test(integration): expect drop-reason header on no-candidates rejection

The director sets x-llm-d-request-dropped-reason: rejected-no-endpoints on
the 503 returned when subsetting leaves no endpoint candidates. Update the
hermetic integration fixtures to expect the header.

Signed-off-by: Luke Van Drie <lukevandrie@google.com>

Author: LukeAVanDrie

test/integration/epp/common_tests.go

@@ -26,6 +26,7 @@ import (
 	extProcPb "github.com/envoyproxy/go-control-plane/envoy/service/ext_proc/v3"
 	envoyTypePb "github.com/envoyproxy/go-control-plane/envoy/type/v3"
 
+	errcommon "github.com/llm-d/llm-d-router/pkg/common/error"
 	reqcommon "github.com/llm-d/llm-d-router/pkg/common/request"
 	integration "github.com/llm-d/llm-d-router/test/integration"
 )
@@ -208,6 +209,17 @@ func ExpectReject(code envoyTypePb.StatusCode, msg string) []*extProcPb.Processi
 	return integration.NewImmediateErrorResponse(code, msg)
 }
 
+// ExpectRejectWithDropReason asserts that the EPP immediately rejected the request with the given code
+// and message, and that the response carries the drop-reason header.
+func ExpectRejectWithDropReason(code envoyTypePb.StatusCode, msg string, reason errcommon.RequestDroppedReason) []*extProcPb.ProcessingResponse {
+	return integration.NewImmediateErrorResponse(code, msg, &envoyCorev3.HeaderValueOption{
+		Header: &envoyCorev3.HeaderValue{
+			Key:      errcommon.RequestDroppedReasonHeaderKey,
+			RawValue: []byte(reason),
+		},
+	})
+}
+
 // ExpectBufferResp asserts that the EPP buffers the response and rewrites the body.
 // This uses the shared primitive but adds EPP-specific headers we expect.
 func ExpectBufferResp(body string, contentType string) []*extProcPb.ProcessingResponse {

test/integration/epp/grpc_test.go

@@ -27,6 +27,7 @@ import (
 	"github.com/stretchr/testify/require"
 	"google.golang.org/protobuf/testing/protocmp"
 
+	errcommon "github.com/llm-d/llm-d-router/pkg/common/error"
 	reqcommon "github.com/llm-d/llm-d-router/pkg/common/request"
 	pb "github.com/llm-d/llm-d-router/pkg/epp/framework/plugins/requesthandling/parsers/vllmgrpc/api/gen"
 	integration "github.com/llm-d/llm-d-router/test/integration"
@@ -204,8 +205,9 @@ func TestFullDuplexStreamed_GRPC_KubeInferenceObjectiveRequest(t *testing.T) {
 				P(0, 0, 0.2, "foo"),
 				P(1, 0, 0.1, "foo", modelSQLLoraTarget),
 			},
-			wantResponses: ExpectReject(envoyTypePb.StatusCode_ServiceUnavailable,
-				"inference error: ServiceUnavailable - failed to find endpoint candidates for serving the request"),
+			wantResponses: ExpectRejectWithDropReason(envoyTypePb.StatusCode_ServiceUnavailable,
+				"inference error: ServiceUnavailable - failed to find endpoint candidates for serving the request",
+				errcommon.RequestDroppedReasonNoEndpoints),
 		},
 
 		// --- Response Processing (Non-streaming) ---

test/integration/epp/hermetic_test.go

@@ -44,6 +44,7 @@ import (
 	v1 "sigs.k8s.io/gateway-api-inference-extension/api/v1"
 
 	"github.com/llm-d/llm-d-router/apix/v1alpha2"
+	errcommon "github.com/llm-d/llm-d-router/pkg/common/error"
 	reqcommon "github.com/llm-d/llm-d-router/pkg/common/request"
 	"github.com/llm-d/llm-d-router/pkg/epp/metadata"
 	"github.com/llm-d/llm-d-router/pkg/epp/metrics"
@@ -293,8 +294,9 @@ dataLayer:
 						P(0, 0, 0.2, "foo"),
 						P(1, 0, 0.1, "foo", modelSQLLoraTarget),
 					},
-					wantResponses: ExpectReject(envoyTypePb.StatusCode_ServiceUnavailable,
-						"inference error: ServiceUnavailable - failed to find endpoint candidates for serving the request"),
+					wantResponses: ExpectRejectWithDropReason(envoyTypePb.StatusCode_ServiceUnavailable,
+						"inference error: ServiceUnavailable - failed to find endpoint candidates for serving the request",
+						errcommon.RequestDroppedReasonNoEndpoints),
 				},
 
 				// --- Request Modification (Passthrough & Rewrite) ---

test/integration/util.go

@@ -441,16 +441,20 @@ func NewResponseStreamChunk(body string, endOfStream bool) *extProcPb.Processing
 	}
 }
 
-// NewImmediateErrorResponse creates a response that immediately terminates the request with a specific HTTP status code
-// and body.
+// NewImmediateErrorResponse creates a response that immediately terminates the request with a specific HTTP status code,
+// body, and optional response headers.
 // Use this for testing Load Shedding (503), Rate Limiting (429), or Bad Request (400) logic.
-func NewImmediateErrorResponse(code envoyTypePb.StatusCode, body string) []*extProcPb.ProcessingResponse {
+func NewImmediateErrorResponse(code envoyTypePb.StatusCode, body string, headers ...*envoyCorev3.HeaderValueOption) []*extProcPb.ProcessingResponse {
+	immediateResponse := &extProcPb.ImmediateResponse{
+		Status: &envoyTypePb.HttpStatus{Code: code},
+		Body:   []byte(body),
+	}
+	if len(headers) > 0 {
+		immediateResponse.Headers = &extProcPb.HeaderMutation{SetHeaders: headers}
+	}
 	return []*extProcPb.ProcessingResponse{{
 		Response: &extProcPb.ProcessingResponse_ImmediateResponse{
-			ImmediateResponse: &extProcPb.ImmediateResponse{
-				Status: &envoyTypePb.HttpStatus{Code: code},
-				Body:   []byte(body),
-			},
+			ImmediateResponse: immediateResponse,
 		},
 	}}
 }