Merge pull request #215 from daimaxiaxie/add-imagefamily

added ImageFamily field

Author: jwcesign

pkg/apis/v1alpha1/ecsnodeclass.go

@@ -18,6 +18,8 @@ package v1alpha1
 
 import (
 	"fmt"
+	"log"
+	"strings"
 
 	"github.com/mitchellh/hashstructure/v2"
 	"github.com/samber/lo"
@@ -315,15 +317,34 @@ func (in *ECSNodeClass) Hash() string {
 	})))
 }
 
-// ImageFamily If an alias is specified, return alias, or be 'Custom' (enforced via validation).
-func (in *ECSNodeClass) ImageFamily() string {
-	if term, ok := lo.Find(in.Spec.ImageSelectorTerms, func(t ImageSelectorTerm) bool {
-		return t.Alias != ""
-	}); ok {
-		return ImageFamilyFromAlias(term.Alias)
+func (in *ECSNodeClass) Alias() *Alias {
+	term, ok := lo.Find(in.Spec.ImageSelectorTerms, func(term ImageSelectorTerm) bool {
+		return term.Alias != ""
+	})
+	if !ok {
+		return nil
 	}
-	// Unreachable: validation enforces that one of the above conditions must be met
-	return ""
+	return NewAlias(term.Alias)
+}
+
+type Alias struct {
+	Family  string
+	Version string
+}
+
+const (
+	AliasVersionLatest = "latest"
+)
+
+func NewAlias(item string) *Alias {
+	return &Alias{
+		Family:  imageFamilyFromAlias(item),
+		Version: imageVersionFromAlias(item),
+	}
+}
+
+func (a *Alias) String() string {
+	return fmt.Sprintf("%s@%s", a.Family, a.Version)
 }
 
 // ECSNodeClassList contains a list of ECSNodeClass
@@ -334,8 +355,29 @@ type ECSNodeClassList struct {
 	Items           []ECSNodeClass `json:"items"`
 }
 
-func ImageFamilyFromAlias(alias string) string {
-	return alias
+func imageFamilyFromAlias(alias string) string {
+	components := strings.Split(alias, "@")
+	if len(components) > 2 {
+		log.Fatalf("failed to parse Image alias %q, invalid format", alias)
+	}
+	family, ok := lo.Find([]string{
+		ImageFamilyAlibabaCloudLinux3,
+		ImageFamilyContainerOS,
+	}, func(family string) bool {
+		return strings.ToLower(family) == components[0]
+	})
+	if !ok {
+		log.Fatalf("%q is an invalid alias family", components[0])
+	}
+	return family
+}
+
+func imageVersionFromAlias(alias string) string {
+	components := strings.Split(alias, "@")
+	if len(components) != 2 {
+		return AliasVersionLatest
+	}
+	return components[1]
 }
 
 func (sd *SystemDisk) GetGiBSize() int32 {

pkg/apis/v1alpha1/labels.go

@@ -70,6 +70,7 @@ var (
 	}
 	ImageFamilyAlibabaCloudLinux3                     = "AlibabaCloudLinux3"
 	ImageFamilyContainerOS                            = "ContainerOS"
+	ImageFamilyCustom                                 = "Custom"
 	ECSAMDCPUModelValue                               = "AMD"
 	ECSIntelCPUModelValue                             = "Intel"
 	ResourceNVIDIAGPU             corev1.ResourceName = "nvidia.com/gpu"

pkg/apis/v1alpha1/zz_generated.deepcopy.go

@@ -30,7 +30,7 @@ import (
 
 	alicache "github.com/cloudpilot-ai/karpenter-provider-alibabacloud/pkg/cache"
 	"github.com/cloudpilot-ai/karpenter-provider-alibabacloud/pkg/operator/options"
-	"github.com/cloudpilot-ai/karpenter-provider-alibabacloud/pkg/providers/ack"
+	"github.com/cloudpilot-ai/karpenter-provider-alibabacloud/pkg/providers/cluster"
 	"github.com/cloudpilot-ai/karpenter-provider-alibabacloud/pkg/providers/imagefamily"
 	"github.com/cloudpilot-ai/karpenter-provider-alibabacloud/pkg/providers/instance"
 	"github.com/cloudpilot-ai/karpenter-provider-alibabacloud/pkg/providers/instancetype"
@@ -81,7 +81,7 @@ func NewOperator(ctx context.Context, operator *operator.Operator) (context.Cont
 		log.FromContext(ctx).Error(err, "Failed to create ACK client")
 		os.Exit(1)
 	}
-	clusterID := options.FromContext(ctx).ClusterID
+
 	region := *ecsClient.RegionId
 
