Merge branch 'optimize/ai-gateway-round2-20260619'

AI Gateway round 2 optimizations:

- perf: parallelize RAG List calls with errgroup (5 calls → concurrent)
- perf: add chat history limit (max 40 messages) to prevent unbounded growth

Author: mahmut-Abi

go.mod

@@ -14,7 +14,7 @@ require (
 	go.opentelemetry.io/otel/sdk v1.43.0
 	go.uber.org/goleak v1.3.0
 	golang.org/x/net v0.55.0
-	golang.org/x/sync v0.20.0
+	golang.org/x/sync v0.21.0
 	google.golang.org/grpc v1.80.0
 	google.golang.org/protobuf v1.36.12-0.20260120151049-f2248ac996af
 	gopkg.in/yaml.v3 v3.0.1

go.sum

@@ -149,8 +149,8 @@ golang.org/x/net v0.55.0 h1:bcvxaJn3e1U6InsFWt1JUq1aSjnRxLzT2rtD2KfkDF8=
 golang.org/x/net v0.55.0/go.mod h1:L5U2KuzuOe1lY7Z+aWVIKK6qEeJXnXV9yzGA+WCHJww=
 golang.org/x/oauth2 v0.34.0 h1:hqK/t4AKgbqWkdkcAeI8XLmbK+4m4G5YeQRrmiotGlw=
 golang.org/x/oauth2 v0.34.0/go.mod h1:lzm5WQJQwKZ3nwavOZ3IS5Aulzxi68dUSgRHujetwEA=
-golang.org/x/sync v0.20.0 h1:e0PTpb7pjO8GAtTs2dQ6jYa5BWYlMuX047Dco/pItO4=
-golang.org/x/sync v0.20.0/go.mod h1:9xrNwdLfx4jkKbNva9FpL6vEN7evnE43NNNJQ2LF3+0=
+golang.org/x/sync v0.21.0 h1:HLII4xRRTtCRkxYp4HNFF0Js/Og6q2i++KXbg0gHCwM=
+golang.org/x/sync v0.21.0/go.mod h1:9xrNwdLfx4jkKbNva9FpL6vEN7evnE43NNNJQ2LF3+0=
 golang.org/x/sys v0.45.0 h1:dO4czNzziLiiXplLQgBCEpCvXQ3dnkn0SdaZSYdQ+FY=
 golang.org/x/sys v0.45.0/go.mod h1:4GL1E5IUh+htKOUEOaiffhrAeqysfVGipDYzABqnCmw=
 golang.org/x/term v0.43.0 h1:S4RLU2sB31O/NCl+zFN9Aru9A/Cq2aqKpTZJ6B+DwT4=

internal/admin/chatbot/rag.go

@@ -6,6 +6,7 @@ import (
 	"strings"
 
 	corev1 "k8s.io/api/core/v1"
+	"golang.org/x/sync/errgroup"
 	"sigs.k8s.io/controller-runtime/pkg/client"
 	gatewayv1 "sigs.k8s.io/gateway-api/apis/v1"
 )
@@ -20,15 +21,26 @@ import (
 //   - HTTPRoutes and GRPCRoutes (namespace/name, rules, backendRefs).
 //   - Available Services (namespace/name, ports).
 func BuildRAGContext(ctx context.Context, cl client.Client, controllerName string) (string, error) {
-	var b strings.Builder
-
-	// 1. GatewayClasses controlled by this controller.
-	gcList := &gatewayv1.GatewayClassList{}
-	if err := cl.List(ctx, gcList); err != nil {
-		return "", fmt.Errorf("rag: list GatewayClasses: %w", err)
+	var (
+		gcList   gatewayv1.GatewayClassList
+		gwList   gatewayv1.GatewayList
+		httpList gatewayv1.HTTPRouteList
+		grpcList gatewayv1.GRPCRouteList
+		svcList  corev1.ServiceList
+	)
+
+	g, ctx := errgroup.WithContext(ctx)
+	g.Go(func() error { return cl.List(ctx, &gcList) })
+	g.Go(func() error { return cl.List(ctx, &gwList) })
+	g.Go(func() error { return cl.List(ctx, &httpList) })
+	g.Go(func() error { return cl.List(ctx, &grpcList) })
+	g.Go(func() error { return cl.List(ctx, &svcList) })
+
+	if err := g.Wait(); err != nil {
+		return "", fmt.Errorf("rag: list resources: %w", err)
 	}
 
-	managedClasses := make(map[string]bool)
+	managedClasses := make(map[string]bool, len(gcList.Items))
 	for _, gc := range gcList.Items {
 		if string(gc.Spec.ControllerName) == controllerName {
 			managedClasses[gc.Name] = true
@@ -39,39 +51,17 @@ func BuildRAGContext(ctx context.Context, cl client.Client, controllerName strin
 		return "No managed GatewayClasses found for controller " + controllerName, nil
 	}
 
