test: extract EPP envtest harness into test/framework/epp/harness

.golangci.yml

@@ -89,6 +89,8 @@ linters:
           alias: fwkgaie
         - pkg: github.com/llm-d/llm-d-router/test/framework/epp
           alias: fwkepp
+        - pkg: github.com/llm-d/llm-d-router/test/framework/epp/harness
+          alias: eppharness
     revive: # see https://github.com/mgechev/revive#available-rules for all options
       rules:
         - name: blank-imports

Makefile

@@ -167,7 +167,10 @@ sidecar_TEST_PACKAGES = ./pkg/sidecar/...
 # block (epp+sidecar only); CI's baseline cache unions all coverage/*.out, so
 # framework enters the CI baseline once this is on main. The PR run reports
 # coverage/framework.out as a new component.
-framework_TEST_PACKAGES = ./test/framework/...
+# The EPP harness is excluded: it is envtest-driven and exercised by
+# make test-integration-hermetic, so unit coverage would report 0% for code
+# that is in fact covered.
+framework_TEST_PACKAGES = $$(go list ./test/framework/... | grep -vE '/test/framework/epp/harness($$|/)' | tr '\n' ' ')
 
 # Internal variables for generic targets
 epp_IMAGE = $(EPP_IMAGE)

test/integration/epp/harness.go → test/framework/epp/harness/harness.go

@@ -14,7 +14,12 @@ See the License for the specific language governing permissions and
 limitations under the License.
 */
 
-package epp
+// Package harness boots a shared envtest environment and per-test EPP servers for
+// integration suites.
+//
+// It sits at the top of the framework layering and may import anything, including
+// pkg/epp/server and cmd/epp/runner; only integration and e2e suites import it.
+package harness
 
 import (
 	"context"
@@ -24,6 +29,7 @@ import (
 	"os"
 	"os/exec"
 	"path/filepath"
+	"strconv"
 	"strings"
 	"testing"
 	"time"
@@ -585,3 +591,60 @@ func (h *TestHarness) ExpectMetrics(expected map[string]string) {
 		}
 	}
 }
+
+// --- Data Structures & Metrics Helpers ---
+
+type PodState struct {
+	index        int
+	queueSize    int
+	kvCacheUsage float64
+	activeModels []string
+}
+
+// P constructs a Pod State: Index, Queue, KV%, Models...
+// Usage: P(0, 5, 0.2, "model-a")
+func P(idx int, q int, kv float64, models ...string) PodState {
+	return PodState{index: idx, queueSize: q, kvCacheUsage: kv, activeModels: models}
+}
+
+type label struct{ name, value string }
+
+func labelsToString(labels []label) string {
+	parts := make([]string, len(labels))
+	for i, l := range labels {
+		parts[i] = fmt.Sprintf("%s=%q", l.name, l.value)
+	}
+	return strings.Join(parts, ",")
+}
+
+// MetricReqTotal renders the expected inference_objective_request_total exposition text.
+func MetricReqTotal(model, target string, priority int) string {
+	return fmt.Sprintf(`
+    # HELP inference_objective_request_total [ALPHA] [Deprecated: Use llm_d_epp_request_total] Counter of inference objective requests broken out for each model and target model.
+    # TYPE inference_objective_request_total counter
+    inference_objective_request_total{%s} 1
+    `, labelsToString([]label{{"model_name", model}, {"priority", strconv.Itoa(priority)}, {"target_model_name", target}}))
+}
+
+// MetricReadyPods renders the expected inference_pool_ready_pods exposition text.
+func MetricReadyPods(count int) string {
+	return fmt.Sprintf(`
+    # HELP inference_pool_ready_pods [ALPHA] [Deprecated: Use llm_d_epp_ready_endpoints] The number of ready pods in the inference server pool.
+    # TYPE inference_pool_ready_pods gauge
+    inference_pool_ready_pods{%s} %d
+    `, labelsToString([]label{{"name", TestPoolName}}), count)
+}
+
+// CleanMetric removes indentation from multiline metric strings and ensures a trailing newline exists, which is
+// required by the Prometheus text parser.
+func CleanMetric(s string) string {
+	lines := strings.Split(s, "\n")
+	var cleaned []string
+	for _, l := range lines {
+		trimmed := strings.TrimSpace(l)
+		if trimmed != "" {
+			cleaned = append(cleaned, trimmed)
+		}
+	}
+	return strings.Join(cleaned, "\n") + "\n"
+}

test/integration/epp/common_tests.go

@@ -18,16 +18,14 @@ package epp
 
 import (
 	"encoding/json"
-	"fmt"
-	"strconv"
-	"strings"
 
 	envoyCorev3 "github.com/envoyproxy/go-control-plane/envoy/config/core/v3"
 	extProcPb "github.com/envoyproxy/go-control-plane/envoy/service/ext_proc/v3"
 	envoyTypePb "github.com/envoyproxy/go-control-plane/envoy/type/v3"
 
 	reqcommon "github.com/llm-d/llm-d-router/pkg/common/request"
 	fwkepp "github.com/llm-d/llm-d-router/test/framework/epp"
+	eppharness "github.com/llm-d/llm-d-router/test/framework/epp/harness"
 )
 
