feat: accont for gpu attributes

allow wildcards for GPU attributes in deployment messages

Signed-off-by: Artur Troian <[email protected]>

go.mod

@@ -3,7 +3,7 @@ module github.com/akash-network/node
 go 1.20
 
 require (
-	github.com/akash-network/akash-api v0.0.16
+	github.com/akash-network/akash-api v0.0.19
 	github.com/blang/semver/v4 v4.0.0
 	github.com/boz/go-lifecycle v0.1.0
 	github.com/cosmos/cosmos-sdk v0.45.16

go.sum

@@ -99,8 +99,8 @@ github.com/aead/siphash v1.0.1/go.mod h1:Nywa3cDsYNNK3gaciGTWPwHt0wlpNV15vwmswBA
 github.com/afex/hystrix-go v0.0.0-20180502004556-fa1af6a1f4f5/go.mod h1:SkGFH1ia65gfNATL8TAiHDNxPzPdmEL5uirI2Uyuz6c=
 github.com/ajg/form v1.5.1/go.mod h1:uL1WgH+h2mgNtvBq0339dVnzXdBETtL2LeUXaIv25UY=
 github.com/ajstarks/svgo v0.0.0-20180226025133-644b8db467af/go.mod h1:K08gAheRH3/J6wwsYMMT4xOr94bZjxIelGM0+d/wbFw=
-github.com/akash-network/akash-api v0.0.16 h1:6XlJ6a0jROL8JwxXjOVtQbjoOVrWtJL2QoL+6l/f44M=
-github.com/akash-network/akash-api v0.0.16/go.mod h1:e1QqkOFwxHKf88I3U5bPOmREdfHHHX2bY27ZZOFnTX4=
+github.com/akash-network/akash-api v0.0.19 h1:D99DqD5ocBkepA5CmXJzdB1Jy0zA00pefdRDYdIQyHM=
+github.com/akash-network/akash-api v0.0.19/go.mod h1:9/uYusyBcZecBQCgZWUbXRu0i1tyxj4/ze45XB2oLIU=
 github.com/akash-network/cometbft v0.34.27-akash h1:V1dApDOr8Ee7BJzYyQ7Z9VBtrAul4+baMeA6C49dje0=
 github.com/akash-network/cometbft v0.34.27-akash/go.mod h1:BcCbhKv7ieM0KEddnYXvQZR+pZykTKReJJYf7YC7qhw=
 github.com/akash-network/ledger-go v0.14.3 h1:LCEFkTfgGA2xFMN2CtiKvXKE7dh0QSM77PJHCpSkaAo=

sdl/_testdata/simple-gpu.yaml

@@ -21,14 +21,19 @@ profiles:
           units: "100m"
         gpu:
           units: 1
+          attributes:
+            vendor:
+              nvidia:
+              - model: a100
         memory:
           size: "128Mi"
         storage:
-          size: "1Gi"
+        - size: "1Gi"
   placement:
     westcoast:
       attributes:
         region: us-west
+        blalbla: foo
       signedBy:
         anyOf:
           - 1

sdl/full_test.go

@@ -33,6 +33,10 @@ profiles:
             arch: amd64
         gpu:
           units: 1
+          attributes:
+            vendor:
+              nvidia:
+                - model: a100
         memory:
           size: 16Mi
         storage:

sdl/gpu.go

@@ -1,49 +1,110 @@
 package sdl
 
 import (
-	"sort"
+	"fmt"
 
 	"gopkg.in/yaml.v3"
 
 	types "github.com/akash-network/akash-api/go/node/types/v1beta3"
 )
 
-type v2GPUAttributes types.Attributes
+type v2GPUNvidia struct {
+	Model string          `yaml:"model"`
+	RAM   *memoryQuantity `yaml:"ram,omitempty"`
+}
+
+func (sdl *v2GPUNvidia) String() string {
+	key := fmt.Sprintf("%s", sdl.Model)
+	if sdl.RAM != nil {
+		key += "/" + sdl.RAM.StringWithSuffix("Gi")
+	}
+
+	return key
+}
+
+type v2GPUsNvidia []v2GPUNvidia
+
+type gpuVendor struct {
+	Nvidia v2GPUsNvidia `yaml:"nvidia,omitempty"`
+}
+
+type v2GPUAttributes struct {
+	attr   types.Attributes
+	Vendor *gpuVendor `yaml:"vendor,omitempty"`
+}
 
 type v2ResourceGPU struct {
 	Units      gpuQuantity     `yaml:"units"`
 	Attributes v2GPUAttributes `yaml:"attributes,omitempty"`
 }
 
