Merge pull request #1193 from Tharsanan1/airl

Add backend based airls from cp to dp

apim-apk-agent/go.mod

@@ -10,7 +10,7 @@ require (
 	github.com/pelletier/go-toml v1.9.5
 	github.com/sirupsen/logrus v1.9.3
 	github.com/stretchr/testify v1.9.0
-	github.com/wso2/apk/common-go-libs v0.0.0-20240920041902-85449a1c0150
+	github.com/wso2/apk/common-go-libs v0.0.0-20240923143402-ff7fdb0366f9
 	google.golang.org/grpc v1.62.0
 	gopkg.in/natefinch/lumberjack.v2 v2.2.1
 	gopkg.in/yaml.v2 v2.4.0

apim-apk-agent/go.sum

@@ -189,8 +189,8 @@ github.com/vektah/gqlparser v1.3.1 h1:8b0IcD3qZKWJQHSzynbDlrtP3IxVydZ2DZepCGofqf
 github.com/vektah/gqlparser v1.3.1/go.mod h1:bkVf0FX+Stjg/MHnm8mEyubuaArhNEqfQhF+OTiAL74=
 github.com/wso2/apk/adapter v0.0.0-20240408123538-86a74d977eee h1:g0ivVkzybfcEkB0vBGTAXTUuMZpsF3zOTVtAgmW851s=
 github.com/wso2/apk/adapter v0.0.0-20240408123538-86a74d977eee/go.mod h1:xYS5auF/YxnyRykw7NBSn/YR2FHD4hTeyav4Nhec8d0=
-github.com/wso2/apk/common-go-libs v0.0.0-20240920041902-85449a1c0150 h1:X3OezAh2UOxmQIRxsAua87nNqmoIGXx1yfQIvc4a+G4=
-github.com/wso2/apk/common-go-libs v0.0.0-20240920041902-85449a1c0150/go.mod h1:SbZVA1jeiVG9dqk9fGcY/bB0JgEaQgtXqFAlxAfN0Lk=
+github.com/wso2/apk/common-go-libs v0.0.0-20240923143402-ff7fdb0366f9 h1:MwQqG+/ODDIfLfc3xNMYk6jM+hB2ttjwZnaDBeiMOJI=
+github.com/wso2/apk/common-go-libs v0.0.0-20240923143402-ff7fdb0366f9/go.mod h1:SbZVA1jeiVG9dqk9fGcY/bB0JgEaQgtXqFAlxAfN0Lk=
 github.com/yuin/goldmark v1.1.27/go.mod h1:3hX8gzYuyVAZsxl0MRgGTJEmQBFcNTphYh9decYSb74=
 github.com/yuin/goldmark v1.2.1/go.mod h1:3hX8gzYuyVAZsxl0MRgGTJEmQBFcNTphYh9decYSb74=
 github.com/yuin/goldmark v1.4.13/go.mod h1:6yULJ656Px+3vBD8DxQVa3kxgyrAnzto9xy5taEt/CY=

apim-apk-agent/internal/eventhub/dataloader.go

@@ -26,7 +26,7 @@ import (
 	"strconv"
 	"time"
 
-	dpv1alpha2 "github.com/wso2/apk/common-go-libs/apis/dp/v1alpha2"
+	dpv1alpha3 "github.com/wso2/apk/common-go-libs/apis/dp/v1alpha3"
 	"github.com/wso2/product-apim-tooling/apim-apk-agent/config"
 	internalk8sClient "github.com/wso2/product-apim-tooling/apim-apk-agent/internal/k8sClient"
 	logger "github.com/wso2/product-apim-tooling/apim-apk-agent/internal/loggers"
@@ -240,7 +240,7 @@ func FetchAPIsOnStartUp(conf *config.Config, k8sClient client.Client) {
 	if err != nil {
 		logger.LoggerEventhub.Errorf("Error occurred while fetching APIs from control plane %v", err)
 	}
-	removeApis := make([]dpv1alpha2.API, 0)
+	removeApis := make([]dpv1alpha3.API, 0)
 	for _, k8sAPI := range k8sAPIS {
 		found := false
 		if apis != nil {

apim-apk-agent/internal/k8sClient/k8s_client.go

@@ -43,8 +43,8 @@ import (
 )
 
