fix: handles weight zero backendRef gracefully (#1755)

**Description**

Envoy Gateway skips the weight zero backendRef when constructing
localityLbEndoints slices. This starts handling such cases gracefully
rather than taking them as errors.

**Related Issues/PRs (if applicable)**
Close #1664
Supersedes #1694

---------

Signed-off-by: Takeshi Yoneda <t.y.mathetake@gmail.com>

Author: mathetake

internal/extensionserver/extensionserver_test.go

@@ -122,6 +122,7 @@ func Test_maybeModifyCluster(t *testing.T) {
 				{
 					BackendRefs: []aigv1a1.AIGatewayRouteRuleBackendRef{
 						{Name: "aaa", Priority: ptr.To[uint32](0)},
+						{Name: "to-be-ignored", Weight: ptr.To[int32](0)},
 						{Name: "bbb", Priority: ptr.To[uint32](1)},
 					},
 				},
@@ -144,10 +145,6 @@ func Test_maybeModifyCluster(t *testing.T) {
 		{c: &clusterv3.Cluster{
 			Name: "httproute/ns/myroute/rule/0",
 		}, errLog: `LoadAssignment is nil`},
-		{c: &clusterv3.Cluster{
-			Name:           "httproute/ns/myroute/rule/0",
-			LoadAssignment: &endpointv3.ClusterLoadAssignment{},
-		}, errLog: `LoadAssignment endpoints length does not match backend refs length`},
 	} {
 		t.Run("error/"+tc.errLog, func(t *testing.T) {
 			var buf bytes.Buffer

internal/extensionserver/post_translate_modify.go

@@ -207,14 +207,16 @@ func (s *Server) maybeModifyCluster(cluster *clusterv3.Cluster) error {
 			s.log.Info("LoadAssignment is nil", "cluster_name", cluster.Name)
 			return nil
 		}
-		if len(cluster.LoadAssignment.Endpoints) != len(httpRouteRule.BackendRefs) {
-			s.log.Info("LoadAssignment endpoints length does not match backend refs length",
-				"cluster_name", cluster.Name, "endpoints_length", len(cluster.LoadAssignment.Endpoints), "backend_refs_length", len(httpRouteRule.BackendRefs))
-			return nil
-		}
 		// Populate the metadata for each endpoint in the LoadAssignment.
-		for i, endpoints := range cluster.LoadAssignment.Endpoints {
-			backendRef := httpRouteRule.BackendRefs[i]
+		var lbEndpointIndex int
+		for i, backendRef := range httpRouteRule.BackendRefs {
+			// The weight of 0 means this backend is disabled and is not included in the LoadAssignment by EG,
+			// so we skip it here.
+			if backendRef.Weight != nil && *backendRef.Weight == 0 {
+				continue
+			}
+			endpoints := cluster.LoadAssignment.Endpoints[lbEndpointIndex]
+			lbEndpointIndex++
 			name := backendRef.Name
 			namespace := aigwRoute.Namespace
 			if backendRef.Priority != nil {

tests/e2e/testdata/testupstream.yaml

@@ -52,6 +52,8 @@ spec:
       kind: Gateway
       group: gateway.networking.k8s.io
   rules:
+    # The backend refs with weight zero should be ignored by the extension server.
+    # See https://github.com/envoyproxy/ai-gateway/issues/1664
     - matches:
         - headers:
             - type: Exact
@@ -65,14 +67,23 @@ spec:
               name: x-ai-eg-model
               value: another-cool-model
       backendRefs:
+        - name: translation-testupstream-cool-model-backend
+          weight: 0
         - name: translation-testupstream-another-cool-model-backend
+          weight: 100
+        - name: translation-testupstream-cool-model-backend
+          weight: 0
+        - name: translation-testupstream-cool-model-backend
+          weight: 0
     - matches:
         - headers:
             - type: Exact
               name: x-ai-eg-model
               value: whatever-model
       backendRefs:
         - name: translation-testupstream-cool-model-backend
+        - name: translation-testupstream-another-cool-model-backend
+          weight: 0
 ---
 apiVersion: aigateway.envoyproxy.io/v1alpha1
 kind: AIServiceBackend