 // Model name constants shared across test suites.
@@ -52,7 +50,7 @@ func buildEnvoyHeaders(headers map[string]string) []*envoyCorev3.HeaderValue {
 // This simulates the "Subset Load Balancing" flow where EPP picks a specific pod IP.
 func ReqSubset(prompt, model, target string, subsets ...string) []*extProcPb.ProcessingRequest {
 	// Uses the shared low-level generator which handles the metadata construction
-	return fwkepp.GenerateStreamedRequestSet(Logger(), prompt, model, target, subsets)
+	return fwkepp.GenerateStreamedRequestSet(eppharness.Logger(), prompt, model, target, subsets)
 }
 
 // ReqResponseOnly creates a sequence simulating only the response phase from Envoy.
@@ -276,7 +274,7 @@ func ExpectGRPCStreamResp(chunks ...string) []*extProcPb.ProcessingResponse {
 type testCase struct {
 	name          string
 	requests      []*extProcPb.ProcessingRequest
-	pods          []PodState
+	pods          []eppharness.PodState
 	configText    string
 	wantResponses []*extProcPb.ProcessingResponse
 	wantMetrics   map[string]string
@@ -294,30 +292,30 @@ func commonTestCases(prio func(int) int) []testCase {
 	return []testCase{
 		{
 			name:     "select lower queue and kv cache",
-			requests: fwkepp.ReqLLM(Logger(), "test1", modelMyModel, modelMyModelTarget),
-			pods: []PodState{
-				P(0, 3, 0.2),
-				P(1, 0, 0.1), // Winner (Low Queue, Low KV)
-				P(2, 10, 0.2),
+			requests: fwkepp.ReqLLM(eppharness.Logger(), "test1", modelMyModel, modelMyModelTarget),
+			pods: []eppharness.PodState{
+				eppharness.P(0, 3, 0.2),
+				eppharness.P(1, 0, 0.1), // Winner (Low Queue, Low KV)
+				eppharness.P(2, 10, 0.2),
 			},
 			wantResponses: ExpectRouteTo("192.168.1.2:8000", modelMyModelTarget, "test1"),
 			wantMetrics: map[string]string{
-				"inference_objective_request_total": CleanMetric(MetricReqTotal(modelMyModel, modelMyModelTarget, prio(2))),
-				"inference_pool_ready_pods":         CleanMetric(MetricReadyPods(3)),
+				"inference_objective_request_total": eppharness.CleanMetric(eppharness.MetricReqTotal(modelMyModel, modelMyModelTarget, prio(2))),
+				"inference_pool_ready_pods":         eppharness.CleanMetric(eppharness.MetricReadyPods(3)),
 			},
 			wantSpans: []string{"gateway.request", "gateway.request_orchestration"},
 		},
 		{
 			name:     "select active lora, low queue",
-			requests: fwkepp.ReqLLM(Logger(), "test2", modelSQLLora, modelSQLLoraTarget),
-			pods: []PodState{
-				P(0, 0, 0.2, "foo", "bar"),
-				P(1, 0, 0.1, "foo", modelSQLLoraTarget), // Winner (Has LoRA)
-				P(2, 10, 0.2, "foo", "bar"),
+			requests: fwkepp.ReqLLM(eppharness.Logger(), "test2", modelSQLLora, modelSQLLoraTarget),
+			pods: []eppharness.PodState{
+				eppharness.P(0, 0, 0.2, "foo", "bar"),
+				eppharness.P(1, 0, 0.1, "foo", modelSQLLoraTarget), // Winner (Has LoRA)
+				eppharness.P(2, 10, 0.2, "foo", "bar"),
 			},
 			wantResponses: ExpectRouteTo("192.168.1.2:8000", modelSQLLoraTarget, "test2"),
 			wantMetrics: map[string]string{
-				"inference_objective_request_total": CleanMetric(MetricReqTotal(modelSQLLora, modelSQLLoraTarget, prio(2))),
+				"inference_objective_request_total": eppharness.CleanMetric(eppharness.MetricReqTotal(modelSQLLora, modelSQLLoraTarget, prio(2))),
 			},
 		},
 		{
@@ -338,60 +336,3 @@ func commonTestCases(prio func(int) int) []testCase {
 		},
 	}
 }
-
-// --- Data Structures & Metrics Helpers ---
-
-type PodState struct {
-	index        int
-	queueSize    int
-	kvCacheUsage float64
-	activeModels []string
-}
-
-// P constructs a Pod State: Index, Queue, KV%, Models...
-// Usage: P(0, 5, 0.2, "model-a")
-func P(idx int, q int, kv float64, models ...string) PodState {
-	return PodState{index: idx, queueSize: q, kvCacheUsage: kv, activeModels: models}
-}
-
-type label struct{ name, value string }
-
-func labelsToString(labels []label) string {
-	parts := make([]string, len(labels))
-	for i, l := range labels {
-		parts[i] = fmt.Sprintf("%s=%q", l.name, l.value)
-	}
-	return strings.Join(parts, ",")
-}
-
-// MetricReqTotal renders the expected inference_objective_request_total exposition text.
-func MetricReqTotal(model, target string, priority int) string {
-	return fmt.Sprintf(`
-    # HELP inference_objective_request_total [ALPHA] [Deprecated: Use llm_d_epp_request_total] Counter of inference objective requests broken out for each model and target model.
-    # TYPE inference_objective_request_total counter
-    inference_objective_request_total{%s} 1
-    `, labelsToString([]label{{"model_name", model}, {"priority", strconv.Itoa(priority)}, {"target_model_name", target}}))
-}
-
-// MetricReadyPods renders the expected inference_pool_ready_pods exposition text.
-func MetricReadyPods(count int) string {
-	return fmt.Sprintf(`
-    # HELP inference_pool_ready_pods [ALPHA] [Deprecated: Use llm_d_epp_ready_endpoints] The number of ready pods in the inference server pool.
-    # TYPE inference_pool_ready_pods gauge
-    inference_pool_ready_pods{%s} %d
-    `, labelsToString([]label{{"name", TestPoolName}}), count)
-}
-
-// CleanMetric removes indentation from multiline metric strings and ensures a trailing newline exists, which is
-// required by the Prometheus text parser.
-func CleanMetric(s string) string {
-	lines := strings.Split(s, "\n")
-	var cleaned []string
-	for _, l := range lines {
-		trimmed := strings.TrimSpace(l)
-		if trimmed != "" {
-			cleaned = append(cleaned, trimmed)
-		}
-	}
-	return strings.Join(cleaned, "\n") + "\n"
-}