-	// 2. Gateways.
-	gwList := &gatewayv1.GatewayList{}
-	if err := cl.List(ctx, gwList); err != nil {
-		return "", fmt.Errorf("rag: list Gateways: %w", err)
-	}
-
 	var managedGateways []gatewayv1.Gateway
 	for _, gw := range gwList.Items {
 		if managedClasses[string(gw.Spec.GatewayClassName)] {
 			managedGateways = append(managedGateways, gw)
 		}
 	}
 
-	// 3. HTTPRoutes.
-	httpList := &gatewayv1.HTTPRouteList{}
-	if err := cl.List(ctx, httpList); err != nil {
-		return "", fmt.Errorf("rag: list HTTPRoutes: %w", err)
-	}
-
-	// 4. GRPCRoutes.
-	grpcList := &gatewayv1.GRPCRouteList{}
-	if err := cl.List(ctx, grpcList); err != nil {
-		return "", fmt.Errorf("rag: list GRPCRoutes: %w", err)
-	}
-
-	// 5. Services.
-	svcList := &corev1.ServiceList{}
-	if err := cl.List(ctx, svcList); err != nil {
-		return "", fmt.Errorf("rag: list Services: %w", err)
-	}
-
 	// ── Format output ────────────────────────────────────────────
 
+	var b strings.Builder
+
 	b.WriteString("## Current Gateway API Topology\n\n")
 
 	// Gateways
@@ -182,36 +172,37 @@ func BuildRAGContext(ctx context.Context, cl client.Client, controllerName strin
 	// Services
 	b.WriteString("### Services\n\n")
 	if len(svcList.Items) == 0 {
-		b.WriteString("(none)\n")
+		b.WriteString("(none)\n\n")
 	} else {
 		for _, svc := range svcList.Items {
-			fmt.Fprintf(&b, "- **%s/%s**", svc.Namespace, svc.Name)
+			fmt.Fprintf(&b, "- **%s/%s** (type=%s)", svc.Namespace, svc.Name, svc.Spec.Type)
 			if len(svc.Spec.Ports) > 0 {
-				ports := make([]string, 0, len(svc.Spec.Ports))
-				for _, p := range svc.Spec.Ports {
-					ports = append(ports, fmt.Sprintf("%s:%d/%s", p.Name, p.Port, p.Protocol))
+				b.WriteString(" [")
+				for j, port := range svc.Spec.Ports {
+					if j > 0 {
+						b.WriteString(", ")
+					}
+					fmt.Fprintf(&b, "%d/%s", port.Port, port.Protocol)
 				}
-				fmt.Fprintf(&b, " ports=[%s]", strings.Join(ports, ", "))
+				b.WriteString("]")
 			}
 			b.WriteString("\n")
 		}
+		b.WriteString("\n")
 	}
 
 	return b.String(), nil
 }
 
 func fmtRouteParents(b *strings.Builder, refs []gatewayv1.ParentReference, defaultNS string) {
-	if len(refs) == 0 {
-		return
-	}
-	parts := make([]string, 0, len(refs))
-	for _, pr := range refs {
+	for j, ref := range refs {
+		if j > 0 {
+			b.WriteString(", ")
+		}
 		ns := defaultNS
-		if pr.Namespace != nil {
-			ns = string(*pr.Namespace)
+		if ref.Namespace != nil {
+			ns = string(*ref.Namespace)
 		}
-		parts = append(parts, fmt.Sprintf("%s/%s", ns, pr.Name))
+		fmt.Fprintf(b, " → %s/%s", ns, ref.Name)
 	}
-	fmt.Fprintf(b, " → gateways: [%s]", strings.Join(parts, ", "))
 }
-

internal/admin/server_chatbot.go

@@ -18,6 +18,7 @@ import (
 const (
 	chatbotConfigNamespace = "nantian-gw"
 	chatbotConfigSecret    = "chatbot-config"
+	maxChatHistoryMessages = 40
 )
 
 // maskAPIKey returns a masked version of the API key for safe display.
@@ -202,8 +203,11 @@ func (s *Server) handleChatbotChat(w http.ResponseWriter, r *http.Request) {
 	// Prep the system prompt with RAG context.
 	systemPrompt := buildSystemPrompt(ragContext)
 
-	// Build the full history.
+	// Build the full history (with a bound to prevent unbounded growth).
 	history := append([]chatbot.Message(nil), req.History...)
+	if len(history) > maxChatHistoryMessages {
+		history = history[len(history)-maxChatHistoryMessages:]
+	}
 
 	// Create the LLM adapter.
 	llm := chatbot.NewOpenAIAdapter(cfg.APIEndpoint, cfg.APIKey, cfg.Model, cfg.Temperature)