feat: 十轮网关优化 - 安全加固/可观测性/性能/可靠性

- SSE Keepalive Ping (15s心跳防止代理断连)
- Timing HTTP 头 (X-Timing-Queue/Inference/Total-Ms)
- Adapter Request-ID 传播到后端
- Session 清理日志回调
- Server 安全加固 (ReadHeaderTimeout/MaxHeaderBytes 防 slowloris)
- Usage Tracker 数据保留清理 (retentionDays + 定期清理)
- Config Reload 后 Adapter Registry 更新 (RegisterIfAbsent + RWMutex)
- Rate Limiter 空闲 Bucket 清理 (30分钟过期)
- Shutdown Drain 超时可配置 (ShutdownDrainSeconds)
- Config 模型字段校验增强 (provider/endpoint/actual_model)
- Auth 过期 Key 自动清理 (5分钟扫描)
- Admin API Rate Limiting
- Adapter Health Check 独立超时 (每个 adapter 3s)
- TCP 连接阶段超时 (DialContext 5s + KeepAlive 30s)
- 幂等键缓存、审计日志、Gzip 中间件、CORS Expose Headers
- Backpressure 响应头、熔断器 Prometheus 指标
- 连接池优化、Trace-ID 全链路传播

internal/adapter/ollama.go

@@ -6,11 +6,30 @@ import (
 	"encoding/json"
 	"fmt"
 	"io"
+	"net"
 	"net/http"
 	"strings"
 	"time"
 )
 
+// newHTTPClient 创建带连接池优化的 HTTP 客户端，复用 TCP 连接以减少延迟。
+// timeout 为整体请求超时，connectTimeout 为 TCP 连接阶段超时。
+func newHTTPClient(timeout time.Duration) *http.Client {
+	return &http.Client{
+		Timeout: timeout,
+		Transport: &http.Transport{
+			MaxIdleConns:        100,
+			MaxIdleConnsPerHost: 20,
+			IdleConnTimeout:     90 * time.Second,
+			MaxConnsPerHost:     0, // 不限制
+			DialContext: (&net.Dialer{
+				Timeout:   5 * time.Second,
+				KeepAlive: 30 * time.Second,
+			}).DialContext,
+		},
+	}
+}
+
 // OllamaAdapter implements ModelAdapter for Ollama inference engine.
 type OllamaAdapter struct {
 	endpoint   string
@@ -20,10 +39,8 @@ type OllamaAdapter struct {
 // NewOllamaAdapter creates a new Ollama adapter.
 func NewOllamaAdapter(endpoint string) *OllamaAdapter {
 	return &OllamaAdapter{
-		endpoint: strings.TrimRight(endpoint, "/"),
-		httpClient: &http.Client{
-			Timeout: 120 * time.Second,
-		},
+		endpoint:   strings.TrimRight(endpoint, "/"),
+		httpClient: newHTTPClient(120 * time.Second),
 	}
 }
 
@@ -52,20 +69,20 @@ type ollamaOptions struct {
 
 // ollamaChatResponse is the Ollama /api/chat non-streaming response.
 type ollamaChatResponse struct {
-	Model     string `json:"model"`
-	Message   ollamaMsg `json:"message"`
-	Done      bool   `json:"done"`
-	PromptEvalCount int `json:"prompt_eval_count"`
-	EvalCount      int `json:"eval_count"`
+	Model           string    `json:"model"`
+	Message         ollamaMsg `json:"message"`
+	Done            bool      `json:"done"`
+	PromptEvalCount int       `json:"prompt_eval_count"`
+	EvalCount       int       `json:"eval_count"`
 }
 
 // ollamaChatStreamResponse is a single chunk in Ollama streaming response.
 type ollamaChatStreamResponse struct {
-	Model     string `json:"model"`
-	Message   ollamaMsg `json:"message"`
-	Done      bool   `json:"done"`
-	PromptEvalCount int `json:"prompt_eval_count,omitempty"`
-	EvalCount      int `json:"eval_count,omitempty"`
+	Model           string    `json:"model"`
+	Message         ollamaMsg `json:"message"`
+	Done            bool      `json:"done"`
+	PromptEvalCount int       `json:"prompt_eval_count,omitempty"`
+	EvalCount       int       `json:"eval_count,omitempty"`
 }
 
 func (a *OllamaAdapter) ChatCompletion(ctx context.Context, req *ChatRequest) (*ChatResponse, error) {
@@ -81,6 +98,9 @@ func (a *OllamaAdapter) ChatCompletion(ctx context.Context, req *ChatRequest) (*
 		return nil, fmt.Errorf("create ollama request: %w", err)
 	}
 	httpReq.Header.Set("Content-Type", "application/json")
+	if req.RequestID != "" {
+		httpReq.Header.Set("X-Request-ID", req.RequestID)
+	}
 
 	resp, err := a.httpClient.Do(httpReq)
 	if err != nil {
@@ -120,6 +140,9 @@ func (a *OllamaAdapter) ChatCompletionStream(ctx context.Context, req *ChatReque
 		return nil, fmt.Errorf("create ollama stream request: %w", err)
 	}
 	httpReq.Header.Set("Content-Type", "application/json")
+	if req.RequestID != "" {
+		httpReq.Header.Set("X-Request-ID", req.RequestID)
+	}
 
 	resp, err := a.httpClient.Do(httpReq)
 	if err != nil {