feat: Add support for karpenter.azure.com/sku-series label/selector

hakman · Ciprian Hacman · commit 5d7f18663876 · 2025-10-28T18:15:30.000+02:00
Signed-off-by: Ciprian Hacman &lt;chacman@microsoft.com&gt;
Signed-off-by: Ciprian Hacman &lt;ciprian@hakman.dev&gt;
diff --git a/charts/karpenter-crd/templates/karpenter.sh_nodeclaims.yaml b/charts/karpenter-crd/templates/karpenter.sh_nodeclaims.yaml
@@ -131,7 +131,7 @@ spec:
                           - message: label "kubernetes.io/hostname" is restricted
                             rule: self != "kubernetes.io/hostname"
                           - message: label domain "karpenter.azure.com" is restricted
-                            rule: self in [ "karpenter.azure.com/aksnodeclass", "karpenter.azure.com/sku-name", "karpenter.azure.com/sku-family", "karpenter.azure.com/sku-version", "karpenter.azure.com/sku-cpu", "karpenter.azure.com/sku-memory", "karpenter.azure.com/sku-networking-accelerated", "karpenter.azure.com/sku-storage-premium-capable", "karpenter.azure.com/sku-storage-ephemeralos-maxsize", "karpenter.azure.com/sku-gpu-name", "karpenter.azure.com/sku-gpu-manufacturer", "karpenter.azure.com/sku-gpu-count" ] || !self.find("^([^/]+)").endsWith("karpenter.azure.com")
+                            rule: self in [ "karpenter.azure.com/aksnodeclass", "karpenter.azure.com/sku-name", "karpenter.azure.com/sku-family", "karpenter.azure.com/sku-series", "karpenter.azure.com/sku-version", "karpenter.azure.com/sku-cpu", "karpenter.azure.com/sku-memory", "karpenter.azure.com/sku-networking-accelerated", "karpenter.azure.com/sku-storage-premium-capable", "karpenter.azure.com/sku-storage-ephemeralos-maxsize", "karpenter.azure.com/sku-gpu-name", "karpenter.azure.com/sku-gpu-manufacturer", "karpenter.azure.com/sku-gpu-count" ] || !self.find("^([^/]+)").endsWith("karpenter.azure.com")
                       minValues:
                         description: |-
                           This field is ALPHA and can be dropped or replaced at any time
diff --git a/charts/karpenter-crd/templates/karpenter.sh_nodepools.yaml b/charts/karpenter-crd/templates/karpenter.sh_nodepools.yaml
@@ -208,7 +208,7 @@ spec:
                             - message: label "kubernetes.io/hostname" is restricted
                               rule: self.all(x, x != "kubernetes.io/hostname")
                             - message: label domain "karpenter.azure.com" is restricted
-                              rule: self.all(x, x in [ "karpenter.azure.com/aksnodeclass", "karpenter.azure.com/sku-name", "karpenter.azure.com/sku-family", "karpenter.azure.com/sku-version", "karpenter.azure.com/sku-cpu", "karpenter.azure.com/sku-memory", "karpenter.azure.com/sku-networking-accelerated", "karpenter.azure.com/sku-storage-premium-capable", "karpenter.azure.com/sku-storage-ephemeralos-maxsize", "karpenter.azure.com/sku-gpu-name", "karpenter.azure.com/sku-gpu-manufacturer", "karpenter.azure.com/sku-gpu-count" ] || !x.find("^([^/]+)").endsWith("karpenter.azure.com"))
+                              rule: self.all(x, x in [ "karpenter.azure.com/aksnodeclass", "karpenter.azure.com/sku-name", "karpenter.azure.com/sku-family", "karpenter.azure.com/sku-series", "karpenter.azure.com/sku-version", "karpenter.azure.com/sku-cpu", "karpenter.azure.com/sku-memory", "karpenter.azure.com/sku-networking-accelerated", "karpenter.azure.com/sku-storage-premium-capable", "karpenter.azure.com/sku-storage-ephemeralos-maxsize", "karpenter.azure.com/sku-gpu-name", "karpenter.azure.com/sku-gpu-manufacturer", "karpenter.azure.com/sku-gpu-count" ] || !x.find("^([^/]+)").endsWith("karpenter.azure.com"))
                       type: object
                     spec:
                       description: |-
@@ -281,7 +281,7 @@ spec:
                                   - message: label "kubernetes.io/hostname" is restricted
                                     rule: self != "kubernetes.io/hostname"
                                   - message: label domain "karpenter.azure.com" is restricted
-                                    rule: self in [ "karpenter.azure.com/aksnodeclass", "karpenter.azure.com/sku-name", "karpenter.azure.com/sku-family", "karpenter.azure.com/sku-version", "karpenter.azure.com/sku-cpu", "karpenter.azure.com/sku-memory", "karpenter.azure.com/sku-networking-accelerated", "karpenter.azure.com/sku-storage-premium-capable", "karpenter.azure.com/sku-storage-ephemeralos-maxsize", "karpenter.azure.com/sku-gpu-name", "karpenter.azure.com/sku-gpu-manufacturer", "karpenter.azure.com/sku-gpu-count" ] || !self.find("^([^/]+)").endsWith("karpenter.azure.com")
+                                    rule: self in [ "karpenter.azure.com/aksnodeclass", "karpenter.azure.com/sku-name", "karpenter.azure.com/sku-family", "karpenter.azure.com/sku-series", "karpenter.azure.com/sku-version", "karpenter.azure.com/sku-cpu", "karpenter.azure.com/sku-memory", "karpenter.azure.com/sku-networking-accelerated", "karpenter.azure.com/sku-storage-premium-capable", "karpenter.azure.com/sku-storage-ephemeralos-maxsize", "karpenter.azure.com/sku-gpu-name", "karpenter.azure.com/sku-gpu-manufacturer", "karpenter.azure.com/sku-gpu-count" ] || !self.find("^([^/]+)").endsWith("karpenter.azure.com")
                               minValues:
                                 description: |-
                                   This field is ALPHA and can be dropped or replaced at any time
diff --git a/designs/gpu-selection-and-bootstrap.md b/designs/gpu-selection-and-bootstrap.md
@@ -107,10 +107,11 @@ The Requirements API needs to provide an interface for selecting GPU SKUs, sizes
 