+func (sdl *v2ResourceGPU) UnmarshalYAML(node *yaml.Node) error {
+	res := v2ResourceGPU{}
+
+	for i := 0; i < len(node.Content); i += 2 {
+		switch node.Content[i].Value {
+		case "units":
+			if err := node.Content[i+1].Decode(&res.Units); err != nil {
+				return err
+			}
+		case "attributes":
+			if err := node.Content[i+1].Decode(&res.Attributes); err != nil {
+				return err
+			}
+		default:
+			return fmt.Errorf("sdl: unsupported field (%s) for GPU resource", node.Content[i].Value)
+		}
+	}
+
+	if res.Units > 0 && len(res.Attributes.attr) == 0 {
+		return fmt.Errorf("sdl: GPU attributes must be present if units > 0")
+	}
+
+	*sdl = res
+
+	return nil
+}
+
 func (sdl *v2GPUAttributes) UnmarshalYAML(node *yaml.Node) error {
-	var attr v2GPUAttributes
+	var res v2GPUAttributes
 
-	for i := 0; i+1 < len(node.Content); i += 2 {
-		var value string
-		if err := node.Content[i+1].Decode(&value); err != nil {
-			return err
+	for i := 0; i < len(node.Content); i += 2 {
+		switch node.Content[i].Value {
+		case "vendor":
+			if err := node.Content[i+1].Decode(&res.Vendor); err != nil {
+				return err
+			}
+		default:
+			return fmt.Errorf("sdl: unsupported attribute (%s) for GPU resource", node.Content[i].Value)
 		}
-		// switch node.Content[i].Value {
-		// case "arch":
-		// 	if err := node.Content[i+1].Decode(&value); err != nil {
-		// 		return err
-		// 	}
-		// default:
-		// 	return errors.Errorf("unsupported cpu attribute \"%s\"", node.Content[i].Value)
-		// }
-
-		attr = append(attr, types.Attribute{
-			Key:   node.Content[i].Value,
-			Value: value,
+	}
+
+	if res.Vendor == nil {
+		return fmt.Errorf("sdl: invalid GPU attributes. at least one vendor must be set")
+	}
+
+	res.attr = make(types.Attributes, 0, len(res.Vendor.Nvidia))
+
+	for _, model := range res.Vendor.Nvidia {
+		res.attr = append(res.attr, types.Attribute{
+			Key:   fmt.Sprintf("vendor/nvidia/model/%s", model.String()),
+			Value: "true",
 		})
 	}
 
-	// keys are unique in attributes parsed from sdl so don't need to use sort.SliceStable
-	sort.Slice(attr, func(i, j int) bool {
-		return attr[i].Key < attr[j].Key
-	})
+	if len(res.attr) == 0 {
+		res.attr = append(res.attr, types.Attribute{
+			Key:   "vendor/nvidia/model/*",
+			Value: "true",
+		})
+	}
+	res.attr.Sort()
+
+	if err := res.attr.Validate(); err != nil {
+		return fmt.Errorf("sdl: invalid GPU attributes: %w", err)
+	}
 