test/integration/epp/datalayer_integration_test.go

@@ -26,6 +26,7 @@ import (
 	"google.golang.org/protobuf/testing/protocmp"
 
 	fwkepp "github.com/llm-d/llm-d-router/test/framework/epp"
+	eppharness "github.com/llm-d/llm-d-router/test/framework/epp/harness"
 )
 
 // TestFullDuplexStreamed_DataLayer runs integration tests through the datalayer metrics pipeline.
@@ -39,7 +40,7 @@ func TestFullDuplexStreamed_DataLayer(t *testing.T) {
 			ctx, cancel := context.WithCancel(context.Background())
 			defer cancel()
 
-			h := NewTestHarness(ctx, t, WithStandardMode())
+			h := eppharness.NewTestHarness(ctx, t, eppharness.WithStandardMode())
 			h.WithBaseResources().WithPods(tc.pods).WaitForSync(len(tc.pods), modelMyModel)
 			if len(tc.pods) > 0 {
 				h.WaitForReadyPodsMetric(len(tc.pods))

test/integration/epp/dynamic_attributes_test.go

@@ -27,6 +27,7 @@ import (
 	reqcommon "github.com/llm-d/llm-d-router/pkg/common/request"
 	"github.com/llm-d/llm-d-router/pkg/epp/metadata"
 	fwkepp "github.com/llm-d/llm-d-router/test/framework/epp"
+	eppharness "github.com/llm-d/llm-d-router/test/framework/epp/harness"
 )
 
 func TestDynamicAttributes_Concurrency(t *testing.T) {
@@ -73,12 +74,12 @@ flowControl:
 `
 
 	ctx := t.Context()
-	h := NewTestHarness(ctx, t, WithConfigText(configText), WithStandardMode())
+	h := eppharness.NewTestHarness(ctx, t, eppharness.WithConfigText(configText), eppharness.WithStandardMode())
 	h = h.WithBaseResources()
 
 	// Add one pod. We need it to support modelMyModelTarget.
-	pods := []PodState{
-		P(0, 0, 0.0, modelMyModelTarget),
+	pods := []eppharness.PodState{
+		eppharness.P(0, 0, 0.0, modelMyModelTarget),
 	}
 	h.WithPods(pods).WaitForSync(len(pods), modelMyModel)
 	h.WaitForReadyPodsMetric(len(pods))

test/integration/epp/e2e_config_smoke_test.go

@@ -10,7 +10,11 @@ You may obtain a copy of the License at
 
 package epp
 
-import "testing"
+import (
+	"testing"
+
+	eppharness "github.com/llm-d/llm-d-router/test/framework/epp/harness"
+)
 
 // e2eConfigsForSmoke mirrors the YAML strings in test/e2e/configs_test.go so
 // the hermetic harness validates that the e2e fixtures actually parse +
@@ -171,9 +175,9 @@ schedulingProfiles:
 func TestE2EConfigs_StartupParseSmoke(t *testing.T) {
 	for name, yaml := range e2eConfigsForSmoke {
 		t.Run(name, func(t *testing.T) {
-			// NewTestHarness fails the test if config-parse / plugin-init
+			// eppharness.NewTestHarness fails the test if config-parse / plugin-init
 			// errors — exactly what we want to catch before pushing.
-			_ = NewTestHarness(t.Context(), t, WithConfigText(yaml))
+			_ = eppharness.NewTestHarness(t.Context(), t, eppharness.WithConfigText(yaml))
 		})
 	}
 }