New("cannot resize cluster to 0 instances"))
 		}
-		_, newVolumes := d.GetChange("volume")
 
-		volumeList := newVolumes.([]interface{})
-		if len(volumeList) > 0 {
-			vs, err := extractVolumesMap(volumeList)
-			if err != nil {
-				return diag.FromErr(err)
-			}
-			resizeOpts.Volumes = vs
+		resizeOpts := ai.ResizeGPUAIClusterOpts{
+			InstancesCount: instancesCount.(int),
 		}
-
-		_, newIface := d.GetChange("interface")
-		interfaceList := newIface.([]interface{})
-		if len(interfaceList) > 0 {
-			ifaces, err := extractAIClusterInterfacesMap(interfaceList)
-			if err != nil {
-				return diag.FromErr(err)
-			}
-			resizeOpts.Interfaces = ifaces
-		}
-		_, newMetadata := d.GetChange("cluster_metadata")
-
-		for metaKey, metaValue := range newMetadata.(map[string]interface{}) {
-			resizeOpts.Metadata[metaKey] = metaValue.(string)
-		}
-
 		log.Printf("[DEBUG] AI cluster resize options: %+v", resizeOpts)
-		results, err := ai.Resize(clientV1, clusterID, resizeOpts).Extract()
+		results, err := ai.Resize(clientV1GPU, clusterID, resizeOpts).Extract()
 		if err != nil {
 			return diag.FromErr(err)
 		}
@@ -943,7 +916,7 @@ func resourceAIClusterUpdate(ctx context.Context, d *schema.ResourceData, m inte
 		oldVolumeList := extractInstanceVolumesMap(oldVolumes.(*schema.Set).List())
 		newVolumeList := extractInstanceVolumesMap(newVolumes.(*schema.Set).List())
 		if isBmFlavor(d.Get("flavor").(string)) && len(newVolumeList) > 0 {
-			return diag.FromErr(errors.New("baremetal servers don't support external voluems"))
+			return diag.FromErr(errors.New("baremetal servers don't support external volumes"))
 		}
 		poplarInstances := d.Get("poplar_servers").([]interface{})
 		if len(poplarInstances) > 1 {

go.mod

@@ -9,7 +9,7 @@ require (
 	github.com/G-Core/gcore-dns-sdk-go v0.2.9
 	github.com/G-Core/gcore-storage-sdk-go v0.1.34
 	github.com/G-Core/gcorelabscdn-go v1.0.25
-	github.com/G-Core/gcorelabscloud-go v0.10.1
+	github.com/G-Core/gcorelabscloud-go v0.11.0
 	github.com/hashicorp/go-cty v1.4.1-0.20200414143053-d3edf31b6320
 	github.com/hashicorp/terraform-plugin-sdk/v2 v2.27.0
 	github.com/mitchellh/mapstructure v1.5.0
@@ -53,7 +53,6 @@ require (
 	github.com/hashicorp/hcl/v2 v2.17.0 // indirect
 	github.com/hashicorp/logutils v1.0.0 // indirect
 	github.com/hashicorp/terraform-json v0.23.0 // indirect
-	github.com/hashicorp/terraform-plugin-docs v0.20.1 // indirect
 	github.com/hashicorp/terraform-plugin-go v0.16.0 // indirect
 	github.com/hashicorp/terraform-plugin-log v0.9.0 // indirect
 	github.com/hashicorp/terraform-registry-address v0.2.1 // indirect