-	*sdl = attr
+	*sdl = res
 
 	return nil
 }

sdl/gpu_test.go

@@ -0,0 +1,155 @@
+package sdl
+
+import (
+	"testing"
+
+	"github.com/stretchr/testify/require"
+	"gopkg.in/yaml.v3"
+)
+
+func TestV2ResourceGPU_EmptyVendor(t *testing.T) {
+	var stream = `
+units: 1
+attributes:
+  vendor:
+`
+	var p v2ResourceGPU
+
+	err := yaml.Unmarshal([]byte(stream), &p)
+	require.Error(t, err)
+}
+
+func TestV2ResourceGPU_Wildcard(t *testing.T) {
+	var stream = `
+units: 1
+attributes:
+  vendor:
+    nvidia:
+`
+	var p v2ResourceGPU
+
+	err := yaml.Unmarshal([]byte(stream), &p)
+	require.NoError(t, err)
+	require.Equal(t, gpuQuantity(1), p.Units)
+	require.Equal(t, 1, len(p.Attributes.attr))
+	require.Equal(t, "vendor/nvidia/model/*", p.Attributes.attr[0].Key)
+	require.Equal(t, "true", p.Attributes.attr[0].Value)
+}
+
+func TestV2ResourceGPU_SingleModel(t *testing.T) {
+	var stream = `
+units: 1
+attributes:
+  vendor:
+    nvidia:
+      - model: a100
+`
+	var p v2ResourceGPU
+
+	err := yaml.Unmarshal([]byte(stream), &p)
+	require.NoError(t, err)
+	require.Equal(t, gpuQuantity(1), p.Units)
+	require.Equal(t, 1, len(p.Attributes.attr))
+	require.Equal(t, "vendor/nvidia/model/a100", p.Attributes.attr[0].Key)
+	require.Equal(t, "true", p.Attributes.attr[0].Value)
+}
+
+func TestV2ResourceGPU_SingleModelWithRAM(t *testing.T) {
+	var stream = `
+units: 1
+attributes:
+  vendor:
+    nvidia:
+      - model: a100
+        ram: 80Gi
+`
+	var p v2ResourceGPU
+
+	err := yaml.Unmarshal([]byte(stream), &p)
+	require.NoError(t, err)
+	require.Equal(t, gpuQuantity(1), p.Units)
+	require.Equal(t, 1, len(p.Attributes.attr))
+	require.Equal(t, "vendor/nvidia/model/a100/80Gi", p.Attributes.attr[0].Key)
+	require.Equal(t, "true", p.Attributes.attr[0].Value)
+}
+
+func TestV2ResourceGPU_InvalidRAMUnit(t *testing.T) {
+	var stream = `
+units: 1
+attributes:
+  vendor:
+    nvidia:
+      - model: a100
+        ram: 80G
+`
+	var p v2ResourceGPU
+
+	err := yaml.Unmarshal([]byte(stream), &p)
+	require.Error(t, err)
+}
+
+func TestV2ResourceGPU_MultipleModels(t *testing.T) {
+	var stream = `
+units: 1
+attributes:
+  vendor:
+    nvidia:
+      - model: a100
+        ram: 80Gi
+      - model: a100
+        ram: 40Gi
+`
+	var p v2ResourceGPU
+
+	err := yaml.Unmarshal([]byte(stream), &p)
+	require.NoError(t, err)
+	require.Equal(t, gpuQuantity(1), p.Units)
+	require.Equal(t, 2, len(p.Attributes.attr))
+	require.Equal(t, "vendor/nvidia/model/a100/40Gi", p.Attributes.attr[0].Key)
+	require.Equal(t, "true", p.Attributes.attr[0].Value)
+	require.Equal(t, "vendor/nvidia/model/a100/80Gi", p.Attributes.attr[1].Key)
+	require.Equal(t, "true", p.Attributes.attr[1].Value)
+}
+
+func TestV2ResourceGPU_MultipleModels2(t *testing.T) {
+	var stream = `
+units: 1
+attributes:
+  vendor:
+    nvidia:
+      - model: a100
+        ram: 80Gi
+      - model: a100
+`
+	var p v2ResourceGPU
+
+	err := yaml.Unmarshal([]byte(stream), &p)
+	require.NoError(t, err)
+	require.Equal(t, gpuQuantity(1), p.Units)
+	require.Equal(t, 2, len(p.Attributes.attr))
+	require.Equal(t, "vendor/nvidia/model/a100", p.Attributes.attr[0].Key)
+	require.Equal(t, "true", p.Attributes.attr[0].Value)
+	require.Equal(t, "vendor/nvidia/model/a100/80Gi", p.Attributes.attr[1].Key)
+	require.Equal(t, "true", p.Attributes.attr[1].Value)
+}
+
+func TestV2ResourceGPU_MultipleModels3(t *testing.T) {
+	var stream = `
+units: 1
+attributes:
+  vendor:
+    nvidia:
+      - model: a6000
+      - model: a40
+`
+	var p v2ResourceGPU
+
+	err := yaml.Unmarshal([]byte(stream), &p)
+	require.NoError(t, err)
+	require.Equal(t, gpuQuantity(1), p.Units)
+	require.Equal(t, 2, len(p.Attributes.attr))
+	require.Equal(t, "vendor/nvidia/model/a40", p.Attributes.attr[0].Key)
+	require.Equal(t, "true", p.Attributes.attr[0].Value)
+	require.Equal(t, "vendor/nvidia/model/a6000", p.Attributes.attr[1].Key)
+	require.Equal(t, "true", p.Attributes.attr[1].Value)
+}

sdl/resources.go

@@ -28,7 +28,7 @@ func (sdl *v2ComputeResources) toDGroupResourceUnits() types.ResourceUnits {
 	if sdl.GPU != nil {
 		units.GPU = &types.GPU{
 			Units:      types.NewResourceValue(uint64(sdl.GPU.Units)),
-			Attributes: types.Attributes(sdl.GPU.Attributes),
+			Attributes: sdl.GPU.Attributes.attr,
 		}
 	} else {
 		units.GPU = &types.GPU{