 Here are some relevant labels:
 
-| Label                                 | Description                               |
-|---------------------------------------|-------------------------------------------|
-| `karpenter.azure.com/sku-family`       | Family of the SKU (N) for GPU          |
-| `karpenter.azure.com/sku-cpu`          | Number of virtual CPUs                    |
+| Label                            | Description                   |
+|----------------------------------|-------------------------------|
+| `karpenter.azure.com/sku-family` | Family of the SKU (N) for GPU |
+| `karpenter.azure.com/sku-series` | Series of the SKU             |
+| `karpenter.azure.com/sku-cpu`    | Number of virtual CPUs        |
 
 **Note:** GPU SKUs usually support only a single hypervisor generation. Explicit image selection is moot in most cases, though there are some exceptions.
 
diff --git a/hack/validation/labels.sh b/hack/validation/labels.sh
@@ -9,6 +9,7 @@ rule=$'self.all(x, x in
         "karpenter.azure.com/aksnodeclass",
         "karpenter.azure.com/sku-name",
         "karpenter.azure.com/sku-family",
+        "karpenter.azure.com/sku-series",
         "karpenter.azure.com/sku-version",
         "karpenter.azure.com/sku-cpu",
         "karpenter.azure.com/sku-memory",
diff --git a/hack/validation/requirements.sh b/hack/validation/requirements.sh
@@ -9,6 +9,7 @@ rule=$'self in
         "karpenter.azure.com/aksnodeclass",
         "karpenter.azure.com/sku-name",
         "karpenter.azure.com/sku-family",
+        "karpenter.azure.com/sku-series",
         "karpenter.azure.com/sku-version",
         "karpenter.azure.com/sku-cpu",
         "karpenter.azure.com/sku-memory",
diff --git a/pkg/apis/crds/karpenter.sh_nodeclaims.yaml b/pkg/apis/crds/karpenter.sh_nodeclaims.yaml
@@ -131,7 +131,7 @@ spec:
                           - message: label "kubernetes.io/hostname" is restricted
                             rule: self != "kubernetes.io/hostname"
                           - message: label domain "karpenter.azure.com" is restricted
-                            rule: self in [ "karpenter.azure.com/aksnodeclass", "karpenter.azure.com/sku-name", "karpenter.azure.com/sku-family", "karpenter.azure.com/sku-version", "karpenter.azure.com/sku-cpu", "karpenter.azure.com/sku-memory", "karpenter.azure.com/sku-networking-accelerated", "karpenter.azure.com/sku-storage-premium-capable", "karpenter.azure.com/sku-storage-ephemeralos-maxsize", "karpenter.azure.com/sku-gpu-name", "karpenter.azure.com/sku-gpu-manufacturer", "karpenter.azure.com/sku-gpu-count" ] || !self.find("^([^/]+)").endsWith("karpenter.azure.com")
+                            rule: self in [ "karpenter.azure.com/aksnodeclass", "karpenter.azure.com/sku-name", "karpenter.azure.com/sku-family", "karpenter.azure.com/sku-series", "karpenter.azure.com/sku-version", "karpenter.azure.com/sku-cpu", "karpenter.azure.com/sku-memory", "karpenter.azure.com/sku-networking-accelerated", "karpenter.azure.com/sku-storage-premium-capable", "karpenter.azure.com/sku-storage-ephemeralos-maxsize", "karpenter.azure.com/sku-gpu-name", "karpenter.azure.com/sku-gpu-manufacturer", "karpenter.azure.com/sku-gpu-count" ] || !self.find("^([^/]+)").endsWith("karpenter.azure.com")
                       minValues:
                         description: |-
                           This field is ALPHA and can be dropped or replaced at any time
diff --git a/pkg/apis/crds/karpenter.sh_nodepools.yaml b/pkg/apis/crds/karpenter.sh_nodepools.yaml
