fix(azure): set DiskControllerType in StorageProfile for RHEL AI VMs

- Add DiskControllerType field to ImageReference; rhelai sets it to "SCSI"
  as a fallback when the gallery API is unavailable
- Wire ImageReference.DiskControllerType into StorageProfile in
  virtual-machine.go via diskControllerTypePtr helper
- Add GetSharedImageDiskControllerTypes to read disk controller types from
  gallery image definition features (not the version)
- Add FilterComputeSizesByDiskControllerType combining location and disk
  controller type filtering in a single SKU enumeration; replaces the
  previous two-pass approach in allocation.go
- resolveImageRef enriches DiskControllerType from gallery API only when
  the image supports exactly one type (capability != requirement)
- Filter both spot and non-spot compute sizes; return a clear error if no
  compatible sizes remain after filtering

Co-Authored-By: Claude <Sonnet 4.6> <noreply@anthropic.com>
Signed-off-by: Rishabh Kothari <rkothari@redhat.com>

Author: rishupk

pkg/provider/azure/action/rhel-ai/rhelai.go

@@ -46,7 +46,8 @@ func Create(mCtxArgs *maptContext.ContextArgs, args *apiRHELAI.RHELAIArgs) (err
 			ComputeRequest: args.ComputeRequest,
 			Spot:           args.Spot,
 			ImageRef: &data.ImageReference{
-				SharedImageID: sharedImageID,
+				SharedImageID:      sharedImageID,
+				DiskControllerType: "SCSI",
 			},
 			Username:         username,
 			ReadinessCommand: command.CommandPing}

pkg/provider/azure/data/compute-request.go

@@ -66,6 +66,81 @@ func FilterComputeSizesByLocation(ctx context.Context, location *string, compute
 	return supportedSizes, nil
 }
 
+// FilterComputeSizesByDiskControllerType returns the subset of computeSizes that are
+// available in location AND support requiredType. Sizes without a DiskControllerTypes
+// capability are assumed to support only SCSI (Azure historical default).
+func FilterComputeSizesByDiskControllerType(ctx context.Context, location *string, computeSizes []string, requiredType string) ([]string, error) {
+	creds, subscriptionID, err := getCredentials()
+	if err != nil {
+		return nil, err
+	}
+	client, err := armcompute.NewResourceSKUsClient(*subscriptionID, creds, nil)
+	if err != nil {
+		return nil, err
+	}
+	pager := client.NewListPager(nil)
+	supported := []string{}
+	for pager.More() {
+		page, err := pager.NextPage(ctx)
+		if err != nil {
+			return nil, err
+		}
+		for _, sku := range page.Value {
+			if sku.ResourceType == nil || *sku.ResourceType != string(RTVirtualMachines) {
+				continue
+			}
+			if sku.Name == nil || !slices.Contains(computeSizes, *sku.Name) {
+				continue
+			}
+			inLocation := false
+			for _, loc := range sku.Locations {
+				if loc != nil && strings.EqualFold(*loc, *location) {
+					inLocation = true
+					break
+				}
+			}
+			if !inLocation {
+				continue
+			}
+			diskTypes := diskControllerTypesFromCapabilities(sku.Capabilities)
+			if diskControllerTypeSupported(diskTypes, requiredType) && !slices.Contains(supported, *sku.Name) {
+				supported = append(supported, *sku.Name)
+			}
+		}
+	}
+	return supported, nil
+}
+
+// diskControllerTypesFromCapabilities extracts the DiskControllerTypes value from SKU
+// capabilities. Returns nil when the capability is absent.
+func diskControllerTypesFromCapabilities(caps []*armcompute.ResourceSKUCapabilities) []string {
+	for _, cap := range caps {
+		if cap.Name != nil && *cap.Name == "DiskControllerTypes" && cap.Value != nil {
+			return splitDiskControllerTypes(*cap.Value)
+		}
+	}
+	return nil
+}
+
+// diskControllerTypeSupported reports whether requiredType is satisfied by the supported
+// set. Empty requiredType means no restriction (always passes). A nil/empty supported
+// set means the capability is absent; Azure sizes that predate NVMe default to SCSI, so
+// absence is treated as SCSI-only.
+func diskControllerTypeSupported(supported []string, requiredType string) bool {
+	if requiredType == "" {
+		return true
+	}
+	if len(supported) == 0 {
+		return strings.EqualFold(requiredType, "SCSI")
+	}
+	for _, t := range supported {
+		if strings.EqualFold(t, requiredType) {
+			return true
+		}
+	}
+	return false
+}
+
 func getAzureVMSKUs(ctx context.Context, args *cr.ComputeRequestArgs) ([]string, error) {
 	cred, err := azidentity.NewDefaultAzureCredential(nil)
 	if err != nil {

pkg/provider/azure/data/imageref.go

@@ -28,8 +28,11 @@ type ImageReference struct {
 	Sku       string
 	// Community
 	CommunityImageID string
-	// // Private Shared
+	// Private Shared
 	SharedImageID string
+	// Required disk controller type for this image (e.g. "SCSI", "NVMe").
+	// Empty means no specific requirement; Azure uses the VM size default.
+	DiskControllerType string
 }
 
 var (

pkg/provider/azure/data/images.go

@@ -85,6 +85,40 @@ func getSharedImage(ctx context.Context, c *armcompute.ClientFactory, id *string
 	return &res, nil
 }
 