 // DeployAPICR applies the given API struct to the Kubernetes cluster.
-func DeployAPICR(api *dpv1alpha2.API, k8sClient client.Client) {
-	crAPI := &dpv1alpha2.API{}
+func DeployAPICR(api *dpv1alpha3.API, k8sClient client.Client) {
+	crAPI := &dpv1alpha3.API{}
 	if err := k8sClient.Get(context.Background(), client.ObjectKey{Namespace: api.ObjectMeta.Namespace, Name: api.Name}, crAPI); err != nil {
 		if !k8error.IsNotFound(err) {
 			loggers.LoggerK8sClient.Error("Unable to get API CR: " + err.Error())
@@ -66,7 +66,7 @@ func DeployAPICR(api *dpv1alpha2.API, k8sClient client.Client) {
 }
 
 // UndeployK8sAPICR removes the API Custom Resource from the Kubernetes cluster based on API ID label.
-func UndeployK8sAPICR(k8sClient client.Client, k8sAPI dpv1alpha2.API) error {
+func UndeployK8sAPICR(k8sClient client.Client, k8sAPI dpv1alpha3.API) error {
 	err := k8sClient.Delete(context.Background(), &k8sAPI, &client.DeleteOptions{})
 	if err != nil {
 		loggers.LoggerK8sClient.Errorf("Unable to delete API CR: %v", err)
@@ -82,7 +82,7 @@ func UndeployAPICR(apiID string, k8sClient client.Client) {
 	if errReadConfig != nil {
 		loggers.LoggerK8sClient.Errorf("Error reading configurations: %v", errReadConfig)
 	}
-	apiList := &dpv1alpha2.APIList{}
+	apiList := &dpv1alpha3.APIList{}
 	err := k8sClient.List(context.Background(), apiList, &client.ListOptions{Namespace: conf.DataPlane.Namespace, LabelSelector: labels.SelectorFromSet(map[string]string{"apiUUID": apiID})})
 	// Retrieve all API CRs from the Kubernetes cluster
 	if err != nil {
@@ -429,6 +429,66 @@ func DeploySubscriptionRateLimitPolicyCR(policy eventhubTypes.SubscriptionPolicy
 
 }
 
+// DeployAIRateLimitPolicyCR applies the given AIRateLimitPolicies struct to the Kubernetes cluster.
+func DeployAIRateLimitPolicyCR(policy eventhubTypes.SubscriptionPolicy, k8sClient client.Client) {
+	conf, _ := config.ReadConfigs()
+	tokenCount := &dpv1alpha3.TokenCount{}
+	requestCount := &dpv1alpha3.RequestCount{}
+	if policy.DefaultLimit.AiAPIQuota.PromptTokenCount != nil &&
+		policy.DefaultLimit.AiAPIQuota.CompletionTokenCount != nil &&
+		policy.DefaultLimit.AiAPIQuota.TotalTokenCount != nil {
+		tokenCount = &dpv1alpha3.TokenCount{
+			Unit:               policy.DefaultLimit.AiAPIQuota.TimeUnit,
+			RequestTokenCount:  uint32(*policy.DefaultLimit.AiAPIQuota.PromptTokenCount),
+			ResponseTokenCount: uint32(*policy.DefaultLimit.AiAPIQuota.CompletionTokenCount),
+			TotalTokenCount:    uint32(*policy.DefaultLimit.AiAPIQuota.TotalTokenCount),
+		}
+	} else {
+		tokenCount = nil
+	}
+	if policy.DefaultLimit.AiAPIQuota.RequestCount != nil {
+		requestCount = &dpv1alpha3.RequestCount{
+			RequestsPerUnit: uint32(*policy.DefaultLimit.AiAPIQuota.RequestCount),
+			Unit:            policy.DefaultLimit.AiAPIQuota.TimeUnit,
+		}
+	} else {
+		requestCount = nil
+	}
+
+	crRateLimitPolicies := dpv1alpha3.AIRateLimitPolicy{
+		ObjectMeta: metav1.ObjectMeta{Name: policy.Name,
+			Namespace: conf.DataPlane.Namespace,
+		},
+		Spec: dpv1alpha3.AIRateLimitPolicySpec{
+			Override: &dpv1alpha3.AIRateLimit{
+				Organization: policy.TenantDomain,
+				TokenCount:   tokenCount,
+				RequestCount: requestCount,
+			},
+			TargetRef: gwapiv1b1.PolicyTargetReference{Group: constants.GatewayGroup, Kind: "Subscription", Name: "default"},
+		},
+	}
+	crRateLimitPolicyFetched := &dpv1alpha3.AIRateLimitPolicy{}
+	if err := k8sClient.Get(context.Background(), client.ObjectKey{Namespace: crRateLimitPolicies.ObjectMeta.Namespace, Name: crRateLimitPolicies.Name}, crRateLimitPolicyFetched); err != nil {
+		if !k8error.IsNotFound(err) {
+			loggers.LoggerK8sClient.Error("Unable to get AiratelimitPolicy CR: " + err.Error())
+		}
+		if err := k8sClient.Create(context.Background(), &crRateLimitPolicies); err != nil {
+			loggers.LoggerK8sClient.Error("Unable to create AIRateLimitPolicies CR: " + err.Error())
+		} else {
+			loggers.LoggerK8sClient.Info("AIRateLimitPolicies CR created: " + crRateLimitPolicies.Name)
+		}
+	} else {
+		crRateLimitPolicyFetched.Spec = crRateLimitPolicies.Spec
+		crRateLimitPolicyFetched.ObjectMeta.Labels = crRateLimitPolicies.ObjectMeta.Labels
+		if err := k8sClient.Update(context.Background(), crRateLimitPolicyFetched); err != nil {
+			loggers.LoggerK8sClient.Error("Unable to update AiRatelimitPolicy CR: " + err.Error())
+		} else {
+			loggers.LoggerK8sClient.Info("AiRatelimitPolicy CR updated: " + crRateLimitPolicyFetched.Name)
+		}
+	}
+}
+
 // DeployBackendCR applies the given Backends struct to the Kubernetes cluster.
 func DeployBackendCR(backends *dpv1alpha2.Backend, k8sClient client.Client) {
 	crBackends := &dpv1alpha2.Backend{}
@@ -625,10 +685,10 @@ func getSha1Value(input string) string {
 }
 