@@ -208,7 +208,7 @@ spec:
                             - message: label "kubernetes.io/hostname" is restricted
                               rule: self.all(x, x != "kubernetes.io/hostname")
                             - message: label domain "karpenter.azure.com" is restricted
-                              rule: self.all(x, x in [ "karpenter.azure.com/aksnodeclass", "karpenter.azure.com/sku-name", "karpenter.azure.com/sku-family", "karpenter.azure.com/sku-version", "karpenter.azure.com/sku-cpu", "karpenter.azure.com/sku-memory", "karpenter.azure.com/sku-networking-accelerated", "karpenter.azure.com/sku-storage-premium-capable", "karpenter.azure.com/sku-storage-ephemeralos-maxsize", "karpenter.azure.com/sku-gpu-name", "karpenter.azure.com/sku-gpu-manufacturer", "karpenter.azure.com/sku-gpu-count" ] || !x.find("^([^/]+)").endsWith("karpenter.azure.com"))
+                              rule: self.all(x, x in [ "karpenter.azure.com/aksnodeclass", "karpenter.azure.com/sku-name", "karpenter.azure.com/sku-family", "karpenter.azure.com/sku-series", "karpenter.azure.com/sku-version", "karpenter.azure.com/sku-cpu", "karpenter.azure.com/sku-memory", "karpenter.azure.com/sku-networking-accelerated", "karpenter.azure.com/sku-storage-premium-capable", "karpenter.azure.com/sku-storage-ephemeralos-maxsize", "karpenter.azure.com/sku-gpu-name", "karpenter.azure.com/sku-gpu-manufacturer", "karpenter.azure.com/sku-gpu-count" ] || !x.find("^([^/]+)").endsWith("karpenter.azure.com"))
                       type: object
                     spec:
                       description: |-
@@ -281,7 +281,7 @@ spec:
                                   - message: label "kubernetes.io/hostname" is restricted
                                     rule: self != "kubernetes.io/hostname"
                                   - message: label domain "karpenter.azure.com" is restricted
-                                    rule: self in [ "karpenter.azure.com/aksnodeclass", "karpenter.azure.com/sku-name", "karpenter.azure.com/sku-family", "karpenter.azure.com/sku-version", "karpenter.azure.com/sku-cpu", "karpenter.azure.com/sku-memory", "karpenter.azure.com/sku-networking-accelerated", "karpenter.azure.com/sku-storage-premium-capable", "karpenter.azure.com/sku-storage-ephemeralos-maxsize", "karpenter.azure.com/sku-gpu-name", "karpenter.azure.com/sku-gpu-manufacturer", "karpenter.azure.com/sku-gpu-count" ] || !self.find("^([^/]+)").endsWith("karpenter.azure.com")
+                                    rule: self in [ "karpenter.azure.com/aksnodeclass", "karpenter.azure.com/sku-name", "karpenter.azure.com/sku-family", "karpenter.azure.com/sku-series", "karpenter.azure.com/sku-version", "karpenter.azure.com/sku-cpu", "karpenter.azure.com/sku-memory", "karpenter.azure.com/sku-networking-accelerated", "karpenter.azure.com/sku-storage-premium-capable", "karpenter.azure.com/sku-storage-ephemeralos-maxsize", "karpenter.azure.com/sku-gpu-name", "karpenter.azure.com/sku-gpu-manufacturer", "karpenter.azure.com/sku-gpu-count" ] || !self.find("^([^/]+)").endsWith("karpenter.azure.com")
                               minValues:
                                 description: |-
                                   This field is ALPHA and can be dropped or replaced at any time
diff --git a/pkg/apis/v1beta1/labels.go b/pkg/apis/v1beta1/labels.go
@@ -29,6 +29,7 @@ func init() {
 	karpv1.WellKnownLabels = karpv1.WellKnownLabels.Insert(
 		LabelSKUName,
 		LabelSKUFamily,
+		LabelSKUSeries,
 		LabelSKUVersion,
 
