Move RequestBody from scheduling to requesthandling (kubernetes-sigs/gateway-api-inference-extension#2808)

Co-authored-by: Ryan Rosario <6713180+RyanRosario@users.noreply.github.com>

Author: RyanRosario

pkg/epp/framework/interface/requestcontrol/types.go

@@ -18,6 +18,8 @@ package requestcontrol
 
 import (
 	"google.golang.org/protobuf/types/known/structpb"
+
+	requesthandling "sigs.k8s.io/gateway-api-inference-extension/pkg/epp/framework/interface/requesthandling"
 )
 
 // Response contains information from the response received to be passed to the Response requestcontrol plugins
@@ -33,19 +35,8 @@ type Response struct {
 	// Currently, this is only used by conformance test.
 	ReqMetadata map[string]any
 	// Token usage counts parsed from the response body.
-	Usage Usage
+	Usage requesthandling.Usage
 	// DynamicMetadata is a map of metadata that can be passed to the Envoy. It is populated into the dynamic
 	// metadata when processing ProcessingResponse_RequestHeaders.
 	DynamicMetadata *structpb.Struct
 }
-
-type Usage struct {
-	PromptTokens       int                 `json:"prompt_tokens"`
-	CompletionTokens   int                 `json:"completion_tokens"`
-	TotalTokens        int                 `json:"total_tokens"`
-	PromptTokenDetails *PromptTokenDetails `json:"prompt_token_details,omitempty"`
-}
-
-type PromptTokenDetails struct {
-	CachedTokens int `json:"cached_tokens"`
-}

pkg/epp/framework/interface/requesthandling/plugins.go

@@ -21,16 +21,13 @@ import (
 
 	v1 "sigs.k8s.io/gateway-api-inference-extension/api/v1"
 	fwkplugin "sigs.k8s.io/gateway-api-inference-extension/pkg/epp/framework/interface/plugin"
-
-	"sigs.k8s.io/gateway-api-inference-extension/pkg/epp/framework/interface/requestcontrol"
-	"sigs.k8s.io/gateway-api-inference-extension/pkg/epp/framework/interface/scheduling"
 )
 
 // Parser defines the interface for parsing payload(requests and responses).
 type Parser interface {
 	fwkplugin.Plugin
 	// ParseRequest parses the request body and headers and returns a map representation.
-	ParseRequest(ctx context.Context, body []byte, headers map[string]string) (*scheduling.InferenceRequestBody, error)
+	ParseRequest(ctx context.Context, body []byte, headers map[string]string) (*InferenceRequestBody, error)
 
 	// ParseResponse parses the response payload.
 	// For streaming responses , this method is invoked multiple times (once per chunk),
@@ -46,5 +43,5 @@ type Parser interface {
 
 type ParsedResponse struct {
 	// Usage is only populate when the raw response has usage.
-	Usage *requestcontrol.Usage
+	Usage *Usage
 }