 // RetrieveAllAPISFromK8s retrieves all the API CRs from the Kubernetes cluster
-func RetrieveAllAPISFromK8s(k8sClient client.Client, nextToken string) ([]dpv1alpha2.API, string, error) {
+func RetrieveAllAPISFromK8s(k8sClient client.Client, nextToken string) ([]dpv1alpha3.API, string, error) {
 	conf, _ := config.ReadConfigs()
-	apiList := dpv1alpha2.APIList{}
-	resolvedAPIList := make([]dpv1alpha2.API, 0)
+	apiList := dpv1alpha3.APIList{}
+	resolvedAPIList := make([]dpv1alpha3.API, 0)
 	var err error
 	if nextToken == "" {
 		err = k8sClient.List(context.Background(), &apiList, &client.ListOptions{Namespace: conf.DataPlane.Namespace})

apim-apk-agent/internal/synchronizer/ratelimit_policy_fetcher.go

@@ -163,7 +163,7 @@ func FetchRateLimitPoliciesOnEvent(ratelimitName string, organization string, c
 
 // FetchSubscriptionRateLimitPoliciesOnEvent fetches the policies from the control plane on the start up and notification event updates
 func FetchSubscriptionRateLimitPoliciesOnEvent(ratelimitName string, organization string, c client.Client) {
-	logger.LoggerSynchronizer.Info("Fetching RateLimit Policies from Control Plane.")
+	logger.LoggerSynchronizer.Info("Fetching Subscription RateLimit Policies from Control Plane.")
 
 	// Read configurations and derive the eventHub details
 	conf, errReadConfig := config.ReadConfigs()
@@ -189,7 +189,7 @@ func FetchSubscriptionRateLimitPoliciesOnEvent(ratelimitName string, organizatio
 		}
 	}
 
-	logger.LoggerSynchronizer.Infof("Fetching RateLimit Policies from the URL %v: ", ehURL)
+	logger.LoggerSynchronizer.Infof("Fetching Subscription RateLimit Policies from the URL %v: ", ehURL)
 
 	ehUname := ehConfigs.Username
 	ehPass := ehConfigs.Password
@@ -201,19 +201,9 @@ func FetchSubscriptionRateLimitPoliciesOnEvent(ratelimitName string, organizatio
 	// Create a HTTP request
 	req, err := http.NewRequest("GET", ehURL, nil)
 	if err != nil {
-		logger.LoggerSynchronizer.Errorf("Error while creating http request for RateLimit Policies Endpoint : %v", err)
+		logger.LoggerSynchronizer.Errorf("Error while creating http request for Subscription RateLimit Policies Endpoint : %v", err)
 	}
 