 		LabelSKUCPU,
@@ -72,8 +73,9 @@ var (
 	HyperVGenerationV2 = "2"
 	ManufacturerNvidia = "nvidia"
 
-	LabelSKUName    = Group + "/sku-name"    // Standard_A1_v2
-	LabelSKUFamily  = Group + "/sku-family"  // A
+	LabelSKUName    = Group + "/sku-name"    // Standard_D4pls_v6
+	LabelSKUFamily  = Group + "/sku-family"  // D
+	LabelSKUSeries  = Group + "/sku-series"  // Dpls_v6
 	LabelSKUVersion = Group + "/sku-version" // numerical (without v), with 1 backfilled
 
 	LabelSKUCPU    = Group + "/sku-cpu"    // sku.vCPUs
diff --git a/pkg/providers/instancetype/instancetype.go b/pkg/providers/instancetype/instancetype.go
@@ -164,6 +164,7 @@ func computeRequirements(sku *skewer.SKU, vmsize *skewer.VMSizeType, architectur
 
 		// size parts
 		scheduling.NewRequirement(v1beta1.LabelSKUFamily, corev1.NodeSelectorOpDoesNotExist),
+		scheduling.NewRequirement(v1beta1.LabelSKUSeries, corev1.NodeSelectorOpDoesNotExist),
 		scheduling.NewRequirement(v1beta1.LabelSKUVersion, corev1.NodeSelectorOpDoesNotExist),
 
 		// SKU capabilities
@@ -179,6 +180,7 @@ func computeRequirements(sku *skewer.SKU, vmsize *skewer.VMSizeType, architectur
 
 	// size parts
 	requirements[v1beta1.LabelSKUFamily].Insert(vmsize.Family)
+	requirements[v1beta1.LabelSKUSeries].Insert(vmsize.Series)
 
 	setRequirementsEphemeralOSDiskSupported(requirements, sku)
 	setRequirementsHyperVGeneration(requirements, sku)
diff --git a/pkg/providers/instancetype/suite_test.go b/pkg/providers/instancetype/suite_test.go
@@ -1482,6 +1482,7 @@ var _ = Describe("InstanceType Provider", func() {
 				// Well Known to AKS
 				v1beta1.LabelSKUName:                      "Standard_NC24ads_A100_v4",
 				v1beta1.LabelSKUFamily:                    "N",
+				v1beta1.LabelSKUSeries:                    "NCads_v4",
 				v1beta1.LabelSKUVersion:                   "4",
 				v1beta1.LabelSKUStorageEphemeralOSMaxSize: "429",
 				v1beta1.LabelSKUAcceleratedNetworking:     "true",