+// GetSharedImageDiskControllerTypes returns the disk controller types listed in the
+// gallery image definition's features (e.g. ["SCSI"] for RHEL AI images). Returns nil
+// when the feature is absent. Uses the caller's subscription, matching GetSharedImage.
+// The image ID must be a full ARM resource ID with 13 slash-separated parts.
+func GetSharedImageDiskControllerTypes(ctx context.Context, id *string) ([]string, error) {
+	cred, err := azidentity.NewDefaultAzureCredential(nil)
+	if err != nil {
+		return nil, err
+	}
+	parts := strings.Split(*id, "/")
+	if len(parts) != 13 {
+		return nil, fmt.Errorf("invalid shared image ID: %s", *id)
+	}
+	subscriptionId := os.Getenv("AZURE_SUBSCRIPTION_ID")
+	c, err := armcompute.NewClientFactory(subscriptionId, cred, nil)
+	if err != nil {
+		return nil, err
+	}
+	// Query the image definition, not the version — Features live on the definition.
+	res, err := c.NewGalleryImagesClient().Get(ctx, parts[4], parts[8], parts[10], nil)
+	if err != nil {
+		return nil, err
+	}
+	if res.Properties == nil {
+		return nil, nil
+	}
+	for _, f := range res.Properties.Features {
+		if f.Name != nil && *f.Name == "DiskControllerTypes" && f.Value != nil {
+			return splitDiskControllerTypes(*f.Value), nil
+		}
+	}
+	return nil, nil
+}
+
 func SkuG2Support(ctx context.Context, location string, publisher string, offer string, sku string) (string, error) {
 	cred, err := azidentity.NewDefaultAzureCredential(nil)
 	if err != nil {

pkg/provider/azure/data/util.go

@@ -2,6 +2,7 @@ package data
 
 import (
 	"os"
+	"strings"
 
 	"github.com/Azure/azure-sdk-for-go/sdk/azidentity"
 	"github.com/Azure/azure-sdk-for-go/sdk/resourcemanager/resourcegraph/armresourcegraph"
@@ -21,6 +22,17 @@ func getCredentials() (cred *azidentity.DefaultAzureCredential, subscriptionID *
 	return
 }
 
+// splitDiskControllerTypes splits a comma-separated disk controller type string
+// (e.g. "SCSI,NVMe") and trims whitespace from each element.
+func splitDiskControllerTypes(s string) []string {
+	raw := strings.Split(s, ",")
+	out := make([]string, 0, len(raw))
+	for _, t := range raw {
+		out = append(out, strings.TrimSpace(t))
+	}
+	return out
+}
+
 func getGraphClientFactory() (*armresourcegraph.ClientFactory, error) {
 	cred, err := azidentity.NewDefaultAzureCredential(nil)
 	if err != nil {

pkg/provider/azure/modules/allocation/allocation.go

@@ -7,6 +7,7 @@ import (
 	cr "github.com/redhat-developer/mapt/pkg/provider/api/compute-request"
 	spotTypes "github.com/redhat-developer/mapt/pkg/provider/api/spot"
 	"github.com/redhat-developer/mapt/pkg/provider/azure/data"
+	"github.com/redhat-developer/mapt/pkg/util/logging"
 )
 