-	var queryParamMap map[string]string
-
-	if queryParamMap != nil && len(queryParamMap) > 0 {
-		q := req.URL.Query()
-		// Making necessary query parameters for the request
-		for queryParamKey, queryParamValue := range queryParamMap {
-			q.Add(queryParamKey, queryParamValue)
-		}
-		req.URL.RawQuery = q.Encode()
-	}
 	// Setting authorization header
 	req.Header.Set(sync.Authorization, basicAuth)
 
@@ -231,45 +221,74 @@ func FetchSubscriptionRateLimitPoliciesOnEvent(ratelimitName string, organizatio
 	var errorMsg string
 	if err != nil {
 		errorMsg = "Error occurred while calling the REST API: " + policiesEndpoint
-		go retryRLPFetchData(conf, errorMsg, err, c)
+		go retrySubscriptionRLPFetchData(conf, errorMsg, err, c)
 		return
 	}
 	responseBytes, err := ioutil.ReadAll(resp.Body)
 	logger.LoggerSynchronizer.Debugf("Response String received for Policies: %v", string(responseBytes))
 
 	if err != nil {
 		errorMsg = "Error occurred while reading the response received for: " + policiesEndpoint
-		go retryRLPFetchData(conf, errorMsg, err, c)
+		go retrySubscriptionRLPFetchData(conf, errorMsg, err, c)
 		return
 	}
 
 	if resp.StatusCode == http.StatusOK {
 		var rateLimitPolicyList eventhubTypes.SubscriptionPolicyList
 		err := json.Unmarshal(responseBytes, &rateLimitPolicyList)
 		if err != nil {
-			logger.LoggerSynchronizer.Errorf("Error occurred while unmarshelling RateLimit Policies event data %v", err)
+			logger.LoggerSynchronizer.Errorf("Error occurred while unmarshelling Subscription RateLimit Policies event data %v", err)
 			return
 		}
 		logger.LoggerSynchronizer.Debugf("Policies received: %v", rateLimitPolicyList.List)
 		var rateLimitPolicies []eventhubTypes.SubscriptionPolicy = rateLimitPolicyList.List
 		for _, policy := range rateLimitPolicies {
-			if policy.DefaultLimit.RequestCount.TimeUnit == "min" {
-				policy.DefaultLimit.RequestCount.TimeUnit = "Minute"
-			} else if policy.DefaultLimit.RequestCount.TimeUnit == "hours" {
-				policy.DefaultLimit.RequestCount.TimeUnit = "Hour"
-			} else if policy.DefaultLimit.RequestCount.TimeUnit == "days" {
-				policy.DefaultLimit.RequestCount.TimeUnit = "Day"
+			if policy.QuotaType == "aiApiQuota" {
+				if policy.DefaultLimit.AiAPIQuota != nil {
+					switch policy.DefaultLimit.AiAPIQuota.TimeUnit {
+					case "min":
+						policy.DefaultLimit.AiAPIQuota.TimeUnit = "Minute"
+					case "hours":
+						policy.DefaultLimit.AiAPIQuota.TimeUnit = "Hour"
+					case "days":
+						policy.DefaultLimit.AiAPIQuota.TimeUnit = "Day"
+					default:
+						logger.LoggerSynchronizer.Errorf("Unsupported timeunit %s", policy.DefaultLimit.AiAPIQuota.TimeUnit)
+						continue
+					}
+					if policy.DefaultLimit.AiAPIQuota.PromptTokenCount == nil && policy.DefaultLimit.AiAPIQuota.TotalTokenCount != nil {
+						policy.DefaultLimit.AiAPIQuota.PromptTokenCount = policy.DefaultLimit.AiAPIQuota.TotalTokenCount
+					}
+					if policy.DefaultLimit.AiAPIQuota.CompletionTokenCount == nil && policy.DefaultLimit.AiAPIQuota.TotalTokenCount != nil {
+						policy.DefaultLimit.AiAPIQuota.CompletionTokenCount = policy.DefaultLimit.AiAPIQuota.TotalTokenCount
+					}
+					if policy.DefaultLimit.AiAPIQuota.TotalTokenCount == nil && policy.DefaultLimit.AiAPIQuota.PromptTokenCount != nil && policy.DefaultLimit.AiAPIQuota.CompletionTokenCount != nil {
+						total := *policy.DefaultLimit.AiAPIQuota.PromptTokenCount + *policy.DefaultLimit.AiAPIQuota.CompletionTokenCount
+						policy.DefaultLimit.AiAPIQuota.TotalTokenCount = &total
+					}
+					managementserver.AddSubscriptionPolicy(policy)
+					k8sclient.DeployAIRateLimitPolicyCR(policy, c)
+				} else {
+					logger.LoggerSynchronizer.Errorf("AIQuota type response recieved but no data found. %+v", policy.DefaultLimit)
+				}
+			} else {
+				if policy.DefaultLimit.RequestCount.TimeUnit == "min" {
+					policy.DefaultLimit.RequestCount.TimeUnit = "Minute"
+				} else if policy.DefaultLimit.RequestCount.TimeUnit == "hours" {
+					policy.DefaultLimit.RequestCount.TimeUnit = "Hour"
+				} else if policy.DefaultLimit.RequestCount.TimeUnit == "days" {
+					policy.DefaultLimit.RequestCount.TimeUnit = "Day"
+				}
+				managementserver.AddSubscriptionPolicy(policy)
+				logger.LoggerSynchronizer.Infof("RateLimit Policy added to internal map: %v", policy)
+				// Update the exisitng rate limit policies with current policy
+				k8sclient.DeploySubscriptionRateLimitPolicyCR(policy, c)
 			}
-			managementserver.AddSubscriptionPolicy(policy)
-			logger.LoggerSynchronizer.Infof("RateLimit Policy added to internal map: %v", policy)
-			// Update the exisitng rate limit policies with current policy
-			k8sclient.DeploySubscriptionRateLimitPolicyCR(policy, c)
-
 		}
 	} else {
 		errorMsg = "Failed to fetch data! " + policiesEndpoint + " responded with " +
 			strconv.Itoa(resp.StatusCode)
-		go retryRLPFetchData(conf, errorMsg, err, c)
+		go retrySubscriptionRLPFetchData(conf, errorMsg, err, c)
 	}
 }
 