 	pricingProvider, err := pricing.NewDefaultProvider(ctx, region)
@@ -93,16 +93,16 @@ func NewOperator(ctx context.Context, operator *operator.Operator) (context.Cont
 	versionProvider := version.NewDefaultProvider(operator.KubernetesInterface, cache.New(alicache.KubernetesVersionTTL, alicache.DefaultCleanupInterval))
 	vSwitchProvider := vswitch.NewDefaultProvider(region, vpcClient, cache.New(alicache.DefaultTTL, alicache.DefaultCleanupInterval), cache.New(alicache.AvailableIPAddressTTL, alicache.DefaultCleanupInterval))
 	securityGroupProvider := securitygroup.NewDefaultProvider(region, ecsClient, cache.New(alicache.DefaultTTL, alicache.DefaultCleanupInterval))
-	imageProvider := imagefamily.NewDefaultProvider(region, ecsClient, ackClient, versionProvider, cache.New(alicache.DefaultTTL, alicache.DefaultCleanupInterval))
+	clusterProvider := cluster.NewClusterProvider(ctx, ackClient, region)
+	imageProvider := imagefamily.NewDefaultProvider(region, ecsClient, clusterProvider, versionProvider, cache.New(alicache.DefaultTTL, alicache.DefaultCleanupInterval))
 	imageResolver := imagefamily.NewDefaultResolver(region, ecsClient, cache.New(alicache.InstanceTypeAvailableDiskTTL, alicache.DefaultCleanupInterval))
-	ackProvider := ack.NewDefaultProvider(clusterID, ackClient, cache.New(alicache.ClusterAttachScriptTTL, alicache.DefaultCleanupInterval))
 
 	unavailableOfferingsCache := alicache.NewUnavailableOfferings()
 	instanceTypeProvider := instancetype.NewDefaultProvider(
 		*ecsClient.RegionId, ecsClient,
 		cache.New(alicache.InstanceTypesAndZonesTTL, alicache.DefaultCleanupInterval),
 		unavailableOfferingsCache,
-		pricingProvider, ackProvider)
+		pricingProvider, clusterProvider)
 
 	instanceProvider := instance.NewDefaultProvider(
 		ctx,
@@ -111,7 +111,7 @@ func NewOperator(ctx context.Context, operator *operator.Operator) (context.Cont
 		unavailableOfferingsCache,
 		imageResolver,
 		vSwitchProvider,
-		ackProvider,
+		clusterProvider,
 	)
 
 	return ctx, &Operator{

pkg/operator/operator.go

@@ -41,6 +41,7 @@ type Options struct {
 	Interruption            bool
 	TelemetryShare          bool
 	APGCreationQPS          int
+	ClusterType             string
 }
 
 func (o *Options) AddFlags(fs *coreoptions.FlagSet) {
@@ -52,6 +53,7 @@ func (o *Options) AddFlags(fs *coreoptions.FlagSet) {
 	fs.BoolVar(&o.Interruption, "interruption", env.WithDefaultBool("INTERRUPTION", true), "Enable interruption handling.")
 	fs.BoolVar(&o.TelemetryShare, "telemetry-share", env.WithDefaultBool("TELEMETRY_SHARE", true), "Enable telemetry sharing.")
 	fs.IntVar(&o.APGCreationQPS, "apg-qps", int(env.WithDefaultInt64("APG_CREATION_QPS", 100)), "The QPS limit for creating AutoProvisionGroup.")
+	fs.StringVar(&o.ClusterType, "cluster-type", env.WithDefaultString("CLUSTER_TYPE", "ACKManaged"), "Type of cluster, which specifies the method to generate userdata. The default is ACKManaged, with an option for Custom configuration.")
 }
 
 func (o *Options) Parse(fs *coreoptions.FlagSet, args ...string) error {