 type AllocationArgs struct {
@@ -34,6 +35,17 @@ func Allocation(mCtx *mc.Context, args *AllocationArgs) (*AllocationResult, erro
 			return nil, err
 		}
 	}
+
+	// Derive the effective image reference. For shared gallery images, resolveImageRef
+	// queries the gallery API and sets DiskControllerType when the image supports
+	// exactly one type; the caller's explicit value is preserved in all other cases.
+	ir := resolveImageRef(mCtx, args.ImageRef)
+
+	diskControllerType := ""
+	if ir != nil {
+		diskControllerType = ir.DiskControllerType
+	}
+
 	if args.Spot != nil && args.Spot.Spot {
 		sArgs := &data.SpotInfoArgs{
 			ComputeSizes:          computeSizes,
@@ -42,15 +54,29 @@ func Allocation(mCtx *mc.Context, args *AllocationArgs) (*AllocationResult, erro
 			ExcludedLocations:     args.Spot.ExcludedHostingPlaces,
 			SpotPriceIncreaseRate: &args.Spot.IncreaseRate,
 		}
-		if args.ImageRef != nil {
-			sArgs.ImageRef = args.ImageRef
+		if ir != nil {
+			sArgs.ImageRef = ir
 		}
 		bsc, err := data.SpotInfo(mCtx, sArgs)
 		if err != nil {
 			return nil, err
 		}
+		// Filter the spot-selected sizes by disk controller type compatibility.
+		if diskControllerType != "" {
+			spotLocation := bsc.HostingPlace
+			bsc.ComputeType, err = data.FilterComputeSizesByDiskControllerType(
+				mCtx.Context(), &spotLocation, bsc.ComputeType, diskControllerType)
+			if err != nil {
+				return nil, err
+			}
+			if len(bsc.ComputeType) == 0 {
+				return nil, fmt.Errorf(
+					"spot compute sizes in location %q do not support disk controller type %q required by the selected image",
+					bsc.HostingPlace, diskControllerType)
+			}
+		}
 		return &AllocationResult{
-			ImageRef:     args.ImageRef,
+			ImageRef:     ir,
 			Location:     &bsc.HostingPlace,
 			Price:        &bsc.Price,
 			ComputeSizes: bsc.ComputeType,
@@ -65,20 +91,48 @@ func Allocation(mCtx *mc.Context, args *AllocationArgs) (*AllocationResult, erro
 			}
 			location = &hp
 		}
-		// Filter for current location the computesizes
-		supportedComputeSizes, err := data.FilterComputeSizesByLocation(
-			mCtx.Context(), location, computeSizes)
+		// Single SKU enumeration: filter by location and disk controller type together.
+		// When diskControllerType is empty the type check is a no-op (any type passes).
+		supportedComputeSizes, err := data.FilterComputeSizesByDiskControllerType(
+			mCtx.Context(), location, computeSizes, diskControllerType)
 		if err != nil {
 			return nil, err
 		}
 		if len(supportedComputeSizes) == 0 {
+			if diskControllerType != "" {
+				return nil, fmt.Errorf(
+					"no compute sizes in location %q support disk controller type %q required by the selected image",
+					*location, diskControllerType)
+			}
 			return nil, fmt.Errorf("no compute sizes available for location %q", *location)
 		}
 		return &AllocationResult{
-			ImageRef:     args.ImageRef,
+			ImageRef:     ir,
 			Location:     location,
 			ComputeSizes: supportedComputeSizes,
 		}, nil
+	}
+}
 
+// resolveImageRef returns a copy of the image reference, optionally enriched with the
+// disk controller type read from the gallery image definition. The gallery Features value
+// lists types the image *supports*, not what it *requires*. We only override the caller's
+// value when the gallery returns exactly one supported type — that uniquely identifies the
+// requirement. When the gallery returns multiple types the image is flexible, so the
+// caller's explicit value (if any) is preserved unchanged. On fetch failure the caller's
+// value is also preserved.
+func resolveImageRef(mCtx *mc.Context, ir *data.ImageReference) *data.ImageReference {
+	if ir == nil || ir.SharedImageID == "" {
+		return ir
+	}
+	enriched := *ir
+	types, err := data.GetSharedImageDiskControllerTypes(mCtx.Context(), &ir.SharedImageID)
+	if err != nil {
+		logging.Debugf("could not fetch disk controller types for image %s: %v", ir.SharedImageID, err)
+		return &enriched
+	}
+	if len(types) == 1 {
+		enriched.DiskControllerType = types[0]
 	}
+	return &enriched
 }

pkg/provider/azure/modules/virtual-machine/virtual-machine.go

@@ -82,6 +82,7 @@ func Create(ctx *pulumi.Context, mCtx *mc.Context, args *VirtualMachineArgs) (Vi
 					StorageAccountType: pulumi.String("Standard_LRS"),
 				},
 			},
+			DiskControllerType: diskControllerTypePtr(args.Image.DiskControllerType),
 		},
 		// Try to improve provisioning time
 		DiagnosticsProfile: compute.DiagnosticsProfileArgs{
@@ -163,3 +164,10 @@ func isSelfOwned(sharedImageId *string) bool {
 	sharedImageParams := strings.Split(*sharedImageId, "/")
 	return os.Getenv("AZURE_SUBSCRIPTION_ID") == sharedImageParams[2]
 }
+
+func diskControllerTypePtr(t string) pulumi.StringPtrInput {
+	if t == "" {
+		return nil
+	}
+	return pulumi.StringPtr(t)
+}