@@ -284,3 +303,15 @@ func retryRLPFetchData(conf *config.Config, errorMessage string, err error, c cl
 		return
 	}
 }
+
+func retrySubscriptionRLPFetchData(conf *config.Config, errorMessage string, err error, c client.Client) {
+	logger.LoggerSynchronizer.Debugf("Time Duration for retrying: %v",
+		conf.ControlPlane.RetryInterval*time.Second)
+	time.Sleep(conf.ControlPlane.RetryInterval * time.Second)
+	FetchSubscriptionRateLimitPoliciesOnEvent("", "", c)
+	retryAttempt++
+	if retryAttempt >= retryCount {
+		logger.LoggerSynchronizer.Errorf(errorMessage, err)
+		return
+	}
+}

apim-apk-agent/pkg/eventhub/types/types.go

@@ -196,7 +196,8 @@ type ConditionGroup struct {
 
 // DefaultLimit represents the default limit within the response.
 type DefaultLimit struct {
-	QuotaType    string `json:"quotaType"`
+	AiAPIQuota   *AiAPIQuota `json:"aiApiQuota"`
+	QuotaType    string      `json:"quotaType"`
 	RequestCount struct {
 		TimeUnit     string `json:"timeUnit"`
 		UnitTime     int    `json:"unitTime"`
@@ -206,6 +207,16 @@ type DefaultLimit struct {
 	EventCount interface{} `json:"eventCount"`
 }
 
+// AiAPIQuota contains the AI ratelimit configurations
+type AiAPIQuota struct {
+	CompletionTokenCount *int   `json:"completionTokenCount"`
+	PromptTokenCount     *int   `json:"promptTokenCount"`
+	RequestCount         *int   `json:"requestCount"`
+	TimeUnit             string `json:"timeUnit"`
+	TotalTokenCount      *int   `json:"totalTokenCount"`
+	UnitTime             int    `json:"unitTime"`
+}
+
 // Scope for struct Scope
 type Scope struct {
 	Name            string `json:"name"`

apim-apk-agent/pkg/synchronizer/apis_fetcher.go

@@ -88,7 +88,7 @@ func SendRequestToControlPlane(req *http.Request, apiID *string, gwLabels []stri
 	if apiID != nil {
 		logger.LoggerSync.Debugf("Sending the control plane request for the API: %q", *apiID)
 	} else {
-		logger.LoggerSync.Debug("Sending the control plane request")
+		logger.LoggerSync.Debugf("Sending the control plane request, url: %s", req.URL.String())
 	}
 	resp, err := client.Do(req)
 

apim-apk-agent/pkg/transformer/api_model.go

@@ -179,6 +179,7 @@ type APIMApi struct {
 	APIThrottlingPolicy string                `yaml:"apiThrottlingPolicy"`
 	APIPolicies         APIMOperationPolicies `yaml:"apiPolicies"`
 	AIConfiguration     APIMAIConfiguration   `yaml:"aiConfiguration"`
+	MaxTps              *MaxTps               `yaml:"maxTps"`
 }
 
 // APIMAIConfiguration holds the configuration details for AI providers
@@ -192,6 +193,46 @@ type APIYaml struct {
 	Data APIMApi `json:"data"`
 }
 
+// MaxTps represents the maximum transactions per second (TPS) settings for both
+// production and sandbox environments. It also includes an optional configuration
+// for token-based throttling.
+//
+// Fields:
+// - Production: Maximum TPS for the production environment.
+// - ProductionTimeUnit: The time unit for the production TPS limit (e.g., seconds, minutes).
+// - Sandbox: Maximum TPS for the sandbox environment.
+// - SandboxTimeUnit: The time unit for the sandbox TPS limit.
+// - TokenBasedThrottlingConfiguration: Configuration for token-based throttling.
+type MaxTps struct {
+	Production                        *int                        `yaml:"production"`
+	ProductionTimeUnit                *string                     `yaml:"productionTimeUnit"`
+	Sandbox                           *int                        `yaml:"sandbox"`
+	SandboxTimeUnit                   *string                     `yaml:"sandboxTimeUnit"`
+	TokenBasedThrottlingConfiguration *TokenBasedThrottlingConfig `yaml:"tokenBasedThrottlingConfiguration"`
+}
+
+// TokenBasedThrottlingConfig defines the token-based throttling limits for
+// both production and sandbox environments. Token-based throttling places
+// a limit on the number of prompt and completion tokens that can be used.
+//
+// Fields:
+// - ProductionMaxPromptTokenCount: Maximum number of prompt tokens for production.
+// - ProductionMaxCompletionTokenCount: Maximum number of completion tokens for production.
+// - ProductionMaxTotalTokenCount: Maximum total token count (prompt + completion) for production.
+// - SandboxMaxPromptTokenCount: Maximum number of prompt tokens for sandbox.
+// - SandboxMaxCompletionTokenCount: Maximum number of completion tokens for sandbox.
+// - SandboxMaxTotalTokenCount: Maximum total token count (prompt + completion) for sandbox.
+// - IsTokenBasedThrottlingEnabled: Flag to enable or disable token-based throttling.
+type TokenBasedThrottlingConfig struct {
+	ProductionMaxPromptTokenCount     *int  `yaml:"productionMaxPromptTokenCount"`
+	ProductionMaxCompletionTokenCount *int  `yaml:"productionMaxCompletionTokenCount"`
+	ProductionMaxTotalTokenCount      *int  `yaml:"productionMaxTotalTokenCount"`
+	SandboxMaxPromptTokenCount        *int  `yaml:"sandboxMaxPromptTokenCount"`
+	SandboxMaxCompletionTokenCount    *int  `yaml:"sandboxMaxCompletionTokenCount"`
+	SandboxMaxTotalTokenCount         *int  `yaml:"sandboxMaxTotalTokenCount"`
+	IsTokenBasedThrottlingEnabled     *bool `yaml:"isTokenBasedThrottlingEnabled"`
+}
+
 // APIArtifact represents the artifact details of an API, including api details, environment configuration,
 // Swagger definition, deployment descriptor, and revision ID extracted from the API Project Zip.
 type APIArtifact struct {